X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=secchan%2Fsecchan.c;h=fa8d6e859502e885825c24bc13beb85c0e3891e4;hb=40fe0d7e038b570a99f0e96ecc363f6186164d49;hp=fb0eb41cd661bba968bd80f2e83ec19140075164;hpb=5fed6531b9146c9d30222062b83d9403ee254e03;p=sliver-openvswitch.git diff --git a/secchan/secchan.c b/secchan/secchan.c index fb0eb41cd..fa8d6e859 100644 --- a/secchan/secchan.c +++ b/secchan/secchan.c @@ -1,6 +1,6 @@ /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford * Junior University - * + * * We are making the OpenFlow specification and associated documentation * (Software) available for public use and benefit with the expectation * that others will use, modify and enhance the Software and contribute @@ -13,10 +13,10 @@ * distribute, sublicense, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: - * + * * The above copyright notice and this permission notice shall be * included in all copies or substantial portions of the Software. - * + * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND @@ -25,40 +25,45 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * + * * The name and trademarks of copyright holder(s) may NOT be used in * advertising or publicity pertaining to the Software or any * derivatives without specific, written prior permission. */ #include +#include "secchan.h" +#include #include #include -#include #include -#include -#include #include +#include #include -#include -#include -#include "buffer.h" #include "command-line.h" #include "compiler.h" #include "daemon.h" -#include "dhcp.h" -#include "dhcp-client.h" +#include "dirs.h" +#include "discovery.h" +#include "executer.h" +#include "fail-open.h" #include "fault.h" -#include "flow.h" -#include "learning-switch.h" +#include "in-band.h" #include "list.h" -#include "mac-learning.h" -#include "netdev.h" -#include "openflow.h" +#include "ofpbuf.h" +#include "openflow/openflow.h" #include "packets.h" +#include "port-watcher.h" #include "poll-loop.h" +#include "ratelimit.h" #include "rconn.h" +#ifdef SUPPORT_SNAT +#include "snat.h" +#endif +#include "stp-secchan.h" +#include "status.h" +#include "timeval.h" #include "util.h" #include "vconn-ssl.h" #include "vconn.h" @@ -67,258 +72,186 @@ #include "vlog.h" #define THIS_MODULE VLM_secchan -static const char *listen_vconn_name; - -struct half { - struct rconn *rconn; - struct buffer *rxbuf; -}; - -/* Behavior when the connection to the controller fails. */ -enum fail_mode { - FAIL_OPEN, /* Act as learning switch. */ - FAIL_CLOSED /* Drop all packets. */ +struct hook { + const struct hook_class *class; + void *aux; }; -struct relay { - struct list node; - -#define HALF_LOCAL 0 -#define HALF_REMOTE 1 - struct half halves[2]; - - bool is_mgmt_conn; - struct lswitch *lswitch; +struct secchan { + struct hook *hooks; + size_t n_hooks, allocated_hooks; }; -static struct list relays = LIST_INITIALIZER(&relays); - -/* Mode of operation. Note that autodiscovery implies in-band - * communication. */ -static bool autodiscovery; /* Discover the controller automatically? */ -static bool in_band; /* Connect to controller in-band? */ - -/* MAC address of local port. */ -static uint8_t local_mac[ETH_ADDR_LEN]; - -/* MAC learning table for local port. */ -static struct mac_learning *local_ml; - -/* Controller vconn name, or null to perform controller autodiscovery. */ -static char *controller_name = NULL; - -/* -f, --fail: Behavior when the connection to the controller fails. */ -static enum fail_mode fail_mode = FAIL_OPEN; - -/* The OpenFlow virtual network device ofX. */ -static struct netdev *of_device; - -/* --inactivity-probe: Number of seconds without receiving a message from the - controller before sending an inactivity probe. */ -static int probe_interval = 15; - -/* --max-idle: Idle time to assign to flows created by learning switch when in - * fail-open mode. */ -static int max_idle = 15; - -/* --max-backoff: Maximum interval between controller connection attempts, in - * seconds. */ -static int max_backoff = 15; - -/* DHCP client, for controller autodiscovery. */ -static struct dhclient *dhcp; +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60); -/* --accept-vconn: Regular expression specifying the class of controller vconns - * that we will accept during autodiscovery. */ -static const char *accept_controller_re; -static regex_t accept_controller_regex; - -/* --no-resolv-conf: Update resolv.conf upon successful controller - * discovery? */ -static bool update_resolv_conf = true; - -static void parse_options(int argc, char *argv[]); +static void parse_options(int argc, char *argv[], struct settings *); static void usage(void) NO_RETURN; -static void new_management_connection(const char *nl_name, struct vconn *new_remote); -static struct relay *relay_create(struct rconn *local, struct rconn *remote, - bool is_mgmt_conn); -static void relay_run(struct relay *); +static char *vconn_name_without_subscription(const char *); +static struct pvconn *open_passive_vconn(const char *name); +static struct vconn *accept_vconn(struct pvconn *pvconn); + +static struct relay *relay_create(struct rconn *async, + struct rconn *local, struct rconn *remote); +static struct relay *relay_accept(const struct settings *, struct pvconn *); +static void relay_run(struct relay *, struct secchan *); static void relay_wait(struct relay *); static void relay_destroy(struct relay *); -static bool local_hook(struct relay *r); -static bool failing_open(struct relay *r); -static bool fail_open_hook(struct relay *r); - -static void modify_dhcp_request(struct dhcp_msg *, void *aux); -static bool validate_dhcp_offer(const struct dhcp_msg *, void *aux); - int main(int argc, char *argv[]) { - struct rconn *local_rconn, *remote_rconn; - struct vconn *listen_vconn; + struct settings s; + + struct list relays = LIST_INITIALIZER(&relays); + + struct secchan secchan; + + struct pvconn *monitor; + + struct pvconn *listeners[MAX_MGMT]; + size_t n_listeners; + + char *local_rconn_name; + struct rconn *async_rconn, *local_rconn, *remote_rconn; struct relay *controller_relay; - const char *nl_name; - char of_name[16]; + struct discovery *discovery; + struct switch_status *switch_status; + struct port_watcher *pw; + int i; int retval; set_program_name(argv[0]); register_fault_handlers(); + time_init(); vlog_init(); - parse_options(argc, argv); - - argc -= optind; - argv += optind; - if (argc < 1 || argc > 2) { - fatal(0, "need one or two non-option arguments; use --help for usage"); - } - nl_name = argv[0]; - if (strncmp(nl_name, "nl:", 3) - || strlen(nl_name) < 4 - || nl_name[strspn(nl_name + 3, "0123456789") + 3]) { - fatal(0, "%s: argument is not of the form \"nl:DP_IDX\"", nl_name); - } - controller_name = argc > 1 ? xstrdup(argv[1]) : NULL; - autodiscovery = controller_name == NULL; + parse_options(argc, argv, &s); + signal(SIGPIPE, SIG_IGN); - if (!accept_controller_re) { - accept_controller_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*"; - } - retval = regcomp(&accept_controller_regex, accept_controller_re, - REG_NOSUB | REG_EXTENDED); - if (retval) { - size_t length = regerror(retval, &accept_controller_regex, NULL, 0); - char *buffer = xmalloc(length); - regerror(retval, &accept_controller_regex, buffer, length); - fatal(0, "%s: %s", accept_controller_re, buffer); - } + secchan.hooks = NULL; + secchan.n_hooks = 0; + secchan.allocated_hooks = 0; - if (listen_vconn_name) { - retval = vconn_open(listen_vconn_name, &listen_vconn); - if (retval && retval != EAGAIN) { - fatal(retval, "opening %s", listen_vconn_name); - } - if (!vconn_is_passive(listen_vconn)) { - fatal(0, "%s is not a passive vconn", listen_vconn_name); - } - } else { - listen_vconn = NULL; + /* Start listening for management and monitoring connections. */ + n_listeners = 0; + for (i = 0; i < s.n_listeners; i++) { + listeners[n_listeners++] = open_passive_vconn(s.listener_names[i]); } + monitor = s.monitor_name ? open_passive_vconn(s.monitor_name) : NULL; - snprintf(of_name, sizeof of_name, "of%s", nl_name + 3); - retval = netdev_open(of_name, NETDEV_ETH_TYPE_NONE, &of_device); - if (!retval) { - enum netdev_flags flags; + /* Initialize switch status hook. */ + switch_status_start(&secchan, &s, &switch_status); - if (autodiscovery) { - retval = netdev_turn_flags_on(of_device, NETDEV_UP, true); - if (retval) { - fatal(retval, "Could not bring %s device up", of_name); - } - } + die_if_already_running(); + daemonize(); - retval = netdev_get_flags(of_device, &flags); - if (!retval) { - if (flags & NETDEV_UP) { - struct in6_addr in6; - - in_band = true; - memcpy(local_mac, netdev_get_etheraddr(of_device), - ETH_ADDR_LEN); - if (netdev_get_in6(of_device, &in6)) { - VLOG_WARN("Ignoring IPv6 address on %s device: " - "IPv6 not supported", of_name); - } - local_ml = mac_learning_create(); - } - } else { - error(retval, "Could not get flags for %s device", of_name); - } - } else { - error(retval, "Could not open %s device", of_name); - } - if (autodiscovery && !in_band) { - fatal(retval, "In autodiscovery mode but failed to configure " - "in-band control"); + /* Start listening for vlogconf requests. */ + retval = vlog_server_listen(NULL, NULL); + if (retval) { + ofp_fatal(retval, "Could not listen for vlog connections"); } - if (autodiscovery) { - retval = dhclient_create(of_name, modify_dhcp_request, - validate_dhcp_offer, NULL, &dhcp); - if (retval) { - fatal(retval, "Failed to initialize DHCP client"); - } - dhclient_init(dhcp, 0); - } + VLOG_WARN("OpenFlow reference implementation version %s", VERSION BUILDNR); + VLOG_WARN("OpenFlow protocol version 0x%02x", OFP_VERSION); - retval = vlog_server_listen(NULL, NULL); - if (retval) { - fatal(retval, "Could not listen for vlog connections"); + /* Check datapath name, to try to catch command-line invocation errors. */ + if (strncmp(s.dp_name, "nl:", 3) && strncmp(s.dp_name, "unix:", 5) + && !s.controller_name) { + VLOG_WARN("Controller not specified and datapath is not nl: or " + "unix:. (Did you forget to specify the datapath?)"); } - daemonize(); + /* Connect to datapath with a subscription for asynchronous events. */ + async_rconn = rconn_create(0, s.max_backoff); + rconn_connect(async_rconn, s.dp_name); + switch_status_register_category(switch_status, "async", + rconn_status_cb, async_rconn); - local_rconn = rconn_create(1, 0, max_backoff); - rconn_connect(local_rconn, nl_name); + /* Connect to datapath without a subscription, for requests and replies. */ + local_rconn_name = vconn_name_without_subscription(s.dp_name); + local_rconn = rconn_create(0, s.max_backoff); + rconn_connect(local_rconn, local_rconn_name); + free(local_rconn_name); + switch_status_register_category(switch_status, "local", + rconn_status_cb, local_rconn); - remote_rconn = rconn_create(1, probe_interval, max_backoff); - if (controller_name) { - retval = rconn_connect(remote_rconn, controller_name); + /* Connect to controller. */ + remote_rconn = rconn_create(s.probe_interval, s.max_backoff); + if (s.controller_name) { + retval = rconn_connect(remote_rconn, s.controller_name); if (retval == EAFNOSUPPORT) { - fatal(0, "No support for %s vconn", controller_name); + ofp_fatal(0, "No support for %s vconn", s.controller_name); } } - controller_relay = relay_create(local_rconn, remote_rconn, false); + switch_status_register_category(switch_status, "remote", + rconn_status_cb, remote_rconn); + + /* Start relaying. */ + controller_relay = relay_create(async_rconn, local_rconn, remote_rconn); + list_push_back(&relays, &controller_relay->node); + + /* Set up hooks. */ + port_watcher_start(&secchan, local_rconn, remote_rconn, &pw); + discovery = s.discovery ? discovery_init(&s, pw, switch_status) : NULL; +#ifdef SUPPORT_SNAT + snat_start(&secchan, pw); +#endif + if (s.enable_stp) { + stp_start(&secchan, &s, pw, local_rconn, remote_rconn); + } + if (s.in_band) { + in_band_start(&secchan, &s, switch_status, pw, remote_rconn); + } + if (s.fail_mode == FAIL_OPEN) { + fail_open_start(&secchan, &s, switch_status, + local_rconn, remote_rconn); + } + if (s.rate_limit) { + rate_limit_start(&secchan, &s, switch_status, remote_rconn); + } + if (s.command_acl[0]) { + executer_start(&secchan, &s); + } + for (;;) { struct relay *r, *n; + size_t i; /* Do work. */ LIST_FOR_EACH_SAFE (r, n, struct relay, node, &relays) { - relay_run(r); + relay_run(r, &secchan); } - if (listen_vconn) { + for (i = 0; i < n_listeners; i++) { for (;;) { - struct vconn *new_remote; - retval = vconn_accept(listen_vconn, &new_remote); - if (retval) { - if (retval != EAGAIN) { - VLOG_WARN("accept failed (%s)", strerror(retval)); - } + struct relay *r = relay_accept(&s, listeners[i]); + if (!r) { break; } - new_management_connection(nl_name, new_remote); + list_push_back(&relays, &r->node); } } - if (controller_relay) { - /* FIXME: should also fail open when controller_relay is NULL. */ - failing_open(controller_relay); + if (monitor) { + struct vconn *new = accept_vconn(monitor); + if (new) { + /* XXX should monitor async_rconn too but rconn_add_monitor() + * takes ownership of the vconn passed in. */ + rconn_add_monitor(local_rconn, new); + } + } + for (i = 0; i < secchan.n_hooks; i++) { + if (secchan.hooks[i].class->periodic_cb) { + secchan.hooks[i].class->periodic_cb(secchan.hooks[i].aux); + } } - if (dhcp) { + if (s.discovery) { + char *controller_name; if (rconn_is_connectivity_questionable(remote_rconn)) { - dhclient_force_renew(dhcp, 15); + discovery_question_connectivity(discovery); } - dhclient_run(dhcp); - if (dhclient_changed(dhcp)) { - dhclient_configure_netdev(dhcp); - if (update_resolv_conf) { - dhclient_update_resolv_conf(dhcp); - } - - if (dhclient_is_bound(dhcp)) { - free(controller_name); - controller_name = dhcp_msg_get_string( - dhclient_get_config(dhcp), - DHCP_CODE_OFP_CONTROLLER_VCONN); - VLOG_WARN("%s: discovered controller", - controller_name); + if (discovery_run(discovery, &controller_name)) { + if (controller_name) { rconn_connect(remote_rconn, controller_name); - } else if (controller_name) { - VLOG_WARN("%s: discover controller no longer available", - controller_name); - free(controller_name); - controller_name = NULL; + } else { rconn_disconnect(remote_rconn); } } @@ -328,11 +261,19 @@ main(int argc, char *argv[]) LIST_FOR_EACH (r, struct relay, node, &relays) { relay_wait(r); } - if (listen_vconn) { - vconn_accept_wait(listen_vconn); + for (i = 0; i < n_listeners; i++) { + pvconn_wait(listeners[i]); + } + if (monitor) { + pvconn_wait(monitor); + } + for (i = 0; i < secchan.n_hooks; i++) { + if (secchan.hooks[i].class->wait_cb) { + secchan.hooks[i].class->wait_cb(secchan.hooks[i].aux); + } } - if (dhcp) { - dhclient_wait(dhcp); + if (discovery) { + discovery_wait(discovery); } poll_block(); } @@ -340,64 +281,183 @@ main(int argc, char *argv[]) return 0; } -static void -new_management_connection(const char *nl_name, struct vconn *new_remote) +static struct pvconn * +open_passive_vconn(const char *name) +{ + struct pvconn *pvconn; + int retval; + + retval = pvconn_open(name, &pvconn); + if (retval && retval != EAGAIN) { + ofp_fatal(retval, "opening %s", name); + } + return pvconn; +} + +static struct vconn * +accept_vconn(struct pvconn *pvconn) +{ + struct vconn *new; + int retval; + + retval = pvconn_accept(pvconn, OFP_VERSION, &new); + if (retval && retval != EAGAIN) { + VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); + } + return new; +} + +void +add_hook(struct secchan *secchan, const struct hook_class *class, void *aux) +{ + struct hook *hook; + + if (secchan->n_hooks >= secchan->allocated_hooks) { + secchan->allocated_hooks = secchan->allocated_hooks * 2 + 1; + secchan->hooks = xrealloc(secchan->hooks, + (sizeof *secchan->hooks + * secchan->allocated_hooks)); + } + hook = &secchan->hooks[secchan->n_hooks++]; + hook->class = class; + hook->aux = aux; +} + +struct ofp_packet_in * +get_ofp_packet_in(struct relay *r) { - char *nl_name_without_subscription; - struct vconn *new_local; + struct ofpbuf *msg = r->halves[HALF_LOCAL].rxbuf; + struct ofp_header *oh = msg->data; + if (oh->type == OFPT_PACKET_IN) { + if (msg->size >= offsetof (struct ofp_packet_in, data)) { + return msg->data; + } else { + VLOG_WARN("packet too short (%zu bytes) for packet_in", + msg->size); + } + } + return NULL; +} + +bool +get_ofp_packet_eth_header(struct relay *r, struct ofp_packet_in **opip, + struct eth_header **ethp) +{ + const int min_len = offsetof(struct ofp_packet_in, data) + ETH_HEADER_LEN; + struct ofp_packet_in *opi = get_ofp_packet_in(r); + if (opi && ntohs(opi->header.length) >= min_len) { + *opip = opi; + *ethp = (void *) opi->data; + return true; + } + return false; +} + +/* OpenFlow message relaying. */ + +/* Returns a malloc'd string containing a copy of 'vconn_name' modified not to + * subscribe to asynchronous messages such as 'ofp_packet_in' events (if + * possible). */ +static char * +vconn_name_without_subscription(const char *vconn_name) +{ + int nl_index; + if (sscanf(vconn_name, "nl:%d", &nl_index) == 1) { + /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. + * nl:123:0 opens a netlink connection to local datapath 123 without + * obtaining a subscription for ofp_packet_in or ofp_flow_expired + * messages. */ + return xasprintf("nl:%d:0", nl_index); + } else { + /* We don't have a way to specify not to subscribe to those messages + * for other transports. (That's a defect: really this should be in + * the OpenFlow protocol, not the Netlink transport). */ + VLOG_WARN_RL(&rl, "new management connection will receive " + "asynchronous messages"); + return xstrdup(vconn_name); + } +} + +static struct relay * +relay_accept(const struct settings *s, struct pvconn *pvconn) +{ + struct vconn *new_remote, *new_local; struct rconn *r1, *r2; + char *vconn_name; int retval; - /* nl:123 or nl:123:1 opens a netlink connection to local datapath 123. We - * only accept the former syntax in main(). - * - * nl:123:0 opens a netlink connection to local datapath 123 without - * obtaining a subscription for ofp_packet_in or ofp_flow_expired - * messages.*/ - nl_name_without_subscription = xasprintf("%s:0", nl_name); - retval = vconn_open(nl_name_without_subscription, &new_local); + new_remote = accept_vconn(pvconn); + if (!new_remote) { + return NULL; + } + + vconn_name = vconn_name_without_subscription(s->dp_name); + retval = vconn_open(vconn_name, OFP_VERSION, &new_local); if (retval) { - VLOG_ERR("could not connect to %s (%s)", - nl_name_without_subscription, strerror(retval)); + VLOG_ERR_RL(&rl, "could not connect to %s (%s)", + vconn_name, strerror(retval)); vconn_close(new_remote); - free(nl_name_without_subscription); - return; + free(vconn_name); + return NULL; } - /* Add it to the relay list. */ - r1 = rconn_create(1, 0, 0); - rconn_connect_unreliably(r1, nl_name_without_subscription, new_local); - r2 = rconn_create(1, 0, 0); + /* Create and return relay. */ + r1 = rconn_create(0, 0); + rconn_connect_unreliably(r1, vconn_name, new_local); + free(vconn_name); + + r2 = rconn_create(0, 0); rconn_connect_unreliably(r2, "passive", new_remote); - relay_create(r1, r2, true); - free(nl_name_without_subscription); + return relay_create(NULL, r1, r2); } static struct relay * -relay_create(struct rconn *local, struct rconn *remote, bool is_mgmt_conn) +relay_create(struct rconn *async, struct rconn *local, struct rconn *remote) { - struct relay *r; - int i; - - r = xmalloc(sizeof *r); + struct relay *r = xcalloc(1, sizeof *r); r->halves[HALF_LOCAL].rconn = local; r->halves[HALF_REMOTE].rconn = remote; - for (i = 0; i < 2; i++) { - r->halves[i].rxbuf = NULL; - } - r->is_mgmt_conn = is_mgmt_conn; - r->lswitch = NULL; - list_push_back(&relays, &r->node); + r->is_mgmt_conn = async == NULL; + r->async_rconn = async; return r; } +static bool +call_local_packet_cbs(struct secchan *secchan, struct relay *r) +{ + const struct hook *h; + for (h = secchan->hooks; h < &secchan->hooks[secchan->n_hooks]; h++) { + bool (*cb)(struct relay *, void *aux) = h->class->local_packet_cb; + if (cb && (cb)(r, h->aux)) { + return true; + } + } + return false; +} + +static bool +call_remote_packet_cbs(struct secchan *secchan, struct relay *r) +{ + const struct hook *h; + for (h = secchan->hooks; h < &secchan->hooks[secchan->n_hooks]; h++) { + bool (*cb)(struct relay *, void *aux) = h->class->remote_packet_cb; + if (cb && (cb)(r, h->aux)) { + return true; + } + } + return false; +} + static void -relay_run(struct relay *r) +relay_run(struct relay *r, struct secchan *secchan) { int iteration; int i; + if (r->async_rconn) { + rconn_run(r->async_rconn); + } for (i = 0; i < 2; i++) { rconn_run(r->halves[i].rconn); } @@ -411,20 +471,30 @@ relay_run(struct relay *r) if (!this->rxbuf) { this->rxbuf = rconn_recv(this->rconn); - if (this->rxbuf && !r->is_mgmt_conn && i == HALF_LOCAL - && (local_hook(r) || fail_open_hook(r))) { - buffer_delete(this->rxbuf); - this->rxbuf = NULL; + if (!this->rxbuf && i == HALF_LOCAL && r->async_rconn) { + this->rxbuf = rconn_recv(r->async_rconn); + } + if (this->rxbuf && (i == HALF_REMOTE || !r->is_mgmt_conn)) { + if (i == HALF_LOCAL + ? call_local_packet_cbs(secchan, r) + : call_remote_packet_cbs(secchan, r)) + { + ofpbuf_delete(this->rxbuf); + this->rxbuf = NULL; + progress = true; + break; + } } } - if (this->rxbuf) { - int retval = rconn_send(peer->rconn, this->rxbuf); + if (this->rxbuf && !this->n_txq) { + int retval = rconn_send(peer->rconn, this->rxbuf, + &this->n_txq); if (retval != EAGAIN) { if (!retval) { progress = true; } else { - buffer_delete(this->rxbuf); + ofpbuf_delete(this->rxbuf); } this->rxbuf = NULL; } @@ -451,12 +521,18 @@ relay_wait(struct relay *r) { int i; + if (r->async_rconn) { + rconn_run_wait(r->async_rconn); + } for (i = 0; i < 2; i++) { struct half *this = &r->halves[i]; rconn_run_wait(this->rconn); if (!this->rxbuf) { rconn_recv_wait(this->rconn); + if (i == HALF_LOCAL && r->async_rconn) { + rconn_recv_wait(r->async_rconn); + } } } } @@ -467,250 +543,83 @@ relay_destroy(struct relay *r) int i; list_remove(&r->node); + rconn_destroy(r->async_rconn); for (i = 0; i < 2; i++) { struct half *this = &r->halves[i]; rconn_destroy(this->rconn); - buffer_delete(this->rxbuf); + ofpbuf_delete(this->rxbuf); } free(r); } + +/* User interface. */ static void -queue_tx(struct rconn *rc, struct buffer *b) -{ - if (rconn_force_send(rc, b)) { - buffer_delete(b); - } -} - -static bool -is_controller_mac(const uint8_t dl_addr[ETH_ADDR_LEN], - struct rconn *controller) -{ - static uint32_t ip, last_nonzero_ip; - static uint8_t mac[ETH_ADDR_LEN], last_nonzero_mac[ETH_ADDR_LEN]; - static time_t next_refresh = 0; - - uint32_t last_ip = ip; - - time_t now = time(0); - - ip = rconn_get_ip(controller); - if (last_ip != ip || !next_refresh || now >= next_refresh) { - bool have_mac; - - /* Look up MAC address. */ - memset(mac, 0, sizeof mac); - if (ip) { - int retval = netdev_arp_lookup(of_device, ip, mac); - if (retval) { - VLOG_DBG("cannot look up controller hw address ("IP_FMT"): %s", - IP_ARGS(&ip), strerror(retval)); - } - } - have_mac = !eth_addr_is_zero(mac); - - /* Log changes in IP, MAC addresses. */ - if (ip && ip != last_nonzero_ip) { - VLOG_DBG("controller IP address changed from "IP_FMT - " to "IP_FMT, IP_ARGS(&last_nonzero_ip), IP_ARGS(&ip)); - last_nonzero_ip = ip; - } - if (have_mac && memcmp(last_nonzero_mac, mac, ETH_ADDR_LEN)) { - VLOG_DBG("controller MAC address changed from "ETH_ADDR_FMT" to " - ETH_ADDR_FMT, - ETH_ADDR_ARGS(last_nonzero_mac), ETH_ADDR_ARGS(mac)); - memcpy(last_nonzero_mac, mac, ETH_ADDR_LEN); - } - - /* Schedule next refresh. - * - * If we have an IP address but not a MAC address, then refresh - * quickly, since we probably will get a MAC address soon (via ARP). - * Otherwise, we can afford to wait a little while. */ - next_refresh = now + (!ip || have_mac ? 10 : 1); - } - return !eth_addr_is_zero(mac) && eth_addr_equals(mac, dl_addr); -} - -static bool -local_hook(struct relay *r) -{ - struct rconn *rc = r->halves[HALF_LOCAL].rconn; - struct buffer *msg = r->halves[HALF_LOCAL].rxbuf; - struct ofp_packet_in *opi; - struct ofp_header *oh; - size_t pkt_ofs, pkt_len; - struct buffer pkt; - struct flow flow; - uint16_t in_port, out_port; - - if (!in_band) { - return false; - } - - oh = msg->data; - if (oh->type != OFPT_PACKET_IN) { - return false; - } - if (msg->size < offsetof (struct ofp_packet_in, data)) { - VLOG_WARN("packet too short (%zu bytes) for packet_in", msg->size); - return false; - } - - /* Extract flow data from 'opi' into 'flow'. */ - opi = msg->data; - in_port = ntohs(opi->in_port); - pkt_ofs = offsetof(struct ofp_packet_in, data); - pkt_len = ntohs(opi->header.length) - pkt_ofs; - pkt.data = opi->data; - pkt.size = pkt_len; - flow_extract(&pkt, in_port, &flow); - - /* Deal with local stuff. */ - if (in_port == OFPP_LOCAL) { - out_port = mac_learning_lookup(local_ml, flow.dl_dst); - } else if (eth_addr_equals(flow.dl_dst, local_mac)) { - out_port = OFPP_LOCAL; - if (mac_learning_learn(local_ml, flow.dl_src, in_port)) { - VLOG_DBG("learned that "ETH_ADDR_FMT" is on port %"PRIu16, - ETH_ADDR_ARGS(flow.dl_src), in_port); - } - } else if (flow.dl_type == htons(ETH_TYPE_ARP) - && eth_addr_is_broadcast(flow.dl_dst) - && is_controller_mac(flow.dl_src, - r->halves[HALF_REMOTE].rconn)) { - out_port = OFPP_FLOOD; - } else { - return false; - } - - if (out_port != OFPP_FLOOD) { - /* The output port is known, so add a new flow. */ - queue_tx(rc, make_add_simple_flow(&flow, ntohl(opi->buffer_id), - out_port, max_idle)); - - /* If the switch didn't buffer the packet, we need to send a copy. */ - if (ntohl(opi->buffer_id) == UINT32_MAX) { - queue_tx(rc, make_unbuffered_packet_out(&pkt, in_port, out_port)); - } - } else { - /* We don't know that MAC. Send along the packet without setting up a - * flow. */ - struct buffer *b; - if (ntohl(opi->buffer_id) == UINT32_MAX) { - b = make_unbuffered_packet_out(&pkt, in_port, out_port); - } else { - b = make_buffered_packet_out(ntohl(opi->buffer_id), - in_port, out_port); - } - queue_tx(rc, b); - } - return true; -} - -/* Causess 'r' to enter or leave fail-open mode, if appropriate. Returns true - * if 'r' is in fail-open fail, false otherwise. */ -static bool -failing_open(struct relay *r) -{ - struct rconn *local = r->halves[HALF_LOCAL].rconn; - struct rconn *remote = r->halves[HALF_REMOTE].rconn; - int disconnected_duration; - - if (fail_mode == FAIL_CLOSED) { - /* We fail closed, so there's never anything to do. */ - return false; - } - - disconnected_duration = rconn_disconnected_duration(remote); - if (disconnected_duration < probe_interval * 3) { - /* It's not time to fail open yet. */ - if (r->lswitch && rconn_is_connected(remote)) { - /* We're connected, so drop the learning switch. */ - VLOG_WARN("No longer in fail-open mode"); - lswitch_destroy(r->lswitch); - r->lswitch = NULL; - } - return false; - } - - if (!r->lswitch) { - VLOG_WARN("Could not connect to controller for %d seconds, " - "failing open", disconnected_duration); - r->lswitch = lswitch_create(local, true, max_idle); - } - return true; -} - -static bool -fail_open_hook(struct relay *r) -{ - if (!failing_open(r)) { - return false; - } else { - struct buffer *msg = r->halves[HALF_LOCAL].rxbuf; - struct rconn *local = r->halves[HALF_LOCAL].rconn; - lswitch_process_packet(r->lswitch, local, msg); - rconn_run(local); - return true; - } -} - -static void -modify_dhcp_request(struct dhcp_msg *msg, void *aux) -{ - dhcp_msg_put_string(msg, DHCP_CODE_VENDOR_CLASS, "OpenFlow"); -} - -static bool -validate_dhcp_offer(const struct dhcp_msg *msg, void *aux) -{ - char *vconn_name; - bool accept; - - vconn_name = dhcp_msg_get_string(msg, DHCP_CODE_OFP_CONTROLLER_VCONN); - if (!vconn_name) { - VLOG_WARN("rejecting DHCP offer missing controller vconn"); - return false; - } - accept = !regexec(&accept_controller_regex, vconn_name, 0, NULL, 0); - if (!accept) { - VLOG_WARN("rejecting controller vconn that fails to match %s", - accept_controller_re); - } - free(vconn_name); - return accept; -} - -static void -parse_options(int argc, char *argv[]) +parse_options(int argc, char *argv[], struct settings *s) { enum { OPT_ACCEPT_VCONN = UCHAR_MAX + 1, OPT_NO_RESOLV_CONF, OPT_INACTIVITY_PROBE, OPT_MAX_IDLE, - OPT_MAX_BACKOFF + OPT_MAX_BACKOFF, + OPT_RATE_LIMIT, + OPT_BURST_LIMIT, + OPT_BOOTSTRAP_CA_CERT, + OPT_STP, + OPT_NO_STP, + OPT_OUT_OF_BAND, + OPT_IN_BAND, + OPT_COMMAND_ACL, + OPT_COMMAND_DIR, + VLOG_OPTION_ENUMS }; static struct option long_options[] = { {"accept-vconn", required_argument, 0, OPT_ACCEPT_VCONN}, {"no-resolv-conf", no_argument, 0, OPT_NO_RESOLV_CONF}, - {"fail", required_argument, 0, 'f'}, + {"fail", required_argument, 0, 'F'}, {"inactivity-probe", required_argument, 0, OPT_INACTIVITY_PROBE}, {"max-idle", required_argument, 0, OPT_MAX_IDLE}, {"max-backoff", required_argument, 0, OPT_MAX_BACKOFF}, {"listen", required_argument, 0, 'l'}, - {"detach", no_argument, 0, 'D'}, - {"pidfile", optional_argument, 0, 'P'}, + {"monitor", required_argument, 0, 'm'}, + {"rate-limit", optional_argument, 0, OPT_RATE_LIMIT}, + {"burst-limit", required_argument, 0, OPT_BURST_LIMIT}, + {"stp", no_argument, 0, OPT_STP}, + {"no-stp", no_argument, 0, OPT_NO_STP}, + {"out-of-band", no_argument, 0, OPT_OUT_OF_BAND}, + {"in-band", no_argument, 0, OPT_IN_BAND}, + {"command-acl", required_argument, 0, OPT_COMMAND_ACL}, + {"command-dir", required_argument, 0, OPT_COMMAND_DIR}, {"verbose", optional_argument, 0, 'v'}, {"help", no_argument, 0, 'h'}, {"version", no_argument, 0, 'V'}, + DAEMON_LONG_OPTIONS, + VLOG_LONG_OPTIONS, +#ifdef HAVE_OPENSSL VCONN_SSL_LONG_OPTIONS + {"bootstrap-ca-cert", required_argument, 0, OPT_BOOTSTRAP_CA_CERT}, +#endif {0, 0, 0, 0}, }; char *short_options = long_options_to_short_options(long_options); - + char *accept_re = NULL; + int retval; + + /* Set defaults that we can figure out before parsing options. */ + s->n_listeners = 0; + s->monitor_name = NULL; + s->fail_mode = FAIL_OPEN; + s->max_idle = 15; + s->probe_interval = 15; + s->max_backoff = 15; + s->update_resolv_conf = true; + s->rate_limit = 0; + s->burst_limit = 0; + s->enable_stp = false; + s->in_band = true; + s->command_acl = ""; + s->command_dir = xasprintf("%s/commands", ofp_pkgdatadir); for (;;) { int c; @@ -721,82 +630,132 @@ parse_options(int argc, char *argv[]) switch (c) { case OPT_ACCEPT_VCONN: - accept_controller_re = (optarg[0] == '^' - ? optarg - : xasprintf("^%s", optarg)); + accept_re = optarg[0] == '^' ? optarg : xasprintf("^%s", optarg); break; case OPT_NO_RESOLV_CONF: - update_resolv_conf = false; + s->update_resolv_conf = false; break; - case 'f': + case 'F': if (!strcmp(optarg, "open")) { - fail_mode = FAIL_OPEN; + s->fail_mode = FAIL_OPEN; } else if (!strcmp(optarg, "closed")) { - fail_mode = FAIL_CLOSED; + s->fail_mode = FAIL_CLOSED; } else { - fatal(0, - "-f or --fail argument must be \"open\" or \"closed\""); + ofp_fatal(0, "-f or --fail argument must be \"open\" " + "or \"closed\""); } break; case OPT_INACTIVITY_PROBE: - probe_interval = atoi(optarg); - if (probe_interval < 5) { - fatal(0, "--inactivity-probe argument must be at least 5"); + s->probe_interval = atoi(optarg); + if (s->probe_interval < 5) { + ofp_fatal(0, "--inactivity-probe argument must be at least 5"); } break; case OPT_MAX_IDLE: if (!strcmp(optarg, "permanent")) { - max_idle = OFP_FLOW_PERMANENT; + s->max_idle = OFP_FLOW_PERMANENT; } else { - max_idle = atoi(optarg); - if (max_idle < 1 || max_idle > 65535) { - fatal(0, "--max-idle argument must be between 1 and " - "65535 or the word 'permanent'"); + s->max_idle = atoi(optarg); + if (s->max_idle < 1 || s->max_idle > 65535) { + ofp_fatal(0, "--max-idle argument must be between 1 and " + "65535 or the word 'permanent'"); } } break; case OPT_MAX_BACKOFF: - max_backoff = atoi(optarg); - if (max_backoff < 1) { - fatal(0, "--max-backoff argument must be at least 1"); - } else if (max_backoff > 3600) { - max_backoff = 3600; + s->max_backoff = atoi(optarg); + if (s->max_backoff < 1) { + ofp_fatal(0, "--max-backoff argument must be at least 1"); + } else if (s->max_backoff > 3600) { + s->max_backoff = 3600; + } + break; + + case OPT_RATE_LIMIT: + if (optarg) { + s->rate_limit = atoi(optarg); + if (s->rate_limit < 1) { + ofp_fatal(0, "--rate-limit argument must be at least 1"); + } + } else { + s->rate_limit = 1000; } break; - case 'D': - set_detach(); + case OPT_BURST_LIMIT: + s->burst_limit = atoi(optarg); + if (s->burst_limit < 1) { + ofp_fatal(0, "--burst-limit argument must be at least 1"); + } + break; + + case OPT_STP: + s->enable_stp = true; + break; + + case OPT_NO_STP: + s->enable_stp = false; + break; + + case OPT_OUT_OF_BAND: + s->in_band = false; break; - case 'P': - set_pidfile(optarg); + case OPT_IN_BAND: + s->in_band = true; + break; + + case OPT_COMMAND_ACL: + s->command_acl = (s->command_acl[0] + ? xasprintf("%s,%s", s->command_acl, optarg) + : optarg); + break; + + case OPT_COMMAND_DIR: + s->command_dir = optarg; break; case 'l': - if (listen_vconn_name) { - fatal(0, "-l or --listen may be only specified once"); + if (s->n_listeners >= MAX_MGMT) { + ofp_fatal(0, + "-l or --listen may be specified at most %d times", + MAX_MGMT); } - listen_vconn_name = optarg; + s->listener_names[s->n_listeners++] = optarg; + break; + + case 'm': + if (s->monitor_name) { + ofp_fatal(0, "-m or --monitor may only be specified once"); + } + s->monitor_name = optarg; break; case 'h': usage(); case 'V': - printf("%s "VERSION" compiled "__DATE__" "__TIME__"\n", argv[0]); + printf("%s %s compiled "__DATE__" "__TIME__"\n", + program_name, VERSION BUILDNR); exit(EXIT_SUCCESS); - case 'v': - vlog_set_verbosity(optarg); - break; + DAEMON_OPTION_HANDLERS + VLOG_OPTION_HANDLERS + +#ifdef HAVE_OPENSSL VCONN_SSL_OPTION_HANDLERS + case OPT_BOOTSTRAP_CA_CERT: + vconn_ssl_set_ca_cert_file(optarg, true); + break; +#endif + case '?': exit(EXIT_FAILURE); @@ -805,6 +764,50 @@ parse_options(int argc, char *argv[]) } } free(short_options); + + argc -= optind; + argv += optind; + if (argc < 1 || argc > 2) { + ofp_fatal(0, "need one or two non-option arguments; " + "use --help for usage"); + } + + /* Local and remote vconns. */ + s->dp_name = argv[0]; + s->controller_name = argc > 1 ? xstrdup(argv[1]) : NULL; + + /* Set accept_controller_regex. */ + if (!accept_re) { + accept_re = vconn_ssl_is_configured() ? "^ssl:.*" : ".*"; + } + retval = regcomp(&s->accept_controller_regex, accept_re, + REG_NOSUB | REG_EXTENDED); + if (retval) { + size_t length = regerror(retval, &s->accept_controller_regex, NULL, 0); + char *buffer = xmalloc(length); + regerror(retval, &s->accept_controller_regex, buffer, length); + ofp_fatal(0, "%s: %s", accept_re, buffer); + } + s->accept_controller_re = accept_re; + + /* Mode of operation. */ + s->discovery = s->controller_name == NULL; + if (s->discovery && !s->in_band) { + ofp_fatal(0, "Cannot perform discovery with out-of-band control"); + } + + /* Rate limiting. */ + if (s->rate_limit) { + if (s->rate_limit < 100) { + VLOG_WARN("Rate limit set to unusually low value %d", + s->rate_limit); + } + if (!s->burst_limit) { + s->burst_limit = s->rate_limit / 4; + } + s->burst_limit = MAX(s->burst_limit, 1); + s->burst_limit = MIN(s->burst_limit, INT_MAX / 1000); + } } static void @@ -814,14 +817,14 @@ usage(void) "usage: %s [OPTIONS] nl:DP_IDX [CONTROLLER]\n" "where nl:DP_IDX is a datapath that has been added with dpctl.\n" "CONTROLLER is an active OpenFlow connection method; if it is\n" - "omitted, then secchan performs controller autodiscovery.\n", + "omitted, then secchan performs controller discovery.\n", program_name, program_name); - vconn_usage(true, true); + vconn_usage(true, true, true); printf("\nController discovery options:\n" " --accept-vconn=REGEX accept matching discovered controllers\n" " --no-resolv-conf do not update /etc/resolv.conf\n" "\nNetworking options:\n" - " -f, --fail=open|closed when controller connection fails:\n" + " -F, --fail=open|closed when controller connection fails:\n" " closed: drop all packets\n" " open (default): act as learning switch\n" " --inactivity-probe=SECS time between inactivity probes\n" @@ -830,13 +833,22 @@ usage(void) " attempts (default: 15 seconds)\n" " -l, --listen=METHOD allow management connections on METHOD\n" " (a passive OpenFlow connection method)\n" - "\nOther options:\n" - " -D, --detach run in background as daemon\n" - " -P, --pidfile[=FILE] create pidfile (default: %s/secchan.pid)\n" - " -v, --verbose=MODULE[:FACILITY[:LEVEL]] set logging levels\n" - " -v, --verbose set maximum verbosity level\n" + " -m, --monitor=METHOD copy traffic to/from kernel to METHOD\n" + " (a passive OpenFlow connection method)\n" + " --out-of-band controller connection is out-of-band\n" + " --stp enable 802.1D Spanning Tree Protocol\n" + " --no-stp disable 802.1D Spanning Tree Protocol\n" + "\nRate-limiting of \"packet-in\" messages to the controller:\n" + " --rate-limit[=PACKETS] max rate, in packets/s (default: 1000)\n" + " --burst-limit=BURST limit on packet credit for idle time\n" + "\nRemote command execution options:\n" + " --command-acl=[!]GLOB[,[!]GLOB...] set allowed/denied commands\n" + " --command-dir=DIR set command dir (default: %s/commands)\n", + ofp_pkgdatadir); + daemon_usage(); + vlog_usage(); + printf("\nOther options:\n" " -h, --help display this help message\n" - " -V, --version display version information\n", - RUNDIR); + " -V, --version display version information\n"); exit(EXIT_SUCCESS); }