/*
- * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include "coverage.h"
#include "dpctl.h"
+#include "dp-packet.h"
+#include "dpif-netdev.h"
#include "dynamic-string.h"
#include "flow.h"
#include "netdev.h"
#include "ofp-print.h"
#include "ofp-util.h"
#include "ofpbuf.h"
-#include "packet-dpif.h"
#include "packets.h"
#include "poll-loop.h"
+#include "route-table.h"
+#include "seq.h"
#include "shash.h"
#include "sset.h"
#include "timeval.h"
+#include "tnl-neigh-cache.h"
+#include "tnl-ports.h"
#include "util.h"
+#include "uuid.h"
#include "valgrind.h"
-#include "vlog.h"
+#include "openvswitch/vlog.h"
VLOG_DEFINE_THIS_MODULE(dpif);
COVERAGE_DEFINE(dpif_execute_with_help);
static const struct dpif_class *base_dpif_classes[] = {
-#ifdef __linux__
- &dpif_linux_class,
+#if defined(__linux__) || defined(_WIN32)
+ &dpif_netlink_class,
#endif
&dpif_netdev_class,
};
const char *operation,
const struct nlattr *key, size_t key_len,
const struct nlattr *mask, size_t mask_len,
+ const ovs_u128 *ufid,
const struct dpif_flow_stats *stats,
const struct nlattr *actions, size_t actions_len);
static void log_operation(const struct dpif *, const char *operation,
int error);
static void log_execute_message(struct dpif *, const struct dpif_execute *,
bool subexecute, int error);
+static void log_flow_get_message(const struct dpif *,
+ const struct dpif_flow_get *, int error);
+
+/* Incremented whenever tnl route, arp, etc changes. */
+struct seq *tnl_conf_seq;
static void
dp_initialize(void)
if (ovsthread_once_start(&once)) {
int i;
+ tnl_conf_seq = seq_create();
+ dpctl_unixctl_register();
+ tnl_port_map_init();
+ tnl_neigh_cache_init();
+ route_table_init();
+
for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
dp_register_provider(base_dpif_classes[i]);
}
- dpctl_unixctl_register();
+
ovsthread_once_done(&once);
}
}
dp_register_provider__(const struct dpif_class *new_class)
{
struct registered_dpif_class *registered_class;
+ int error;
if (sset_contains(&dpif_blacklist, new_class->type)) {
VLOG_DBG("attempted to register blacklisted provider: %s",
return EEXIST;
}
+ error = new_class->init ? new_class->init() : 0;
+ if (error) {
+ VLOG_WARN("failed to initialize %s datapath class: %s",
+ new_class->type, ovs_strerror(error));
+ return error;
+ }
+
registered_class = xmalloc(sizeof *registered_class);
registered_class->dpif_class = new_class;
registered_class->refcount = 0;
ovs_mutex_unlock(&dpif_mutex);
}
-/* Clears 'types' and enumerates the types of all currently registered datapath
- * providers into it. The caller must first initialize the sset. */
+/* Adds the types of all currently registered datapath providers to 'types'.
+ * The caller must first initialize the sset. */
void
dp_enumerate_types(struct sset *types)
{
struct shash_node *node;
dp_initialize();
- sset_clear(types);
ovs_mutex_lock(&dpif_mutex);
SHASH_FOR_EACH(node, &dpif_classes) {
}
/* Performs periodic work needed by 'dpif'. */
-void
+bool
dpif_run(struct dpif *dpif)
{
if (dpif->dpif_class->run) {
- dpif->dpif_class->run(dpif);
+ return dpif->dpif_class->run(dpif);
}
+ return false;
}
/* Arranges for poll_block() to wake up when dp_run() needs to be called for
* arguments must have been initialized through a call to flow_extract().
* 'used' is stored into stats->used. */
void
-dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet,
+dpif_flow_stats_extract(const struct flow *flow, const struct dp_packet *packet,
long long int used, struct dpif_flow_stats *stats)
{
stats->tcp_flags = ntohs(flow->tcp_flags);
- stats->n_bytes = ofpbuf_size(packet);
+ stats->n_bytes = dp_packet_size(packet);
stats->n_packets = 1;
stats->used = used;
}
}
}
+/* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
+void
+dpif_flow_hash(const struct dpif *dpif OVS_UNUSED,
+ const void *key, size_t key_len, ovs_u128 *hash)
+{
+ static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
+ static uint32_t secret;
+
+ if (ovsthread_once_start(&once)) {
+ secret = random_uint32();
+ ovsthread_once_done(&once);
+ }
+ hash_bytes128(key, key_len, secret, hash);
+ uuid_set_bits_v4((struct uuid *)hash);
+}
+
/* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
* positive errno value. */
int
return error;
}
-/* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
- * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
- * 'key'.
- *
- * Returns 0 if successful. If no flow matches, returns ENOENT. On other
- * failure, returns a positive errno value.
- *
- * On success, '*bufp' will be set to an ofpbuf owned by the caller that
- * contains the response for 'flow->mask' and 'flow->actions'. The caller must
- * supply a valid pointer, and must free the ofpbuf (with ofpbuf_delete()) when
- * it is no longer needed.
- *
- * On success, 'flow' will be populated with the mask, actions and stats for
- * the datapath flow corresponding to 'key'. The mask and actions will point
- * within '*bufp'.
- *
- * Implementations may opt to point 'flow->mask' and/or 'flow->actions' at
- * RCU-protected data rather than making a copy of them. Therefore, callers
- * that wish to hold these over quiescent periods must make a copy of these
- * fields before quiescing. */
-int
-dpif_flow_get(const struct dpif *dpif,
- const struct nlattr *key, size_t key_len,
- struct ofpbuf **bufp, struct dpif_flow *flow)
+/* Attempts to install 'key' into the datapath, fetches it, then deletes it.
+ * Returns true if the datapath supported installing 'flow', false otherwise.
+ */
+bool
+dpif_probe_feature(struct dpif *dpif, const char *name,
+ const struct ofpbuf *key, const ovs_u128 *ufid)
{
+ struct dpif_flow flow;
+ struct ofpbuf reply;
+ uint64_t stub[DPIF_FLOW_BUFSIZE / 8];
+ bool enable_feature = false;
int error;
- struct nlattr *mask, *actions;
- size_t mask_len, actions_len;
- struct dpif_flow_stats stats;
-
- COVERAGE_INC(dpif_flow_get);
- *bufp = NULL;
- error = dpif->dpif_class->flow_get(dpif, key, key_len, bufp,
- &mask, &mask_len,
- &actions, &actions_len, &stats);
+ /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
+ * restarted) at just the right time such that feature probes from the
+ * previous run are still present in the datapath. */
+ error = dpif_flow_put(dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY | DPIF_FP_PROBE,
+ key->data, key->size, NULL, 0, NULL, 0,
+ ufid, PMD_ID_NULL, NULL);
if (error) {
- memset(flow, 0, sizeof *flow);
- ofpbuf_delete(*bufp);
- *bufp = NULL;
- } else {
- flow->mask = mask;
- flow->mask_len = mask_len;
- flow->actions = actions;
- flow->actions_len = actions_len;
- flow->stats = stats;
+ if (error != EINVAL) {
+ VLOG_WARN("%s: %s flow probe failed (%s)",
+ dpif_name(dpif), name, ovs_strerror(error));
+ }
+ return false;
}
- if (should_log_flow_message(error)) {
- log_flow_message(dpif, error, "flow_get", key, key_len,
- NULL, 0, &flow->stats,
- flow->actions, flow->actions_len);
+
+ ofpbuf_use_stack(&reply, &stub, sizeof stub);
+ error = dpif_flow_get(dpif, key->data, key->size, ufid,
+ PMD_ID_NULL, &reply, &flow);
+ if (!error
+ && (!ufid || (flow.ufid_present
+ && ovs_u128_equals(ufid, &flow.ufid)))) {
+ enable_feature = true;
}
- return error;
+
+ error = dpif_flow_del(dpif, key->data, key->size, ufid,
+ PMD_ID_NULL, NULL);
+ if (error) {
+ VLOG_WARN("%s: failed to delete %s feature probe flow",
+ dpif_name(dpif), name);
+ }
+
+ return enable_feature;
+}
+
+/* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
+int
+dpif_flow_get(struct dpif *dpif,
+ const struct nlattr *key, size_t key_len, const ovs_u128 *ufid,
+ const unsigned pmd_id, struct ofpbuf *buf, struct dpif_flow *flow)
+{
+ struct dpif_op *opp;
+ struct dpif_op op;
+
+ op.type = DPIF_OP_FLOW_GET;
+ op.u.flow_get.key = key;
+ op.u.flow_get.key_len = key_len;
+ op.u.flow_get.ufid = ufid;
+ op.u.flow_get.pmd_id = pmd_id;
+ op.u.flow_get.buffer = buf;
+
+ memset(flow, 0, sizeof *flow);
+ op.u.flow_get.flow = flow;
+ op.u.flow_get.flow->key = key;
+ op.u.flow_get.flow->key_len = key_len;
+
+ opp = &op;
+ dpif_operate(dpif, &opp, 1);
+
+ return op.error;
}
/* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
const struct nlattr *key, size_t key_len,
const struct nlattr *mask, size_t mask_len,
const struct nlattr *actions, size_t actions_len,
+ const ovs_u128 *ufid, const unsigned pmd_id,
struct dpif_flow_stats *stats)
{
struct dpif_op *opp;
op.u.flow_put.mask_len = mask_len;
op.u.flow_put.actions = actions;
op.u.flow_put.actions_len = actions_len;
+ op.u.flow_put.ufid = ufid;
+ op.u.flow_put.pmd_id = pmd_id;
op.u.flow_put.stats = stats;
opp = &op;
/* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
int
dpif_flow_del(struct dpif *dpif,
- const struct nlattr *key, size_t key_len,
- struct dpif_flow_stats *stats)
+ const struct nlattr *key, size_t key_len, const ovs_u128 *ufid,
+ const unsigned pmd_id, struct dpif_flow_stats *stats)
{
struct dpif_op *opp;
struct dpif_op op;
op.type = DPIF_OP_FLOW_DEL;
op.u.flow_del.key = key;
op.u.flow_del.key_len = key_len;
+ op.u.flow_del.ufid = ufid;
+ op.u.flow_del.pmd_id = pmd_id;
op.u.flow_del.stats = stats;
+ op.u.flow_del.terse = false;
opp = &op;
dpif_operate(dpif, &opp, 1);
}
/* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
- * flows in 'dpif'.
+ * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
+ * be returned in the dump. Otherwise, all fields will be returned.
*
* This function always successfully returns a dpif_flow_dump. Error
* reporting is deferred to dpif_flow_dump_destroy(). */
struct dpif_flow_dump *
-dpif_flow_dump_create(const struct dpif *dpif)
+dpif_flow_dump_create(const struct dpif *dpif, bool terse)
{
- return dpif->dpif_class->flow_dump_create(dpif);
+ return dpif->dpif_class->flow_dump_create(dpif, terse);
}
/* Destroys 'dump', which must have been created with dpif_flow_dump_create().
for (f = flows; f < &flows[n] && should_log_flow_message(0); f++) {
log_flow_message(dpif, 0, "flow_dump",
f->key, f->key_len, f->mask, f->mask_len,
- &f->stats, f->actions, f->actions_len);
+ &f->ufid, &f->stats, f->actions, f->actions_len);
}
} else {
VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
/* This is called for actions that need the context of the datapath to be
* meaningful. */
static void
-dpif_execute_helper_cb(void *aux_, struct dpif_packet **packets, int cnt,
- struct pkt_metadata *md,
+dpif_execute_helper_cb(void *aux_, struct dp_packet **packets, int cnt,
const struct nlattr *action, bool may_steal OVS_UNUSED)
{
struct dpif_execute_helper_aux *aux = aux_;
int type = nl_attr_type(action);
- struct ofpbuf * packet = &packets[0]->ofpbuf;
+ struct dp_packet *packet = *packets;
ovs_assert(cnt == 1);
switch ((enum ovs_action_attr)type) {
+ case OVS_ACTION_ATTR_CT:
case OVS_ACTION_ATTR_OUTPUT:
+ case OVS_ACTION_ATTR_TUNNEL_PUSH:
+ case OVS_ACTION_ATTR_TUNNEL_POP:
case OVS_ACTION_ATTR_USERSPACE:
case OVS_ACTION_ATTR_RECIRC: {
struct dpif_execute execute;
struct ofpbuf execute_actions;
uint64_t stub[256 / 8];
+ struct pkt_metadata *md = &packet->md;
+ bool dst_set;
- if (md->tunnel.ip_dst) {
+ dst_set = flow_tnl_dst_is_set(&md->tunnel);
+ if (dst_set) {
/* The Linux kernel datapath throws away the tunnel information
* that we supply as metadata. We have to use a "set" action to
* supply it. */
odp_put_tunnel_action(&md->tunnel, &execute_actions);
ofpbuf_put(&execute_actions, action, NLA_ALIGN(action->nla_len));
- execute.actions = ofpbuf_data(&execute_actions);
- execute.actions_len = ofpbuf_size(&execute_actions);
+ execute.actions = execute_actions.data;
+ execute.actions_len = execute_actions.size;
} else {
execute.actions = action;
execute.actions_len = NLA_ALIGN(action->nla_len);
}
execute.packet = packet;
- execute.md = *md;
execute.needs_help = false;
+ execute.probe = false;
+ execute.mtu = 0;
aux->error = dpif_execute(aux->dpif, &execute);
log_execute_message(aux->dpif, &execute, true, aux->error);
- if (md->tunnel.ip_dst) {
+ if (dst_set) {
ofpbuf_uninit(&execute_actions);
}
break;
case OVS_ACTION_ATTR_PUSH_MPLS:
case OVS_ACTION_ATTR_POP_MPLS:
case OVS_ACTION_ATTR_SET:
+ case OVS_ACTION_ATTR_SET_MASKED:
case OVS_ACTION_ATTR_SAMPLE:
case OVS_ACTION_ATTR_UNSPEC:
case __OVS_ACTION_ATTR_MAX:
dpif_execute_with_help(struct dpif *dpif, struct dpif_execute *execute)
{
struct dpif_execute_helper_aux aux = {dpif, 0};
- struct dpif_packet packet, *pp;
+ struct dp_packet *pp;
COVERAGE_INC(dpif_execute_with_help);
- packet.ofpbuf = *execute->packet;
- pp = &packet;
-
- odp_execute_actions(&aux, &pp, 1, false, &execute->md, execute->actions,
+ pp = execute->packet;
+ odp_execute_actions(&aux, &pp, 1, false, execute->actions,
execute->actions_len, dpif_execute_helper_cb);
-
- /* Even though may_steal is set to false, some actions could modify or
- * reallocate the ofpbuf memory. We need to pass those changes to the
- * caller */
- *execute->packet = packet.ofpbuf;
-
return aux.error;
}
break;
}
+ case DPIF_OP_FLOW_GET: {
+ struct dpif_flow_get *get = &op->u.flow_get;
+
+ COVERAGE_INC(dpif_flow_get);
+ if (error) {
+ memset(get->flow, 0, sizeof *get->flow);
+ }
+ log_flow_get_message(dpif, get, error);
+
+ break;
+ }
+
case DPIF_OP_FLOW_DEL: {
struct dpif_flow_del *del = &op->u.flow_del;
}
void
-dpif_register_upcall_cb(struct dpif *dpif, exec_upcall_cb *cb)
+dpif_register_dp_purge_cb(struct dpif *dpif, dp_purge_callback *cb, void *aux)
+{
+ if (dpif->dpif_class->register_dp_purge_cb) {
+ dpif->dpif_class->register_dp_purge_cb(dpif, cb, aux);
+ }
+}
+
+void
+dpif_register_upcall_cb(struct dpif *dpif, upcall_callback *cb, void *aux)
{
if (dpif->dpif_class->register_upcall_cb) {
- dpif->dpif_class->register_upcall_cb(dpif, cb);
+ dpif->dpif_class->register_upcall_cb(dpif, cb, aux);
}
}
struct ds flow;
char *packet;
- packet = ofp_packet_to_string(ofpbuf_data(&upcall->packet),
- ofpbuf_size(&upcall->packet));
+ packet = ofp_packet_to_string(dp_packet_data(&upcall->packet),
+ dp_packet_size(&upcall->packet));
ds_init(&flow);
odp_flow_key_format(upcall->key, upcall->key_len, &flow);
}
}
+/* If 'dpif' creates its own I/O polling threads, refreshes poll threads
+ * configuration. */
+int
+dpif_poll_threads_set(struct dpif *dpif, const char *cmask)
+{
+ int error = 0;
+
+ if (dpif->dpif_class->poll_threads_set) {
+ error = dpif->dpif_class->poll_threads_set(dpif, cmask);
+ if (error) {
+ log_operation(dpif, "poll_threads_set", error);
+ }
+ }
+
+ return error;
+}
+
/* Polls for an upcall from 'dpif' for an upcall handler. Since there
* there can be multiple poll loops, 'handler_id' is needed as index to
* identify the corresponding poll loop. If successful, stores the upcall
}
}
+/*
+ * Return the datapath version. Caller is responsible for freeing
+ * the string.
+ */
+char *
+dpif_get_dp_version(const struct dpif *dpif)
+{
+ char *version = NULL;
+
+ if (dpif->dpif_class->get_datapath_version) {
+ version = dpif->dpif_class->get_datapath_version();
+ }
+
+ return version;
+}
+
/* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
* and '*engine_id', respectively. */
void
static bool
should_log_flow_message(int error)
{
- return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
+ return !vlog_should_drop(&this_module, flow_message_log_level(error),
error ? &error_rl : &dpmsg_rl);
}
log_flow_message(const struct dpif *dpif, int error, const char *operation,
const struct nlattr *key, size_t key_len,
const struct nlattr *mask, size_t mask_len,
- const struct dpif_flow_stats *stats,
+ const ovs_u128 *ufid, const struct dpif_flow_stats *stats,
const struct nlattr *actions, size_t actions_len)
{
struct ds ds = DS_EMPTY_INITIALIZER;
if (error) {
ds_put_format(&ds, "(%s) ", ovs_strerror(error));
}
+ if (ufid) {
+ odp_format_ufid(ufid, &ds);
+ ds_put_cstr(&ds, " ");
+ }
odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, true);
if (stats) {
ds_put_cstr(&ds, ", ");
ds_put_cstr(&ds, ", actions:");
format_odp_actions(&ds, actions, actions_len);
}
- vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
+ vlog(&this_module, flow_message_log_level(error), "%s", ds_cstr(&ds));
ds_destroy(&ds);
}
log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put,
int error)
{
- if (should_log_flow_message(error)) {
+ if (should_log_flow_message(error) && !(put->flags & DPIF_FP_PROBE)) {
struct ds s;
ds_init(&s);
}
log_flow_message(dpif, error, ds_cstr(&s),
put->key, put->key_len, put->mask, put->mask_len,
- put->stats, put->actions, put->actions_len);
+ put->ufid, put->stats, put->actions,
+ put->actions_len);
ds_destroy(&s);
}
}
{
if (should_log_flow_message(error)) {
log_flow_message(dpif, error, "flow_del", del->key, del->key_len,
- NULL, 0, !error ? del->stats : NULL, NULL, 0);
+ NULL, 0, del->ufid, !error ? del->stats : NULL,
+ NULL, 0);
}
}
* called after the dpif_provider's '->execute' function, which is allowed to
* modify execute->packet and execute->md. In practice, though:
*
- * - dpif-linux doesn't modify execute->packet or execute->md.
+ * - dpif-netlink doesn't modify execute->packet or execute->md.
*
* - dpif-netdev does modify them but it is less likely to have problems
* because it is built into ovs-vswitchd and cannot have version skew,
log_execute_message(struct dpif *dpif, const struct dpif_execute *execute,
bool subexecute, int error)
{
- if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
+ if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))
+ && !execute->probe) {
struct ds ds = DS_EMPTY_INITIALIZER;
char *packet;
- packet = ofp_packet_to_string(ofpbuf_data(execute->packet),
- ofpbuf_size(execute->packet));
+ packet = ofp_packet_to_string(dp_packet_data(execute->packet),
+ dp_packet_size(execute->packet));
ds_put_format(&ds, "%s: %sexecute ",
dpif_name(dpif),
(subexecute ? "sub-"
ds_put_format(&ds, " failed (%s)", ovs_strerror(error));
}
ds_put_format(&ds, " on packet %s", packet);
- vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
+ ds_put_format(&ds, " mtu %d", execute->mtu);
+ vlog(&this_module, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
ds_destroy(&ds);
free(packet);
}
}
+
+static void
+log_flow_get_message(const struct dpif *dpif, const struct dpif_flow_get *get,
+ int error)
+{
+ if (should_log_flow_message(error)) {
+ log_flow_message(dpif, error, "flow_get",
+ get->key, get->key_len,
+ get->flow->mask, get->flow->mask_len,
+ get->ufid, &get->flow->stats,
+ get->flow->actions, get->flow->actions_len);
+ }
+}
+
+bool
+dpif_supports_tnl_push_pop(const struct dpif *dpif)
+{
+ return dpif_is_netdev(dpif);
+}
projects / cascardo / ovs.git / blobdiff