#include "bitmap.h"
#include "byte-order.h"
#include "classifier.h"
+#include "connectivity.h"
#include "connmgr.h"
#include "coverage.h"
#include "dynamic-string.h"
#include "pktbuf.h"
#include "poll-loop.h"
#include "random.h"
+#include "seq.h"
#include "shash.h"
#include "simap.h"
+#include "smap.h"
#include "sset.h"
#include "timeval.h"
#include "unaligned.h"
VLOG_DEFINE_THIS_MODULE(ofproto);
-COVERAGE_DEFINE(ofproto_error);
COVERAGE_DEFINE(ofproto_flush);
-COVERAGE_DEFINE(ofproto_no_packet_in);
COVERAGE_DEFINE(ofproto_packet_out);
COVERAGE_DEFINE(ofproto_queue_req);
COVERAGE_DEFINE(ofproto_recv_openflow);
COVERAGE_DEFINE(ofproto_reinit_ports);
-COVERAGE_DEFINE(ofproto_uninstallable);
COVERAGE_DEFINE(ofproto_update_port);
enum ofproto_state {
static int init_ports(struct ofproto *);
static void reinit_ports(struct ofproto *);
+static long long int ofport_get_usage(const struct ofproto *,
+ ofp_port_t ofp_port);
+static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
+ long long int last_used);
+static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
+
+/* Ofport usage.
+ *
+ * Keeps track of the currently used and recently used ofport values and is
+ * used to prevent immediate recycling of ofport values. */
+struct ofport_usage {
+ struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
+ ofp_port_t ofp_port; /* OpenFlow port number. */
+ long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
+ represents in-use ofports. */
+};
+
/* rule. */
static void ofproto_rule_destroy__(struct rule *);
static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
static void delete_flow__(struct rule *rule, struct ofopgroup *,
enum ofp_flow_removed_reason)
OVS_REQUIRES(ofproto_mutex);
+static bool ofproto_group_exists__(const struct ofproto *ofproto,
+ uint32_t group_id)
+ OVS_REQ_RDLOCK(ofproto->groups_rwlock);
+static bool ofproto_group_exists(const struct ofproto *ofproto,
+ uint32_t group_id)
+ OVS_EXCLUDED(ofproto->groups_rwlock);
static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
/* Global lock that protects all flow table operations. */
struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
-unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
-unsigned n_handler_threads;
+unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
+size_t n_handlers, n_revalidators;
+
/* Map from datapath name to struct ofproto, for use by unixctl commands. */
static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
{
size_t i;
+ sset_clear(types);
for (i = 0; i < n_ofproto_classes; i++) {
ofproto_classes[i]->enumerate_types(types);
}
{
const struct ofproto_class *class = ofproto_class_find__(type);
return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
- }
+}
int
ofproto_create(const char *datapath_name, const char *datapath_type,
ofproto->dp_desc = NULL;
ofproto->frag_handling = OFPC_FRAG_NORMAL;
hmap_init(&ofproto->ports);
+ hmap_init(&ofproto->ofport_usage);
shash_init(&ofproto->port_by_name);
simap_init(&ofproto->ofp_requests);
ofproto->max_ports = ofp_to_u16(OFPP_MAX);
ovs_rwlock_init(&ofproto->groups_rwlock);
hmap_init(&ofproto->groups);
ovs_mutex_unlock(&ofproto_mutex);
+ ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
+ OFPGFC_SELECT_WEIGHT;
+ ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX;
+ ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX;
+ ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX;
+ ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX;
+ ofproto->ogf.actions[0] =
+ (1 << OFPAT11_OUTPUT) |
+ (1 << OFPAT11_COPY_TTL_OUT) |
+ (1 << OFPAT11_COPY_TTL_IN) |
+ (1 << OFPAT11_SET_MPLS_TTL) |
+ (1 << OFPAT11_DEC_MPLS_TTL) |
+ (1 << OFPAT11_PUSH_VLAN) |
+ (1 << OFPAT11_POP_VLAN) |
+ (1 << OFPAT11_PUSH_MPLS) |
+ (1 << OFPAT11_POP_MPLS) |
+ (1 << OFPAT11_SET_QUEUE) |
+ (1 << OFPAT11_GROUP) |
+ (1 << OFPAT11_SET_NW_TTL) |
+ (1 << OFPAT11_DEC_NW_TTL) |
+ (1 << OFPAT12_SET_FIELD);
+/* not supported:
+ * (1 << OFPAT13_PUSH_PBB) |
+ * (1 << OFPAT13_POP_PBB) */
error = ofproto->ofproto_class->construct(ofproto);
if (error) {
return error;
}
- /* The "max_ports" member should have been set by ->construct(ofproto).
- * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
- ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
- bitmap_set1(ofproto->ofp_port_ids, 0);
-
/* Check that hidden tables, if any, are at the end. */
ovs_assert(ofproto->n_tables);
for (i = 0; i + 1 < ofproto->n_tables; i++) {
/* Sets the number of flows at which eviction from the kernel flow table
* will occur. */
void
-ofproto_set_flow_eviction_threshold(unsigned threshold)
+ofproto_set_flow_limit(unsigned limit)
{
- flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
- threshold);
+ ofproto_flow_limit = limit;
}
/* Sets the path for handling flow misses. */
}
}
-/* Sets number of upcall handler threads. The default is
- * (number of online cores - 2). */
void
-ofproto_set_n_handler_threads(unsigned limit)
+ofproto_set_threads(size_t n_handlers_, size_t n_revalidators_)
{
- if (limit) {
- n_handler_threads = limit;
- } else {
- int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
- n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
+ int threads = MAX(count_cpu_cores(), 2);
+
+ n_revalidators = n_revalidators_;
+ n_handlers = n_handlers_;
+
+ if (!n_revalidators) {
+ n_revalidators = n_handlers
+ ? MAX(threads - (int) n_handlers, 1)
+ : threads / 4 + 1;
+ }
+
+ if (!n_handlers) {
+ n_handlers = MAX(threads - (int) n_revalidators, 1);
}
}
? ofproto->ofproto_class->get_stp_port_status(ofport, s)
: EOPNOTSUPP);
}
+
+/* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
+ * 's'. If the 'enabled' member in 's' is false, then the other members
+ * are not meaningful.
+ *
+ * Returns 0 if successful, otherwise a positive errno value.*/
+int
+ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
+ struct ofproto_port_stp_stats *s)
+{
+ struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+ if (!ofport) {
+ VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
+ "port %"PRIu16, ofproto->name, ofp_port);
+ return ENODEV;
+ }
+
+ return (ofproto->ofproto_class->get_stp_port_stats
+ ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
+ : EOPNOTSUPP);
+}
\f
/* Queue DSCP configuration. */
}
table->max_flows = s->max_flows;
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_wrlock(&table->cls.rwlock);
if (classifier_count(&table->cls) > table->max_flows
&& table->eviction_fields) {
/* 'table' contains more flows than allowed. We might not be able to
break;
}
}
- ovs_rwlock_unlock(&table->cls.rwlock);
+
+ classifier_set_prefix_fields(&table->cls,
+ s->prefix_fields, s->n_prefix_fields);
+
+ fat_rwlock_unlock(&table->cls.rwlock);
}
\f
bool
}
static void
-ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule)
+ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule,
+ uint8_t reason)
OVS_REQUIRES(ofproto_mutex)
{
struct ofopgroup *group;
ovs_assert(!rule->pending);
group = ofopgroup_create_unattached(ofproto);
- delete_flow__(rule, group, OFPRR_DELETE);
+ delete_flow__(rule, group, reason);
ofopgroup_submit(group);
}
continue;
}
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, NULL);
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
if (!rule->pending) {
- ofproto_rule_delete__(ofproto, rule);
+ ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE);
}
}
}
free(ofproto->serial_desc);
free(ofproto->dp_desc);
hmap_destroy(&ofproto->ports);
+ hmap_destroy(&ofproto->ofport_usage);
shash_destroy(&ofproto->port_by_name);
- bitmap_free(ofproto->ofp_port_ids);
simap_destroy(&ofproto->ofp_requests);
OFPROTO_FOR_EACH_TABLE (table, ofproto) {
OVS_EXCLUDED(ofproto_mutex)
{
struct ofport *ofport, *next_ofport;
+ struct ofport_usage *usage, *next_usage;
if (!p) {
return;
ofport_destroy(ofport);
}
+ HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
+ hmap_remove(&p->ofport_usage, &usage->hmap_node);
+ free(usage);
+ }
+
p->ofproto_class->destruct(p);
ofproto_destroy__(p);
}
return error;
}
-int
-ofproto_type_run_fast(const char *datapath_type)
-{
- const struct ofproto_class *class;
- int error;
-
- datapath_type = ofproto_normalize_type(datapath_type);
- class = ofproto_class_find__(datapath_type);
-
- error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
- if (error && error != EAGAIN) {
- VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
- datapath_type, ovs_strerror(error));
- }
- return error;
-}
-
void
ofproto_type_wait(const char *datapath_type)
{
int
ofproto_run(struct ofproto *p)
{
- struct sset changed_netdevs;
- const char *changed_netdev;
- struct ofport *ofport;
int error;
+ uint64_t new_seq;
error = p->ofproto_class->run(p);
if (error && error != EAGAIN) {
heap_rebuild(&evg->rules);
}
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
if (!rule->eviction_group
eviction_group_add_rule(rule);
}
}
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
ovs_mutex_unlock(&ofproto_mutex);
}
}
}
}
- /* Update OpenFlow port status for any port whose netdev has changed.
- *
- * Refreshing a given 'ofport' can cause an arbitrary ofport to be
- * destroyed, so it's not safe to update ports directly from the
- * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
- * need this two-phase approach. */
- sset_init(&changed_netdevs);
- HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
- unsigned int change_seq = netdev_change_seq(ofport->netdev);
- if (ofport->change_seq != change_seq) {
- ofport->change_seq = change_seq;
- sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
+ new_seq = seq_read(connectivity_seq_get());
+ if (new_seq != p->change_seq) {
+ struct sset devnames;
+ const char *devname;
+ struct ofport *ofport;
+
+ /* Update OpenFlow port status for any port whose netdev has changed.
+ *
+ * Refreshing a given 'ofport' can cause an arbitrary ofport to be
+ * destroyed, so it's not safe to update ports directly from the
+ * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
+ * need this two-phase approach. */
+ sset_init(&devnames);
+ HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
+ sset_add(&devnames, netdev_get_name(ofport->netdev));
}
+ SSET_FOR_EACH (devname, &devnames) {
+ update_port(p, devname);
+ }
+ sset_destroy(&devnames);
+
+ p->change_seq = new_seq;
}
- SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
- update_port(p, changed_netdev);
- }
- sset_destroy(&changed_netdevs);
switch (p->state) {
case S_OPENFLOW:
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
if (time_msec() >= p->next_op_report) {
return error;
}
-/* Performs periodic activity required by 'ofproto' that needs to be done
- * with the least possible latency.
- *
- * It makes sense to call this function a couple of times per poll loop, to
- * provide a significant performance boost on some benchmarks with the
- * ofproto-dpif implementation. */
-int
-ofproto_run_fast(struct ofproto *p)
-{
- int error;
-
- error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
- if (error && error != EAGAIN) {
- VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
- p->name, ovs_strerror(error));
- }
- return error;
-}
-
void
ofproto_wait(struct ofproto *p)
{
- struct ofport *ofport;
-
p->ofproto_class->wait(p);
if (p->ofproto_class->port_poll_wait) {
p->ofproto_class->port_poll_wait(p);
}
-
- HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
- if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
- poll_immediate_wake();
- }
- }
+ seq_wait(connectivity_seq_get(), p->change_seq);
switch (p->state) {
case S_OPENFLOW:
n_rules = 0;
OFPROTO_FOR_EACH_TABLE (table, ofproto) {
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
n_rules += classifier_count(&table->cls);
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
}
simap_increase(usage, "rules", n_rules);
connmgr_get_memory_usage(ofproto->connmgr, usage);
}
+void
+ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
+{
+ const struct ofproto_class *class;
+
+ datapath_type = ofproto_normalize_type(datapath_type);
+ class = ofproto_class_find__(datapath_type);
+
+ if (class && class->type_get_memory_usage) {
+ class->type_get_memory_usage(datapath_type, usage);
+ }
+}
+
void
ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
struct shash *info)
update_port(ofproto, netdev_name);
}
if (ofp_portp) {
- struct ofproto_port ofproto_port;
-
- ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
- &ofproto_port);
- *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
- ofproto_port_destroy(&ofproto_port);
+ *ofp_portp = OFPP_NONE;
+ if (!error) {
+ struct ofproto_port ofproto_port;
+
+ error = ofproto_port_query_by_name(ofproto,
+ netdev_get_name(netdev),
+ &ofproto_port);
+ if (!error) {
+ *ofp_portp = ofproto_port.ofp_port;
+ ofproto_port_destroy(&ofproto_port);
+ }
+ }
}
return error;
}
return error;
}
+static void
+flow_mod_init(struct ofputil_flow_mod *fm,
+ const struct match *match, unsigned int priority,
+ const struct ofpact *ofpacts, size_t ofpacts_len,
+ enum ofp_flow_mod_command command)
+{
+ memset(fm, 0, sizeof *fm);
+ fm->match = *match;
+ fm->priority = priority;
+ fm->cookie = 0;
+ fm->new_cookie = 0;
+ fm->modify_cookie = false;
+ fm->table_id = 0;
+ fm->command = command;
+ fm->idle_timeout = 0;
+ fm->hard_timeout = 0;
+ fm->buffer_id = UINT32_MAX;
+ fm->out_port = OFPP_ANY;
+ fm->out_group = OFPG_ANY;
+ fm->flags = 0;
+ fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
+ fm->ofpacts_len = ofpacts_len;
+}
+
static int
simple_flow_mod(struct ofproto *ofproto,
const struct match *match, unsigned int priority,
{
struct ofputil_flow_mod fm;
- memset(&fm, 0, sizeof fm);
- fm.match = *match;
- fm.priority = priority;
- fm.cookie = 0;
- fm.new_cookie = 0;
- fm.modify_cookie = false;
- fm.table_id = 0;
- fm.command = command;
- fm.idle_timeout = 0;
- fm.hard_timeout = 0;
- fm.buffer_id = UINT32_MAX;
- fm.out_port = OFPP_ANY;
- fm.out_group = OFPG_ANY;
- fm.flags = 0;
- fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts);
- fm.ofpacts_len = ofpacts_len;
+ flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
return handle_flow_mod__(ofproto, NULL, &fm, NULL);
}
/* First do a cheap check whether the rule we're looking for already exists
* with the actions that we want. If it does, then we're done. */
- ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
+ fat_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
rule = rule_from_cls_rule(classifier_find_match_exactly(
&ofproto->tables[0].cls, match, priority));
if (rule) {
} else {
must_add = true;
}
- ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
+ fat_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
/* If there's no such rule or the rule doesn't have the actions we want,
* fall back to a executing a full flow mod. We can't optimize this at
/* First do a cheap check whether the rule we're looking for has already
* been deleted. If so, then we're done. */
- ovs_rwlock_rdlock(&cls->rwlock);
+ fat_rwlock_rdlock(&cls->rwlock);
rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
priority));
- ovs_rwlock_unlock(&cls->rwlock);
+ fat_rwlock_unlock(&cls->rwlock);
if (!rule) {
return true;
}
port_idx = port_idx ? port_idx : UINT16_MAX;
if (port_idx >= ofproto->max_ports
- || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
- uint16_t end_port_no = ofproto->alloc_port_no;
+ || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
+ uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
+ long long int last_used_at, lru = LLONG_MAX;
/* Search for a free OpenFlow port number. We try not to
* immediately reuse them to prevent problems due to old
- * flows. */
+ * flows.
+ *
+ * We limit the automatically assigned port numbers to the lower half
+ * of the port range, to reserve the upper half for assignment by
+ * controllers. */
for (;;) {
- if (++ofproto->alloc_port_no >= ofproto->max_ports) {
- ofproto->alloc_port_no = 0;
+ if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
+ ofproto->alloc_port_no = 1;
}
- if (!bitmap_is_set(ofproto->ofp_port_ids,
- ofproto->alloc_port_no)) {
+ last_used_at = ofport_get_usage(ofproto,
+ u16_to_ofp(ofproto->alloc_port_no));
+ if (!last_used_at) {
+ port_idx = ofproto->alloc_port_no;
+ break;
+ } else if ( last_used_at < time_msec() - 60*60*1000) {
+ /* If the port with ofport 'ofproto->alloc_port_no' was deleted
+ * more than an hour ago, consider it usable. */
+ ofport_remove_usage(ofproto,
+ u16_to_ofp(ofproto->alloc_port_no));
port_idx = ofproto->alloc_port_no;
break;
+ } else if (last_used_at < lru) {
+ lru = last_used_at;
+ lru_ofport = ofproto->alloc_port_no;
}
+
if (ofproto->alloc_port_no == end_port_no) {
+ if (lru_ofport) {
+ port_idx = lru_ofport;
+ break;
+ }
return OFPP_NONE;
}
}
}
- bitmap_set1(ofproto->ofp_port_ids, port_idx);
+ ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
return u16_to_ofp(port_idx);
}
static void
-dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
+dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
{
if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
- bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
+ ofport_set_usage(ofproto, ofp_port, time_msec());
}
}
}
ofport->ofproto = p;
ofport->netdev = netdev;
- ofport->change_seq = netdev_change_seq(netdev);
ofport->pp = *pp;
ofport->ofp_port = pp->port_no;
ofport->created = time_msec();
return NULL;
}
+static long long int
+ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
+{
+ struct ofport_usage *usage;
+
+ HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
+ &ofproto->ofport_usage) {
+ if (usage->ofp_port == ofp_port) {
+ return usage->last_used;
+ }
+ }
+ return 0;
+}
+
+static void
+ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
+ long long int last_used)
+{
+ struct ofport_usage *usage;
+ HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
+ &ofproto->ofport_usage) {
+ if (usage->ofp_port == ofp_port) {
+ usage->last_used = last_used;
+ return;
+ }
+ }
+ ovs_assert(last_used == LLONG_MAX);
+
+ usage = xmalloc(sizeof *usage);
+ usage->ofp_port = ofp_port;
+ usage->last_used = last_used;
+ hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
+ hash_ofp_port(ofp_port));
+}
+
+static void
+ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
+{
+ struct ofport_usage *usage;
+ HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
+ &ofproto->ofport_usage) {
+ if (usage->ofp_port == ofp_port) {
+ hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
+ free(usage);
+ break;
+ }
+ }
+}
+
int
ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
{
* Don't close the old netdev yet in case port_modified has to
* remove a retained reference to it.*/
port->netdev = netdev;
- port->change_seq = netdev_change_seq(netdev);
if (port->ofproto->ofproto_class->port_modified) {
port->ofproto->ofproto_class->port_modified(port);
atomic_sub(&actions->ref_count, 1, &orig);
if (orig == 1) {
+ free(actions->ofpacts);
free(actions);
} else {
ovs_assert(orig != 0);
op->actions->ofpacts_len, out_port);
}
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
static void
}
}
-/* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
- * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
- * 'flow' may be temporarily modified, but is restored at return.
- */
+/* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
+ * for 'ofproto':
+ *
+ * - If they use a meter, then 'ofproto' has that meter configured.
+ *
+ * - If they use any groups, then 'ofproto' has that group configured.
+ *
+ * Returns 0 if successful, otherwise an OpenFlow error. */
static enum ofperr
ofproto_check_ofpacts(struct ofproto *ofproto,
- const struct ofpact ofpacts[], size_t ofpacts_len,
- struct flow *flow, uint8_t table_id)
+ const struct ofpact ofpacts[], size_t ofpacts_len)
{
- enum ofperr error;
+ const struct ofpact *a;
uint32_t mid;
- error = ofpacts_check(ofpacts, ofpacts_len, flow,
- u16_to_ofp(ofproto->max_ports), table_id);
- if (error) {
- return error;
- }
-
mid = ofpacts_get_meter(ofpacts, ofpacts_len);
if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
return OFPERR_OFPMMFC_INVALID_METER;
}
+
+ OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
+ if (a->type == OFPACT_GROUP
+ && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
+ return OFPERR_OFPBAC_BAD_OUT_GROUP;
+ }
+ }
+
return 0;
}
goto exit_free_ofpacts;
}
-
/* Get payload. */
if (po.buffer_id != UINT32_MAX) {
error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
/* Verify actions against packet, then send packet if successful. */
in_port_.ofp_port = po.in_port;
flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
- error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
+ error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
if (!error) {
error = p->ofproto_class->packet_out(p, payload, &flow,
po.ofpacts, po.ofpacts_len);
ots = xcalloc(p->n_tables, sizeof *ots);
for (i = 0; i < p->n_tables; i++) {
ots[i].table_id = i;
- sprintf(ots[i].name, "table%zu", i);
+ sprintf(ots[i].name, "table%"PRIuSIZE, i);
ots[i].match = htonll(OFPXMT13_MASK);
ots[i].wildcards = htonll(OFPXMT13_MASK);
ots[i].write_actions = htonl(OFPAT11_OUTPUT);
ots[i].instructions = htonl(OFPIT11_ALL);
ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
- ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
+ fat_rwlock_rdlock(&p->tables[i].cls.rwlock);
ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
- ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
+ fat_rwlock_unlock(&p->tables[i].cls.rwlock);
}
p->ofproto_class->get_tables(p, ots);
struct rule_collection *rules)
OVS_REQUIRES(ofproto_mutex)
{
- if (ofproto_rule_is_hidden(rule)) {
+ /* We ordinarily want to skip hidden rules, but there has to be a way for
+ * code internal to OVS to modify and delete them, so if the criteria
+ * specify a priority that can only be for a hidden flow, then allow hidden
+ * rules to be selected. (This doesn't allow OpenFlow clients to meddle
+ * with hidden flows because OpenFlow uses only a 16-bit field to specify
+ * priority.) */
+ if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
return 0;
} else if (rule->pending) {
return OFPROTO_POSTPONE;
struct cls_cursor cursor;
struct rule *rule;
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, &criteria->cr);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
error = collect_rule(rule, criteria, rules);
break;
}
}
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
}
}
FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
struct rule *rule;
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
rule = rule_from_cls_rule(classifier_find_rule_exactly(
&table->cls, &criteria->cr));
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
if (rule) {
error = collect_rule(rule, criteria, rules);
if (error) {
ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
}
ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
- ds_put_format(results, "priority=%u, ", rule->cr.priority);
ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
cls_rule_format(&rule->cr, results);
ds_put_char(results, ',');
+ ds_put_cstr(results, "actions=");
ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
ds_put_cstr(results, "\n");
struct cls_cursor cursor;
struct rule *rule;
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
flow_stats_ds(rule, results);
}
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
}
}
cls_rule_init(&cr, &fm->match, fm->priority);
/* Transform "add" into "modify" if there's an existing identical flow. */
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
if (rule) {
cls_rule_destroy(&cr);
if (!rule_is_modifiable(rule)) {
}
}
- /* Verify actions. */
- error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
- &fm->match.flow, table_id);
- if (error) {
- cls_rule_destroy(&cr);
- return error;
- }
-
/* Serialize against pending deletion. */
if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
cls_rule_destroy(&cr);
if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
bool overlaps;
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
overlaps = classifier_rule_overlaps(&table->cls, &cr);
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
if (overlaps) {
cls_rule_destroy(&cr);
continue;
}
- /* Verify actions. */
- error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
- u16_to_ofp(ofproto->max_ports), rule->table_id);
- if (error) {
- return error;
- }
-
actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
rule->actions->ofpacts,
rule->actions->ofpacts_len);
ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
|| reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
- ofproto_rule_send_removed(rule, reason);
- ofproto_rule_delete__(ofproto, rule);
+ ofproto_rule_delete__(ofproto, rule, reason);
}
/* Reduces '*timeout' to no more than 'max'. A value of zero in either case
ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
- &ofpacts);
+ &ofpacts,
+ u16_to_ofp(ofproto->max_ports),
+ ofproto->n_tables);
+ if (!error) {
+ error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
+ }
if (!error) {
error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
}
* actions, so that when the operation commits we report the change. */
switch (op->type) {
case OFOPERATION_ADD:
- NOT_REACHED();
+ OVS_NOT_REACHED();
case OFOPERATION_MODIFY:
case OFOPERATION_REPLACE:
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
}
fu.ofpacts = actions ? actions->ofpacts : NULL;
struct cls_cursor cursor;
struct rule *rule;
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, &target);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
ovs_assert(!rule->pending); /* XXX */
ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
}
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
}
HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
return 0;
error:
- ovs_mutex_unlock(&ofproto_mutex);
-
for (i = 0; i < n_monitors; i++) {
ofmonitor_destroy(monitors[i]);
}
free(monitors);
+ ovs_mutex_unlock(&ofproto_mutex);
+
return error;
}
}
static bool
-ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
+ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
OVS_REQ_RDLOCK(ofproto->groups_rwlock)
{
struct ofgroup *grp;
return false;
}
+static bool
+ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
+ OVS_EXCLUDED(ofproto->groups_rwlock)
+{
+ bool exists;
+
+ ovs_rwlock_rdlock(&ofproto->groups_rwlock);
+ exists = ofproto_group_exists__(ofproto, group_id);
+ ovs_rwlock_unlock(&ofproto->groups_rwlock);
+
+ return exists;
+}
+
+static uint32_t
+group_get_ref_count(struct ofgroup *group)
+ OVS_EXCLUDED(ofproto_mutex)
+{
+ struct ofproto *ofproto = group->ofproto;
+ struct rule_criteria criteria;
+ struct rule_collection rules;
+ struct match match;
+ enum ofperr error;
+ uint32_t count;
+
+ match_init_catchall(&match);
+ rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
+ OFPP_ANY, group->group_id);
+ ovs_mutex_lock(&ofproto_mutex);
+ error = collect_rules_loose(ofproto, &criteria, &rules);
+ ovs_mutex_unlock(&ofproto_mutex);
+ rule_criteria_destroy(&criteria);
+
+ count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
+
+ rule_collection_destroy(&rules);
+ return count;
+}
+
static void
append_group_stats(struct ofgroup *group, struct list *replies)
OVS_REQ_RDLOCK(group->rwlock)
ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
+ /* Provider sets the packet and byte counts, we do the rest. */
+ ogs.ref_count = group_get_ref_count(group);
+ ogs.n_buckets = group->n_buckets;
+
error = (ofproto->ofproto_class->group_get_stats
? ofproto->ofproto_class->group_get_stats(group, &ogs)
: EOPNOTSUPP);
if (error) {
- ogs.ref_count = UINT32_MAX;
ogs.packet_count = UINT64_MAX;
ogs.byte_count = UINT64_MAX;
- ogs.n_buckets = group->n_buckets;
memset(ogs.bucket_stats, 0xff,
ogs.n_buckets * sizeof *ogs.bucket_stats);
}
return 0;
}
+static enum ofperr
+handle_queue_get_config_request(struct ofconn *ofconn,
+ const struct ofp_header *oh)
+{
+ struct ofproto *p = ofconn_get_ofproto(ofconn);
+ struct netdev_queue_dump queue_dump;
+ struct ofport *ofport;
+ unsigned int queue_id;
+ struct ofpbuf *reply;
+ struct smap details;
+ ofp_port_t request;
+ enum ofperr error;
+
+ error = ofputil_decode_queue_get_config_request(oh, &request);
+ if (error) {
+ return error;
+ }
+
+ ofport = ofproto_get_port(p, request);
+ if (!ofport) {
+ return OFPERR_OFPQOFC_BAD_PORT;
+ }
+
+ reply = ofputil_encode_queue_get_config_reply(oh);
+
+ smap_init(&details);
+ NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
+ struct ofputil_queue_config queue;
+
+ /* None of the existing queues have compatible properties, so we
+ * hard-code omitting min_rate and max_rate. */
+ queue.queue_id = queue_id;
+ queue.min_rate = UINT16_MAX;
+ queue.max_rate = UINT16_MAX;
+ ofputil_append_queue_get_config_reply(reply, &queue);
+ }
+ smap_destroy(&details);
+
+ ofconn_send_reply(ofconn, reply);
+
+ return 0;
+}
+
/* Implements OFPGC11_ADD
* in which no matching flow already exists in the flow table.
*
goto unlock_out;
}
- if (ofproto_group_exists(ofproto, gm->group_id)) {
+ if (ofproto_group_exists__(ofproto, gm->group_id)) {
error = OFPERR_OFPGMFC_GROUP_EXISTS;
goto unlock_out;
}
delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
OVS_RELEASES(ofproto->groups_rwlock)
{
+ struct match match;
+ struct ofputil_flow_mod fm;
+
+ /* Delete all flow entries containing this group in a group action */
+ match_init_catchall(&match);
+ flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
+ fm.out_group = ofgroup->group_id;
+ handle_flow_mod__(ofproto, NULL, &fm, NULL);
+
/* Must wait until existing readers are done,
* while holding the container's write lock at the same time. */
ovs_rwlock_wrlock(&ofgroup->rwlock);
}
}
+static enum ofperr
+table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
+{
+ /* XXX Reject all configurations because none are currently supported */
+ return OFPERR_OFPTMFC_BAD_CONFIG;
+
+ if (tm->table_id == OFPTT_ALL) {
+ int i;
+ for (i = 0; i < ofproto->n_tables; i++) {
+ atomic_store(&ofproto->tables[i].config,
+ (unsigned int)tm->config);
+ }
+ } else if (!check_table_id(ofproto, tm->table_id)) {
+ return OFPERR_OFPTMFC_BAD_TABLE;
+ } else {
+ atomic_store(&ofproto->tables[tm->table_id].config,
+ (unsigned int)tm->config);
+ }
+
+ return 0;
+}
+
static enum ofperr
handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
{
+ struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
struct ofputil_table_mod tm;
enum ofperr error;
return error;
}
- /* XXX Actual table mod support is not implemented yet. */
- return 0;
+ return table_mod(ofproto, &tm);
}
static enum ofperr
if (error) {
return error;
}
+ if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
+ && ofpmp_more(oh)) {
+ /* We have no buffer implementation for multipart requests.
+ * Report overflow for requests which consists of multiple
+ * messages. */
+ return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
+ }
switch (type) {
/* OpenFlow requests. */
case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
return handle_group_features_stats_request(ofconn, oh);
- /* FIXME: Change the following once they are implemented: */
case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
- case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
- return OFPERR_OFPBRC_BAD_TYPE;
+ return handle_queue_get_config_request(ofconn, oh);
case OFPTYPE_HELLO:
case OFPTYPE_ERROR:
case OFPTYPE_METER_STATS_REPLY:
case OFPTYPE_METER_CONFIG_STATS_REPLY:
case OFPTYPE_METER_FEATURES_STATS_REPLY:
+ case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
+ case OFPTYPE_ROLE_STATUS:
default:
- return OFPERR_OFPBRC_BAD_TYPE;
+ if (ofpmsg_is_stat_request(oh)) {
+ return OFPERR_OFPBRC_BAD_STAT;
+ } else {
+ return OFPERR_OFPBRC_BAD_TYPE;
+ }
}
}
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
ofmonitor_report(ofproto->connmgr, rule, event_type,
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
ofoperation_destroy(op);
oftable_init(struct oftable *table)
{
memset(table, 0, sizeof *table);
- classifier_init(&table->cls);
+ classifier_init(&table->cls, flow_segment_u32s);
table->max_flows = UINT_MAX;
+ atomic_init(&table->config, (unsigned int)OFPTC11_TABLE_MISS_CONTROLLER);
}
/* Destroys 'table', including its classifier and eviction groups.
static void
oftable_destroy(struct oftable *table)
{
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
ovs_assert(classifier_is_empty(&table->cls));
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
oftable_disable_eviction(table);
classifier_destroy(&table->cls);
free(table->name);
hmap_init(&table->eviction_groups_by_id);
heap_init(&table->eviction_groups_by_size);
- ovs_rwlock_rdlock(&table->cls.rwlock);
+ fat_rwlock_rdlock(&table->cls.rwlock);
cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
eviction_group_add_rule(rule);
}
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
}
/* Removes 'rule' from the oftable that contains it. */
{
struct classifier *cls = &ofproto->tables[rule->table_id].cls;
- ovs_rwlock_wrlock(&cls->rwlock);
+ fat_rwlock_wrlock(&cls->rwlock);
classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
- ovs_rwlock_unlock(&cls->rwlock);
+ fat_rwlock_unlock(&cls->rwlock);
cookies_remove(ofproto, rule);
struct meter *meter = ofproto->meters[meter_id];
list_insert(&meter->rules, &rule->meter_list_node);
}
- ovs_rwlock_wrlock(&table->cls.rwlock);
+ fat_rwlock_wrlock(&table->cls.rwlock);
classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
- ovs_rwlock_unlock(&table->cls.rwlock);
+ fat_rwlock_unlock(&table->cls.rwlock);
eviction_group_add_rule(rule);
}
\f
ofproto->vlans_changed = false;
OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
- const struct cls_table *table;
+ const struct cls_subtable *table;
- ovs_rwlock_rdlock(&oftable->cls.rwlock);
- HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
+ fat_rwlock_rdlock(&oftable->cls.rwlock);
+ HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) {
if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
const struct cls_rule *rule;
}
}
}
- ovs_rwlock_unlock(&oftable->cls.rwlock);
+ fat_rwlock_unlock(&oftable->cls.rwlock);
}
}
projects / cascardo / ovs.git / blobdiff