/*
- * Copyright (c) 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include "lacp.h"
#include "learn.h"
#include "mac-learning.h"
+#include "meta-flow.h"
#include "multipath.h"
#include "netdev.h"
#include "netlink.h"
struct rule_dpif {
struct rule up;
- long long int used; /* Time last used; time created if not used. */
-
/* These statistics:
*
* - Do include packets and bytes from facets that have been deleted or
*
* - Do not include packet or bytes that can be obtained from any facet's
* packet_count or byte_count member or that can be obtained from the
- * datapath by, e.g., dpif_flow_get() for any facet.
+ * datapath by, e.g., dpif_flow_get() for any subfacet.
*/
uint64_t packet_count; /* Number of packets received. */
uint64_t byte_count; /* Number of bytes received. */
static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
const struct flow *, uint8_t table);
+static void flow_push_stats(struct rule_dpif *, const struct flow *,
+ uint64_t packets, uint64_t bytes,
+ long long int used);
+
+static uint32_t rule_calculate_tag(const struct flow *,
+ const struct flow_wildcards *,
+ uint32_t basis);
+static void rule_invalidate(const struct rule_dpif *);
+
#define MAX_MIRRORS 32
typedef uint32_t mirror_mask_t;
#define MIRROR_MASK_C(X) UINT32_C(X)
struct ofbundle *out; /* Output port or NULL. */
int out_vlan; /* Output VLAN or -1. */
mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
+
+ /* Counters. */
+ int64_t packet_count; /* Number of packets sent. */
+ int64_t byte_count; /* Number of bytes sent. */
};
static void mirror_destroy(struct ofmirror *);
+static void update_mirror_stats(struct ofproto_dpif *ofproto,
+ mirror_mask_t mirrors,
+ uint64_t packets, uint64_t bytes);
-/* A group of one or more OpenFlow ports. */
-#define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
struct ofbundle {
struct ofproto_dpif *ofproto; /* Owning ofproto. */
struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
static void bundle_del_port(struct ofport_dpif *);
static void bundle_run(struct ofbundle *);
static void bundle_wait(struct ofbundle *);
+static struct ofbundle *lookup_input_bundle(struct ofproto_dpif *,
+ uint16_t in_port, bool warn);
+
+/* A controller may use OFPP_NONE as the ingress port to indicate that
+ * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
+ * when an input bundle is needed for validation (e.g., mirroring or
+ * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
+ * any 'port' structs, so care must be taken when dealing with it. */
+static struct ofbundle ofpp_none_bundle = {
+ .name = "OFPP_NONE",
+ .vlan_mode = PORT_VLAN_TRUNK
+};
static void stp_run(struct ofproto_dpif *ofproto);
static void stp_wait(struct ofproto_dpif *ofproto);
+static int set_stp_port(struct ofport *,
+ const struct ofproto_port_stp_settings *);
static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
* we are just revalidating. */
bool may_learn;
+ /* The rule that we are currently translating, or NULL. */
+ struct rule_dpif *rule;
+
+ /* Union of the set of TCP flags seen so far in this flow. (Used only by
+ * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
+ * timeouts.) */
+ uint8_t tcp_flags;
+
/* If nonnull, called just before executing a resubmit action.
*
* This is normally null so the client has to set it manually after
* be reassessed for every packet. */
bool has_learn; /* Actions include NXAST_LEARN? */
bool has_normal; /* Actions output to OFPP_NORMAL? */
+ bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
uint16_t nf_output_iface; /* Output interface index for NetFlow. */
+ mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
/* xlate_actions() initializes and uses these members, but the client has no
* reason to look at them. */
int recurse; /* Recursion level, via xlate_table_action. */
struct flow base_flow; /* Flow at the last commit. */
- uint32_t original_priority; /* Priority when packet arrived. */
+ uint32_t orig_skb_priority; /* Priority when packet arrived. */
uint8_t table_id; /* OpenFlow table ID where flow was found. */
uint32_t sflow_n_outputs; /* Number of output ports. */
uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
static void action_xlate_ctx_init(struct action_xlate_ctx *,
struct ofproto_dpif *, const struct flow *,
- const struct ofpbuf *);
+ ovs_be16 initial_tci, struct rule_dpif *,
+ uint8_t tcp_flags, const struct ofpbuf *);
static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
const union ofp_action *in, size_t n_in);
-/* An exact-match instantiation of an OpenFlow flow. */
+/* An exact-match instantiation of an OpenFlow flow.
+ *
+ * A facet associates a "struct flow", which represents the Open vSwitch
+ * userspace idea of an exact-match flow, with one or more subfacets. Each
+ * subfacet tracks the datapath's idea of the exact-match flow equivalent to
+ * the facet. When the kernel module (or other dpif implementation) and Open
+ * vSwitch userspace agree on the definition of a flow key, there is exactly
+ * one subfacet per facet. If the dpif implementation supports more-specific
+ * flow matching than userspace, however, a facet can have more than one
+ * subfacet, each of which corresponds to some distinction in flow that
+ * userspace simply doesn't understand.
+ *
+ * Flow expiration works in terms of subfacets, so a facet must have at least
+ * one subfacet or it will never expire, leaking memory. */
struct facet {
+ /* Owners. */
+ struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
+ struct list list_node; /* In owning rule's 'facets' list. */
+ struct rule_dpif *rule; /* Owning rule. */
+
+ /* Owned data. */
+ struct list subfacets;
long long int used; /* Time last used; time created if not used. */
+ /* Key. */
+ struct flow flow;
+
/* These statistics:
*
* - Do include packets and bytes sent "by hand", e.g. with
* dpif_execute().
*
* - Do include packets and bytes that were obtained from the datapath
- * when its statistics were reset (e.g. dpif_flow_put() with
+ * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
* DPIF_FP_ZERO_STATS).
+ *
+ * - Do not include packets or bytes that can be obtained from the
+ * datapath for any existing subfacet.
*/
uint64_t packet_count; /* Number of packets received. */
uint64_t byte_count; /* Number of bytes received. */
- uint64_t dp_packet_count; /* Last known packet count in the datapath. */
- uint64_t dp_byte_count; /* Last known byte count in the datapath. */
-
- uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
- uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
- long long int rs_used; /* Used time pushed to resubmit children. */
+ /* Resubmit statistics. */
+ uint64_t prev_packet_count; /* Number of packets from last stats push. */
+ uint64_t prev_byte_count; /* Number of bytes from last stats push. */
+ long long int prev_used; /* Used time from last stats push. */
+ /* Accounting. */
uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
+ struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
+ uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
- struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
- struct list list_node; /* In owning rule's 'facets' list. */
- struct rule_dpif *rule; /* Owning rule. */
- struct flow flow; /* Exact-match flow. */
- bool installed; /* Installed in datapath? */
- bool may_install; /* True ordinarily; false if actions must
- * be reassessed for every packet. */
+ /* Properties of datapath actions.
+ *
+ * Every subfacet has its own actions because actions can differ slightly
+ * between splintered and non-splintered subfacets due to the VLAN tag
+ * being initially different (present vs. absent). All of them have these
+ * properties in common so we just store one copy of them here. */
+ bool may_install; /* Reassess actions for every packet? */
bool has_learn; /* Actions include NXAST_LEARN? */
bool has_normal; /* Actions output to OFPP_NORMAL? */
- size_t actions_len; /* Number of bytes in actions[]. */
- struct nlattr *actions; /* Datapath actions. */
- tag_type tags; /* Tags. */
- struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
+ bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
+ tag_type tags; /* Tags that would require revalidation. */
+ mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
};
static struct facet *facet_create(struct rule_dpif *, const struct flow *);
-static void facet_remove(struct ofproto_dpif *, struct facet *);
+static void facet_remove(struct facet *);
static void facet_free(struct facet *);
static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
static struct facet *facet_lookup_valid(struct ofproto_dpif *,
const struct flow *);
-static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
-
-static bool execute_controller_action(struct ofproto_dpif *,
- const struct flow *,
- const struct nlattr *odp_actions,
- size_t actions_len,
- struct ofpbuf *packet);
-static void facet_execute(struct ofproto_dpif *, struct facet *,
- struct ofpbuf *packet);
-
-static int facet_put__(struct ofproto_dpif *, struct facet *,
- const struct nlattr *actions, size_t actions_len,
- struct dpif_flow_stats *);
-static void facet_install(struct ofproto_dpif *, struct facet *,
- bool zero_stats);
-static void facet_uninstall(struct ofproto_dpif *, struct facet *);
-static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
-
-static void facet_make_actions(struct ofproto_dpif *, struct facet *,
- const struct ofpbuf *packet);
-static void facet_update_time(struct ofproto_dpif *, struct facet *,
- long long int used);
-static void facet_update_stats(struct ofproto_dpif *, struct facet *,
- const struct dpif_flow_stats *);
+static bool facet_revalidate(struct facet *);
+static bool facet_check_consistency(struct facet *);
+
+static void facet_flush_stats(struct facet *);
+
+static void facet_update_time(struct facet *, long long int used);
static void facet_reset_counters(struct facet *);
-static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
static void facet_push_stats(struct facet *);
-static void facet_account(struct ofproto_dpif *, struct facet *);
+static void facet_account(struct facet *);
static bool facet_is_controller_flow(struct facet *);
-static void flow_push_stats(const struct rule_dpif *,
- struct flow *, uint64_t packets, uint64_t bytes,
- long long int used);
+/* A dpif flow and actions associated with a facet.
+ *
+ * See also the large comment on struct facet. */
+struct subfacet {
+ /* Owners. */
+ struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
+ struct list list_node; /* In struct facet's 'facets' list. */
+ struct facet *facet; /* Owning facet. */
+
+ /* Key.
+ *
+ * To save memory in the common case, 'key' is NULL if 'key_fitness' is
+ * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
+ * regenerate the ODP flow key from ->facet->flow. */
+ enum odp_key_fitness key_fitness;
+ struct nlattr *key;
+ int key_len;
-static uint32_t rule_calculate_tag(const struct flow *,
- const struct flow_wildcards *,
- uint32_t basis);
-static void rule_invalidate(const struct rule_dpif *);
+ long long int used; /* Time last used; time created if not used. */
+
+ uint64_t dp_packet_count; /* Last known packet count in the datapath. */
+ uint64_t dp_byte_count; /* Last known byte count in the datapath. */
+
+ /* Datapath actions.
+ *
+ * These should be essentially identical for every subfacet in a facet, but
+ * may differ in trivial ways due to VLAN splinters. */
+ size_t actions_len; /* Number of bytes in actions[]. */
+ struct nlattr *actions; /* Datapath actions. */
+
+ bool installed; /* Installed in datapath? */
+
+ /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
+ * splinters can cause it to differ. This value should be removed when
+ * the VLAN splinters feature is no longer needed. */
+ ovs_be16 initial_tci; /* Initial VLAN TCI value. */
+};
+
+static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
+ const struct nlattr *key,
+ size_t key_len, ovs_be16 initial_tci);
+static struct subfacet *subfacet_find(struct ofproto_dpif *,
+ const struct nlattr *key, size_t key_len);
+static void subfacet_destroy(struct subfacet *);
+static void subfacet_destroy__(struct subfacet *);
+static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
+ struct ofpbuf *key);
+static void subfacet_reset_dp_stats(struct subfacet *,
+ struct dpif_flow_stats *);
+static void subfacet_update_time(struct subfacet *, long long int used);
+static void subfacet_update_stats(struct subfacet *,
+ const struct dpif_flow_stats *);
+static void subfacet_make_actions(struct subfacet *,
+ const struct ofpbuf *packet);
+static int subfacet_install(struct subfacet *,
+ const struct nlattr *actions, size_t actions_len,
+ struct dpif_flow_stats *);
+static void subfacet_uninstall(struct subfacet *);
struct ofport_dpif {
struct ofport up;
uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
bool may_enable; /* May be enabled in bonds. */
+ /* Spanning tree. */
struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
long long int stp_state_entered;
+
+ struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
+
+ /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device
+ * drivers in old versions of Linux that do not properly support VLANs when
+ * VLAN devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
+ uint16_t realdev_ofp_port;
+ int vlandev_vid;
+};
+
+/* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
+ * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
+ * traffic egressing the 'ofport' with that priority should be marked with. */
+struct priority_to_dscp {
+ struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
+ uint32_t priority; /* Priority of this queue (see struct flow). */
+
+ uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
+};
+
+/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device drivers
+ * in old versions of Linux that do not properly support VLANs when VLAN
+ * devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
+struct vlan_splinter {
+ struct hmap_node realdev_vid_node;
+ struct hmap_node vlandev_node;
+ uint16_t realdev_ofp_port;
+ uint16_t vlandev_ofp_port;
+ int vid;
};
+static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
+ uint32_t realdev, ovs_be16 vlan_tci);
+static uint16_t vsp_vlandev_to_realdev(const struct ofproto_dpif *,
+ uint16_t vlandev, int *vid);
+static void vsp_remove(struct ofport_dpif *);
+static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
+
static struct ofport_dpif *
ofport_dpif_cast(const struct ofport *ofport)
{
static void port_run(struct ofport_dpif *);
static void port_wait(struct ofport_dpif *);
static int set_cfm(struct ofport *, const struct cfm_settings *);
+static void ofport_clear_priorities(struct ofport_dpif *);
struct dpif_completion {
struct list list_node;
};
struct ofproto_dpif {
+ struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
struct ofproto up;
struct dpif *dpif;
int max_ports;
/* Facets. */
struct hmap facets;
+ struct hmap subfacets;
/* Revalidation. */
struct table_dpif tables[N_TABLES];
struct list completions;
bool has_bundle_action; /* True when the first bundle action appears. */
+ struct netdev_stats stats; /* To account packets generated and consumed in
+ * userspace. */
/* Spanning tree. */
struct stp *stp;
long long int stp_last_tick;
+
+ /* VLAN splinters. */
+ struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
+ struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
};
/* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
* for debugging the asynchronous flow_mod implementation.) */
static bool clogged;
+/* All existing ofproto_dpif instances, indexed by ->up.name. */
+static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
+
static void ofproto_dpif_unixctl_init(void);
static struct ofproto_dpif *
struct ofbundle *);
/* Upcalls. */
#define FLOW_MISS_MAX_BATCH 50
-
-static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
-static void handle_miss_upcalls(struct ofproto_dpif *,
- struct dpif_upcall *, size_t n);
+static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
/* Flow expiration. */
static int expire(struct ofproto_dpif *);
+/* NetFlow. */
+static void send_netflow_active_timeouts(struct ofproto_dpif *);
+
/* Utilities. */
-static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
- const struct ofpbuf *packet);
+static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
static size_t
compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
const struct flow *, uint32_t odp_port);
+static void add_mirror_actions(struct action_xlate_ctx *ctx,
+ const struct flow *flow);
/* Global variables. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
\f
}
static int
-construct(struct ofproto *ofproto_, int *n_tablesp)
+construct(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
const char *name = ofproto->up.name;
dpif_flow_flush(ofproto->dpif);
dpif_recv_purge(ofproto->dpif);
- error = dpif_recv_set_mask(ofproto->dpif,
- ((1u << DPIF_UC_MISS) |
- (1u << DPIF_UC_ACTION)));
+ error = dpif_recv_set(ofproto->dpif, true);
if (error) {
VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
dpif_close(ofproto->dpif);
ofproto->sflow = NULL;
ofproto->stp = NULL;
hmap_init(&ofproto->bundles);
- ofproto->ml = mac_learning_create();
+ ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
for (i = 0; i < MAX_MIRRORS; i++) {
ofproto->mirrors[i] = NULL;
}
timer_set_duration(&ofproto->next_expiration, 1000);
hmap_init(&ofproto->facets);
+ hmap_init(&ofproto->subfacets);
for (i = 0; i < N_TABLES; i++) {
struct table_dpif *table = &ofproto->tables[i];
ofproto->has_bundle_action = false;
- *n_tablesp = N_TABLES;
+ hmap_init(&ofproto->vlandev_map);
+ hmap_init(&ofproto->realdev_vid_map);
+
+ hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
+ hash_string(ofproto->up.name, 0));
+ memset(&ofproto->stats, 0, sizeof ofproto->stats);
+
+ ofproto_init_tables(ofproto_, N_TABLES);
+
return 0;
}
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
struct rule_dpif *rule, *next_rule;
- struct classifier *table;
+ struct oftable *table;
int i;
+ hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
complete_operations(ofproto);
OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
struct cls_cursor cursor;
- cls_cursor_init(&cursor, table, NULL);
+ cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
ofproto_rule_destroy(&rule->up);
}
mac_learning_destroy(ofproto->ml);
hmap_destroy(&ofproto->facets);
+ hmap_destroy(&ofproto->subfacets);
+
+ hmap_destroy(&ofproto->vlandev_map);
+ hmap_destroy(&ofproto->realdev_vid_map);
dpif_close(ofproto->dpif);
}
+static int
+run_fast(struct ofproto *ofproto_)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ unsigned int work;
+
+ /* Handle one or more batches of upcalls, until there's nothing left to do
+ * or until we do a fixed total amount of work.
+ *
+ * We do work in batches because it can be much cheaper to set up a number
+ * of flows and fire off their patches all at once. We do multiple batches
+ * because in some cases handling a packet can cause another packet to be
+ * queued almost immediately as part of the return flow. Both
+ * optimizations can make major improvements on some benchmarks and
+ * presumably for real traffic as well. */
+ work = 0;
+ while (work < FLOW_MISS_MAX_BATCH) {
+ int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
+ if (retval <= 0) {
+ return -retval;
+ }
+ work += retval;
+ }
+ return 0;
+}
+
static int
run(struct ofproto *ofproto_)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
struct ofport_dpif *ofport;
struct ofbundle *bundle;
- size_t n_misses;
- int i;
+ int error;
if (!clogged) {
complete_operations(ofproto);
}
dpif_run(ofproto->dpif);
- n_misses = 0;
- for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) {
- struct dpif_upcall *upcall = &misses[n_misses];
- int error;
-
- error = dpif_recv(ofproto->dpif, upcall);
- if (error) {
- if (error == ENODEV && n_misses == 0) {
- return error;
- }
- break;
- }
-
- if (upcall->type == DPIF_UC_MISS) {
- /* Handle it later. */
- n_misses++;
- } else {
- handle_upcall(ofproto, upcall);
- }
+ error = run_fast(ofproto_);
+ if (error) {
+ return error;
}
- handle_miss_upcalls(ofproto, misses, n_misses);
-
if (timer_expired(&ofproto->next_expiration)) {
int delay = expire(ofproto);
timer_set_duration(&ofproto->next_expiration, delay);
}
if (ofproto->netflow) {
- netflow_run(ofproto->netflow);
+ if (netflow_run(ofproto->netflow)) {
+ send_netflow_active_timeouts(ofproto);
+ }
}
if (ofproto->sflow) {
dpif_sflow_run(ofproto->sflow);
HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
if (revalidate_all
|| tag_set_intersects(&revalidate_set, facet->tags)) {
- facet_revalidate(ofproto, facet);
+ facet_revalidate(facet);
+ }
+ }
+ }
+
+ /* Check the consistency of a random facet, to aid debugging. */
+ if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
+ struct facet *facet;
+
+ facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
+ struct facet, hmap_node);
+ if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
+ if (!facet_check_consistency(facet)) {
+ ofproto->need_revalidate = true;
}
}
}
HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
bundle_wait(bundle);
}
+ if (ofproto->netflow) {
+ netflow_wait(ofproto->netflow);
+ }
mac_learning_wait(ofproto->ml);
stp_wait(ofproto);
if (ofproto->need_revalidate) {
* bother trying to uninstall it. There is no point in uninstalling it
* individually since we are about to blow away all the facets with
* dpif_flow_flush(). */
- facet->installed = false;
- facet->dp_packet_count = 0;
- facet->dp_byte_count = 0;
- facet_remove(ofproto, facet);
+ struct subfacet *subfacet;
+
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ subfacet->installed = false;
+ subfacet->dp_packet_count = 0;
+ subfacet->dp_byte_count = 0;
+ }
+ facet_remove(facet);
}
dpif_flow_flush(ofproto->dpif);
}
htonll(s.n_hit + ofproto->n_matches));
}
-static int
-set_netflow(struct ofproto *ofproto_,
- const struct netflow_options *netflow_options)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
-
- if (netflow_options) {
- if (!ofproto->netflow) {
- ofproto->netflow = netflow_create();
- }
- return netflow_set_options(ofproto->netflow, netflow_options);
- } else {
- netflow_destroy(ofproto->netflow);
- ofproto->netflow = NULL;
- return 0;
- }
-}
-
static struct ofport *
port_alloc(void)
{
port->may_enable = true;
port->stp_port = NULL;
port->stp_state = STP_DISABLED;
+ hmap_init(&port->priorities);
+ port->realdev_ofp_port = 0;
+ port->vlandev_vid = 0;
if (ofproto->sflow) {
- dpif_sflow_add_port(ofproto->sflow, port->odp_port,
- netdev_get_name(port->up.netdev));
+ dpif_sflow_add_port(ofproto->sflow, port_);
}
return 0;
if (ofproto->sflow) {
dpif_sflow_del_port(ofproto->sflow, port->odp_port);
}
+
+ ofport_clear_priorities(port);
+ hmap_destroy(&port->priorities);
}
static void
ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
- dpif_sflow_add_port(ds, ofport->odp_port,
- netdev_get_name(ofport->up.netdev));
+ dpif_sflow_add_port(ds, &ofport->up);
}
ofproto->need_revalidate = true;
}
VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
"with unknown MAC", ofproto->up.name, port_num);
} else {
- send_packet(ofproto_dpif_cast(ofport->up.ofproto),
- ofport->odp_port, pkt);
+ send_packet(ofport, pkt);
}
}
ofpbuf_delete(pkt);
stp_set_max_age(ofproto->stp, s->max_age);
stp_set_forward_delay(ofproto->stp, s->fwd_delay);
} else {
+ struct ofport *ofport;
+
+ HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
+ set_stp_port(ofport, NULL);
+ }
+
stp_destroy(ofproto->stp);
ofproto->stp = NULL;
}
if (stp_learn_in_state(ofport->stp_state)
!= stp_learn_in_state(state)) {
/* xxx Learning action flows should also be flushed. */
- mac_learning_flush(ofproto->ml);
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
}
fwd_change = stp_forward_in_state(ofport->stp_state)
!= stp_forward_in_state(state);
update_stp_port_state(ofport);
}
}
+
+ if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ }
}
}
}
}
\f
+static struct priority_to_dscp *
+get_priority(const struct ofport_dpif *ofport, uint32_t priority)
+{
+ struct priority_to_dscp *pdscp;
+ uint32_t hash;
+
+ hash = hash_int(priority, 0);
+ HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
+ if (pdscp->priority == priority) {
+ return pdscp;
+ }
+ }
+ return NULL;
+}
+
+static void
+ofport_clear_priorities(struct ofport_dpif *ofport)
+{
+ struct priority_to_dscp *pdscp, *next;
+
+ HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
+ hmap_remove(&ofport->priorities, &pdscp->hmap_node);
+ free(pdscp);
+ }
+}
+
+static int
+set_queues(struct ofport *ofport_,
+ const struct ofproto_port_queue *qdscp_list,
+ size_t n_qdscp)
+{
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+ struct hmap new = HMAP_INITIALIZER(&new);
+ size_t i;
+
+ for (i = 0; i < n_qdscp; i++) {
+ struct priority_to_dscp *pdscp;
+ uint32_t priority;
+ uint8_t dscp;
+
+ dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
+ if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
+ &priority)) {
+ continue;
+ }
+
+ pdscp = get_priority(ofport, priority);
+ if (pdscp) {
+ hmap_remove(&ofport->priorities, &pdscp->hmap_node);
+ } else {
+ pdscp = xmalloc(sizeof *pdscp);
+ pdscp->priority = priority;
+ pdscp->dscp = dscp;
+ ofproto->need_revalidate = true;
+ }
+
+ if (pdscp->dscp != dscp) {
+ pdscp->dscp = dscp;
+ ofproto->need_revalidate = true;
+ }
+
+ hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
+ }
+
+ if (!hmap_is_empty(&ofport->priorities)) {
+ ofport_clear_priorities(ofport);
+ ofproto->need_revalidate = true;
+ }
+
+ hmap_swap(&new, &ofport->priorities);
+ hmap_destroy(&new);
+
+ return 0;
+}
+\f
/* Bundles. */
-/* Expires all MAC learning entries associated with 'port' and forces ofproto
- * to revalidate every flow. */
+/* Expires all MAC learning entries associated with 'bundle' and forces its
+ * ofproto to revalidate every flow.
+ *
+ * Normally MAC learning entries are removed only from the ofproto associated
+ * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
+ * are removed from every ofproto. When patch ports and SLB bonds are in use
+ * and a VM migration happens and the gratuitous ARPs are somehow lost, this
+ * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
+ * with the host from which it migrated. */
static void
-bundle_flush_macs(struct ofbundle *bundle)
+bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
{
struct ofproto_dpif *ofproto = bundle->ofproto;
struct mac_learning *ml = ofproto->ml;
ofproto->need_revalidate = true;
LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
if (mac->port.p == bundle) {
+ if (all_ofprotos) {
+ struct ofproto_dpif *o;
+
+ HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ if (o != ofproto) {
+ struct mac_entry *e;
+
+ e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
+ NULL);
+ if (e) {
+ tag_set_add(&o->revalidate_set, e->tag);
+ mac_learning_expire(o->ml, e);
+ }
+ }
+ }
+ }
+
mac_learning_expire(ml, mac);
}
}
bundle_del_port(port);
}
- bundle_flush_macs(bundle);
+ bundle_flush_macs(bundle, true);
hmap_remove(&ofproto->bundles, &bundle->hmap_node);
free(bundle->name);
free(bundle->trunks);
/* If we changed something that would affect MAC learning, un-learn
* everything on this port and force flow revalidation. */
if (need_flush) {
- bundle_flush_macs(bundle);
+ bundle_flush_macs(bundle, false);
}
return 0;
pdu_size);
memcpy(packet_pdu, pdu, pdu_size);
- send_packet(ofproto_dpif_cast(port->up.ofproto), port->odp_port,
- &packet);
+ send_packet(port, &packet);
ofpbuf_uninit(&packet);
} else {
VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
if (e->port.p != bundle) {
struct ofpbuf *learning_packet;
struct ofport_dpif *port;
+ void *port_void;
int ret;
- learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
- e->vlan,
- (void **)&port);
- ret = send_packet(ofproto_dpif_cast(port->up.ofproto),
- port->odp_port, learning_packet);
+ /* The assignment to "port" is unnecessary but makes "grep"ing for
+ * struct ofport_dpif more effective. */
+ learning_packet = bond_compose_learning_packet(bundle->bond,
+ e->mac, e->vlan,
+ &port_void);
+ port = port_void;
+ ret = send_packet(port, learning_packet);
ofpbuf_delete(learning_packet);
if (ret) {
error = ret;
}
bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
- lacp_negotiated(bundle->lacp));
+ lacp_status(bundle->lacp));
if (bond_should_send_learning_packets(bundle->bond)) {
bundle_send_learning_packets(bundle);
}
}
ofproto->need_revalidate = true;
- mac_learning_flush(ofproto->ml);
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
mirror_update_dups(ofproto);
return 0;
ofproto = mirror->ofproto;
ofproto->need_revalidate = true;
- mac_learning_flush(ofproto->ml);
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
mirror_update_dups(ofproto);
}
+static int
+mirror_get_stats(struct ofproto *ofproto_, void *aux,
+ uint64_t *packets, uint64_t *bytes)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ struct ofmirror *mirror = mirror_lookup(ofproto, aux);
+
+ if (!mirror) {
+ *packets = *bytes = UINT64_MAX;
+ return 0;
+ }
+
+ *packets = mirror->packet_count;
+ *bytes = mirror->byte_count;
+
+ return 0;
+}
+
static int
set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
- ofproto->need_revalidate = true;
- mac_learning_flush(ofproto->ml);
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
}
return 0;
}
/* Revalidate cached flows whenever forward_bpdu option changes. */
ofproto->need_revalidate = true;
}
+
+static void
+set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+ mac_learning_set_idle_time(ofproto->ml, idle_time);
+}
\f
/* Ports. */
ofpbuf_init(&packet, 0);
cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
- send_packet(ofproto_dpif_cast(ofport->up.ofproto),
- ofport->odp_port, &packet);
+ send_packet(ofport, &packet);
ofpbuf_uninit(&packet);
}
return error;
}
+static int
+port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
+{
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+ int error;
+
+ error = netdev_get_stats(ofport->up.netdev, stats);
+
+ if (!error && ofport->odp_port == OVSP_LOCAL) {
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
+
+ /* ofproto->stats.tx_packets represents packets that we created
+ * internally and sent to some port (e.g. packets sent with
+ * send_packet()). Account for them as if they had come from
+ * OFPP_LOCAL and got forwarded. */
+
+ if (stats->rx_packets != UINT64_MAX) {
+ stats->rx_packets += ofproto->stats.tx_packets;
+ }
+
+ if (stats->rx_bytes != UINT64_MAX) {
+ stats->rx_bytes += ofproto->stats.tx_bytes;
+ }
+
+ /* ofproto->stats.rx_packets represents packets that were received on
+ * some port and we processed internally and dropped (e.g. STP).
+ * Account fro them as if they had been forwarded to OFPP_LOCAL. */
+
+ if (stats->tx_packets != UINT64_MAX) {
+ stats->tx_packets += ofproto->stats.rx_packets;
+ }
+
+ if (stats->tx_bytes != UINT64_MAX) {
+ stats->tx_bytes += ofproto->stats.rx_bytes;
+ }
+ }
+
+ return error;
+}
+
+/* Account packets for LOCAL port. */
+static void
+ofproto_update_local_port_stats(const struct ofproto *ofproto_,
+ size_t tx_size, size_t rx_size)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+
+ if (rx_size) {
+ ofproto->stats.rx_packets++;
+ ofproto->stats.rx_bytes += rx_size;
+ }
+ if (tx_size) {
+ ofproto->stats.tx_packets++;
+ ofproto->stats.tx_bytes += tx_size;
+ }
+}
+
struct port_dump_state {
struct dpif_port_dump dump;
bool done;
struct flow_miss {
struct hmap_node hmap_node;
struct flow flow;
+ enum odp_key_fitness key_fitness;
const struct nlattr *key;
size_t key_len;
+ ovs_be16 initial_tci;
struct list packets;
};
struct flow_miss_op {
- union dpif_op dpif_op;
- struct facet *facet;
+ struct dpif_op dpif_op;
+ struct subfacet *subfacet;
};
/* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
* OpenFlow controller as necessary according to their individual
- * configurations.
- *
- * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
- * ownership is transferred to this function. */
+ * configurations. */
static void
-send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
- const struct flow *flow, bool clone)
+send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
+ const struct flow *flow)
{
struct ofputil_packet_in pin;
- pin.packet = packet;
- pin.in_port = flow->in_port;
+ pin.packet = packet->data;
+ pin.packet_len = packet->size;
+ pin.total_len = packet->size;
pin.reason = OFPR_NO_MATCH;
+ pin.controller_id = 0;
+
+ pin.table_id = 0;
+ pin.cookie = 0;
+
pin.buffer_id = 0; /* not yet known */
pin.send_len = 0; /* not used for flow table misses */
- connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
- clone ? NULL : packet);
-}
-/* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
- * OpenFlow controller as necessary according to their individual
- * configurations.
- *
- * 'send_len' should be the number of bytes of 'packet' to send to the
- * controller, as specified in the action that caused the packet to be sent.
- *
- * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
- * Otherwise, ownership is transferred to this function. */
-static void
-send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
- uint64_t userdata, const struct flow *flow, bool clone)
-{
- struct ofputil_packet_in pin;
- struct user_action_cookie cookie;
+ flow_get_metadata(flow, &pin.fmd);
- memcpy(&cookie, &userdata, sizeof(cookie));
+ /* Registers aren't meaningful on a miss. */
+ memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
- pin.packet = packet;
- pin.in_port = flow->in_port;
- pin.reason = OFPR_ACTION;
- pin.buffer_id = 0; /* not yet known */
- pin.send_len = cookie.data;
- connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
- clone ? NULL : packet);
+ connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow);
}
static bool
static struct flow_miss *
flow_miss_create(struct hmap *todo, const struct flow *flow,
- const struct nlattr *key, size_t key_len)
+ enum odp_key_fitness key_fitness,
+ const struct nlattr *key, size_t key_len,
+ ovs_be16 initial_tci)
{
uint32_t hash = flow_hash(flow, 0);
struct flow_miss *miss;
miss = xmalloc(sizeof *miss);
hmap_insert(todo, &miss->hmap_node, hash);
miss->flow = *flow;
+ miss->key_fitness = key_fitness;
miss->key = key;
miss->key_len = key_len;
+ miss->initial_tci = initial_tci;
list_init(&miss->packets);
return miss;
}
{
const struct flow *flow = &miss->flow;
struct ofpbuf *packet, *next_packet;
+ struct subfacet *subfacet;
struct facet *facet;
facet = facet_lookup_valid(ofproto, flow);
flow->in_port);
}
- LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
- &miss->packets) {
- list_remove(&packet->list_node);
- send_packet_in_miss(ofproto, packet, flow, false);
+ LIST_FOR_EACH (packet, list_node, &miss->packets) {
+ send_packet_in_miss(ofproto, packet, flow);
}
return;
facet = facet_create(rule, flow);
}
+ subfacet = subfacet_create(facet,
+ miss->key_fitness, miss->key, miss->key_len,
+ miss->initial_tci);
+
LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
- list_remove(&packet->list_node);
+ struct dpif_flow_stats stats;
+ struct flow_miss_op *op;
+ struct dpif_execute *execute;
+
ofproto->n_matches++;
if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
*
* See the top-level comment in fail-open.c for more information.
*/
- send_packet_in_miss(ofproto, packet, flow, true);
+ send_packet_in_miss(ofproto, packet, flow);
+ }
+
+ if (!facet->may_install || !subfacet->actions) {
+ subfacet_make_actions(subfacet, packet);
}
- if (!facet->may_install) {
- facet_make_actions(ofproto, facet, packet);
+ dpif_flow_stats_extract(&facet->flow, packet, &stats);
+ subfacet_update_stats(subfacet, &stats);
+
+ if (!subfacet->actions_len) {
+ /* No actions to execute, so skip talking to the dpif. */
+ continue;
}
- if (!execute_controller_action(ofproto, &facet->flow,
- facet->actions, facet->actions_len,
- packet)) {
- struct flow_miss_op *op = &ops[(*n_ops)++];
- struct dpif_execute *execute = &op->dpif_op.execute;
- op->facet = facet;
- execute->type = DPIF_OP_EXECUTE;
- execute->key = miss->key;
- execute->key_len = miss->key_len;
- execute->actions
- = (facet->may_install
- ? facet->actions
- : xmemdup(facet->actions, facet->actions_len));
- execute->actions_len = facet->actions_len;
- execute->packet = packet;
+ if (flow->vlan_tci != subfacet->initial_tci) {
+ /* This packet was received on a VLAN splinter port. We added
+ * a VLAN to the packet to make the packet resemble the flow,
+ * but the actions were composed assuming that the packet
+ * contained no VLAN. So, we must remove the VLAN header from
+ * the packet before trying to execute the actions. */
+ eth_pop_vlan(packet);
}
- }
- if (facet->may_install) {
+ op = &ops[(*n_ops)++];
+ execute = &op->dpif_op.u.execute;
+ op->subfacet = subfacet;
+ op->dpif_op.type = DPIF_OP_EXECUTE;
+ execute->key = miss->key;
+ execute->key_len = miss->key_len;
+ execute->actions = (facet->may_install
+ ? subfacet->actions
+ : xmemdup(subfacet->actions,
+ subfacet->actions_len));
+ execute->actions_len = subfacet->actions_len;
+ execute->packet = packet;
+ }
+
+ if (facet->may_install && subfacet->key_fitness != ODP_FIT_TOO_LITTLE) {
struct flow_miss_op *op = &ops[(*n_ops)++];
- struct dpif_flow_put *put = &op->dpif_op.flow_put;
+ struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
- op->facet = facet;
- put->type = DPIF_OP_FLOW_PUT;
+ op->subfacet = subfacet;
+ op->dpif_op.type = DPIF_OP_FLOW_PUT;
put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
put->key = miss->key;
put->key_len = miss->key_len;
- put->actions = facet->actions;
- put->actions_len = facet->actions_len;
+ put->actions = subfacet->actions;
+ put->actions_len = subfacet->actions_len;
put->stats = NULL;
}
}
-static void
-handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
- size_t n_upcalls)
-{
- struct dpif_upcall *upcall;
- struct flow_miss *miss, *next_miss;
- struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
- union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
- struct hmap todo;
- size_t n_ops;
+/* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
+ * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
+ * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
+ * what a flow key should contain.
+ *
+ * This function also includes some logic to help make VLAN splinters
+ * transparent to the rest of the upcall processing logic. In particular, if
+ * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
+ * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
+ * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
+ *
+ * Sets '*initial_tci' to the VLAN TCI with which the packet was really
+ * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
+ * (This differs from the value returned in flow->vlan_tci only for packets
+ * received on VLAN splinters.)
+ */
+static enum odp_key_fitness
+ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
+ const struct nlattr *key, size_t key_len,
+ struct flow *flow, ovs_be16 *initial_tci,
+ struct ofpbuf *packet)
+{
+ enum odp_key_fitness fitness;
+ uint16_t realdev;
+ int vid;
+
+ fitness = odp_flow_key_to_flow(key, key_len, flow);
+ if (fitness == ODP_FIT_ERROR) {
+ return fitness;
+ }
+ *initial_tci = flow->vlan_tci;
+
+ realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
+ if (realdev) {
+ /* Cause the flow to be processed as if it came in on the real device
+ * with the VLAN device's VLAN ID. */
+ flow->in_port = realdev;
+ flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
+ if (packet) {
+ /* Make the packet resemble the flow, so that it gets sent to an
+ * OpenFlow controller properly, so that it looks correct for
+ * sFlow, and so that flow_extract() will get the correct vlan_tci
+ * if it is called on 'packet'.
+ *
+ * The allocated space inside 'packet' probably also contains
+ * 'key', that is, both 'packet' and 'key' are probably part of a
+ * struct dpif_upcall (see the large comment on that structure
+ * definition), so pushing data on 'packet' is in general not a
+ * good idea since it could overwrite 'key' or free it as a side
+ * effect. However, it's OK in this special case because we know
+ * that 'packet' is inside a Netlink attribute: pushing 4 bytes
+ * will just overwrite the 4-byte "struct nlattr", which is fine
+ * since we don't need that header anymore. */
+ eth_push_vlan(packet, flow->vlan_tci);
+ }
+
+ /* Let the caller know that we can't reproduce 'key' from 'flow'. */
+ if (fitness == ODP_FIT_PERFECT) {
+ fitness = ODP_FIT_TOO_MUCH;
+ }
+ }
+
+ return fitness;
+}
+
+static void
+handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
+ size_t n_upcalls)
+{
+ struct dpif_upcall *upcall;
+ struct flow_miss *miss, *next_miss;
+ struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
+ struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
+ struct hmap todo;
+ size_t n_ops;
size_t i;
if (!n_upcalls) {
* that we can process them together. */
hmap_init(&todo);
for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
+ enum odp_key_fitness fitness;
struct flow_miss *miss;
+ ovs_be16 initial_tci;
struct flow flow;
- /* Obtain in_port and tun_id, at least, then set 'flow''s header
- * pointers. */
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- flow_extract(upcall->packet, flow.priority, flow.tun_id,
+ /* Obtain metadata and check userspace/kernel agreement on flow match,
+ * then set 'flow''s header pointers. */
+ fitness = ofproto_dpif_extract_flow_key(ofproto,
+ upcall->key, upcall->key_len,
+ &flow, &initial_tci,
+ upcall->packet);
+ if (fitness == ODP_FIT_ERROR) {
+ ofpbuf_delete(upcall->packet);
+ continue;
+ }
+ flow_extract(upcall->packet, flow.skb_priority, flow.tun_id,
flow.in_port, &flow);
/* Handle 802.1ag, LACP, and STP specially. */
if (process_special(ofproto, &flow, upcall->packet)) {
+ ofproto_update_local_port_stats(&ofproto->up,
+ 0, upcall->packet->size);
ofpbuf_delete(upcall->packet);
ofproto->n_matches++;
continue;
}
/* Add other packets to a to-do list. */
- miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
+ miss = flow_miss_create(&todo, &flow, fitness,
+ upcall->key, upcall->key_len, initial_tci);
list_push_back(&miss->packets, &upcall->packet->list_node);
}
/* Process each element in the to-do list, constructing the set of
* operations to batch. */
n_ops = 0;
- HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
+ HMAP_FOR_EACH (miss, hmap_node, &todo) {
handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
- ofpbuf_list_delete(&miss->packets);
- hmap_remove(&todo, &miss->hmap_node);
- free(miss);
}
assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
- hmap_destroy(&todo);
/* Execute batch. */
for (i = 0; i < n_ops; i++) {
for (i = 0; i < n_ops; i++) {
struct flow_miss_op *op = &flow_miss_ops[i];
struct dpif_execute *execute;
- struct dpif_flow_put *put;
switch (op->dpif_op.type) {
case DPIF_OP_EXECUTE:
- execute = &op->dpif_op.execute;
- if (op->facet->actions != execute->actions) {
+ execute = &op->dpif_op.u.execute;
+ if (op->subfacet->actions != execute->actions) {
free((struct nlattr *) execute->actions);
}
- ofpbuf_delete((struct ofpbuf *) execute->packet);
break;
case DPIF_OP_FLOW_PUT:
- put = &op->dpif_op.flow_put;
- if (!put->error) {
- op->facet->installed = true;
+ if (!op->dpif_op.error) {
+ op->subfacet->installed = true;
}
break;
}
}
+ HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
+ ofpbuf_list_delete(&miss->packets);
+ hmap_remove(&todo, &miss->hmap_node);
+ free(miss);
+ }
+ hmap_destroy(&todo);
}
static void
handle_userspace_upcall(struct ofproto_dpif *ofproto,
struct dpif_upcall *upcall)
{
- struct flow flow;
struct user_action_cookie cookie;
+ enum odp_key_fitness fitness;
+ ovs_be16 initial_tci;
+ struct flow flow;
memcpy(&cookie, &upcall->userdata, sizeof(cookie));
+ fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
+ upcall->key_len, &flow,
+ &initial_tci, upcall->packet);
+ if (fitness == ODP_FIT_ERROR) {
+ ofpbuf_delete(upcall->packet);
+ return;
+ }
+
if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
if (ofproto->sflow) {
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
+ dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
+ &cookie);
}
- ofpbuf_delete(upcall->packet);
-
- } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
- COVERAGE_INC(ofproto_dpif_ctlr_action);
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
- send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
- &flow, false);
} else {
VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
}
+ ofpbuf_delete(upcall->packet);
}
-static void
-handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
+static int
+handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
{
- switch (upcall->type) {
- case DPIF_UC_ACTION:
- handle_userspace_upcall(ofproto, upcall);
- break;
+ struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
+ int n_misses;
+ int i;
- case DPIF_UC_MISS:
- /* The caller handles these. */
- NOT_REACHED();
+ assert (max_batch <= FLOW_MISS_MAX_BATCH);
- case DPIF_N_UC_TYPES:
- default:
- VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
- break;
+ n_misses = 0;
+ for (i = 0; i < max_batch; i++) {
+ struct dpif_upcall *upcall = &misses[n_misses];
+ int error;
+
+ error = dpif_recv(ofproto->dpif, upcall);
+ if (error) {
+ break;
+ }
+
+ switch (upcall->type) {
+ case DPIF_UC_ACTION:
+ handle_userspace_upcall(ofproto, upcall);
+ break;
+
+ case DPIF_UC_MISS:
+ /* Handle it later. */
+ n_misses++;
+ break;
+
+ case DPIF_N_UC_TYPES:
+ default:
+ VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
+ upcall->type);
+ break;
+ }
}
+
+ handle_miss_upcalls(ofproto, misses, n_misses);
+
+ return i;
}
\f
/* Flow expiration. */
-static int facet_max_idle(const struct ofproto_dpif *);
+static int subfacet_max_idle(const struct ofproto_dpif *);
static void update_stats(struct ofproto_dpif *);
static void rule_expire(struct rule_dpif *);
-static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
+static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
/* This function is called periodically by run(). Its job is to collect
* updates for the flows that have been installed into the datapath, most
expire(struct ofproto_dpif *ofproto)
{
struct rule_dpif *rule, *next_rule;
- struct classifier *table;
+ struct oftable *table;
int dp_max_idle;
/* Update stats for each flow in the datapath. */
update_stats(ofproto);
- /* Expire facets that have been idle too long. */
- dp_max_idle = facet_max_idle(ofproto);
- expire_facets(ofproto, dp_max_idle);
+ /* Expire subfacets that have been idle too long. */
+ dp_max_idle = subfacet_max_idle(ofproto);
+ expire_subfacets(ofproto, dp_max_idle);
/* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
struct cls_cursor cursor;
- cls_cursor_init(&cursor, table, NULL);
+ cls_cursor_init(&cursor, &table->cls, NULL);
CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
rule_expire(rule);
}
dpif_flow_dump_start(&dump, p->dpif);
while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
- struct facet *facet;
- struct flow flow;
+ struct subfacet *subfacet;
- if (odp_flow_key_to_flow(key, key_len, &flow)) {
- struct ds s;
+ subfacet = subfacet_find(p, key, key_len);
+ if (subfacet && subfacet->installed) {
+ struct facet *facet = subfacet->facet;
- ds_init(&s);
- odp_flow_key_format(key, key_len, &s);
- VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
- ds_cstr(&s));
- ds_destroy(&s);
-
- continue;
- }
- facet = facet_find(p, &flow);
-
- if (facet && facet->installed) {
-
- if (stats->n_packets >= facet->dp_packet_count) {
- uint64_t extra = stats->n_packets - facet->dp_packet_count;
+ if (stats->n_packets >= subfacet->dp_packet_count) {
+ uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
facet->packet_count += extra;
} else {
VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
}
- if (stats->n_bytes >= facet->dp_byte_count) {
- facet->byte_count += stats->n_bytes - facet->dp_byte_count;
+ if (stats->n_bytes >= subfacet->dp_byte_count) {
+ facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
} else {
VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
}
- facet->dp_packet_count = stats->n_packets;
- facet->dp_byte_count = stats->n_bytes;
+ subfacet->dp_packet_count = stats->n_packets;
+ subfacet->dp_byte_count = stats->n_bytes;
- facet_update_time(p, facet, stats->used);
- facet_account(p, facet);
+ facet->tcp_flags |= stats->tcp_flags;
+
+ subfacet_update_time(subfacet, stats->used);
+ facet_account(facet);
facet_push_stats(facet);
} else {
- /* There's a flow in the datapath that we know nothing about.
- * Delete it. */
+ if (!VLOG_DROP_WARN(&rl)) {
+ struct ds s;
+
+ ds_init(&s);
+ odp_flow_key_format(key, key_len, &s);
+ VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+
COVERAGE_INC(facet_unexpected);
+ /* There's a flow in the datapath that we know nothing about, or a
+ * flow that shouldn't be installed but was anyway. Delete it. */
dpif_flow_del(p->dpif, key, key_len, NULL);
}
}
}
/* Calculates and returns the number of milliseconds of idle time after which
- * facets should expire from the datapath and we should fold their statistics
- * into their parent rules in userspace. */
+ * subfacets should expire from the datapath. When a subfacet expires, we fold
+ * its statistics into its facet, and when a facet's last subfacet expires, we
+ * fold its statistic into its rule. */
static int
-facet_max_idle(const struct ofproto_dpif *ofproto)
+subfacet_max_idle(const struct ofproto_dpif *ofproto)
{
/*
* Idle time histogram.
*
- * Most of the time a switch has a relatively small number of facets. When
- * this is the case we might as well keep statistics for all of them in
- * userspace and to cache them in the kernel datapath for performance as
+ * Most of the time a switch has a relatively small number of subfacets.
+ * When this is the case we might as well keep statistics for all of them
+ * in userspace and to cache them in the kernel datapath for performance as
* well.
*
- * As the number of facets increases, the memory required to maintain
+ * As the number of subfacets increases, the memory required to maintain
* statistics about them in userspace and in the kernel becomes
- * significant. However, with a large number of facets it is likely that
- * only a few of them are "heavy hitters" that consume a large amount of
- * bandwidth. At this point, only heavy hitters are worth caching in the
- * kernel and maintaining in userspaces; other facets we can discard.
+ * significant. However, with a large number of subfacets it is likely
+ * that only a few of them are "heavy hitters" that consume a large amount
+ * of bandwidth. At this point, only heavy hitters are worth caching in
+ * the kernel and maintaining in userspaces; other subfacets we can
+ * discard.
*
* The technique used to compute the idle time is to build a histogram with
- * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
+ * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
* that is installed in the kernel gets dropped in the appropriate bucket.
* After the histogram has been built, we compute the cutoff so that only
- * the most-recently-used 1% of facets (but at least
+ * the most-recently-used 1% of subfacets (but at least
* ofproto->up.flow_eviction_threshold flows) are kept cached. At least
- * the most-recently-used bucket of facets is kept, so actually an
- * arbitrary number of facets can be kept in any given expiration run
+ * the most-recently-used bucket of subfacets is kept, so actually an
+ * arbitrary number of subfacets can be kept in any given expiration run
* (though the next run will delete most of those unless they receive
* additional data).
*
- * This requires a second pass through the facets, in addition to the pass
- * made by update_stats(), because the former function never looks
- * at uninstallable facets.
+ * This requires a second pass through the subfacets, in addition to the
+ * pass made by update_stats(), because the former function never looks at
+ * uninstallable subfacets.
*/
enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
int buckets[N_BUCKETS] = { 0 };
int total, subtotal, bucket;
- struct facet *facet;
+ struct subfacet *subfacet;
long long int now;
int i;
- total = hmap_count(&ofproto->facets);
+ total = hmap_count(&ofproto->subfacets);
if (total <= ofproto->up.flow_eviction_threshold) {
return N_BUCKETS * BUCKET_WIDTH;
}
/* Build histogram. */
now = time_msec();
- HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
- long long int idle = now - facet->used;
+ HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
+ long long int idle = now - subfacet->used;
int bucket = (idle <= 0 ? 0
: idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
: (unsigned int) idle / BUCKET_WIDTH);
}
static void
-facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
-{
- if (ofproto->netflow && !facet_is_controller_flow(facet) &&
- netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
- struct ofexpired expired;
-
- if (facet->installed) {
- struct dpif_flow_stats stats;
-
- facet_put__(ofproto, facet, facet->actions, facet->actions_len,
- &stats);
- facet_update_stats(ofproto, facet, &stats);
- }
-
- expired.flow = facet->flow;
- expired.packet_count = facet->packet_count;
- expired.byte_count = facet->byte_count;
- expired.used = facet->used;
- netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
- }
-}
-
-static void
-expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
+expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
{
long long int cutoff = time_msec() - dp_max_idle;
- struct facet *facet, *next_facet;
+ struct subfacet *subfacet, *next_subfacet;
- HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
- facet_active_timeout(ofproto, facet);
- if (facet->used < cutoff) {
- facet_remove(ofproto, facet);
+ HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
+ &ofproto->subfacets) {
+ if (subfacet->used < cutoff) {
+ subfacet_destroy(subfacet);
}
}
}
static void
rule_expire(struct rule_dpif *rule)
{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct facet *facet, *next_facet;
long long int now;
uint8_t reason;
if (rule->up.hard_timeout
&& now > rule->up.modified + rule->up.hard_timeout * 1000) {
reason = OFPRR_HARD_TIMEOUT;
- } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
- && now > rule->used + rule->up.idle_timeout * 1000) {
+ } else if (rule->up.idle_timeout
+ && now > rule->up.used + rule->up.idle_timeout * 1000) {
reason = OFPRR_IDLE_TIMEOUT;
} else {
return;
/* Update stats. (This is a no-op if the rule expired due to an idle
* timeout, because that only happens when the rule has no facets left.) */
LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_remove(ofproto, facet);
+ facet_remove(facet);
}
/* Get rid of the rule. */
* 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
* the ofproto's classifier table.
*
- * The facet will initially have no ODP actions. The caller should fix that
- * by calling facet_make_actions(). */
+ * The facet will initially have no subfacets. The caller should create (at
+ * least) one subfacet with subfacet_create(). */
static struct facet *
facet_create(struct rule_dpif *rule, const struct flow *flow)
{
list_push_back(&rule->facets, &facet->list_node);
facet->rule = rule;
facet->flow = *flow;
+ list_init(&facet->subfacets);
netflow_flow_init(&facet->nf_flow);
netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
static void
facet_free(struct facet *facet)
{
- free(facet->actions);
free(facet);
}
-static bool
-execute_controller_action(struct ofproto_dpif *ofproto,
- const struct flow *flow,
- const struct nlattr *odp_actions, size_t actions_len,
- struct ofpbuf *packet)
-{
- if (actions_len
- && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
- && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
- /* As an optimization, avoid a round-trip from userspace to kernel to
- * userspace. This also avoids possibly filling up kernel packet
- * buffers along the way.
- *
- * This optimization will not accidentally catch sFlow
- * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
- * inside OVS_ACTION_ATTR_SAMPLE. */
- const struct nlattr *nla;
-
- nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
- send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
- false);
- return true;
- } else {
- return false;
- }
-}
-
/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
* 'packet', which arrived on 'in_port'.
*
struct ofpbuf key;
int error;
- if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
- packet)) {
- return true;
- }
-
ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
odp_flow_key_from_flow(&key, flow);
return !error;
}
-/* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
- * statistics appropriately. 'packet' must have at least sizeof(struct
- * ofp_packet_in) bytes of headroom.
- *
- * For correct results, 'packet' must actually be in 'facet''s flow; that is,
- * applying flow_extract() to 'packet' would yield the same flow as
- * 'facet->flow'.
- *
- * 'facet' must have accurately composed datapath actions; that is, it must
- * not be in need of revalidation.
- *
- * Takes ownership of 'packet'. */
-static void
-facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
- struct ofpbuf *packet)
-{
- struct dpif_flow_stats stats;
-
- assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
-
- dpif_flow_stats_extract(&facet->flow, packet, &stats);
- stats.used = time_msec();
- if (execute_odp_actions(ofproto, &facet->flow,
- facet->actions, facet->actions_len, packet)) {
- facet_update_stats(ofproto, facet, &stats);
- }
-}
-
/* Remove 'facet' from 'ofproto' and free up the associated memory:
*
* - If 'facet' was installed in the datapath, uninstalls it and updates its
- * rule's statistics, via facet_uninstall().
+ * rule's statistics, via subfacet_uninstall().
*
* - Removes 'facet' from its rule and from ofproto->facets.
*/
static void
-facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
-{
- facet_uninstall(ofproto, facet);
- facet_flush_stats(ofproto, facet);
- hmap_remove(&ofproto->facets, &facet->hmap_node);
- list_remove(&facet->list_node);
- facet_free(facet);
-}
-
-/* Composes the datapath actions for 'facet' based on its rule's actions. */
-static void
-facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
- const struct ofpbuf *packet)
-{
- const struct rule_dpif *rule = facet->rule;
- struct ofpbuf *odp_actions;
- struct action_xlate_ctx ctx;
-
- action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
- odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
- facet->tags = ctx.tags;
- facet->may_install = ctx.may_set_up_flow;
- facet->has_learn = ctx.has_learn;
- facet->has_normal = ctx.has_normal;
- facet->nf_flow.output_iface = ctx.nf_output_iface;
-
- if (facet->actions_len != odp_actions->size
- || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
- free(facet->actions);
- facet->actions_len = odp_actions->size;
- facet->actions = xmemdup(odp_actions->data, odp_actions->size);
- }
-
- ofpbuf_delete(odp_actions);
-}
-
-/* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
- * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
- * in the datapath will be zeroed and 'stats' will be updated with traffic new
- * since 'facet' was last updated.
- *
- * Returns 0 if successful, otherwise a positive errno value.*/
-static int
-facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
- const struct nlattr *actions, size_t actions_len,
- struct dpif_flow_stats *stats)
+facet_remove(struct facet *facet)
{
- struct odputil_keybuf keybuf;
- enum dpif_flow_put_flags flags;
- struct ofpbuf key;
- int ret;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct subfacet *subfacet, *next_subfacet;
- flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
- if (stats) {
- flags |= DPIF_FP_ZERO_STATS;
- }
-
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, &facet->flow);
-
- ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
- actions, actions_len, stats);
+ assert(!list_is_empty(&facet->subfacets));
- if (stats) {
- facet_reset_dp_stats(facet, stats);
+ /* First uninstall all of the subfacets to get final statistics. */
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ subfacet_uninstall(subfacet);
}
- return ret;
-}
-
-/* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
- * 'zero_stats' is true, clears any existing statistics from the datapath for
- * 'facet'. */
-static void
-facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
-{
- struct dpif_flow_stats stats;
+ /* Flush the final stats to the rule.
+ *
+ * This might require us to have at least one subfacet around so that we
+ * can use its actions for accounting in facet_account(), which is why we
+ * have uninstalled but not yet destroyed the subfacets. */
+ facet_flush_stats(facet);
- if (facet->may_install
- && !facet_put__(p, facet, facet->actions, facet->actions_len,
- zero_stats ? &stats : NULL)) {
- facet->installed = true;
+ /* Now we're really all done so destroy everything. */
+ LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
+ &facet->subfacets) {
+ subfacet_destroy__(subfacet);
}
+ hmap_remove(&ofproto->facets, &facet->hmap_node);
+ list_remove(&facet->list_node);
+ facet_free(facet);
}
static void
-facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
+facet_account(struct facet *facet)
{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
uint64_t n_bytes;
+ struct subfacet *subfacet;
const struct nlattr *a;
unsigned int left;
ovs_be16 vlan_tci;
/* Feed information from the active flows back into the learning table to
* ensure that table is always in sync with what is actually flowing
* through the datapath. */
- if (facet->has_learn || facet->has_normal) {
+ if (facet->has_learn || facet->has_normal
+ || (facet->has_fin_timeout
+ && facet->tcp_flags & (TCP_FIN | TCP_RST))) {
struct action_xlate_ctx ctx;
- action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+ facet->flow.vlan_tci,
+ facet->rule, facet->tcp_flags, NULL);
ctx.may_learn = true;
ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
facet->rule->up.n_actions));
* as a basis. We also need to track the actual VLAN on which the packet
* is going to be sent to ensure that it matches the one passed to
* bond_choose_output_slave(). (Otherwise, we will account to the wrong
- * hash bucket.) */
+ * hash bucket.)
+ *
+ * We use the actions from an arbitrary subfacet because they should all
+ * be equally valid for our purpose. */
+ subfacet = CONTAINER_OF(list_front(&facet->subfacets),
+ struct subfacet, list_node);
vlan_tci = facet->flow.vlan_tci;
- NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
+ NL_ATTR_FOR_EACH_UNSAFE (a, left,
+ subfacet->actions, subfacet->actions_len) {
const struct ovs_action_push_vlan *vlan;
struct ofport_dpif *port;
}
}
-/* If 'rule' is installed in the datapath, uninstalls it. */
-static void
-facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
-{
- if (facet->installed) {
- struct odputil_keybuf keybuf;
- struct dpif_flow_stats stats;
- struct ofpbuf key;
- int error;
-
- ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
- odp_flow_key_from_flow(&key, &facet->flow);
-
- error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
- facet_reset_dp_stats(facet, &stats);
- if (!error) {
- facet_update_stats(p, facet, &stats);
- }
- facet->installed = false;
- } else {
- assert(facet->dp_packet_count == 0);
- assert(facet->dp_byte_count == 0);
- }
-}
-
/* Returns true if the only action for 'facet' is to send to the controller.
* (We don't report NetFlow expiration messages for such facets because they
* are just part of the control logic for the network, not real traffic). */
htons(OFPP_CONTROLLER)));
}
-/* Resets 'facet''s datapath statistics counters. This should be called when
- * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
- * it should contain the statistics returned by dpif when 'facet' was reset in
- * the datapath. 'stats' will be modified to only included statistics new
- * since 'facet' was last updated. */
-static void
-facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
-{
- if (stats && facet->dp_packet_count <= stats->n_packets
- && facet->dp_byte_count <= stats->n_bytes) {
- stats->n_packets -= facet->dp_packet_count;
- stats->n_bytes -= facet->dp_byte_count;
- }
-
- facet->dp_packet_count = 0;
- facet->dp_byte_count = 0;
-}
-
/* Folds all of 'facet''s statistics into its rule. Also updates the
* accounting ofhook and emits a NetFlow expiration if appropriate. All of
* 'facet''s statistics in the datapath should have been zeroed and folded into
* its packet and byte counts before this function is called. */
static void
-facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
+facet_flush_stats(struct facet *facet)
{
- assert(!facet->dp_byte_count);
- assert(!facet->dp_packet_count);
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct subfacet *subfacet;
+
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ assert(!subfacet->dp_byte_count);
+ assert(!subfacet->dp_packet_count);
+ }
facet_push_stats(facet);
- facet_account(ofproto, facet);
+ facet_account(facet);
if (ofproto->netflow && !facet_is_controller_flow(facet)) {
struct ofexpired expired;
facet_reset_counters(facet);
netflow_flow_clear(&facet->nf_flow);
+ facet->tcp_flags = 0;
}
/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
if (facet
&& (ofproto->need_revalidate
|| tag_set_intersects(&ofproto->revalidate_set, facet->tags))
- && !facet_revalidate(ofproto, facet)) {
+ && !facet_revalidate(facet)) {
COVERAGE_INC(facet_invalidated);
return NULL;
}
return facet;
}
-/* Re-searches 'ofproto''s classifier for a rule matching 'facet':
+static bool
+facet_check_consistency(struct facet *facet)
+{
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
+
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+
+ struct rule_dpif *rule;
+ struct subfacet *subfacet;
+ bool may_log = false;
+ bool ok;
+
+ /* Check the rule for consistency. */
+ rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
+ if (!rule) {
+ if (!VLOG_DROP_WARN(&rl)) {
+ char *s = flow_to_string(&facet->flow);
+ VLOG_WARN("%s: facet should not exist", s);
+ free(s);
+ }
+ return false;
+ } else if (rule != facet->rule) {
+ may_log = !VLOG_DROP_WARN(&rl);
+ ok = false;
+ if (may_log) {
+ struct ds s;
+
+ ds_init(&s);
+ flow_format(&s, &facet->flow);
+ ds_put_format(&s, ": facet associated with wrong rule (was "
+ "table=%"PRIu8",", facet->rule->up.table_id);
+ cls_rule_format(&facet->rule->up.cr, &s);
+ ds_put_format(&s, ") (should have been table=%"PRIu8",",
+ rule->up.table_id);
+ cls_rule_format(&rule->up.cr, &s);
+ ds_put_char(&s, ')');
+
+ VLOG_WARN("%s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+ } else {
+ ok = true;
+ }
+
+ /* Check the datapath actions for consistency. */
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ struct action_xlate_ctx ctx;
+ struct ofpbuf *odp_actions;
+ bool actions_changed;
+ bool should_install;
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+ subfacet->initial_tci, rule, 0, NULL);
+ odp_actions = xlate_actions(&ctx, rule->up.actions,
+ rule->up.n_actions);
+
+ should_install = (ctx.may_set_up_flow
+ && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
+ if (!should_install && !subfacet->installed) {
+ /* The actions for uninstallable flows may vary from one packet to
+ * the next, so don't compare the actions. */
+ goto next;
+ }
+
+ actions_changed = (subfacet->actions_len != odp_actions->size
+ || memcmp(subfacet->actions, odp_actions->data,
+ subfacet->actions_len));
+ if (should_install != subfacet->installed || actions_changed) {
+ if (ok) {
+ may_log = !VLOG_DROP_WARN(&rl);
+ ok = false;
+ }
+
+ if (may_log) {
+ struct odputil_keybuf keybuf;
+ struct ofpbuf key;
+ struct ds s;
+
+ ds_init(&s);
+ subfacet_get_key(subfacet, &keybuf, &key);
+ odp_flow_key_format(key.data, key.size, &s);
+
+ ds_put_cstr(&s, ": inconsistency in subfacet");
+ if (should_install != subfacet->installed) {
+ enum odp_key_fitness fitness = subfacet->key_fitness;
+
+ ds_put_format(&s, " (should%s have been installed)",
+ should_install ? "" : " not");
+ ds_put_format(&s, " (may_set_up_flow=%s, fitness=%s)",
+ ctx.may_set_up_flow ? "true" : "false",
+ odp_key_fitness_to_string(fitness));
+ }
+ if (actions_changed) {
+ ds_put_cstr(&s, " (actions were: ");
+ format_odp_actions(&s, subfacet->actions,
+ subfacet->actions_len);
+ ds_put_cstr(&s, ") (correct actions: ");
+ format_odp_actions(&s, odp_actions->data,
+ odp_actions->size);
+ ds_put_char(&s, ')');
+ } else {
+ ds_put_cstr(&s, " (actions: ");
+ format_odp_actions(&s, subfacet->actions,
+ subfacet->actions_len);
+ ds_put_char(&s, ')');
+ }
+ VLOG_WARN("%s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+ }
+
+ next:
+ ofpbuf_delete(odp_actions);
+ }
+
+ return ok;
+}
+
+/* Re-searches the classifier for 'facet':
*
* - If the rule found is different from 'facet''s current rule, moves
* 'facet' to the new rule and recompiles its actions.
*
* Returns true if 'facet' still exists, false if it has been destroyed. */
static bool
-facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
+facet_revalidate(struct facet *facet)
{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct actions {
+ struct nlattr *odp_actions;
+ size_t actions_len;
+ };
+ struct actions *new_actions;
+
struct action_xlate_ctx ctx;
- struct ofpbuf *odp_actions;
struct rule_dpif *new_rule;
+ struct subfacet *subfacet;
bool actions_changed;
+ int i;
COVERAGE_INC(facet_revalidate);
new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
if (!new_rule) {
/* No new rule, so delete the facet. */
- facet_remove(ofproto, facet);
+ facet_remove(facet);
return false;
}
* We do not modify any 'facet' state yet, because we might need to, e.g.,
* emit a NetFlow expiration and, if so, we need to have the old state
* around to properly compose it. */
- action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
- odp_actions = xlate_actions(&ctx,
- new_rule->up.actions, new_rule->up.n_actions);
- actions_changed = (facet->actions_len != odp_actions->size
- || memcmp(facet->actions, odp_actions->data,
- facet->actions_len));
/* If the datapath actions changed or the installability changed,
* then we need to talk to the datapath. */
- if (actions_changed || ctx.may_set_up_flow != facet->installed) {
- if (ctx.may_set_up_flow) {
- struct dpif_flow_stats stats;
+ i = 0;
+ new_actions = NULL;
+ memset(&ctx, 0, sizeof ctx);
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ struct ofpbuf *odp_actions;
+ bool should_install;
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+ subfacet->initial_tci, new_rule, 0, NULL);
+ odp_actions = xlate_actions(&ctx, new_rule->up.actions,
+ new_rule->up.n_actions);
+ actions_changed = (subfacet->actions_len != odp_actions->size
+ || memcmp(subfacet->actions, odp_actions->data,
+ subfacet->actions_len));
+
+ should_install = (ctx.may_set_up_flow
+ && subfacet->key_fitness != ODP_FIT_TOO_LITTLE);
+ if (actions_changed || should_install != subfacet->installed) {
+ if (should_install) {
+ struct dpif_flow_stats stats;
+
+ subfacet_install(subfacet,
+ odp_actions->data, odp_actions->size, &stats);
+ subfacet_update_stats(subfacet, &stats);
+ } else {
+ subfacet_uninstall(subfacet);
+ }
- facet_put__(ofproto, facet,
- odp_actions->data, odp_actions->size, &stats);
- facet_update_stats(ofproto, facet, &stats);
- } else {
- facet_uninstall(ofproto, facet);
+ if (!new_actions) {
+ new_actions = xcalloc(list_size(&facet->subfacets),
+ sizeof *new_actions);
+ }
+ new_actions[i].odp_actions = xmemdup(odp_actions->data,
+ odp_actions->size);
+ new_actions[i].actions_len = odp_actions->size;
}
- /* The datapath flow is gone or has zeroed stats, so push stats out of
- * 'facet' into 'rule'. */
- facet_flush_stats(ofproto, facet);
+ ofpbuf_delete(odp_actions);
+ i++;
+ }
+ if (new_actions) {
+ facet_flush_stats(facet);
}
/* Update 'facet' now that we've taken care of all the old state. */
facet->may_install = ctx.may_set_up_flow;
facet->has_learn = ctx.has_learn;
facet->has_normal = ctx.has_normal;
- if (actions_changed) {
- free(facet->actions);
- facet->actions_len = odp_actions->size;
- facet->actions = xmemdup(odp_actions->data, odp_actions->size);
+ facet->has_fin_timeout = ctx.has_fin_timeout;
+ facet->mirrors = ctx.mirrors;
+ if (new_actions) {
+ i = 0;
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ if (new_actions[i].odp_actions) {
+ free(subfacet->actions);
+ subfacet->actions = new_actions[i].odp_actions;
+ subfacet->actions_len = new_actions[i].actions_len;
+ }
+ i++;
+ }
+ free(new_actions);
}
if (facet->rule != new_rule) {
COVERAGE_INC(facet_changed_rule);
list_push_back(&new_rule->facets, &facet->list_node);
facet->rule = new_rule;
facet->used = new_rule->up.created;
- facet->rs_used = facet->used;
+ facet->prev_used = facet->used;
+ }
+
+ return true;
+}
+
+/* Updates 'facet''s used time. Caller is responsible for calling
+ * facet_push_stats() to update the flows which 'facet' resubmits into. */
+static void
+facet_update_time(struct facet *facet, long long int used)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ if (used > facet->used) {
+ facet->used = used;
+ ofproto_rule_update_used(&facet->rule->up, used);
+ netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
+ }
+}
+
+static void
+facet_reset_counters(struct facet *facet)
+{
+ facet->packet_count = 0;
+ facet->byte_count = 0;
+ facet->prev_packet_count = 0;
+ facet->prev_byte_count = 0;
+ facet->accounted_bytes = 0;
+}
+
+static void
+facet_push_stats(struct facet *facet)
+{
+ uint64_t new_packets, new_bytes;
+
+ assert(facet->packet_count >= facet->prev_packet_count);
+ assert(facet->byte_count >= facet->prev_byte_count);
+ assert(facet->used >= facet->prev_used);
+
+ new_packets = facet->packet_count - facet->prev_packet_count;
+ new_bytes = facet->byte_count - facet->prev_byte_count;
+
+ if (new_packets || new_bytes || facet->used > facet->prev_used) {
+ facet->prev_packet_count = facet->packet_count;
+ facet->prev_byte_count = facet->byte_count;
+ facet->prev_used = facet->used;
+
+ flow_push_stats(facet->rule, &facet->flow,
+ new_packets, new_bytes, facet->used);
+
+ update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
+ facet->mirrors, new_packets, new_bytes);
+ }
+}
+
+struct ofproto_push {
+ struct action_xlate_ctx ctx;
+ uint64_t packets;
+ uint64_t bytes;
+ long long int used;
+};
+
+static void
+push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
+{
+ struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
+
+ if (rule) {
+ rule->packet_count += push->packets;
+ rule->byte_count += push->bytes;
+ ofproto_rule_update_used(&rule->up, push->used);
+ }
+}
+
+/* Pushes flow statistics to the rules which 'flow' resubmits into given
+ * 'rule''s actions and mirrors. */
+static void
+flow_push_stats(struct rule_dpif *rule,
+ const struct flow *flow, uint64_t packets, uint64_t bytes,
+ long long int used)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+ struct ofproto_push push;
+
+ push.packets = packets;
+ push.bytes = bytes;
+ push.used = used;
+
+ ofproto_rule_update_used(&rule->up, used);
+
+ action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, rule,
+ 0, NULL);
+ push.ctx.resubmit_hook = push_resubmit;
+ ofpbuf_delete(xlate_actions(&push.ctx,
+ rule->up.actions, rule->up.n_actions));
+}
+\f
+/* Subfacets. */
+
+static struct subfacet *
+subfacet_find__(struct ofproto_dpif *ofproto,
+ const struct nlattr *key, size_t key_len, uint32_t key_hash,
+ const struct flow *flow)
+{
+ struct subfacet *subfacet;
+
+ HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
+ &ofproto->subfacets) {
+ if (subfacet->key
+ ? (subfacet->key_len == key_len
+ && !memcmp(key, subfacet->key, key_len))
+ : flow_equal(flow, &subfacet->facet->flow)) {
+ return subfacet;
+ }
+ }
+
+ return NULL;
+}
+
+/* Searches 'facet' (within 'ofproto') for a subfacet with the specified
+ * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
+ * there is one, otherwise creates and returns a new subfacet.
+ *
+ * If the returned subfacet is new, then subfacet->actions will be NULL, in
+ * which case the caller must populate the actions with
+ * subfacet_make_actions(). */
+static struct subfacet *
+subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
+ const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ uint32_t key_hash = odp_flow_key_hash(key, key_len);
+ struct subfacet *subfacet;
+
+ subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
+ if (subfacet) {
+ if (subfacet->facet == facet) {
+ return subfacet;
+ }
+
+ /* This shouldn't happen. */
+ VLOG_ERR_RL(&rl, "subfacet with wrong facet");
+ subfacet_destroy(subfacet);
+ }
+
+ subfacet = xzalloc(sizeof *subfacet);
+ hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
+ list_push_back(&facet->subfacets, &subfacet->list_node);
+ subfacet->facet = facet;
+ subfacet->used = time_msec();
+ subfacet->key_fitness = key_fitness;
+ if (key_fitness != ODP_FIT_PERFECT) {
+ subfacet->key = xmemdup(key, key_len);
+ subfacet->key_len = key_len;
+ }
+ subfacet->installed = false;
+ subfacet->initial_tci = initial_tci;
+
+ return subfacet;
+}
+
+/* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
+ * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
+static struct subfacet *
+subfacet_find(struct ofproto_dpif *ofproto,
+ const struct nlattr *key, size_t key_len)
+{
+ uint32_t key_hash = odp_flow_key_hash(key, key_len);
+ enum odp_key_fitness fitness;
+ struct flow flow;
+
+ fitness = odp_flow_key_to_flow(key, key_len, &flow);
+ if (fitness == ODP_FIT_ERROR) {
+ return NULL;
+ }
+
+ return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
+}
+
+/* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
+ * its facet within 'ofproto', and frees it. */
+static void
+subfacet_destroy__(struct subfacet *subfacet)
+{
+ struct facet *facet = subfacet->facet;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+
+ subfacet_uninstall(subfacet);
+ hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
+ list_remove(&subfacet->list_node);
+ free(subfacet->key);
+ free(subfacet->actions);
+ free(subfacet);
+}
+
+/* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
+ * last remaining subfacet in its facet destroys the facet too. */
+static void
+subfacet_destroy(struct subfacet *subfacet)
+{
+ struct facet *facet = subfacet->facet;
+
+ if (list_is_singleton(&facet->subfacets)) {
+ /* facet_remove() needs at least one subfacet (it will remove it). */
+ facet_remove(facet);
+ } else {
+ subfacet_destroy__(subfacet);
+ }
+}
+
+/* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
+ * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
+ * for use as temporary storage. */
+static void
+subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
+ struct ofpbuf *key)
+{
+ if (!subfacet->key) {
+ ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
+ odp_flow_key_from_flow(key, &subfacet->facet->flow);
+ } else {
+ ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
+ }
+}
+
+/* Composes the datapath actions for 'subfacet' based on its rule's actions. */
+static void
+subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
+{
+ struct facet *facet = subfacet->facet;
+ struct rule_dpif *rule = facet->rule;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+ struct ofpbuf *odp_actions;
+ struct action_xlate_ctx ctx;
+
+ action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
+ rule, 0, packet);
+ odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
+ facet->tags = ctx.tags;
+ facet->may_install = ctx.may_set_up_flow;
+ facet->has_learn = ctx.has_learn;
+ facet->has_normal = ctx.has_normal;
+ facet->has_fin_timeout = ctx.has_fin_timeout;
+ facet->nf_flow.output_iface = ctx.nf_output_iface;
+ facet->mirrors = ctx.mirrors;
+
+ if (subfacet->actions_len != odp_actions->size
+ || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
+ free(subfacet->actions);
+ subfacet->actions_len = odp_actions->size;
+ subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
+ }
+
+ ofpbuf_delete(odp_actions);
+}
+
+/* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
+ * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
+ * in the datapath will be zeroed and 'stats' will be updated with traffic new
+ * since 'subfacet' was last updated.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
+static int
+subfacet_install(struct subfacet *subfacet,
+ const struct nlattr *actions, size_t actions_len,
+ struct dpif_flow_stats *stats)
+{
+ struct facet *facet = subfacet->facet;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+ struct odputil_keybuf keybuf;
+ enum dpif_flow_put_flags flags;
+ struct ofpbuf key;
+ int ret;
+
+ flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
+ if (stats) {
+ flags |= DPIF_FP_ZERO_STATS;
}
- ofpbuf_delete(odp_actions);
+ subfacet_get_key(subfacet, &keybuf, &key);
+ ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
+ actions, actions_len, stats);
- return true;
+ if (stats) {
+ subfacet_reset_dp_stats(subfacet, stats);
+ }
+
+ return ret;
}
-/* Updates 'facet''s used time. Caller is responsible for calling
- * facet_push_stats() to update the flows which 'facet' resubmits into. */
+/* If 'subfacet' is installed in the datapath, uninstalls it. */
static void
-facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
- long long int used)
+subfacet_uninstall(struct subfacet *subfacet)
{
- if (used > facet->used) {
- facet->used = used;
- if (used > facet->rule->used) {
- facet->rule->used = used;
+ if (subfacet->installed) {
+ struct rule_dpif *rule = subfacet->facet->rule;
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+ struct odputil_keybuf keybuf;
+ struct dpif_flow_stats stats;
+ struct ofpbuf key;
+ int error;
+
+ subfacet_get_key(subfacet, &keybuf, &key);
+ error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
+ subfacet_reset_dp_stats(subfacet, &stats);
+ if (!error) {
+ subfacet_update_stats(subfacet, &stats);
}
- netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
+ subfacet->installed = false;
+ } else {
+ assert(subfacet->dp_packet_count == 0);
+ assert(subfacet->dp_byte_count == 0);
}
}
-/* Folds the statistics from 'stats' into the counters in 'facet'.
- *
- * Because of the meaning of a facet's counters, it only makes sense to do this
- * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
- * packet that was sent by hand or if it represents statistics that have been
- * cleared out of the datapath. */
+/* Resets 'subfacet''s datapath statistics counters. This should be called
+ * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
+ * non-null, it should contain the statistics returned by dpif when 'subfacet'
+ * was reset in the datapath. 'stats' will be modified to include only
+ * statistics new since 'subfacet' was last updated. */
static void
-facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
- const struct dpif_flow_stats *stats)
+subfacet_reset_dp_stats(struct subfacet *subfacet,
+ struct dpif_flow_stats *stats)
{
- if (stats->n_packets || stats->used > facet->used) {
- facet_update_time(ofproto, facet, stats->used);
- facet->packet_count += stats->n_packets;
- facet->byte_count += stats->n_bytes;
- facet_push_stats(facet);
- netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
+ if (stats
+ && subfacet->dp_packet_count <= stats->n_packets
+ && subfacet->dp_byte_count <= stats->n_bytes) {
+ stats->n_packets -= subfacet->dp_packet_count;
+ stats->n_bytes -= subfacet->dp_byte_count;
}
-}
-static void
-facet_reset_counters(struct facet *facet)
-{
- facet->packet_count = 0;
- facet->byte_count = 0;
- facet->rs_packet_count = 0;
- facet->rs_byte_count = 0;
- facet->accounted_bytes = 0;
+ subfacet->dp_packet_count = 0;
+ subfacet->dp_byte_count = 0;
}
+/* Updates 'subfacet''s used time. The caller is responsible for calling
+ * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
static void
-facet_push_stats(struct facet *facet)
+subfacet_update_time(struct subfacet *subfacet, long long int used)
{
- uint64_t rs_packets, rs_bytes;
-
- assert(facet->packet_count >= facet->rs_packet_count);
- assert(facet->byte_count >= facet->rs_byte_count);
- assert(facet->used >= facet->rs_used);
-
- rs_packets = facet->packet_count - facet->rs_packet_count;
- rs_bytes = facet->byte_count - facet->rs_byte_count;
-
- if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
- facet->rs_packet_count = facet->packet_count;
- facet->rs_byte_count = facet->byte_count;
- facet->rs_used = facet->used;
-
- flow_push_stats(facet->rule, &facet->flow,
- rs_packets, rs_bytes, facet->used);
+ if (used > subfacet->used) {
+ subfacet->used = used;
+ facet_update_time(subfacet->facet, used);
}
}
-struct ofproto_push {
- struct action_xlate_ctx ctx;
- uint64_t packets;
- uint64_t bytes;
- long long int used;
-};
-
+/* Folds the statistics from 'stats' into the counters in 'subfacet'.
+ *
+ * Because of the meaning of a subfacet's counters, it only makes sense to do
+ * this if 'stats' are not tracked in the datapath, that is, if 'stats'
+ * represents a packet that was sent by hand or if it represents statistics
+ * that have been cleared out of the datapath. */
static void
-push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
+subfacet_update_stats(struct subfacet *subfacet,
+ const struct dpif_flow_stats *stats)
{
- struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
+ if (stats->n_packets || stats->used > subfacet->used) {
+ struct facet *facet = subfacet->facet;
- if (rule) {
- rule->packet_count += push->packets;
- rule->byte_count += push->bytes;
- rule->used = MAX(push->used, rule->used);
+ subfacet_update_time(subfacet, stats->used);
+ facet->packet_count += stats->n_packets;
+ facet->byte_count += stats->n_bytes;
+ facet->tcp_flags |= stats->tcp_flags;
+ facet_push_stats(facet);
+ netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
}
}
-
-/* Pushes flow statistics to the rules which 'flow' resubmits into given
- * 'rule''s actions. */
-static void
-flow_push_stats(const struct rule_dpif *rule,
- struct flow *flow, uint64_t packets, uint64_t bytes,
- long long int used)
-{
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- struct ofproto_push push;
-
- push.packets = packets;
- push.bytes = bytes;
- push.used = used;
-
- action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
- push.ctx.resubmit_hook = push_resubmit;
- ofpbuf_delete(xlate_actions(&push.ctx,
- rule->up.actions, rule->up.n_actions));
-}
\f
/* Rules. */
return NULL;
}
- cls = &ofproto->up.tables[table_id];
+ cls = &ofproto->up.tables[table_id].cls;
if (flow->nw_frag & FLOW_NW_FRAG_ANY
&& ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
/* For OFPC_NORMAL frag_handling, we must pretend that transport ports
free(rule);
}
-static int
+static enum ofperr
rule_construct(struct rule *rule_)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct rule_dpif *victim;
uint8_t table_id;
- int error;
+ enum ofperr error;
error = validate_actions(rule->up.actions, rule->up.n_actions,
&rule->up.cr.flow, ofproto->max_ports);
return error;
}
- rule->used = rule->up.created;
rule->packet_count = 0;
rule->byte_count = 0;
rule_destruct(struct rule *rule_)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
- struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct facet *facet, *next_facet;
LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
- facet_revalidate(ofproto, facet);
+ facet_revalidate(facet);
}
complete_operation(rule);
}
}
-static int
-rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
+static enum ofperr
+rule_execute(struct rule *rule_, const struct flow *flow,
+ struct ofpbuf *packet)
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
struct action_xlate_ctx ctx;
struct ofpbuf *odp_actions;
- struct facet *facet;
size_t size;
- /* First look for a related facet. If we find one, account it to that. */
- facet = facet_lookup_valid(ofproto, flow);
- if (facet && facet->rule == rule) {
- if (!facet->may_install) {
- facet_make_actions(ofproto, facet, packet);
- }
- facet_execute(ofproto, facet, packet);
- return 0;
- }
-
- /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
- * create a new facet for it and use that. */
- if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
- facet = facet_create(rule, flow);
- facet_make_actions(ofproto, facet, packet);
- facet_execute(ofproto, facet, packet);
- facet_install(ofproto, facet, true);
- return 0;
- }
-
- /* We can't account anything to a facet. If we were to try, then that
- * facet would have a non-matching rule, busting our invariants. */
- action_xlate_ctx_init(&ctx, ofproto, flow, packet);
+ action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
+ rule, packet_get_tcp_flags(packet, flow), packet);
odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
size = packet->size;
if (execute_odp_actions(ofproto, flow, odp_actions->data,
odp_actions->size, packet)) {
- rule->used = time_msec();
rule->packet_count++;
rule->byte_count += size;
- flow_push_stats(rule, flow, 1, size, rule->used);
+ flow_push_stats(rule, flow, 1, size, time_msec());
}
ofpbuf_delete(odp_actions);
{
struct rule_dpif *rule = rule_dpif_cast(rule_);
struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
- int error;
+ enum ofperr error;
error = validate_actions(rule->up.actions, rule->up.n_actions,
&rule->up.cr.flow, ofproto->max_ports);
complete_operation(rule);
}
\f
-/* Sends 'packet' out of port 'odp_port' within 'ofproto'.
+/* Sends 'packet' out 'ofport'.
+ * May modify 'packet'.
* Returns 0 if successful, otherwise a positive errno value. */
static int
-send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
- const struct ofpbuf *packet)
+send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
{
+ const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
struct ofpbuf key, odp_actions;
struct odputil_keybuf keybuf;
+ uint16_t odp_port;
struct flow flow;
int error;
flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
+ odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
+ flow.vlan_tci);
+ if (odp_port != ofport->odp_port) {
+ eth_pop_vlan(packet);
+ flow.vlan_tci = htons(0);
+ }
+
ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
odp_flow_key_from_flow(&key, &flow);
VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
ofproto->up.name, odp_port, strerror(error));
}
+ ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
return error;
}
\f
cookie->vlan_tci = base->vlan_tci;
}
-static void
-commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
- const void *key, size_t key_size)
-{
- size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
- nl_msg_put_unspec(odp_actions, key_type, key, key_size);
- nl_msg_end_nested(odp_actions, offset);
-}
-
-static void
-commit_set_tun_id_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- if (base->tun_id == flow->tun_id) {
- return;
- }
- base->tun_id = flow->tun_id;
-
- commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID,
- &base->tun_id, sizeof(base->tun_id));
-}
-
-static void
-commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- struct ovs_key_ethernet eth_key;
-
- if (eth_addr_equals(base->dl_src, flow->dl_src) &&
- eth_addr_equals(base->dl_dst, flow->dl_dst)) {
- return;
- }
-
- memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
- memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
-
- memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
- memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
-
- commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
- ð_key, sizeof(eth_key));
-}
-
-static void
-commit_vlan_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- if (base->vlan_tci == flow->vlan_tci) {
- return;
- }
-
- if (base->vlan_tci & htons(VLAN_CFI)) {
- nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
- }
-
- if (flow->vlan_tci & htons(VLAN_CFI)) {
- struct ovs_action_push_vlan vlan;
-
- vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
- vlan.vlan_tci = flow->vlan_tci;
- nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
- &vlan, sizeof vlan);
- }
- base->vlan_tci = flow->vlan_tci;
-}
-
-static void
-commit_set_nw_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- struct ovs_key_ipv4 ipv4_key;
-
- if (base->dl_type != htons(ETH_TYPE_IP) ||
- !base->nw_src || !base->nw_dst) {
- return;
- }
-
- if (base->nw_src == flow->nw_src &&
- base->nw_dst == flow->nw_dst &&
- base->nw_tos == flow->nw_tos &&
- base->nw_ttl == flow->nw_ttl &&
- base->nw_frag == flow->nw_frag) {
- return;
- }
-
- ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
- ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
- ipv4_key.ipv4_tos = base->nw_tos = flow->nw_tos;
- ipv4_key.ipv4_ttl = base->nw_ttl = flow->nw_ttl;
- ipv4_key.ipv4_proto = base->nw_proto;
- ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE
- : base->nw_frag == FLOW_NW_FRAG_ANY
- ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER);
-
- commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
- &ipv4_key, sizeof(ipv4_key));
-}
-
-static void
-commit_set_port_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- if (!base->tp_src || !base->tp_dst) {
- return;
- }
-
- if (base->tp_src == flow->tp_src &&
- base->tp_dst == flow->tp_dst) {
- return;
- }
-
- if (flow->nw_proto == IPPROTO_TCP) {
- struct ovs_key_tcp port_key;
-
- port_key.tcp_src = base->tp_src = flow->tp_src;
- port_key.tcp_dst = base->tp_dst = flow->tp_dst;
-
- commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
- &port_key, sizeof(port_key));
-
- } else if (flow->nw_proto == IPPROTO_UDP) {
- struct ovs_key_udp port_key;
-
- port_key.udp_src = base->tp_src = flow->tp_src;
- port_key.udp_dst = base->tp_dst = flow->tp_dst;
-
- commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
- &port_key, sizeof(port_key));
- }
-}
-
-static void
-commit_set_priority_action(const struct flow *flow, struct flow *base,
- struct ofpbuf *odp_actions)
-{
- if (base->priority == flow->priority) {
- return;
- }
- base->priority = flow->priority;
-
- commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
- &base->priority, sizeof(base->priority));
-}
-
-static void
-commit_odp_actions(struct action_xlate_ctx *ctx)
-{
- const struct flow *flow = &ctx->flow;
- struct flow *base = &ctx->base_flow;
- struct ofpbuf *odp_actions = ctx->odp_actions;
-
- commit_set_tun_id_action(flow, base, odp_actions);
- commit_set_ether_addr_action(flow, base, odp_actions);
- commit_vlan_action(flow, base, odp_actions);
- commit_set_nw_action(flow, base, odp_actions);
- commit_set_port_action(flow, base, odp_actions);
- commit_set_priority_action(flow, base, odp_actions);
-}
-
static void
compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
bool check_stp)
{
const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
+ ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
+ uint8_t flow_nw_tos = ctx->flow.nw_tos;
+ uint16_t out_port;
if (ofport) {
+ struct priority_to_dscp *pdscp;
+
if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
|| (check_stp && !stp_forward_in_state(ofport->stp_state))) {
return;
}
+
+ pdscp = get_priority(ofport, ctx->flow.skb_priority);
+ if (pdscp) {
+ ctx->flow.nw_tos &= ~IP_DSCP_MASK;
+ ctx->flow.nw_tos |= pdscp->dscp;
+ }
} else {
/* We may not have an ofport record for this port, but it doesn't hurt
* to allow forwarding to it anyhow. Maybe such a port will appear
* later and we're pre-populating the flow table. */
}
- commit_odp_actions(ctx);
- nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
+ out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
+ ctx->flow.vlan_tci);
+ if (out_port != odp_port) {
+ ctx->flow.vlan_tci = htons(0);
+ }
+ commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
+ nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
+
ctx->sflow_odp_port = odp_port;
ctx->sflow_n_outputs++;
ctx->nf_output_iface = ofp_port;
+ ctx->flow.vlan_tci = flow_vlan_tci;
+ ctx->flow.nw_tos = flow_nw_tos;
}
static void
}
if (rule) {
+ struct rule_dpif *old_rule = ctx->rule;
+
ctx->recurse++;
+ ctx->rule = rule;
do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
+ ctx->rule = old_rule;
ctx->recurse--;
}
{
struct ofport_dpif *ofport;
- commit_odp_actions(ctx);
HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
uint16_t ofp_port = ofport->up.ofp_port;
}
static void
-compose_controller_action(struct action_xlate_ctx *ctx, int len)
+execute_controller_action(struct action_xlate_ctx *ctx, int len,
+ enum ofp_packet_in_reason reason,
+ uint16_t controller_id)
{
- struct user_action_cookie cookie;
+ struct ofputil_packet_in pin;
+ struct ofpbuf *packet;
- commit_odp_actions(ctx);
- cookie.type = USER_ACTION_COOKIE_CONTROLLER;
- cookie.data = len;
- cookie.n_output = 0;
- cookie.vlan_tci = 0;
- put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
+ ctx->may_set_up_flow = false;
+ if (!ctx->packet) {
+ return;
+ }
+
+ packet = ofpbuf_clone(ctx->packet);
+
+ if (packet->l2 && packet->l3) {
+ struct eth_header *eh;
+
+ eth_pop_vlan(packet);
+ eh = packet->l2;
+ assert(eh->eth_type == ctx->flow.dl_type);
+ memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
+ memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
+
+ if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
+ eth_push_vlan(packet, ctx->flow.vlan_tci);
+ }
+
+ if (packet->l4) {
+ if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
+ packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
+ ctx->flow.nw_tos, ctx->flow.nw_ttl);
+ }
+
+ if (packet->l7) {
+ if (ctx->flow.nw_proto == IPPROTO_TCP) {
+ packet_set_tcp_port(packet, ctx->flow.tp_src,
+ ctx->flow.tp_dst);
+ } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
+ packet_set_udp_port(packet, ctx->flow.tp_src,
+ ctx->flow.tp_dst);
+ }
+ }
+ }
+ }
+
+ pin.packet = packet->data;
+ pin.packet_len = packet->size;
+ pin.reason = reason;
+ pin.controller_id = controller_id;
+ pin.table_id = ctx->table_id;
+ pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
+
+ pin.buffer_id = 0;
+ pin.send_len = len;
+ pin.total_len = packet->size;
+ flow_get_metadata(&ctx->flow, &pin.fmd);
+
+ connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin, &ctx->flow);
+ ofpbuf_delete(packet);
+}
+
+static bool
+compose_dec_ttl(struct action_xlate_ctx *ctx)
+{
+ if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
+ ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
+ return false;
+ }
+
+ if (ctx->flow.nw_ttl > 1) {
+ ctx->flow.nw_ttl--;
+ return false;
+ } else {
+ execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
+
+ /* Stop processing for current table. */
+ return true;
+ }
}
static void
flood_packets(ctx, true);
break;
case OFPP_CONTROLLER:
- compose_controller_action(ctx, max_len);
- break;
- case OFPP_LOCAL:
- compose_output_action(ctx, OFPP_LOCAL);
+ execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
break;
case OFPP_NONE:
break;
+ case OFPP_LOCAL:
default:
if (port != ctx->flow.in_port) {
compose_output_action(ctx, port);
xlate_output_reg_action(struct action_xlate_ctx *ctx,
const struct nx_action_output_reg *naor)
{
+ struct mf_subfield src;
uint64_t ofp_port;
- ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
+ nxm_decode(&src, naor->src, naor->ofs_nbits);
+ ofp_port = mf_get_subfield(&src, &ctx->flow);
if (ofp_port <= UINT16_MAX) {
xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
}
/* Add datapath actions. */
- flow_priority = ctx->flow.priority;
- ctx->flow.priority = priority;
+ flow_priority = ctx->flow.skb_priority;
+ ctx->flow.skb_priority = priority;
compose_output_action(ctx, ofp_port);
- ctx->flow.priority = flow_priority;
+ ctx->flow.skb_priority = flow_priority;
/* Update NetFlow output port. */
if (ctx->nf_output_iface == NF_OUT_DROP) {
return;
}
- ctx->flow.priority = priority;
+ ctx->flow.skb_priority = priority;
}
struct xlate_reg_state {
error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
if (error && !VLOG_DROP_WARN(&rl)) {
- char *msg = ofputil_error_to_string(error);
- VLOG_WARN("learning action failed to modify flow table (%s)", msg);
- free(msg);
+ VLOG_WARN("learning action failed to modify flow table (%s)",
+ ofperr_get_name(error));
}
free(fm.actions);
}
+/* Reduces '*timeout' to no more than 'max'. A value of zero in either case
+ * means "infinite". */
+static void
+reduce_timeout(uint16_t max, uint16_t *timeout)
+{
+ if (max && (!*timeout || *timeout > max)) {
+ *timeout = max;
+ }
+}
+
+static void
+xlate_fin_timeout(struct action_xlate_ctx *ctx,
+ const struct nx_action_fin_timeout *naft)
+{
+ if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
+ struct rule_dpif *rule = ctx->rule;
+
+ reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
+ reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
+ }
+}
+
static bool
may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
{
{
const struct ofport_dpif *port;
const union ofp_action *ia;
+ bool was_evictable = true;
size_t left;
port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
return;
}
+ if (ctx->rule) {
+ /* Don't let the rule we're working on get evicted underneath us. */
+ was_evictable = ctx->rule->up.evictable;
+ ctx->rule->up.evictable = false;
+ }
OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
const struct ofp_action_dl_addr *oada;
const struct nx_action_resubmit *nar;
const struct nx_action_autopath *naa;
const struct nx_action_bundle *nab;
const struct nx_action_output_reg *naor;
+ const struct nx_action_controller *nac;
enum ofputil_action_code code;
ovs_be64 tun_id;
break;
case OFPUTIL_OFPAT_SET_NW_TOS:
- ctx->flow.nw_tos &= ~IP_DSCP_MASK;
- ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
+ /* OpenFlow 1.0 only supports IPv4. */
+ if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
+ ctx->flow.nw_tos &= ~IP_DSCP_MASK;
+ ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
+ }
break;
case OFPUTIL_OFPAT_SET_TP_SRC:
break;
case OFPUTIL_NXAST_POP_QUEUE:
- ctx->flow.priority = ctx->original_priority;
+ ctx->flow.skb_priority = ctx->orig_skb_priority;
break;
case OFPUTIL_NXAST_REG_MOVE:
}
break;
+ case OFPUTIL_NXAST_DEC_TTL:
+ if (compose_dec_ttl(ctx)) {
+ goto out;
+ }
+ break;
+
case OFPUTIL_NXAST_EXIT:
ctx->exit = true;
break;
+
+ case OFPUTIL_NXAST_FIN_TIMEOUT:
+ ctx->has_fin_timeout = true;
+ xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
+ break;
+
+ case OFPUTIL_NXAST_CONTROLLER:
+ nac = (const struct nx_action_controller *) ia;
+ execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
+ ntohs(nac->controller_id));
+ break;
}
}
+out:
/* We've let OFPP_NORMAL and the learning action look at the packet,
* so drop it now if forwarding is disabled. */
if (port && !stp_forward_in_state(port->stp_state)) {
ofpbuf_clear(ctx->odp_actions);
add_sflow_action(ctx);
}
+ if (ctx->rule) {
+ ctx->rule->up.evictable = was_evictable;
+ }
}
static void
action_xlate_ctx_init(struct action_xlate_ctx *ctx,
struct ofproto_dpif *ofproto, const struct flow *flow,
- const struct ofpbuf *packet)
+ ovs_be16 initial_tci, struct rule_dpif *rule,
+ uint8_t tcp_flags, const struct ofpbuf *packet)
{
ctx->ofproto = ofproto;
ctx->flow = *flow;
+ ctx->base_flow = ctx->flow;
+ ctx->base_flow.tun_id = 0;
+ ctx->base_flow.vlan_tci = initial_tci;
+ ctx->rule = rule;
ctx->packet = packet;
ctx->may_learn = packet != NULL;
+ ctx->tcp_flags = tcp_flags;
ctx->resubmit_hook = NULL;
}
xlate_actions(struct action_xlate_ctx *ctx,
const union ofp_action *in, size_t n_in)
{
+ struct flow orig_flow = ctx->flow;
+
COVERAGE_INC(ofproto_dpif_xlate);
ctx->odp_actions = ofpbuf_new(512);
ctx->may_set_up_flow = true;
ctx->has_learn = false;
ctx->has_normal = false;
+ ctx->has_fin_timeout = false;
ctx->nf_output_iface = NF_OUT_DROP;
+ ctx->mirrors = 0;
ctx->recurse = 0;
- ctx->original_priority = ctx->flow.priority;
- ctx->base_flow = ctx->flow;
- ctx->base_flow.tun_id = 0;
+ ctx->orig_skb_priority = ctx->flow.skb_priority;
ctx->table_id = 0;
ctx->exit = false;
case OFPC_FRAG_NX_MATCH:
/* Nothing to do. */
break;
+
+ case OFPC_INVALID_TTL_TO_CONTROLLER:
+ NOT_REACHED();
}
}
compose_output_action(ctx, OFPP_LOCAL);
}
}
+ add_mirror_actions(ctx, &orig_flow);
fix_sflow_action(ctx);
}
static bool
input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
{
+ /* Allow any VID on the OFPP_NONE port. */
+ if (in_bundle == &ofpp_none_bundle) {
+ return true;
+ }
+
switch (in_bundle->vlan_mode) {
case PORT_VLAN_ACCESS:
if (vid) {
ofbundle_get_a_port(const struct ofbundle *bundle)
{
return CONTAINER_OF(list_front(&bundle->ports),
- struct ofport_dpif, bundle_node);
-}
-
-static mirror_mask_t
-compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
- const struct ofbundle *in_bundle,
- const struct ofbundle *out_bundle)
-{
- mirror_mask_t dst_mirrors = 0;
-
- if (out_bundle == OFBUNDLE_FLOOD) {
- struct ofbundle *bundle;
-
- HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
- if (bundle != in_bundle
- && ofbundle_includes_vlan(bundle, vlan)
- && bundle->floodable
- && !bundle->mirror_out) {
- output_normal(ctx, bundle, vlan);
- dst_mirrors |= bundle->dst_mirrors;
- }
- }
- ctx->nf_output_iface = NF_OUT_FLOOD;
- } else if (out_bundle) {
- output_normal(ctx, out_bundle, vlan);
- dst_mirrors = out_bundle->dst_mirrors;
- }
-
- return dst_mirrors;
+ struct ofport_dpif, bundle_node);
}
static bool
}
static void
-output_mirrors(struct action_xlate_ctx *ctx,
- uint16_t vlan, const struct ofbundle *in_bundle,
- mirror_mask_t dst_mirrors)
+add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
{
struct ofproto_dpif *ofproto = ctx->ofproto;
mirror_mask_t mirrors;
+ struct ofbundle *in_bundle;
+ uint16_t vlan;
+ uint16_t vid;
+ const struct nlattr *a;
+ size_t left;
+
+ in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
+ ctx->packet != NULL);
+ if (!in_bundle) {
+ return;
+ }
+ mirrors = in_bundle->src_mirrors;
+
+ /* Drop frames on bundles reserved for mirroring. */
+ if (in_bundle->mirror_out) {
+ if (ctx->packet != NULL) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
+ "%s, which is reserved exclusively for mirroring",
+ ctx->ofproto->up.name, in_bundle->name);
+ }
+ return;
+ }
+
+ /* Check VLAN. */
+ vid = vlan_tci_to_vid(orig_flow->vlan_tci);
+ if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
+ return;
+ }
+ vlan = input_vid_to_vlan(in_bundle, vid);
+
+ /* Look at the output ports to check for destination selections. */
+
+ NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
+ ctx->odp_actions->size) {
+ enum ovs_action_attr type = nl_attr_type(a);
+ struct ofport_dpif *ofport;
+
+ if (type != OVS_ACTION_ATTR_OUTPUT) {
+ continue;
+ }
+
+ ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
+ if (ofport && ofport->bundle) {
+ mirrors |= ofport->bundle->dst_mirrors;
+ }
+ }
- mirrors = in_bundle->src_mirrors | dst_mirrors;
if (!mirrors) {
return;
}
+ /* Restore the original packet before adding the mirror actions. */
+ ctx->flow = *orig_flow;
+
while (mirrors) {
struct ofmirror *m;
}
mirrors &= ~m->dup_mirrors;
+ ctx->mirrors |= m->dup_mirrors;
if (m->out) {
output_normal(ctx, m->out, vlan);
- } else if (eth_dst_may_rspan(ctx->flow.dl_dst)
+ } else if (eth_dst_may_rspan(orig_flow->dl_dst)
&& vlan != m->out_vlan) {
struct ofbundle *bundle;
}
}
+static void
+update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
+ uint64_t packets, uint64_t bytes)
+{
+ if (!mirrors) {
+ return;
+ }
+
+ for (; mirrors; mirrors &= mirrors - 1) {
+ struct ofmirror *m;
+
+ m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
+
+ if (!m) {
+ /* In normal circumstances 'm' will not be NULL. However,
+ * if mirrors are reconfigured, we can temporarily get out
+ * of sync in facet_revalidate(). We could "correct" the
+ * mirror list before reaching here, but doing that would
+ * not properly account the traffic stats we've currently
+ * accumulated for previous mirror configuration. */
+ continue;
+ }
+
+ m->packet_count += packets;
+ m->byte_count += bytes;
+ }
+}
+
/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
* migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
* indicate this; newer upstream kernels use gratuitous ARP requests. */
{
struct mac_entry *mac;
+ /* Don't learn the OFPP_NONE port. */
+ if (in_bundle == &ofpp_none_bundle) {
+ return;
+ }
+
if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
return;
}
}
}
-static struct ofport_dpif *
+static struct ofbundle *
lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
{
struct ofport_dpif *ofport;
+ /* Special-case OFPP_NONE, which a controller may use as the ingress
+ * port for traffic that it is sourcing. */
+ if (in_port == OFPP_NONE) {
+ return &ofpp_none_bundle;
+ }
+
/* Find the port and bundle for the received packet. */
ofport = get_ofp_port(ofproto, in_port);
if (ofport && ofport->bundle) {
- return ofport;
+ return ofport->bundle;
}
/* Odd. A few possible reasons here:
static void
xlate_normal(struct action_xlate_ctx *ctx)
{
- mirror_mask_t dst_mirrors = 0;
struct ofport_dpif *in_port;
struct ofbundle *in_bundle;
- struct ofbundle *out_bundle;
struct mac_entry *mac;
uint16_t vlan;
uint16_t vid;
ctx->has_normal = true;
- /* Obtain in_port from ctx->flow.in_port.
- *
- * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
- in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
+ in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
ctx->packet != NULL);
- if (!in_port) {
+ if (!in_bundle) {
return;
}
- in_bundle = in_port->bundle;
+
+ /* We know 'in_port' exists unless it is "ofpp_none_bundle",
+ * since lookup_input_bundle() succeeded. */
+ in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
/* Drop malformed frames. */
if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
vlan = input_vid_to_vlan(in_bundle, vid);
/* Check other admissibility requirements. */
- if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
- output_mirrors(ctx, vlan, in_bundle, 0);
+ if (in_port &&
+ !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
return;
}
mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
&ctx->tags);
if (mac) {
- out_bundle = mac->port.p;
- } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
- /* If we are revalidating but don't have a learning entry then eject
- * the flow. Installing a flow that floods packets opens up a window
- * of time where we could learn from a packet reflected on a bond and
- * blackhole packets before the learning table is updated to reflect
- * the correct port. */
- ctx->may_set_up_flow = false;
- return;
+ if (mac->port.p != in_bundle) {
+ output_normal(ctx, mac->port.p, vlan);
+ }
} else {
- out_bundle = OFBUNDLE_FLOOD;
- }
+ struct ofbundle *bundle;
- /* Don't send packets out their input bundles. */
- if (in_bundle != out_bundle) {
- dst_mirrors = compose_dsts(ctx, vlan, in_bundle, out_bundle);
+ HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
+ if (bundle != in_bundle
+ && ofbundle_includes_vlan(bundle, vlan)
+ && bundle->floodable
+ && !bundle->mirror_out) {
+ output_normal(ctx, bundle, vlan);
+ }
+ }
+ ctx->nf_output_iface = NF_OUT_FLOOD;
}
- output_mirrors(ctx, vlan, in_bundle, dst_mirrors);
}
\f
/* Optimized flow revalidation.
table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
{
struct table_dpif *table = &ofproto->tables[table_id];
- const struct classifier *cls = &ofproto->up.tables[table_id];
+ const struct oftable *oftable = &ofproto->up.tables[table_id];
struct cls_table *catchall, *other;
struct cls_table *t;
catchall = other = NULL;
- switch (hmap_count(&cls->tables)) {
+ switch (hmap_count(&oftable->cls.tables)) {
case 0:
/* We could tag this OpenFlow table but it would make the logic a
* little harder and it's a corner case that doesn't seem worth it
case 1:
case 2:
- HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
+ HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
if (cls_table_is_catchall(t)) {
catchall = t;
} else if (!other) {
}
}
-static int
+static enum ofperr
packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
const struct flow *flow,
const union ofp_action *ofp_actions, size_t n_ofp_actions)
{
struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
- int error;
+ enum ofperr error;
if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
- return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
+ return OFPERR_NXBRC_BAD_IN_PORT;
}
error = validate_actions(ofp_actions, n_ofp_actions, flow,
ofproto->max_ports);
if (!error) {
struct odputil_keybuf keybuf;
- struct action_xlate_ctx ctx;
struct ofpbuf *odp_actions;
+ struct ofproto_push push;
struct ofpbuf key;
ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
odp_flow_key_from_flow(&key, flow);
- action_xlate_ctx_init(&ctx, ofproto, flow, packet);
- odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
+ action_xlate_ctx_init(&push.ctx, ofproto, flow, flow->vlan_tci, NULL,
+ packet_get_tcp_flags(packet, flow), packet);
+
+ /* Ensure that resubmits in 'ofp_actions' get accounted to their
+ * matching rules. */
+ push.packets = 1;
+ push.bytes = packet->size;
+ push.used = time_msec();
+ push.ctx.resubmit_hook = push_resubmit;
+
+ odp_actions = xlate_actions(&push.ctx, ofp_actions, n_ofp_actions);
dpif_execute(ofproto->dpif, key.data, key.size,
odp_actions->data, odp_actions->size, packet);
ofpbuf_delete(odp_actions);
}
return error;
}
+\f
+/* NetFlow. */
+
+static int
+set_netflow(struct ofproto *ofproto_,
+ const struct netflow_options *netflow_options)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+
+ if (netflow_options) {
+ if (!ofproto->netflow) {
+ ofproto->netflow = netflow_create();
+ }
+ return netflow_set_options(ofproto->netflow, netflow_options);
+ } else {
+ netflow_destroy(ofproto->netflow);
+ ofproto->netflow = NULL;
+ return 0;
+ }
+}
static void
get_netflow_ids(const struct ofproto *ofproto_,
dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
}
+
+static void
+send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
+{
+ if (!facet_is_controller_flow(facet) &&
+ netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
+ struct subfacet *subfacet;
+ struct ofexpired expired;
+
+ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
+ if (subfacet->installed) {
+ struct dpif_flow_stats stats;
+
+ subfacet_install(subfacet, subfacet->actions,
+ subfacet->actions_len, &stats);
+ subfacet_update_stats(subfacet, &stats);
+ }
+ }
+
+ expired.flow = facet->flow;
+ expired.packet_count = facet->packet_count;
+ expired.byte_count = facet->byte_count;
+ expired.used = facet->used;
+ netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
+ }
+}
+
+static void
+send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
+{
+ struct facet *facet;
+
+ HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
+ send_active_timeout(ofproto, facet);
+ }
+}
\f
static struct ofproto_dpif *
ofproto_dpif_lookup(const char *name)
{
- struct ofproto *ofproto = ofproto_lookup(name);
- return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
- ? ofproto_dpif_cast(ofproto)
- : NULL);
+ struct ofproto_dpif *ofproto;
+
+ HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
+ hash_string(name, 0), &all_ofproto_dpifs) {
+ if (!strcmp(ofproto->up.name, name)) {
+ return ofproto;
+ }
+ }
+ return NULL;
}
static void
-ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
- const char *args, void *aux OVS_UNUSED)
+ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
+ const char *argv[], void *aux OVS_UNUSED)
{
- const struct ofproto_dpif *ofproto;
+ struct ofproto_dpif *ofproto;
- ofproto = ofproto_dpif_lookup(args);
- if (!ofproto) {
- unixctl_command_reply(conn, 501, "no such bridge");
- return;
+ if (argc > 1) {
+ ofproto = ofproto_dpif_lookup(argv[1]);
+ if (!ofproto) {
+ unixctl_command_reply(conn, 501, "no such bridge");
+ return;
+ }
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ } else {
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
+ }
}
- mac_learning_flush(ofproto->ml);
unixctl_command_reply(conn, 200, "table successfully flushed");
}
static void
-ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
- const char *args, void *aux OVS_UNUSED)
+ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
+ const char *argv[], void *aux OVS_UNUSED)
{
struct ds ds = DS_EMPTY_INITIALIZER;
const struct ofproto_dpif *ofproto;
const struct mac_entry *e;
- ofproto = ofproto_dpif_lookup(args);
+ ofproto = ofproto_dpif_lookup(argv[1]);
if (!ofproto) {
unixctl_command_reply(conn, 501, "no such bridge");
return;
struct ofbundle *bundle = e->port.p;
ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
ofbundle_get_a_port(bundle)->odp_port,
- e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
+ e->vlan, ETH_ADDR_ARGS(e->mac),
+ mac_entry_age(ofproto->ml, e));
}
unixctl_command_reply(conn, 200, ds_cstr(&ds));
ds_destroy(&ds);
ds_put_char(result, '\n');
}
+static void
+trace_format_odp(struct ds *result, int level, const char *title,
+ struct ofproto_trace *trace)
+{
+ struct ofpbuf *odp_actions = trace->ctx.odp_actions;
+
+ ds_put_char_multiple(result, '\t', level);
+ ds_put_format(result, "%s: ", title);
+ format_odp_actions(result, odp_actions->data, odp_actions->size);
+ ds_put_char(result, '\n');
+}
+
static void
trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
{
ds_put_char(result, '\n');
trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
+ trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
}
static void
-ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
+ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
void *aux OVS_UNUSED)
{
- char *dpname, *arg1, *arg2, *arg3, *arg4;
- char *args = xstrdup(args_);
- char *save_ptr = NULL;
+ const char *dpname = argv[1];
struct ofproto_dpif *ofproto;
struct ofpbuf odp_key;
struct ofpbuf *packet;
struct rule_dpif *rule;
+ ovs_be16 initial_tci;
struct ds result;
struct flow flow;
char *s;
ofpbuf_init(&odp_key, 0);
ds_init(&result);
- dpname = strtok_r(args, " ", &save_ptr);
- arg1 = strtok_r(NULL, " ", &save_ptr);
- arg2 = strtok_r(NULL, " ", &save_ptr);
- arg3 = strtok_r(NULL, " ", &save_ptr);
- arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
- if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
+ ofproto = ofproto_dpif_lookup(dpname);
+ if (!ofproto) {
+ unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
+ "for help)");
+ goto exit;
+ }
+ if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
/* ofproto/trace dpname flow [-generate] */
+ const char *flow_s = argv[2];
+ const char *generate_s = argv[3];
int error;
/* Convert string to datapath key. */
ofpbuf_init(&odp_key, 0);
- error = odp_flow_key_from_string(arg1, NULL, &odp_key);
+ error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
if (error) {
unixctl_command_reply(conn, 501, "Bad flow syntax");
goto exit;
}
/* Convert odp_key to flow. */
- error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
- if (error) {
+ error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
+ odp_key.size, &flow,
+ &initial_tci, NULL);
+ if (error == ODP_FIT_ERROR) {
unixctl_command_reply(conn, 501, "Invalid flow");
goto exit;
}
/* Generate a packet, if requested. */
- if (arg2) {
+ if (generate_s) {
packet = ofpbuf_new(0);
flow_compose(packet, &flow);
}
- } else if (dpname && arg1 && arg2 && arg3 && arg4) {
+ } else if (argc == 6) {
/* ofproto/trace dpname priority tun_id in_port packet */
- uint16_t in_port;
- ovs_be64 tun_id;
- uint32_t priority;
-
- priority = atoi(arg1);
- tun_id = htonll(strtoull(arg2, NULL, 0));
- in_port = ofp_port_to_odp_port(atoi(arg3));
-
- packet = ofpbuf_new(strlen(args) / 2);
- arg4 = ofpbuf_put_hex(packet, arg4, NULL);
- arg4 += strspn(arg4, " ");
- if (*arg4 != '\0') {
- unixctl_command_reply(conn, 501, "Trailing garbage in command");
- goto exit;
- }
- if (packet->size < ETH_HEADER_LEN) {
- unixctl_command_reply(conn, 501,
- "Packet data too short for Ethernet");
+ const char *priority_s = argv[2];
+ const char *tun_id_s = argv[3];
+ const char *in_port_s = argv[4];
+ const char *packet_s = argv[5];
+ uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
+ ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
+ uint32_t priority = atoi(priority_s);
+ const char *msg;
+
+ msg = eth_from_hex(packet_s, &packet);
+ if (msg) {
+ unixctl_command_reply(conn, 501, msg);
goto exit;
}
ds_put_cstr(&result, "Packet: ");
- s = ofp_packet_to_string(packet->data, packet->size, packet->size);
+ s = ofp_packet_to_string(packet->data, packet->size);
ds_put_cstr(&result, s);
free(s);
flow_extract(packet, priority, tun_id, in_port, &flow);
+ initial_tci = flow.vlan_tci;
} else {
unixctl_command_reply(conn, 501, "Bad command syntax");
goto exit;
}
- ofproto = ofproto_dpif_lookup(dpname);
- if (!ofproto) {
- unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
- "for help)");
- goto exit;
- }
-
ds_put_cstr(&result, "Flow: ");
flow_format(&result, &flow);
ds_put_char(&result, '\n');
if (rule) {
struct ofproto_trace trace;
struct ofpbuf *odp_actions;
+ uint8_t tcp_flags;
+ tcp_flags = packet ? packet_get_tcp_flags(packet, &flow) : 0;
trace.result = &result;
trace.flow = flow;
- action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
+ action_xlate_ctx_init(&trace.ctx, ofproto, &flow, initial_tci,
+ rule, tcp_flags, packet);
trace.ctx.resubmit_hook = trace_resubmit;
odp_actions = xlate_actions(&trace.ctx,
rule->up.actions, rule->up.n_actions);
ds_destroy(&result);
ofpbuf_delete(packet);
ofpbuf_uninit(&odp_key);
- free(args);
}
static void
-ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
- const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
+ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
clogged = true;
unixctl_command_reply(conn, 200, NULL);
}
static void
-ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
- const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
+ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
+ const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
{
clogged = false;
unixctl_command_reply(conn, 200, NULL);
}
+/* Runs a self-check of flow translations in 'ofproto'. Appends a message to
+ * 'reply' describing the results. */
+static void
+ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
+{
+ struct facet *facet;
+ int errors;
+
+ errors = 0;
+ HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
+ if (!facet_check_consistency(facet)) {
+ errors++;
+ }
+ }
+ if (errors) {
+ ofproto->need_revalidate = true;
+ }
+
+ if (errors) {
+ ds_put_format(reply, "%s: self-check failed (%d errors)\n",
+ ofproto->up.name, errors);
+ } else {
+ ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
+ }
+}
+
+static void
+ofproto_dpif_self_check(struct unixctl_conn *conn,
+ int argc, const char *argv[], void *aux OVS_UNUSED)
+{
+ struct ds reply = DS_EMPTY_INITIALIZER;
+ struct ofproto_dpif *ofproto;
+
+ if (argc > 1) {
+ ofproto = ofproto_dpif_lookup(argv[1]);
+ if (!ofproto) {
+ unixctl_command_reply(conn, 501, "Unknown ofproto (use "
+ "ofproto/list for help)");
+ return;
+ }
+ ofproto_dpif_self_check__(ofproto, &reply);
+ } else {
+ HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
+ ofproto_dpif_self_check__(ofproto, &reply);
+ }
+ }
+
+ unixctl_command_reply(conn, 200, ds_cstr(&reply));
+ ds_destroy(&reply);
+}
+
static void
ofproto_dpif_unixctl_init(void)
{
}
registered = true;
- unixctl_command_register("ofproto/trace",
- "bridge {tun_id in_port packet | odp_flow [-generate]}",
- ofproto_unixctl_trace, NULL);
- unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
- NULL);
- unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
- NULL);
- unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
- unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
+ unixctl_command_register(
+ "ofproto/trace",
+ "bridge {tun_id in_port packet | odp_flow [-generate]}",
+ 2, 5, ofproto_unixctl_trace, NULL);
+ unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
+ ofproto_unixctl_fdb_flush, NULL);
+ unixctl_command_register("fdb/show", "bridge", 1, 1,
+ ofproto_unixctl_fdb_show, NULL);
+ unixctl_command_register("ofproto/clog", "", 0, 0,
+ ofproto_dpif_clog, NULL);
+ unixctl_command_register("ofproto/unclog", "", 0, 0,
+ ofproto_dpif_unclog, NULL);
+ unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
+ ofproto_dpif_self_check, NULL);
+}
+\f
+/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
+ *
+ * This is deprecated. It is only for compatibility with broken device drivers
+ * in old versions of Linux that do not properly support VLANs when VLAN
+ * devices are not used. When broken device drivers are no longer in
+ * widespread use, we will delete these interfaces. */
+
+static int
+set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
+ struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+
+ if (realdev_ofp_port == ofport->realdev_ofp_port
+ && vid == ofport->vlandev_vid) {
+ return 0;
+ }
+
+ ofproto->need_revalidate = true;
+
+ if (ofport->realdev_ofp_port) {
+ vsp_remove(ofport);
+ }
+ if (realdev_ofp_port && ofport->bundle) {
+ /* vlandevs are enslaved to their realdevs, so they are not allowed to
+ * themselves be part of a bundle. */
+ bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
+ }
+
+ ofport->realdev_ofp_port = realdev_ofp_port;
+ ofport->vlandev_vid = vid;
+
+ if (realdev_ofp_port) {
+ vsp_add(ofport, realdev_ofp_port, vid);
+ }
+
+ return 0;
+}
+
+static uint32_t
+hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
+{
+ return hash_2words(realdev_ofp_port, vid);
+}
+
+static uint32_t
+vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
+ uint32_t realdev_odp_port, ovs_be16 vlan_tci)
+{
+ if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
+ uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
+ int vid = vlan_tci_to_vid(vlan_tci);
+ const struct vlan_splinter *vsp;
+
+ HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
+ hash_realdev_vid(realdev_ofp_port, vid),
+ &ofproto->realdev_vid_map) {
+ if (vsp->realdev_ofp_port == realdev_ofp_port
+ && vsp->vid == vid) {
+ return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
+ }
+ }
+ }
+ return realdev_odp_port;
+}
+
+static struct vlan_splinter *
+vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
+{
+ struct vlan_splinter *vsp;
+
+ HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
+ &ofproto->vlandev_map) {
+ if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
+ return vsp;
+ }
+ }
+
+ return NULL;
+}
+
+static uint16_t
+vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
+ uint16_t vlandev_ofp_port, int *vid)
+{
+ if (!hmap_is_empty(&ofproto->vlandev_map)) {
+ const struct vlan_splinter *vsp;
+
+ vsp = vlandev_find(ofproto, vlandev_ofp_port);
+ if (vsp) {
+ if (vid) {
+ *vid = vsp->vid;
+ }
+ return vsp->realdev_ofp_port;
+ }
+ }
+ return 0;
+}
+
+static void
+vsp_remove(struct ofport_dpif *port)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+ struct vlan_splinter *vsp;
+
+ vsp = vlandev_find(ofproto, port->up.ofp_port);
+ if (vsp) {
+ hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
+ hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
+ free(vsp);
+
+ port->realdev_ofp_port = 0;
+ } else {
+ VLOG_ERR("missing vlan device record");
+ }
+}
+
+static void
+vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
+{
+ struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+
+ if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
+ && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
+ == realdev_ofp_port)) {
+ struct vlan_splinter *vsp;
+
+ vsp = xmalloc(sizeof *vsp);
+ hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
+ hash_int(port->up.ofp_port, 0));
+ hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
+ hash_realdev_vid(realdev_ofp_port, vid));
+ vsp->realdev_ofp_port = realdev_ofp_port;
+ vsp->vlandev_ofp_port = port->up.ofp_port;
+ vsp->vid = vid;
+
+ port->realdev_ofp_port = realdev_ofp_port;
+ } else {
+ VLOG_ERR("duplicate vlan device record");
+ }
}
\f
const struct ofproto_class ofproto_dpif_class = {
destruct,
dealloc,
run,
+ run_fast,
wait,
flush,
get_features,
port_query_by_name,
port_add,
port_del,
+ port_get_stats,
port_dump_start,
port_dump_next,
port_dump_done,
get_stp_status,
set_stp_port,
get_stp_port_status,
+ set_queues,
bundle_set,
bundle_remove,
mirror_set,
+ mirror_get_stats,
set_flood_vlans,
is_mirror_output_bundle,
forward_bpdu_changed,
+ set_mac_idle_time,
+ set_realdev,
};
git://git.onelab.eu / sliver-openvswitch.git / blobdiff