/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <config.h>
#include "dpif-provider.h"
-#include <assert.h>
#include <ctype.h>
#include <errno.h>
#include <inttypes.h>
#include "flow.h"
#include "netdev.h"
#include "netlink.h"
+#include "odp-execute.h"
#include "odp-util.h"
+#include "ofp-errors.h"
#include "ofp-print.h"
#include "ofp-util.h"
#include "ofpbuf.h"
#include "packets.h"
#include "poll-loop.h"
#include "shash.h"
-#include "svec.h"
+#include "sset.h"
+#include "timeval.h"
#include "util.h"
#include "valgrind.h"
#include "vlog.h"
COVERAGE_DEFINE(dpif_flow_get);
COVERAGE_DEFINE(dpif_flow_put);
COVERAGE_DEFINE(dpif_flow_del);
-COVERAGE_DEFINE(dpif_flow_query_list);
-COVERAGE_DEFINE(dpif_flow_query_list_n);
COVERAGE_DEFINE(dpif_execute);
COVERAGE_DEFINE(dpif_purge);
+COVERAGE_DEFINE(dpif_execute_with_help);
static const struct dpif_class *base_dpif_classes[] = {
-#ifdef HAVE_NETLINK
+#ifdef __linux__
&dpif_linux_class,
#endif
&dpif_netdev_class,
int refcount;
};
static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes);
+static struct sset dpif_blacklist = SSET_INITIALIZER(&dpif_blacklist);
+
+/* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
+static struct ovs_mutex dpif_mutex = OVS_MUTEX_INITIALIZER;
/* Rate limit for individual messages going to or from the datapath, output at
* DBG level. This is very high because, if these are enabled, it is because
/* Not really much point in logging many dpif errors. */
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
+static void log_flow_message(const struct dpif *dpif, int error,
+ const char *operation,
+ const struct nlattr *key, size_t key_len,
+ const struct nlattr *mask, size_t mask_len,
+ const struct dpif_flow_stats *stats,
+ const struct nlattr *actions, size_t actions_len);
static void log_operation(const struct dpif *, const char *operation,
int error);
-static void log_flow_operation(const struct dpif *, const char *operation,
- int error, struct odp_flow *flow);
-static void log_flow_put(struct dpif *, int error,
- const struct odp_flow_put *);
static bool should_log_flow_message(int error);
-static void check_rw_flow_actions(struct odp_flow *);
-static void check_rw_flow_key(struct odp_flow *);
+static void log_flow_put_message(struct dpif *, const struct dpif_flow_put *,
+ int error);
+static void log_flow_del_message(struct dpif *, const struct dpif_flow_del *,
+ int error);
+static void log_execute_message(struct dpif *, const struct dpif_execute *,
+ int error);
static void
dp_initialize(void)
{
- static int status = -1;
+ static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
- if (status < 0) {
+ if (ovsthread_once_start(&once)) {
int i;
- status = 0;
for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
dp_register_provider(base_dpif_classes[i]);
}
+ ovsthread_once_done(&once);
}
}
-/* Performs periodic work needed by all the various kinds of dpifs.
- *
- * If your program opens any dpifs, it must call both this function and
- * netdev_run() within its main poll loop. */
-void
-dp_run(void)
+static int
+dp_register_provider__(const struct dpif_class *new_class)
{
- struct shash_node *node;
- SHASH_FOR_EACH(node, &dpif_classes) {
- const struct registered_dpif_class *registered_class = node->data;
- if (registered_class->dpif_class->run) {
- registered_class->dpif_class->run();
- }
- }
-}
+ struct registered_dpif_class *registered_class;
-/* Arranges for poll_block() to wake up when dp_run() needs to be called.
- *
- * If your program opens any dpifs, it must call both this function and
- * netdev_wait() within its main poll loop. */
-void
-dp_wait(void)
-{
- struct shash_node *node;
- SHASH_FOR_EACH(node, &dpif_classes) {
- const struct registered_dpif_class *registered_class = node->data;
- if (registered_class->dpif_class->wait) {
- registered_class->dpif_class->wait();
- }
+ if (sset_contains(&dpif_blacklist, new_class->type)) {
+ VLOG_DBG("attempted to register blacklisted provider: %s",
+ new_class->type);
+ return EINVAL;
}
-}
-
-/* Registers a new datapath provider. After successful registration, new
- * datapaths of that type can be opened using dpif_open(). */
-int
-dp_register_provider(const struct dpif_class *new_class)
-{
- struct registered_dpif_class *registered_class;
if (shash_find(&dpif_classes, new_class->type)) {
VLOG_WARN("attempted to register duplicate datapath provider: %s",
return 0;
}
+/* Registers a new datapath provider. After successful registration, new
+ * datapaths of that type can be opened using dpif_open(). */
+int
+dp_register_provider(const struct dpif_class *new_class)
+{
+ int error;
+
+ ovs_mutex_lock(&dpif_mutex);
+ error = dp_register_provider__(new_class);
+ ovs_mutex_unlock(&dpif_mutex);
+
+ return error;
+}
+
/* Unregisters a datapath provider. 'type' must have been previously
* registered and not currently be in use by any dpifs. After unregistration
* new datapaths of that type cannot be opened using dpif_open(). */
-int
-dp_unregister_provider(const char *type)
+static int
+dp_unregister_provider__(const char *type)
{
struct shash_node *node;
struct registered_dpif_class *registered_class;
return 0;
}
+/* Unregisters a datapath provider. 'type' must have been previously
+ * registered and not currently be in use by any dpifs. After unregistration
+ * new datapaths of that type cannot be opened using dpif_open(). */
+int
+dp_unregister_provider(const char *type)
+{
+ int error;
+
+ dp_initialize();
+
+ ovs_mutex_lock(&dpif_mutex);
+ error = dp_unregister_provider__(type);
+ ovs_mutex_unlock(&dpif_mutex);
+
+ return error;
+}
+
+/* Blacklists a provider. Causes future calls of dp_register_provider() with
+ * a dpif_class which implements 'type' to fail. */
+void
+dp_blacklist_provider(const char *type)
+{
+ ovs_mutex_lock(&dpif_mutex);
+ sset_add(&dpif_blacklist, type);
+ ovs_mutex_unlock(&dpif_mutex);
+}
+
/* Clears 'types' and enumerates the types of all currently registered datapath
- * providers into it. The caller must first initialize the svec. */
+ * providers into it. The caller must first initialize the sset. */
void
-dp_enumerate_types(struct svec *types)
+dp_enumerate_types(struct sset *types)
{
struct shash_node *node;
dp_initialize();
- svec_clear(types);
+ sset_clear(types);
+ ovs_mutex_lock(&dpif_mutex);
SHASH_FOR_EACH(node, &dpif_classes) {
const struct registered_dpif_class *registered_class = node->data;
- svec_add(types, registered_class->dpif_class->type);
+ sset_add(types, registered_class->dpif_class->type);
+ }
+ ovs_mutex_unlock(&dpif_mutex);
+}
+
+static void
+dp_class_unref(struct registered_dpif_class *rc)
+{
+ ovs_mutex_lock(&dpif_mutex);
+ ovs_assert(rc->refcount);
+ rc->refcount--;
+ ovs_mutex_unlock(&dpif_mutex);
+}
+
+static struct registered_dpif_class *
+dp_class_lookup(const char *type)
+{
+ struct registered_dpif_class *rc;
+
+ ovs_mutex_lock(&dpif_mutex);
+ rc = shash_find_data(&dpif_classes, type);
+ if (rc) {
+ rc->refcount++;
}
+ ovs_mutex_unlock(&dpif_mutex);
+
+ return rc;
}
/* Clears 'names' and enumerates the names of all known created datapaths with
- * the given 'type'. The caller must first initialize the svec. Returns 0 if
+ * the given 'type'. The caller must first initialize the sset. Returns 0 if
* successful, otherwise a positive errno value.
*
* Some kinds of datapaths might not be practically enumerable. This is not
* considered an error. */
int
-dp_enumerate_names(const char *type, struct svec *names)
+dp_enumerate_names(const char *type, struct sset *names)
{
- const struct registered_dpif_class *registered_class;
+ struct registered_dpif_class *registered_class;
const struct dpif_class *dpif_class;
int error;
dp_initialize();
- svec_clear(names);
+ sset_clear(names);
- registered_class = shash_find_data(&dpif_classes, type);
+ registered_class = dp_class_lookup(type);
if (!registered_class) {
VLOG_WARN("could not enumerate unknown type: %s", type);
return EAFNOSUPPORT;
dpif_class = registered_class->dpif_class;
error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
-
if (error) {
VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
- strerror(error));
+ ovs_strerror(error));
}
+ dp_class_unref(registered_class);
return error;
}
-/* Parses 'datapath name', which is of the form type@name into its
- * component pieces. 'name' and 'type' must be freed by the caller. */
+/* Parses 'datapath_name_', which is of the form [type@]name into its
+ * component pieces. 'name' and 'type' must be freed by the caller.
+ *
+ * The returned 'type' is normalized, as if by dpif_normalize_type(). */
void
dp_parse_name(const char *datapath_name_, char **name, char **type)
{
if (separator) {
*separator = '\0';
*type = datapath_name;
- *name = xstrdup(separator + 1);
+ *name = xstrdup(dpif_normalize_type(separator + 1));
} else {
*name = datapath_name;
- *type = NULL;
+ *type = xstrdup(dpif_normalize_type(NULL));
}
}
dp_initialize();
- if (!type || *type == '\0') {
- type = "system";
- }
-
- registered_class = shash_find_data(&dpif_classes, type);
+ type = dpif_normalize_type(type);
+ registered_class = dp_class_lookup(type);
if (!registered_class) {
VLOG_WARN("could not create datapath %s of unknown type %s", name,
type);
error = registered_class->dpif_class->open(registered_class->dpif_class,
name, create, &dpif);
if (!error) {
- assert(dpif->dpif_class == registered_class->dpif_class);
- registered_class->refcount++;
+ ovs_assert(dpif->dpif_class == registered_class->dpif_class);
+ } else {
+ dp_class_unref(registered_class);
}
exit:
error = dpif_open(name, type, dpifp);
if (error) {
VLOG_WARN("datapath %s already exists but cannot be opened: %s",
- name, strerror(error));
+ name, ovs_strerror(error));
}
} else if (error) {
- VLOG_WARN("failed to create datapath %s: %s", name, strerror(error));
+ VLOG_WARN("failed to create datapath %s: %s",
+ name, ovs_strerror(error));
}
return error;
}
dpif_close(struct dpif *dpif)
{
if (dpif) {
- struct registered_dpif_class *registered_class;
-
- registered_class = shash_find_data(&dpif_classes,
- dpif->dpif_class->type);
- assert(registered_class);
- assert(registered_class->refcount);
+ struct registered_dpif_class *rc;
- registered_class->refcount--;
+ rc = shash_find_data(&dpif_classes, dpif->dpif_class->type);
dpif_uninit(dpif, true);
+ dp_class_unref(rc);
+ }
+}
+
+/* Performs periodic work needed by 'dpif'. */
+void
+dpif_run(struct dpif *dpif)
+{
+ if (dpif->dpif_class->run) {
+ dpif->dpif_class->run(dpif);
+ }
+}
+
+/* Arranges for poll_block() to wake up when dp_run() needs to be called for
+ * 'dpif'. */
+void
+dpif_wait(struct dpif *dpif)
+{
+ if (dpif->dpif_class->wait) {
+ dpif->dpif_class->wait(dpif);
}
}
return dpif->base_name;
}
-/* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
- * Linux datapath, for example, supports opening a datapath both by number,
- * e.g. "dp0", and by the name of the datapath's local port. For some
- * datapaths, this might be an infinite set (e.g. in a file name, slashes may
- * be duplicated any number of times), in which case only the names most likely
- * to be used will be enumerated.
- *
- * The caller must already have initialized 'all_names'. Any existing names in
- * 'all_names' will not be disturbed. */
-int
-dpif_get_all_names(const struct dpif *dpif, struct svec *all_names)
+/* Returns the type of datapath 'dpif'. */
+const char *
+dpif_type(const struct dpif *dpif)
{
- if (dpif->dpif_class->get_all_names) {
- int error = dpif->dpif_class->get_all_names(dpif, all_names);
- if (error) {
- VLOG_WARN_RL(&error_rl,
- "failed to retrieve names for datpath %s: %s",
- dpif_name(dpif), strerror(error));
- }
- return error;
- } else {
- svec_add(all_names, dpif_base_name(dpif));
- return 0;
- }
+ return dpif->dpif_class->type;
}
+/* Returns the fully spelled out name for the given datapath 'type'.
+ *
+ * Normalized type string can be compared with strcmp(). Unnormalized type
+ * string might be the same even if they have different spellings. */
+const char *
+dpif_normalize_type(const char *type)
+{
+ return type && type[0] ? type : "system";
+}
/* Destroys the datapath that 'dpif' is connected to, first removing all of its
* ports. After calling this function, it does not make sense to pass 'dpif'
/* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
* otherwise a positive errno value. */
int
-dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
+dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
{
int error = dpif->dpif_class->get_stats(dpif, stats);
if (error) {
return error;
}
-/* Retrieves the current IP fragment handling policy for 'dpif' into
- * '*drop_frags': true indicates that fragments are dropped, false indicates
- * that fragments are treated in the same way as other IP packets (except that
- * the L4 header cannot be read). Returns 0 if successful, otherwise a
- * positive errno value. */
-int
-dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
+const char *
+dpif_port_open_type(const char *datapath_type, const char *port_type)
{
- int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags);
- if (error) {
- *drop_frags = false;
+ struct registered_dpif_class *rc;
+
+ datapath_type = dpif_normalize_type(datapath_type);
+
+ ovs_mutex_lock(&dpif_mutex);
+ rc = shash_find_data(&dpif_classes, datapath_type);
+ if (rc && rc->dpif_class->port_open_type) {
+ port_type = rc->dpif_class->port_open_type(rc->dpif_class, port_type);
}
- log_operation(dpif, "get_drop_frags", error);
- return error;
-}
+ ovs_mutex_unlock(&dpif_mutex);
-/* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
- * the same as for the get_drop_frags member function. Returns 0 if
- * successful, otherwise a positive errno value. */
-int
-dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
-{
- int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags);
- log_operation(dpif, "set_drop_frags", error);
- return error;
+ return port_type;
}
-/* Attempts to add 'netdev' as a port on 'dpif'. If successful, returns 0 and
- * sets '*port_nop' to the new port's port number (if 'port_nop' is non-null).
- * On failure, returns a positive errno value and sets '*port_nop' to
- * UINT16_MAX (if 'port_nop' is non-null). */
+/* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
+ * non-null and its value is not ODPP_NONE, then attempts to use the
+ * value as the port number.
+ *
+ * If successful, returns 0 and sets '*port_nop' to the new port's port
+ * number (if 'port_nop' is non-null). On failure, returns a positive
+ * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
+ * non-null). */
int
-dpif_port_add(struct dpif *dpif, struct netdev *netdev, uint16_t *port_nop)
+dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop)
{
const char *netdev_name = netdev_get_name(netdev);
- uint16_t port_no;
+ odp_port_t port_no = ODPP_NONE;
int error;
COVERAGE_INC(dpif_port_add);
+ if (port_nop) {
+ port_no = *port_nop;
+ }
+
error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
if (!error) {
- VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
+ VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32,
dpif_name(dpif), netdev_name, port_no);
} else {
VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
- dpif_name(dpif), netdev_name, strerror(error));
- port_no = UINT16_MAX;
+ dpif_name(dpif), netdev_name, ovs_strerror(error));
+ port_no = ODPP_NONE;
}
if (port_nop) {
*port_nop = port_no;
/* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
* otherwise a positive errno value. */
int
-dpif_port_del(struct dpif *dpif, uint16_t port_no)
+dpif_port_del(struct dpif *dpif, odp_port_t port_no)
{
int error;
error = dpif->dpif_class->port_del(dpif, port_no);
if (!error) {
- VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu16")",
+ VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu32")",
dpif_name(dpif), port_no);
} else {
log_operation(dpif, "port_del", error);
free(dpif_port->type);
}
+/* Checks if port named 'devname' exists in 'dpif'. If so, returns
+ * true; otherwise, returns false. */
+bool
+dpif_port_exists(const struct dpif *dpif, const char *devname)
+{
+ int error = dpif->dpif_class->port_query_by_name(dpif, devname, NULL);
+ if (error != 0 && error != ENOENT && error != ENODEV) {
+ VLOG_WARN_RL(&error_rl, "%s: failed to query port %s: %s",
+ dpif_name(dpif), devname, ovs_strerror(error));
+ }
+
+ return !error;
+}
+
/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
* initializes '*port' appropriately; on failure, returns a positive errno
* value.
* The caller owns the data in 'port' and must free it with
* dpif_port_destroy() when it is no longer needed. */
int
-dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
+dpif_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
struct dpif_port *port)
{
int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
if (!error) {
- VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
+ VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu32" is device %s",
dpif_name(dpif), port_no, port->name);
} else {
memset(port, 0, sizeof *port);
- VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
- dpif_name(dpif), port_no, strerror(error));
+ VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu32": %s",
+ dpif_name(dpif), port_no, ovs_strerror(error));
}
return error;
}
{
int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
if (!error) {
- VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
+ VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu32,
dpif_name(dpif), devname, port->port_no);
} else {
memset(port, 0, sizeof *port);
- /* Log level is DBG here because all the current callers are interested
- * in whether 'dpif' actually has a port 'devname', so that it's not an
- * issue worth logging if it doesn't. */
- VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
- dpif_name(dpif), devname, strerror(error));
+ /* For ENOENT or ENODEV we use DBG level because the caller is probably
+ * interested in whether 'dpif' actually has a port 'devname', so that
+ * it's not an issue worth logging if it doesn't. Other errors are
+ * uncommon and more likely to indicate a real problem. */
+ VLOG_RL(&error_rl,
+ error == ENOENT || error == ENODEV ? VLL_DBG : VLL_WARN,
+ "%s: failed to query port %s: %s",
+ dpif_name(dpif), devname, ovs_strerror(error));
}
return error;
}
-/* Returns one greater than the maximum port number accepted in flow
- * actions. */
-int
-dpif_get_max_ports(const struct dpif *dpif)
+/* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
+ * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
+ * flows whose packets arrived on port 'port_no'. In the case where the
+ * provider allocates multiple Netlink PIDs to a single port, it may use
+ * 'hash' to spread load among them. The caller need not use a particular
+ * hash function; a 5-tuple hash is suitable.
+ *
+ * (The datapath implementation might use some different hash function for
+ * distributing packets received via flow misses among PIDs. This means
+ * that packets received via flow misses might be reordered relative to
+ * packets received via userspace actions. This is not ordinarily a
+ * problem.)
+ *
+ * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
+ * allocated to any port, that the client may use for special purposes.
+ *
+ * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
+ * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
+ * disabled and then re-enabled, so a client that does that must be prepared to
+ * update all of the flows that it installed that contain
+ * OVS_ACTION_ATTR_USERSPACE actions. */
+uint32_t
+dpif_port_get_pid(const struct dpif *dpif, odp_port_t port_no, uint32_t hash)
{
- return dpif->dpif_class->get_max_ports(dpif);
+ return (dpif->dpif_class->port_get_pid
+ ? (dpif->dpif_class->port_get_pid)(dpif, port_no, hash)
+ : 0);
}
/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
* result is null-terminated. On failure, returns a positive errno value and
* makes 'name' the empty string. */
int
-dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
+dpif_port_get_name(struct dpif *dpif, odp_port_t port_no,
char *name, size_t name_size)
{
struct dpif_port port;
int error;
- assert(name_size > 0);
+ ovs_assert(name_size > 0);
error = dpif_port_query_by_number(dpif, port_no, &port);
if (!error) {
dpif->dpif_class->port_poll_wait(dpif);
}
+/* Extracts the flow stats for a packet. The 'flow' and 'packet'
+ * arguments must have been initialized through a call to flow_extract().
+ * 'used' is stored into stats->used. */
+void
+dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet,
+ long long int used, struct dpif_flow_stats *stats)
+{
+ stats->tcp_flags = ntohs(flow->tcp_flags);
+ stats->n_bytes = ofpbuf_size(packet);
+ stats->n_packets = 1;
+ stats->used = used;
+}
+
+/* Appends a human-readable representation of 'stats' to 's'. */
+void
+dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s)
+{
+ ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:",
+ stats->n_packets, stats->n_bytes);
+ if (stats->used) {
+ ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0);
+ } else {
+ ds_put_format(s, "never");
+ }
+ if (stats->tcp_flags) {
+ ds_put_cstr(s, ", flags:");
+ packet_format_tcp_flags(s, stats->tcp_flags);
+ }
+}
+
/* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
* positive errno value. */
int
return error;
}
-/* Queries 'dpif' for a flow entry matching 'flow->key'.
+/* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
+ * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
+ * 'key'.
*
- * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
- * flow into 'flow->stats'. If 'flow->actions_len' is zero, then
- * 'flow->actions' is ignored. If 'flow->actions_len' is nonzero, then
- * 'flow->actions' should point to an array of the specified number of bytes.
- * At most that many bytes of the flow's actions will be copied into that
- * array. 'flow->actions_len' will be updated to the number of bytes of
- * actions actually present in the flow, which may be greater than the amount
- * stored if the flow has more actions than space available in the array.
+ * Returns 0 if successful. If no flow matches, returns ENOENT. On other
+ * failure, returns a positive errno value.
*
- * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
- * failure, returns a positive errno value. */
+ * If 'actionsp' is nonnull, then on success '*actionsp' will be set to an
+ * ofpbuf owned by the caller that contains the Netlink attributes for the
+ * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete()) when
+ * it is no longer needed.
+ *
+ * If 'stats' is nonnull, then on success it will be updated with the flow's
+ * statistics. */
int
-dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
+dpif_flow_get(const struct dpif *dpif,
+ const struct nlattr *key, size_t key_len,
+ struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
{
int error;
COVERAGE_INC(dpif_flow_get);
- check_rw_flow_actions(flow);
- error = dpif->dpif_class->flow_get(dpif, flow, 1);
- if (!error) {
- error = flow->stats.error;
- }
+ error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats);
if (error) {
- /* Make the results predictable on error. */
- memset(&flow->stats, 0, sizeof flow->stats);
- flow->actions_len = 0;
+ if (actionsp) {
+ *actionsp = NULL;
+ }
+ if (stats) {
+ memset(stats, 0, sizeof *stats);
+ }
}
if (should_log_flow_message(error)) {
- log_flow_operation(dpif, "flow_get", error, flow);
+ const struct nlattr *actions;
+ size_t actions_len;
+
+ if (!error && actionsp) {
+ actions = ofpbuf_data(*actionsp);
+ actions_len = ofpbuf_size(*actionsp);
+ } else {
+ actions = NULL;
+ actions_len = 0;
+ }
+ log_flow_message(dpif, error, "flow_get", key, key_len,
+ NULL, 0, stats, actions, actions_len);
}
return error;
}
-/* For each flow 'flow' in the 'n' flows in 'flows':
- *
- * - If a flow matching 'flow->key' exists in 'dpif':
- *
- * Stores 0 into 'flow->stats.error' and stores statistics for the flow
- * into 'flow->stats'.
- *
- * If 'flow->actions_len' is zero, then 'flow->actions' is ignored. If
- * 'flow->actions_len' is nonzero, then 'flow->actions' should point to an
- * array of the specified number of bytes. At most that amount of flow's
- * actions will be copied into that array. 'flow->actions_len' will be
- * updated to the number of bytes of actions actually present in the flow,
- * which may be greater than the amount stored if the flow's actions are
- * longer than the available space.
- *
- * - Flow-specific errors are indicated by a positive errno value in
- * 'flow->stats.error'. In particular, ENOENT indicates that no flow
- * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
- * contents of 'flow->key' are preserved but other members of 'flow' should
- * be treated as indeterminate.
- *
- * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
- * individually successful or not is indicated by 'flow->stats.error',
- * however). Returns a positive errno value if an error that prevented this
- * update occurred, in which the caller must not depend on any elements in
- * 'flows' being updated or not updated.
- */
-int
-dpif_flow_get_multiple(const struct dpif *dpif,
- struct odp_flow flows[], size_t n)
+static int
+dpif_flow_put__(struct dpif *dpif, const struct dpif_flow_put *put)
{
int error;
- size_t i;
- COVERAGE_ADD(dpif_flow_get, n);
+ COVERAGE_INC(dpif_flow_put);
+ ovs_assert(!(put->flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY
+ | DPIF_FP_ZERO_STATS)));
- for (i = 0; i < n; i++) {
- check_rw_flow_actions(&flows[i]);
+ error = dpif->dpif_class->flow_put(dpif, put);
+ if (error && put->stats) {
+ memset(put->stats, 0, sizeof *put->stats);
}
-
- error = dpif->dpif_class->flow_get(dpif, flows, n);
- log_operation(dpif, "flow_get_multiple", error);
+ log_flow_put_message(dpif, put, error);
return error;
}
-/* Adds or modifies a flow in 'dpif' as specified in 'put':
+/* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
+ * attribute OVS_FLOW_ATTR_KEY with types OVS_KEY_ATTR_* in the 'key_len' bytes
+ * starting at 'key', and OVS_FLOW_ATTR_MASK with types of OVS_KEY_ATTR_* in
+ * the 'mask_len' bytes starting at 'mask'. The associated actions are
+ * specified by the Netlink attributes with types OVS_ACTION_ATTR_* in the
+ * 'actions_len' bytes starting at 'actions'.
*
- * - If the flow specified in 'put->flow' does not exist in 'dpif', then
- * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
- * it is, the flow will be added, otherwise the operation will fail with
+ * - If the flow's key does not exist in 'dpif', then the flow will be added if
+ * 'flags' includes DPIF_FP_CREATE. Otherwise the operation will fail with
* ENOENT.
*
- * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
- * Behavior in this case depends on whether ODPPF_MODIFY is specified in
- * 'put->flags': if it is, the flow's actions will be updated, otherwise the
- * operation will fail with EEXIST. If the flow's actions are updated, then
- * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
- * left as-is otherwise.
+ * The datapath may reject attempts to insert overlapping flows with EINVAL
+ * or EEXIST, but clients should not rely on this: avoiding overlapping flows
+ * is primarily the client's responsibility.
+ *
+ * If the operation succeeds, then 'stats', if nonnull, will be zeroed.
*
- * Returns 0 if successful, otherwise a positive errno value.
+ * - If the flow's key does exist in 'dpif', then the flow's actions will be
+ * updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the operation will
+ * fail with EEXIST. If the flow's actions are updated, then its statistics
+ * will be zeroed if 'flags' includes DPIF_FP_ZERO_STATS, and left as-is
+ * otherwise.
+ *
+ * If the operation succeeds, then 'stats', if nonnull, will be set to the
+ * flow's statistics before the update.
*/
int
-dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
+dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
+ const struct nlattr *key, size_t key_len,
+ const struct nlattr *mask, size_t mask_len,
+ const struct nlattr *actions, size_t actions_len,
+ struct dpif_flow_stats *stats)
+{
+ struct dpif_flow_put put;
+
+ put.flags = flags;
+ put.key = key;
+ put.key_len = key_len;
+ put.mask = mask;
+ put.mask_len = mask_len;
+ put.actions = actions;
+ put.actions_len = actions_len;
+ put.stats = stats;
+ return dpif_flow_put__(dpif, &put);
+}
+
+static int
+dpif_flow_del__(struct dpif *dpif, struct dpif_flow_del *del)
{
int error;
- COVERAGE_INC(dpif_flow_put);
+ COVERAGE_INC(dpif_flow_del);
- error = dpif->dpif_class->flow_put(dpif, put);
- if (should_log_flow_message(error)) {
- log_flow_put(dpif, error, put);
+ error = dpif->dpif_class->flow_del(dpif, del);
+ if (error && del->stats) {
+ memset(del->stats, 0, sizeof *del->stats);
}
+ log_flow_del_message(dpif, del, error);
return error;
}
-/* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
- * does not contain such a flow.
+/* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif' does
+ * not contain such a flow. The flow is specified by the Netlink attributes
+ * with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at 'key'.
*
- * If successful, updates 'flow->stats', 'flow->actions_len', and
- * 'flow->actions' as described for dpif_flow_get(). */
+ * If the operation succeeds, then 'stats', if nonnull, will be set to the
+ * flow's statistics before its deletion. */
int
-dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
+dpif_flow_del(struct dpif *dpif,
+ const struct nlattr *key, size_t key_len,
+ struct dpif_flow_stats *stats)
{
- int error;
+ struct dpif_flow_del del;
- COVERAGE_INC(dpif_flow_del);
-
- check_rw_flow_actions(flow);
- memset(&flow->stats, 0, sizeof flow->stats);
+ del.key = key;
+ del.key_len = key_len;
+ del.stats = stats;
+ return dpif_flow_del__(dpif, &del);
+}
- error = dpif->dpif_class->flow_del(dpif, flow);
- if (should_log_flow_message(error)) {
- log_flow_operation(dpif, "delete flow", error, flow);
- }
- return error;
+/* Allocates thread-local state for use with the 'flow_dump_next' function for
+ * 'dpif'. On return, initializes '*statep' with any private data needed for
+ * iteration. */
+void
+dpif_flow_dump_state_init(const struct dpif *dpif, void **statep)
+{
+ dpif->dpif_class->flow_dump_state_init(statep);
}
-/* Initializes 'dump' to begin dumping the flows in a dpif.
- *
- * This function provides no status indication. An error status for the entire
- * dump operation is provided when it is completed by calling
- * dpif_flow_dump_done().
- */
+/* Releases 'state' which was initialized by a call to the
+ * 'flow_dump_state_init' function for 'dpif'. */
void
+dpif_flow_dump_state_uninit(const struct dpif *dpif, void *state)
+{
+ dpif->dpif_class->flow_dump_state_uninit(state);
+}
+
+/* Initializes 'dump' to begin dumping the flows in a dpif. On sucess,
+ * initializes 'dump' with any data needed for iteration and returns 0.
+ * Otherwise, returns a positive errno value describing the problem. */
+int
dpif_flow_dump_start(struct dpif_flow_dump *dump, const struct dpif *dpif)
{
+ int error;
dump->dpif = dpif;
- dump->error = dpif->dpif_class->flow_dump_start(dpif, &dump->state);
- log_operation(dpif, "flow_dump_start", dump->error);
+ error = dpif->dpif_class->flow_dump_start(dpif, &dump->iter);
+ log_operation(dpif, "flow_dump_start", error);
+ return error;
}
-/* Attempts to retrieve another flow from 'dump', which must have been
- * initialized with dpif_flow_dump_start(). On success, stores a new odp_flow
- * into 'flow' and returns true. Failure might indicate an actual error or
- * merely the end of the flow table. An error status for the entire dump
+/* Attempts to retrieve another flow from 'dump', using 'state' for
+ * thread-local storage. 'dump' must have been initialized with a successful
+ * call to dpif_flow_dump_start(), and 'state' must have been initialized with
+ * dpif_flow_state_init().
+ *
+ * On success, updates the output parameters as described below and returns
+ * true. Otherwise, returns false. Failure might indicate an actual error or
+ * merely the end of the flow table. An error status for the entire dump
* operation is provided when it is completed by calling dpif_flow_dump_done().
+ * Multiple threads may use the same 'dump' with this function, but all other
+ * parameters must not be shared.
*
- * Dumping flow actions is optional. To avoid dumping actions initialize
- * 'flow->actions' to NULL and 'flow->actions_len' to 0. Otherwise, point
- * 'flow->actions' to an array of struct nlattr and initialize
- * 'flow->actions_len' with the number of bytes of Netlink attributes.
- * dpif_flow_dump_next() will fill in as many actions as will fit into the
- * provided array and update 'flow->actions_len' with the number of bytes
- * required (regardless of whether they fit in the provided space). */
+ * On success, if 'key' and 'key_len' are nonnull then '*key' and '*key_len'
+ * will be set to Netlink attributes with types OVS_KEY_ATTR_* representing the
+ * dumped flow's key. If 'actions' and 'actions_len' are nonnull then they are
+ * set to Netlink attributes with types OVS_ACTION_ATTR_* representing the
+ * dumped flow's actions. If 'stats' is nonnull then it will be set to the
+ * dumped flow's statistics.
+ *
+ * All of the returned data is owned by 'dpif', not by the caller, and the
+ * caller must not modify or free it. 'dpif' guarantees that it remains
+ * accessible and unchanging until at least the next call to 'flow_dump_next'
+ * or 'flow_dump_done' for 'dump' and 'state'. */
bool
-dpif_flow_dump_next(struct dpif_flow_dump *dump, struct odp_flow *flow)
+dpif_flow_dump_next(struct dpif_flow_dump *dump, void *state,
+ const struct nlattr **key, size_t *key_len,
+ const struct nlattr **mask, size_t *mask_len,
+ const struct nlattr **actions, size_t *actions_len,
+ const struct dpif_flow_stats **stats)
{
const struct dpif *dpif = dump->dpif;
+ int error;
- check_rw_flow_actions(flow);
- check_rw_flow_key(flow);
-
- if (dump->error) {
- return false;
- }
-
- dump->error = dpif->dpif_class->flow_dump_next(dpif, dump->state, flow);
- if (dump->error == EOF) {
- VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
- } else {
- if (dump->error) {
- flow->key_len = 0;
+ error = dpif->dpif_class->flow_dump_next(dpif, dump->iter, state,
+ key, key_len, mask, mask_len,
+ actions, actions_len, stats);
+ if (error) {
+ if (key) {
+ *key = NULL;
+ *key_len = 0;
+ }
+ if (mask) {
+ *mask = NULL;
+ *mask_len = 0;
}
- if (should_log_flow_message(dump->error)) {
- log_flow_operation(dpif, "flow_dump_next", dump->error, flow);
+ if (actions) {
+ *actions = NULL;
+ *actions_len = 0;
+ }
+ if (stats) {
+ *stats = NULL;
}
}
-
- if (dump->error) {
- dpif->dpif_class->flow_dump_done(dpif, dump->state);
- return false;
+ if (error == EOF) {
+ VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
+ } else if (should_log_flow_message(error)) {
+ log_flow_message(dpif, error, "flow_dump",
+ key ? *key : NULL, key ? *key_len : 0,
+ mask ? *mask : NULL, mask ? *mask_len : 0,
+ stats ? *stats : NULL, actions ? *actions : NULL,
+ actions ? *actions_len : 0);
}
- return true;
+ return !error;
+}
+
+/* Determines whether the next call to 'dpif_flow_dump_next' for 'dump' and
+ * 'state' will modify or free the keys that it previously returned. 'state'
+ * must have been initialized by a call to 'dpif_flow_dump_state_init' for
+ * 'dump'.
+ *
+ * 'dpif' guarantees that data returned by flow_dump_next() will remain
+ * accessible and unchanging until the next call. This function provides a way
+ * for callers to determine whether that guarantee extends beyond the next
+ * call.
+ *
+ * Returns true if the next call to flow_dump_next() is expected to be
+ * destructive to previously returned keys for 'state', false otherwise. */
+bool
+dpif_flow_dump_next_may_destroy_keys(struct dpif_flow_dump *dump, void *state)
+{
+ const struct dpif *dpif = dump->dpif;
+ return (dpif->dpif_class->flow_dump_next_may_destroy_keys
+ ? dpif->dpif_class->flow_dump_next_may_destroy_keys(state)
+ : true);
}
/* Completes flow table dump operation 'dump', which must have been initialized
- * with dpif_flow_dump_start(). Returns 0 if the dump operation was
- * error-free, otherwise a positive errno value describing the problem. */
+ * with a successful call to dpif_flow_dump_start(). Returns 0 if the dump
+ * operation was error-free, otherwise a positive errno value describing the
+ * problem. */
int
dpif_flow_dump_done(struct dpif_flow_dump *dump)
{
const struct dpif *dpif = dump->dpif;
- if (!dump->error) {
- dump->error = dpif->dpif_class->flow_dump_done(dpif, dump->state);
- log_operation(dpif, "flow_dump_done", dump->error);
+ int error = dpif->dpif_class->flow_dump_done(dpif, dump->iter);
+ log_operation(dpif, "flow_dump_done", error);
+ return error == EOF ? 0 : error;
+}
+
+struct dpif_execute_helper_aux {
+ struct dpif *dpif;
+ int error;
+};
+
+/* This is called for actions that need the context of the datapath to be
+ * meaningful. */
+static void
+dpif_execute_helper_cb(void *aux_, struct ofpbuf *packet,
+ struct pkt_metadata *md,
+ const struct nlattr *action, bool may_steal OVS_UNUSED)
+{
+ struct dpif_execute_helper_aux *aux = aux_;
+ struct dpif_execute execute;
+ int type = nl_attr_type(action);
+
+ switch ((enum ovs_action_attr)type) {
+ case OVS_ACTION_ATTR_OUTPUT:
+ case OVS_ACTION_ATTR_USERSPACE:
+ execute.actions = action;
+ execute.actions_len = NLA_ALIGN(action->nla_len);
+ execute.packet = packet;
+ execute.md = *md;
+ execute.needs_help = false;
+ aux->error = aux->dpif->dpif_class->execute(aux->dpif, &execute);
+ break;
+
+ case OVS_ACTION_ATTR_PUSH_VLAN:
+ case OVS_ACTION_ATTR_POP_VLAN:
+ case OVS_ACTION_ATTR_PUSH_MPLS:
+ case OVS_ACTION_ATTR_POP_MPLS:
+ case OVS_ACTION_ATTR_SET:
+ case OVS_ACTION_ATTR_SAMPLE:
+ case OVS_ACTION_ATTR_UNSPEC:
+ case OVS_ACTION_ATTR_RECIRC:
+ case OVS_ACTION_ATTR_HASH:
+ case __OVS_ACTION_ATTR_MAX:
+ OVS_NOT_REACHED();
}
- return dump->error == EOF ? 0 : dump->error;
}
-/* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on
- * the Ethernet frame specified in 'packet'.
+/* Executes 'execute' by performing most of the actions in userspace and
+ * passing the fully constructed packets to 'dpif' for output and userspace
+ * actions.
+ *
+ * This helps with actions that a given 'dpif' doesn't implement directly. */
+static int
+dpif_execute_with_help(struct dpif *dpif, struct dpif_execute *execute)
+{
+ struct dpif_execute_helper_aux aux = {dpif, 0};
+
+ COVERAGE_INC(dpif_execute_with_help);
+
+ odp_execute_actions(&aux, execute->packet, false, &execute->md,
+ execute->actions, execute->actions_len,
+ dpif_execute_helper_cb);
+ return aux.error;
+}
+
+/* Causes 'dpif' to perform the 'execute->actions_len' bytes of actions in
+ * 'execute->actions' on the Ethernet frame in 'execute->packet' and on packet
+ * metadata in 'execute->md'. The implementation is allowed to modify both the
+ * '*execute->packet' and 'execute->md'.
+ *
+ * Some dpif providers do not implement every action. The Linux kernel
+ * datapath, in particular, does not implement ARP field modification. If
+ * 'needs_help' is true, the dpif layer executes in userspace all of the
+ * actions that it can, and for OVS_ACTION_ATTR_OUTPUT and
+ * OVS_ACTION_ATTR_USERSPACE actions it passes the packet through to the dpif
+ * implementation.
+ *
+ * This works even if 'execute->actions_len' is too long for a Netlink
+ * attribute.
*
* Returns 0 if successful, otherwise a positive errno value. */
int
-dpif_execute(struct dpif *dpif,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *buf)
+dpif_execute(struct dpif *dpif, struct dpif_execute *execute)
{
int error;
COVERAGE_INC(dpif_execute);
- if (actions_len > 0) {
- error = dpif->dpif_class->execute(dpif, actions, actions_len, buf);
+ if (execute->actions_len > 0) {
+ error = (execute->needs_help || nl_attr_oversized(execute->actions_len)
+ ? dpif_execute_with_help(dpif, execute)
+ : dpif->dpif_class->execute(dpif, execute));
} else {
error = 0;
}
- if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
- struct ds ds = DS_EMPTY_INITIALIZER;
- char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
- ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
- format_odp_actions(&ds, actions, actions_len);
- if (error) {
- ds_put_format(&ds, " failed (%s)", strerror(error));
- }
- ds_put_format(&ds, " on packet %s", packet);
- vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
- ds_destroy(&ds);
- free(packet);
- }
+ log_execute_message(dpif, execute, error);
+
return error;
}
-/* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
- * in '*listen_mask' indicates that dpif_recv() will receive messages of that
- * type. Returns 0 if successful, otherwise a positive errno value. */
-int
-dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask)
+/* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
+ * which they are specified, placing each operation's results in the "output"
+ * members documented in comments.
+ *
+ * This function exists because some datapaths can perform batched operations
+ * faster than individual operations. */
+void
+dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
{
- int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask);
- if (error) {
- *listen_mask = 0;
+ if (dpif->dpif_class->operate) {
+ while (n_ops > 0) {
+ size_t chunk;
+
+ /* Count 'chunk', the number of ops that can be executed without
+ * needing any help. Ops that need help should be rare, so we
+ * expect this to ordinarily be 'n_ops', that is, all the ops. */
+ for (chunk = 0; chunk < n_ops; chunk++) {
+ struct dpif_op *op = ops[chunk];
+
+ if (op->type == DPIF_OP_EXECUTE && op->u.execute.needs_help) {
+ break;
+ }
+ }
+
+ if (chunk) {
+ /* Execute a chunk full of ops that the dpif provider can
+ * handle itself, without help. */
+ size_t i;
+
+ dpif->dpif_class->operate(dpif, ops, chunk);
+
+ for (i = 0; i < chunk; i++) {
+ struct dpif_op *op = ops[i];
+
+ switch (op->type) {
+ case DPIF_OP_FLOW_PUT:
+ log_flow_put_message(dpif, &op->u.flow_put, op->error);
+ break;
+
+ case DPIF_OP_FLOW_DEL:
+ log_flow_del_message(dpif, &op->u.flow_del, op->error);
+ break;
+
+ case DPIF_OP_EXECUTE:
+ log_execute_message(dpif, &op->u.execute, op->error);
+ break;
+ }
+ }
+
+ ops += chunk;
+ n_ops -= chunk;
+ } else {
+ /* Help the dpif provider to execute one op. */
+ struct dpif_op *op = ops[0];
+
+ op->error = dpif_execute(dpif, &op->u.execute);
+ ops++;
+ n_ops--;
+ }
+ }
+ } else {
+ size_t i;
+
+ for (i = 0; i < n_ops; i++) {
+ struct dpif_op *op = ops[i];
+
+ switch (op->type) {
+ case DPIF_OP_FLOW_PUT:
+ op->error = dpif_flow_put__(dpif, &op->u.flow_put);
+ break;
+
+ case DPIF_OP_FLOW_DEL:
+ op->error = dpif_flow_del__(dpif, &op->u.flow_del);
+ break;
+
+ case DPIF_OP_EXECUTE:
+ op->error = dpif_execute(dpif, &op->u.execute);
+ break;
+
+ default:
+ OVS_NOT_REACHED();
+ }
+ }
}
- log_operation(dpif, "recv_get_mask", error);
- return error;
}
-/* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
- * '*listen_mask' requests that dpif_recv() receive messages of that type.
- * Returns 0 if successful, otherwise a positive errno value. */
-int
-dpif_recv_set_mask(struct dpif *dpif, int listen_mask)
+/* Returns a string that represents 'type', for use in log messages. */
+const char *
+dpif_upcall_type_to_string(enum dpif_upcall_type type)
{
- int error = dpif->dpif_class->recv_set_mask(dpif, listen_mask);
- log_operation(dpif, "recv_set_mask", error);
- return error;
+ switch (type) {
+ case DPIF_UC_MISS: return "miss";
+ case DPIF_UC_ACTION: return "action";
+ case DPIF_N_UC_TYPES: default: return "<unknown>";
+ }
}
-/* Retrieve the sFlow sampling probability. '*probability' is expressed as the
- * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
- * the probability of sampling a given packet.
+/* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
+ * if successful, otherwise a positive errno value.
*
- * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
- * indicates that 'dpif' does not support sFlow sampling. */
+ * Turning packet receive off and then back on may change the Netlink PID
+ * assignments returned by dpif_port_get_pid(). If the client does this, it
+ * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
+ * using the new PID assignment. */
int
-dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability)
+dpif_recv_set(struct dpif *dpif, bool enable)
{
- int error = (dpif->dpif_class->get_sflow_probability
- ? dpif->dpif_class->get_sflow_probability(dpif, probability)
- : EOPNOTSUPP);
- if (error) {
- *probability = 0;
- }
- log_operation(dpif, "get_sflow_probability", error);
+ int error = dpif->dpif_class->recv_set(dpif, enable);
+ log_operation(dpif, "recv_set", error);
return error;
}
-/* Set the sFlow sampling probability. 'probability' is expressed as the
- * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
- * the probability of sampling a given packet.
+/* Refreshes the poll loops and Netlink sockets associated to each port,
+ * when the number of upcall handlers (upcall receiving thread) is changed
+ * to 'n_handlers' and receiving packets for 'dpif' is enabled by
+ * recv_set().
+ *
+ * Since multiple upcall handlers can read upcalls simultaneously from
+ * 'dpif', each port can have multiple Netlink sockets, one per upcall
+ * handler. So, handlers_set() is responsible for the following tasks:
+ *
+ * When receiving upcall is enabled, extends or creates the
+ * configuration to support:
*
- * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
- * indicates that 'dpif' does not support sFlow sampling. */
+ * - 'n_handlers' Netlink sockets for each port.
+ *
+ * - 'n_handlers' poll loops, one for each upcall handler.
+ *
+ * - registering the Netlink sockets for the same upcall handler to
+ * the corresponding poll loop.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. */
int
-dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability)
+dpif_handlers_set(struct dpif *dpif, uint32_t n_handlers)
{
- int error = (dpif->dpif_class->set_sflow_probability
- ? dpif->dpif_class->set_sflow_probability(dpif, probability)
- : EOPNOTSUPP);
- log_operation(dpif, "set_sflow_probability", error);
+ int error = dpif->dpif_class->handlers_set(dpif, n_handlers);
+ log_operation(dpif, "handlers_set", error);
return error;
}
-/* Polls for an upcall from 'dpif'. If successful, stores the upcall into
- * '*upcall'. Only upcalls of the types selected with the set_listen_mask
- * member function will ordinarily be received (but if a message type is
- * enabled and then later disabled, some stragglers might pop up).
+/* Polls for an upcall from 'dpif' for an upcall handler. Since there
+ * there can be multiple poll loops, 'handler_id' is needed as index to
+ * identify the corresponding poll loop. If successful, stores the upcall
+ * into '*upcall', using 'buf' for storage. Should only be called if
+ * 'recv_set' has been used to enable receiving packets from 'dpif'.
+ *
+ * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
+ * 'buf', so their memory cannot be freed separately from 'buf'.
*
- * The caller takes ownership of the data that 'upcall' points to.
- * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned by
- * 'upcall->packet', so their memory cannot be freed separately. (This is
- * hardly a great way to do things but it works out OK for the dpif providers
- * and clients that exist so far.)
+ * The caller owns the data of 'upcall->packet' and may modify it. If
+ * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
+ * will be reallocated. This requires the data of 'upcall->packet' to be
+ * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
+ * when an error is returned, the 'upcall->packet' may be uninitialized
+ * and should not be released.
*
* Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
* if no upcall is immediately available. */
int
-dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall)
+dpif_recv(struct dpif *dpif, uint32_t handler_id, struct dpif_upcall *upcall,
+ struct ofpbuf *buf)
{
- int error = dpif->dpif_class->recv(dpif, upcall);
+ int error = dpif->dpif_class->recv(dpif, handler_id, upcall, buf);
if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) {
- struct flow flow;
- char *s;
-
- s = ofp_packet_to_string(upcall->packet->data,
- upcall->packet->size, upcall->packet->size);
- odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
-
- VLOG_DBG("%s: %s upcall on port %"PRIu16": %s", dpif_name(dpif),
- (upcall->type == _ODPL_MISS_NR ? "miss"
- : upcall->type == _ODPL_ACTION_NR ? "action"
- : upcall->type == _ODPL_SFLOW_NR ? "sFlow"
- : "<unknown>"),
- flow.in_port, s);
- free(s);
+ struct ds flow;
+ char *packet;
+
+ packet = ofp_packet_to_string(ofpbuf_data(&upcall->packet),
+ ofpbuf_size(&upcall->packet));
+
+ ds_init(&flow);
+ odp_flow_key_format(upcall->key, upcall->key_len, &flow);
+
+ VLOG_DBG("%s: %s upcall:\n%s\n%s",
+ dpif_name(dpif), dpif_upcall_type_to_string(upcall->type),
+ ds_cstr(&flow), packet);
+
+ ds_destroy(&flow);
+ free(packet);
+ } else if (error && error != EAGAIN) {
+ log_operation(dpif, "recv", error);
}
return error;
}
}
}
-/* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
- * received with dpif_recv(). */
+/* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
+ * 'dpif' has a message queued to be received with the recv member
+ * function. Since there can be multiple poll loops, 'handler_id' is
+ * needed as index to identify the corresponding poll loop. */
void
-dpif_recv_wait(struct dpif *dpif)
+dpif_recv_wait(struct dpif *dpif, uint32_t handler_id)
{
- dpif->dpif_class->recv_wait(dpif);
+ dpif->dpif_class->recv_wait(dpif, handler_id);
}
/* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
}
/* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
- * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
- * stores the priority into '*priority'. On failure, returns a positive errno
- * value and stores 0 into '*priority'. */
+ * value used for setting packet priority.
+ * On success, returns 0 and stores the priority into '*priority'.
+ * On failure, returns a positive errno value and stores 0 into '*priority'. */
int
dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
uint32_t *priority)
{
if (!error) {
VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
- } else if (is_errno(error)) {
+ } else if (ofperr_is_valid(error)) {
VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
- dpif_name(dpif), operation, strerror(error));
+ dpif_name(dpif), operation, ofperr_get_name(error));
} else {
- VLOG_WARN_RL(&error_rl, "%s: %s failed (%d/%d)",
- dpif_name(dpif), operation,
- get_ofp_err_type(error), get_ofp_err_code(error));
+ VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
+ dpif_name(dpif), operation, ovs_strerror(error));
}
}
static enum vlog_level
flow_message_log_level(int error)
{
- return error ? VLL_WARN : VLL_DBG;
+ /* If flows arrive in a batch, userspace may push down multiple
+ * unique flow definitions that overlap when wildcards are applied.
+ * Kernels that support flow wildcarding will reject these flows as
+ * duplicates (EEXIST), so lower the log level to debug for these
+ * types of messages. */
+ return (error && error != EEXIST) ? VLL_WARN : VLL_DBG;
}
static bool
static void
log_flow_message(const struct dpif *dpif, int error, const char *operation,
const struct nlattr *key, size_t key_len,
- const struct odp_flow_stats *stats,
+ const struct nlattr *mask, size_t mask_len,
+ const struct dpif_flow_stats *stats,
const struct nlattr *actions, size_t actions_len)
{
struct ds ds = DS_EMPTY_INITIALIZER;
}
ds_put_format(&ds, "%s ", operation);
if (error) {
- ds_put_format(&ds, "(%s) ", strerror(error));
+ ds_put_format(&ds, "(%s) ", ovs_strerror(error));
}
- odp_flow_key_format(key, key_len, &ds);
+ odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, true);
if (stats) {
ds_put_cstr(&ds, ", ");
- format_odp_flow_stats(&ds, stats);
+ dpif_flow_stats_format(stats, &ds);
}
if (actions || actions_len) {
ds_put_cstr(&ds, ", actions:");
}
static void
-log_flow_operation(const struct dpif *dpif, const char *operation, int error,
- struct odp_flow *flow)
+log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put,
+ int error)
{
- if (error) {
- flow->key_len = 0;
- flow->actions_len = 0;
- }
- log_flow_message(dpif, error, operation,
- flow->key, flow->key_len, !error ? &flow->stats : NULL,
- flow->actions, flow->actions_len);
-}
-
-static void
-log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put)
-{
- enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS };
- struct ds s;
+ if (should_log_flow_message(error)) {
+ struct ds s;
- ds_init(&s);
- ds_put_cstr(&s, "put");
- if (put->flags & ODPPF_CREATE) {
- ds_put_cstr(&s, "[create]");
- }
- if (put->flags & ODPPF_MODIFY) {
- ds_put_cstr(&s, "[modify]");
- }
- if (put->flags & ODPPF_ZERO_STATS) {
- ds_put_cstr(&s, "[zero]");
- }
- if (put->flags & ~ODPPF_ALL) {
- ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL);
+ ds_init(&s);
+ ds_put_cstr(&s, "put");
+ if (put->flags & DPIF_FP_CREATE) {
+ ds_put_cstr(&s, "[create]");
+ }
+ if (put->flags & DPIF_FP_MODIFY) {
+ ds_put_cstr(&s, "[modify]");
+ }
+ if (put->flags & DPIF_FP_ZERO_STATS) {
+ ds_put_cstr(&s, "[zero]");
+ }
+ log_flow_message(dpif, error, ds_cstr(&s),
+ put->key, put->key_len, put->mask, put->mask_len,
+ put->stats, put->actions, put->actions_len);
+ ds_destroy(&s);
}
- log_flow_message(dpif, error, ds_cstr(&s),
- put->flow.key, put->flow.key_len,
- !error ? &put->flow.stats : NULL,
- put->flow.actions, put->flow.actions_len);
- ds_destroy(&s);
}
-/* There is a tendency to construct odp_flow objects on the stack and to
- * forget to properly initialize their "actions" and "actions_len" members.
- * When this happens, we get memory corruption because the kernel
- * writes through the random pointer that is in the "actions" member.
- *
- * This function attempts to combat the problem by:
- *
- * - Forcing a segfault if "actions" points to an invalid region (instead
- * of just getting back EFAULT, which can be easily missed in the log).
- *
- * - Storing a distinctive value that is likely to cause an
- * easy-to-identify error later if it is dereferenced, etc.
- *
- * - Triggering a warning on uninitialized memory from Valgrind if
- * "actions" or "actions_len" was not initialized.
- */
static void
-check_rw_flow_actions(struct odp_flow *flow)
+log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del,
+ int error)
{
- if (flow->actions_len) {
- memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);
+ if (should_log_flow_message(error)) {
+ log_flow_message(dpif, error, "flow_del", del->key, del->key_len,
+ NULL, 0, !error ? del->stats : NULL, NULL, 0);
}
}
-/* Same as check_rw_flow_actions() but for flow->key. */
static void
-check_rw_flow_key(struct odp_flow *flow)
+log_execute_message(struct dpif *dpif, const struct dpif_execute *execute,
+ int error)
{
- if (flow->key_len) {
- memset(&flow->key[0], 0xcc, sizeof flow->key[0]);
+ if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
+ struct ds ds = DS_EMPTY_INITIALIZER;
+ char *packet;
+
+ packet = ofp_packet_to_string(ofpbuf_data(execute->packet),
+ ofpbuf_size(execute->packet));
+ ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
+ format_odp_actions(&ds, execute->actions, execute->actions_len);
+ if (error) {
+ ds_put_format(&ds, " failed (%s)", ovs_strerror(error));
+ }
+ ds_put_format(&ds, " on packet %s", packet);
+ vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
+ ds_destroy(&ds);
+ free(packet);
}
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff