/depcomp
/distfiles
/install-sh
+ /libtool
/manpage-check
/missing
/missing-distfiles
/package.m4
/stamp-h1
+ /_build-gcc
+ /_build-clang
Module.symvers
TAGS
cscope.*
tags
+myexp/
_debian
- # Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ # Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
#
# Copying and distribution of this file, with or without modification,
# are permitted in any medium without royalty provided the copyright
SUBDIRS = datapath
AM_CPPFLAGS = $(SSL_CFLAGS)
+
+ if WIN32
+ AM_CPPFLAGS += -I $(top_srcdir)/include/windows
+ endif
+
AM_CPPFLAGS += -I $(top_srcdir)/include
AM_CPPFLAGS += -I $(top_srcdir)/lib
AM_CPPFLAGS += -I $(top_builddir)/lib
INSTALL.Fedora \
INSTALL.KVM \
INSTALL.Libvirt \
+ INSTALL.NetBSD \
INSTALL.RHEL \
INSTALL.SSL \
INSTALL.XenServer \
MAN_ROOTS =
noinst_DATA =
noinst_HEADERS =
- noinst_LIBRARIES =
+ lib_LTLIBRARIES =
noinst_man_MANS =
noinst_PROGRAMS =
noinst_SCRIPTS =
SUFFIXES += .in
.in:
+ @mkdir -p $$(dirname $@)
$(PERL) $(srcdir)/build-aux/soexpand.pl -I$(srcdir) < $< | \
sed \
-e 's,[@]PKIDIR[@],$(PKIDIR),g' \
-e 's,[@]bindir[@],$(bindir),g' \
-e 's,[@]sbindir[@],$(sbindir),g' \
-e 's,[@]abs_top_srcdir[@],$(abs_top_srcdir),g' \
- -e 's,[@]ovsdbmonitordir[@],$(ovsdbmonitordir),g' \
> $@.tmp
@if head -n 1 $@.tmp | grep '#!' > /dev/null; then \
echo chmod +x $@.tmp; \
@if test -e $(srcdir)/.git && (git --version) >/dev/null 2>&1; then \
(cd datapath && $(MAKE) distfiles); \
(cat distfiles; sed 's|^|datapath/|' datapath/distfiles) | \
- sort -u > all-distfiles; \
- LC_ALL=C sort -u > all-distfiles; \
- (cd $(srcdir) && git ls-files) | grep -v '\.gitignore$$' | \
- LC_ALL=C sort -u > all-gitfiles; \
- LC_ALL=C comm -1 -3 all-distfiles all-gitfiles > missing-distfiles; \
++ LC_ALL=C sort -u > all-distfiles; \
+ (cd $(srcdir) && git ls-files) | grep -vFf $(srcdir)/.non-distfiles | \
- sort -u > all-gitfiles; \
- comm -1 -3 all-distfiles all-gitfiles > missing-distfiles; \
++ LC_ALL=C sort -u > all-gitfiles; \
++ LC_ALL=C comm -1 -3 all-distfiles all-gitfiles > missing-distfiles; \
if test -s missing-distfiles; then \
echo "The distribution is missing the following files:"; \
cat missing-distfiles; \
list='$(DISTFILES)'; \
for file in $$list; do echo $$file; done | \
sed -e "s|^$$srcdirstrip/||;t" \
- -e "s|^$$topsrcdirstrip/|$(top_builddir)/|;t" | sort -u > $@
+ -e "s|^$$topsrcdirstrip/|$(top_builddir)/|;t" | \
+ LC_ALL=C sort -u > $@
CLEANFILES += distfiles
endif
.PHONY: dist-hook-git
cd datapath/linux && $(MAKE) modules_install
endif
+ include m4/automake.mk
include lib/automake.mk
include ofproto/automake.mk
include utilities/automake.mk
include xenserver/automake.mk
include python/automake.mk
include python/compat/automake.mk
+include planetlab/automake.mk
include tutorial/automake.mk
include vtep/automake.mk
- # Copyright (C) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ # Copyright (C) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
#
# Copying and distribution of this file, with or without modification,
# are permitted in any medium without royalty provided the copyright
# notice and this notice are preserved. This file is offered as-is,
# without warranty of any kind.
- noinst_LIBRARIES += lib/libopenvswitch.a
+ lib_LTLIBRARIES += lib/libopenvswitch.la
- lib_libopenvswitch_a_SOURCES = \
+ lib_libopenvswitch_la_LIBADD = $(SSL_LIBS)
+ lib_libopenvswitch_la_LDFLAGS = -release $(VERSION)
+
+ lib_libopenvswitch_la_SOURCES = \
lib/aes128.c \
lib/aes128.h \
lib/async-append.h \
lib/command-line.c \
lib/command-line.h \
lib/compiler.h \
+ lib/connectivity.c \
+ lib/connectivity.h \
lib/coverage.c \
lib/coverage.h \
lib/crc32c.c \
lib/dynamic-string.h \
lib/entropy.c \
lib/entropy.h \
+ lib/fat-rwlock.c \
+ lib/fat-rwlock.h \
lib/fatal-signal.c \
lib/fatal-signal.h \
lib/flow.c \
lib/jsonrpc.h \
lib/lacp.c \
lib/lacp.h \
- lib/latch.c \
lib/latch.h \
lib/learn.c \
lib/learn.h \
lib/multipath.c \
lib/multipath.h \
lib/netdev-dummy.c \
+ lib/netdev-tunnel.c \
+ lib/netdev-pltap.c \
lib/netdev-provider.h \
lib/netdev-vport.c \
lib/netdev-vport.h \
lib/stream-unix.c \
lib/stream.c \
lib/stream.h \
+ lib/stdio.c \
lib/string.c \
- lib/string.h \
lib/svec.c \
lib/svec.h \
lib/table.c \
lib/timeval.h \
lib/token-bucket.c \
lib/token-bucket.h \
+ lib/tunalloc.c \
+ lib/tunalloc.h \
lib/type-props.h \
lib/unaligned.h \
lib/unicode.c \
lib/vswitch-idl.h \
lib/vtep-idl.c \
lib/vtep-idl.h
+ if WIN32
+ lib_libopenvswitch_la_SOURCES += lib/latch-windows.c
+ else
+ lib_libopenvswitch_la_SOURCES += lib/latch.c
+ endif
+
+ EXTRA_DIST += \
+ lib/stdio.h.in \
+ lib/string.h.in
- nodist_lib_libopenvswitch_a_SOURCES = \
+ nodist_lib_libopenvswitch_la_SOURCES = \
lib/dirs.c
- CLEANFILES += $(nodist_lib_libopenvswitch_a_SOURCES)
+ CLEANFILES += $(nodist_lib_libopenvswitch_la_SOURCES)
- noinst_LIBRARIES += lib/libsflow.a
- lib_libsflow_a_SOURCES = \
+ lib_LTLIBRARIES += lib/libsflow.la
+ lib_libsflow_la_LDFLAGS = -release $(VERSION)
+ lib_libsflow_la_SOURCES = \
lib/sflow_api.h \
lib/sflow.h \
lib/sflow_agent.c \
lib/sflow_sampler.c \
lib/sflow_poller.c \
lib/sflow_receiver.c
- lib_libsflow_a_CFLAGS = $(AM_CFLAGS)
+ lib_libsflow_la_CPPFLAGS = $(AM_CPPFLAGS)
+ lib_libsflow_la_CFLAGS = $(AM_CFLAGS)
if HAVE_WNO_UNUSED
- lib_libsflow_a_CFLAGS += -Wno-unused
+ lib_libsflow_la_CFLAGS += -Wno-unused
endif
if HAVE_WNO_UNUSED_PARAMETER
- lib_libsflow_a_CFLAGS += -Wno-unused-parameter
+ lib_libsflow_la_CFLAGS += -Wno-unused-parameter
endif
if LINUX_DATAPATH
- lib_libopenvswitch_a_SOURCES += \
+ lib_libopenvswitch_la_SOURCES += \
lib/dpif-linux.c \
lib/dpif-linux.h \
lib/netdev-linux.c \
endif
if HAVE_POSIX_AIO
- lib_libopenvswitch_a_SOURCES += lib/async-append-aio.c
+ lib_libopenvswitch_la_SOURCES += lib/async-append-aio.c
else
- lib_libopenvswitch_a_SOURCES += lib/async-append-null.c
+ lib_libopenvswitch_la_SOURCES += lib/async-append-null.c
endif
if ESX
- lib_libopenvswitch_a_SOURCES += \
+ lib_libopenvswitch_la_SOURCES += \
lib/route-table-stub.c
endif
if HAVE_IF_DL
- lib_libopenvswitch_a_SOURCES += \
+ lib_libopenvswitch_la_SOURCES += \
lib/netdev-bsd.c \
lib/rtbsd.c \
lib/rtbsd.h \
endif
if HAVE_OPENSSL
- lib_libopenvswitch_a_SOURCES += lib/stream-ssl.c
- nodist_lib_libopenvswitch_a_SOURCES += lib/dhparams.c
+ lib_libopenvswitch_la_SOURCES += lib/stream-ssl.c
+ nodist_lib_libopenvswitch_la_SOURCES += lib/dhparams.c
lib/dhparams.c: lib/dh1024.pem lib/dh2048.pem lib/dh4096.pem
(echo '#include "lib/dhparams.h"' && \
openssl dhparam -C -in $(srcdir)/lib/dh1024.pem -noout && \
| sed 's/\(get_dh[0-9]*\)()/\1(void)/' > lib/dhparams.c.tmp
mv lib/dhparams.c.tmp lib/dhparams.c
else
- lib_libopenvswitch_a_SOURCES += lib/stream-nossl.c
+ lib_libopenvswitch_la_SOURCES += lib/stream-nossl.c
endif
EXTRA_DIST += \
$(MKDIR_P) $(DESTDIR)$(LOGDIR)
$(MKDIR_P) $(DESTDIR)$(DBDIR)
- if !USE_LINKER_SECTIONS
- # All distributed sources, with names adjust properly for referencing
- # from $(builddir).
- all_sources = \
- `for file in $(DIST_SOURCES); do \
- if test -f $$file; then \
- echo $$file; \
- else \
- echo $(VPATH)/$$file; \
- fi; \
- done`
-
- lib/coverage.$(OBJEXT): lib/coverage.def
- lib/coverage.def: $(DIST_SOURCES)
- sed -n 's|^COVERAGE_DEFINE(\([_a-zA-Z0-9]\{1,\}\)).*$$|COVERAGE_COUNTER(\1)|p' $(all_sources) | LC_ALL=C sort -u > $@
- CLEANFILES += lib/coverage.def
-
- lib/vlog.$(OBJEXT): lib/vlog-modules.def
- lib/vlog-modules.def: $(DIST_SOURCES)
- sed -n 's|^VLOG_DEFINE_\(THIS_\)\{0,1\}MODULE(\([_a-zA-Z0-9]\{1,\}\)).*$$|VLOG_MODULE(\2)|p' $(all_sources) | LC_ALL=C sort -u > $@
- CLEANFILES += lib/vlog-modules.def
- endif
/*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 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 "dynamic-string.h"
#include "flow.h"
#include "hmap.h"
+ #include "latch.h"
#include "list.h"
+ #include "meta-flow.h"
#include "netdev.h"
#include "netdev-vport.h"
#include "netlink.h"
#define NETDEV_RULE_PRIORITY 0x8000
/* Configuration parameters. */
- enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
/* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
+ /* Protects against changes to 'dp_netdevs'. */
+ static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
+
+ /* Contains all 'struct dp_netdev's. */
+ static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex)
+ = SHASH_INITIALIZER(&dp_netdevs);
+
struct dp_netdev_upcall {
struct dpif_upcall upcall; /* Queued upcall information. */
struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
};
+ /* A queue passing packets from a struct dp_netdev to its clients.
+ *
+ *
+ * Thread-safety
+ * =============
+ *
+ * Any access at all requires the owning 'dp_netdev''s queue_mutex. */
struct dp_netdev_queue {
- struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
- unsigned int head, tail;
+ struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN] OVS_GUARDED;
+ unsigned int head OVS_GUARDED;
+ unsigned int tail OVS_GUARDED;
};
- /* Datapath based on the network device interface from netdev.h. */
+ /* Datapath based on the network device interface from netdev.h.
+ *
+ *
+ * Thread-safety
+ * =============
+ *
+ * Some members, marked 'const', are immutable. Accessing other members
+ * requires synchronization, as noted in more detail below.
+ *
+ * Acquisition order is, from outermost to innermost:
+ *
+ * dp_netdev_mutex (global)
+ * port_rwlock
+ * flow_mutex
+ * cls.rwlock
+ * queue_mutex
+ */
struct dp_netdev {
- const struct dpif_class *class;
- char *name;
- int open_cnt;
- bool destroyed;
- int max_mtu; /* Maximum MTU of any port added so far. */
+ const struct dpif_class *const class;
+ const char *const name;
+ struct ovs_refcount ref_cnt;
+ atomic_flag destroyed;
+ /* Flows.
+ *
+ * Readers of 'cls' and 'flow_table' must take a 'cls->rwlock' read lock.
+ *
+ * Writers of 'cls' and 'flow_table' must take the 'flow_mutex' and then
+ * the 'cls->rwlock' write lock. (The outer 'flow_mutex' allows writers to
+ * atomically perform multiple operations on 'cls' and 'flow_table'.)
+ */
+ struct ovs_mutex flow_mutex;
+ struct classifier cls; /* Classifier. Protected by cls.rwlock. */
+ struct hmap flow_table OVS_GUARDED; /* Flow table. */
+
+ /* Queues.
+ *
+ * Everything in 'queues' is protected by 'queue_mutex'. */
+ struct ovs_mutex queue_mutex;
struct dp_netdev_queue queues[N_QUEUES];
- struct classifier cls; /* Classifier. */
- struct hmap flow_table; /* Flow table. */
struct seq *queue_seq; /* Incremented whenever a packet is queued. */
- /* Statistics. */
- long long int n_hit; /* Number of flow table matches. */
- long long int n_missed; /* Number of flow table misses. */
- long long int n_lost; /* Number of misses not passed to client. */
+ /* Statistics.
+ *
+ * ovsthread_counter is internally synchronized. */
+ struct ovsthread_counter *n_hit; /* Number of flow table matches. */
+ struct ovsthread_counter *n_missed; /* Number of flow table misses. */
+ struct ovsthread_counter *n_lost; /* Number of misses not passed up. */
- /* Ports. */
- struct dp_netdev_port *ports[MAX_PORTS];
- struct list port_list;
+ /* Ports.
+ *
+ * Any lookup into 'ports' or any access to the dp_netdev_ports found
+ * through 'ports' requires taking 'port_rwlock'. */
+ struct ovs_rwlock port_rwlock;
+ struct hmap ports OVS_GUARDED;
struct seq *port_seq; /* Incremented whenever a port changes. */
+
+ /* Forwarding threads. */
+ struct latch exit_latch;
+ struct dp_forwarder *forwarders;
+ size_t n_forwarders;
};
+ static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp,
+ odp_port_t)
+ OVS_REQ_RDLOCK(dp->port_rwlock);
+
/* A port in a netdev-based datapath. */
struct dp_netdev_port {
- odp_port_t port_no; /* Index into dp_netdev's 'ports'. */
- struct list node; /* Element in dp_netdev's 'port_list'. */
+ struct hmap_node node; /* Node in dp_netdev's 'ports'. */
+ odp_port_t port_no;
struct netdev *netdev;
struct netdev_saved_flags *sf;
struct netdev_rx *rx;
char *type; /* Port type as requested by user. */
};
- /* A flow in dp_netdev's 'flow_table'. */
+ /* A flow in dp_netdev's 'flow_table'.
+ *
+ *
+ * Thread-safety
+ * =============
+ *
+ * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
+ * its dp_netdev's classifier. The text below calls this classifier 'cls'.
+ *
+ * Motivation
+ * ----------
+ *
+ * The thread safety rules described here for "struct dp_netdev_flow" are
+ * motivated by two goals:
+ *
+ * - Prevent threads that read members of "struct dp_netdev_flow" from
+ * reading bad data due to changes by some thread concurrently modifying
+ * those members.
+ *
+ * - Prevent two threads making changes to members of a given "struct
+ * dp_netdev_flow" from interfering with each other.
+ *
+ *
+ * Rules
+ * -----
+ *
+ * A flow 'flow' may be accessed without a risk of being freed by code that
+ * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
+ * 'flow->ref_cnt' (or both). Code that needs to hold onto a flow for a while
+ * should take 'cls->rwlock', find the flow it needs, increment 'flow->ref_cnt'
+ * with dpif_netdev_flow_ref(), and drop 'cls->rwlock'.
+ *
+ * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the
+ * flow from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
+ * protect members of 'flow' from modification (that's 'flow->mutex').
+ *
+ * 'flow->mutex' protects the members of 'flow' from modification. It doesn't
+ * protect the flow from being deleted from 'cls' (that's 'cls->rwlock') and it
+ * doesn't prevent the flow from being freed (that's 'flow->ref_cnt').
+ *
+ * Some members, marked 'const', are immutable. Accessing other members
+ * requires synchronization, as noted in more detail below.
+ */
struct dp_netdev_flow {
/* Packet classification. */
- struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
+ const struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
+
+ /* Hash table index by unmasked flow. */
+ const struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
+ const struct flow flow; /* The flow that created this entry. */
+
+ /* Number of references.
+ * The classifier owns one reference.
+ * Any thread trying to keep a rule from being freed should hold its own
+ * reference. */
+ struct ovs_refcount ref_cnt;
+
+ /* Protects members marked OVS_GUARDED.
+ *
+ * Acquire after datapath's flow_mutex. */
+ struct ovs_mutex mutex OVS_ACQ_AFTER(dp_netdev_mutex);
+
+ /* Statistics.
+ *
+ * Reading or writing these members requires 'mutex'. */
+ long long int used OVS_GUARDED; /* Last used time, in monotonic msecs. */
+ long long int packet_count OVS_GUARDED; /* Number of packets matched. */
+ long long int byte_count OVS_GUARDED; /* Number of bytes matched. */
+ uint16_t tcp_flags OVS_GUARDED; /* Bitwise-OR of seen tcp_flags values. */
+
+ /* Actions.
+ *
+ * Reading 'actions' requires 'mutex'.
+ * Writing 'actions' requires 'mutex' and (to allow for transactions) the
+ * datapath's flow_mutex. */
+ struct dp_netdev_actions *actions OVS_GUARDED;
+ };
+
+ static struct dp_netdev_flow *dp_netdev_flow_ref(
+ const struct dp_netdev_flow *);
+ static void dp_netdev_flow_unref(struct dp_netdev_flow *);
- /* Hash table index by unmasked flow.*/
- struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
- struct flow flow; /* The flow that created this entry. */
+ /* A set of datapath actions within a "struct dp_netdev_flow".
+ *
+ *
+ * Thread-safety
+ * =============
+ *
+ * A struct dp_netdev_actions 'actions' may be accessed without a risk of being
+ * freed by code that holds a read-lock or write-lock on 'flow->mutex' (where
+ * 'flow' is the dp_netdev_flow for which 'flow->actions == actions') or that
+ * owns a reference to 'actions->ref_cnt' (or both). */
+ struct dp_netdev_actions {
+ struct ovs_refcount ref_cnt;
+
+ /* These members are immutable: they do not change during the struct's
+ * lifetime. */
+ struct nlattr *actions; /* Sequence of OVS_ACTION_ATTR_* attributes. */
+ unsigned int size; /* Size of 'actions', in bytes. */
+ };
- /* Statistics. */
- long long int used; /* Last used time, in monotonic msecs. */
- long long int packet_count; /* Number of packets matched. */
- long long int byte_count; /* Number of bytes matched. */
- uint16_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
+ struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *,
+ size_t);
+ struct dp_netdev_actions *dp_netdev_actions_ref(
+ const struct dp_netdev_actions *);
+ void dp_netdev_actions_unref(struct dp_netdev_actions *);
- /* Actions. */
- struct nlattr *actions;
- size_t actions_len;
+ /* A thread that receives packets from some ports, looks them up in the flow
+ * table, and executes the actions it finds. */
+ struct dp_forwarder {
+ struct dp_netdev *dp;
+ pthread_t thread;
+ char *name;
+ uint32_t min_hash, max_hash;
};
/* Interface to netdev-based datapath. */
uint64_t last_port_seq;
};
- /* All netdev-based datapaths. */
- static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
-
- /* Global lock for all data. */
- static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
-
- static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
- struct dp_netdev_port **portp);
- static int get_port_by_name(struct dp_netdev *, const char *devname,
- struct dp_netdev_port **portp);
- static void dp_netdev_free(struct dp_netdev *);
+ static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no,
+ struct dp_netdev_port **portp)
+ OVS_REQ_RDLOCK(dp->port_rwlock);
+ static int get_port_by_name(struct dp_netdev *dp, const char *devname,
+ struct dp_netdev_port **portp)
+ OVS_REQ_RDLOCK(dp->port_rwlock);
+ static void dp_netdev_free(struct dp_netdev *)
+ OVS_REQUIRES(dp_netdev_mutex);
static void dp_netdev_flow_flush(struct dp_netdev *);
- static int do_add_port(struct dp_netdev *, const char *devname,
- const char *type, odp_port_t port_no);
- static int do_del_port(struct dp_netdev *, odp_port_t port_no);
+ static int do_add_port(struct dp_netdev *dp, const char *devname,
+ const char *type, odp_port_t port_no)
+ OVS_REQ_WRLOCK(dp->port_rwlock);
+ static int do_del_port(struct dp_netdev *dp, odp_port_t port_no)
+ OVS_REQ_WRLOCK(dp->port_rwlock);
static int dpif_netdev_open(const struct dpif_class *, const char *name,
bool create, struct dpif **);
- static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
+ static int dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *,
int queue_no, const struct flow *,
- const struct nlattr *userdata);
- static void dp_netdev_execute_actions(struct dp_netdev *, const struct flow *,
- struct ofpbuf *,
+ const struct nlattr *userdata)
+ OVS_EXCLUDED(dp->queue_mutex);
+ static void dp_netdev_execute_actions(struct dp_netdev *dp,
+ const struct flow *, struct ofpbuf *,
+ struct pkt_metadata *,
const struct nlattr *actions,
- size_t actions_len);
- static void dp_netdev_port_input(struct dp_netdev *dp,
- struct dp_netdev_port *port,
- struct ofpbuf *packet, uint32_t skb_priority,
- uint32_t pkt_mark, const struct flow_tnl *tnl);
+ size_t actions_len)
+ OVS_REQ_RDLOCK(dp->port_rwlock);
+ static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
+ struct pkt_metadata *)
+ OVS_REQ_RDLOCK(dp->port_rwlock);
+ static void dp_netdev_set_threads(struct dp_netdev *, int n);
static struct dpif_netdev *
dpif_netdev_cast(const struct dpif *dpif)
return class != &dpif_netdev_class;
}
+static bool
+dpif_netdev_class_is_planetlab(const struct dpif_class *class)
+{
+ return class == &dpif_planetlab_class;
+}
+
static const char *
dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
{
return strcmp(type, "internal") ? type
+ : dpif_netdev_class_is_planetlab(class) ? "pltap"
: dpif_netdev_class_is_dummy(class) ? "dummy"
: "tap";
}
uint16_t netflow_id = hash_string(dp->name, 0);
struct dpif_netdev *dpif;
- dp->open_cnt++;
+ ovs_refcount_ref(&dp->ref_cnt);
dpif = xmalloc(sizeof *dpif);
dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
* Return ODPP_NONE on failure. */
static odp_port_t
choose_port(struct dp_netdev *dp, const char *name)
+ OVS_REQ_RDLOCK(dp->port_rwlock)
{
uint32_t port_no;
- if (dp->class != &dpif_netdev_class) {
+ if (dp->class != &dpif_netdev_class &&
+ dp->class != &dpif_planetlab_class) {
const char *p;
int start_no = 0;
for (p = name; *p != '\0'; p++) {
if (isdigit((unsigned char) *p)) {
port_no = start_no + strtol(p, NULL, 10);
- if (port_no > 0 && port_no < MAX_PORTS
- && !dp->ports[port_no]) {
+ if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
+ && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
return u32_to_odp(port_no);
}
break;
}
}
- for (port_no = 1; port_no < MAX_PORTS; port_no++) {
- if (!dp->ports[port_no]) {
+ for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
+ if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
return u32_to_odp(port_no);
}
}
static int
create_dp_netdev(const char *name, const struct dpif_class *class,
struct dp_netdev **dpp)
+ OVS_REQUIRES(dp_netdev_mutex)
{
struct dp_netdev *dp;
int error;
int i;
dp = xzalloc(sizeof *dp);
- dp->class = class;
- dp->name = xstrdup(name);
- dp->open_cnt = 0;
- dp->max_mtu = ETH_PAYLOAD_MAX;
+ shash_add(&dp_netdevs, name, dp);
+
+ *CONST_CAST(const struct dpif_class **, &dp->class) = class;
+ *CONST_CAST(const char **, &dp->name) = xstrdup(name);
+ ovs_refcount_init(&dp->ref_cnt);
+ atomic_flag_init(&dp->destroyed);
+
+ ovs_mutex_init(&dp->flow_mutex);
+ classifier_init(&dp->cls, NULL);
+ hmap_init(&dp->flow_table);
+
+ ovs_mutex_init(&dp->queue_mutex);
+ ovs_mutex_lock(&dp->queue_mutex);
for (i = 0; i < N_QUEUES; i++) {
dp->queues[i].head = dp->queues[i].tail = 0;
}
+ ovs_mutex_unlock(&dp->queue_mutex);
dp->queue_seq = seq_create();
- classifier_init(&dp->cls, NULL);
- hmap_init(&dp->flow_table);
- list_init(&dp->port_list);
+
+ dp->n_hit = ovsthread_counter_create();
+ dp->n_missed = ovsthread_counter_create();
+ dp->n_lost = ovsthread_counter_create();
+
+ ovs_rwlock_init(&dp->port_rwlock);
+ hmap_init(&dp->ports);
dp->port_seq = seq_create();
+ latch_init(&dp->exit_latch);
+ ovs_rwlock_wrlock(&dp->port_rwlock);
error = do_add_port(dp, name, "internal", ODPP_LOCAL);
+ ovs_rwlock_unlock(&dp->port_rwlock);
if (error) {
dp_netdev_free(dp);
return error;
}
-
- shash_add(&dp_netdevs, name, dp);
+ dp_netdev_set_threads(dp, 2);
*dpp = dp;
return 0;
{
int i;
+ ovs_mutex_lock(&dp->queue_mutex);
for (i = 0; i < N_QUEUES; i++) {
struct dp_netdev_queue *q = &dp->queues[i];
while (q->tail != q->head) {
struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
+ ofpbuf_uninit(&u->upcall.packet);
ofpbuf_uninit(&u->buf);
}
}
+ ovs_mutex_unlock(&dp->queue_mutex);
}
+ /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp'
+ * through the 'dp_netdevs' shash while freeing 'dp'. */
static void
dp_netdev_free(struct dp_netdev *dp)
+ OVS_REQUIRES(dp_netdev_mutex)
{
struct dp_netdev_port *port, *next;
+ shash_find_and_delete(&dp_netdevs, dp->name);
+
+ dp_netdev_set_threads(dp, 0);
+ free(dp->forwarders);
+
dp_netdev_flow_flush(dp);
- LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
+ ovs_rwlock_wrlock(&dp->port_rwlock);
+ HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
do_del_port(dp, port->port_no);
}
+ ovs_rwlock_unlock(&dp->port_rwlock);
+ ovsthread_counter_destroy(dp->n_hit);
+ ovsthread_counter_destroy(dp->n_missed);
+ ovsthread_counter_destroy(dp->n_lost);
+
dp_netdev_purge_queues(dp);
seq_destroy(dp->queue_seq);
+ ovs_mutex_destroy(&dp->queue_mutex);
+
classifier_destroy(&dp->cls);
hmap_destroy(&dp->flow_table);
+ ovs_mutex_destroy(&dp->flow_mutex);
seq_destroy(dp->port_seq);
- free(dp->name);
+ hmap_destroy(&dp->ports);
+ atomic_flag_destroy(&dp->destroyed);
+ ovs_refcount_destroy(&dp->ref_cnt);
+ latch_destroy(&dp->exit_latch);
+ free(CONST_CAST(char *, dp->name));
free(dp);
}
+ static void
+ dp_netdev_unref(struct dp_netdev *dp)
+ {
+ if (dp) {
+ /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't
+ * get a new reference to 'dp' through the 'dp_netdevs' shash. */
+ ovs_mutex_lock(&dp_netdev_mutex);
+ if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
+ dp_netdev_free(dp);
+ }
+ ovs_mutex_unlock(&dp_netdev_mutex);
+ }
+ }
+
static void
dpif_netdev_close(struct dpif *dpif)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- ovs_mutex_lock(&dp_netdev_mutex);
-
- ovs_assert(dp->open_cnt > 0);
- if (--dp->open_cnt == 0 && dp->destroyed) {
- shash_find_and_delete(&dp_netdevs, dp->name);
- dp_netdev_free(dp);
- }
+ dp_netdev_unref(dp);
free(dpif);
-
- ovs_mutex_unlock(&dp_netdev_mutex);
}
static int
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- ovs_mutex_lock(&dp_netdev_mutex);
- dp->destroyed = true;
- ovs_mutex_unlock(&dp_netdev_mutex);
+ if (!atomic_flag_test_and_set(&dp->destroyed)) {
+ if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
+ /* Can't happen: 'dpif' still owns a reference to 'dp'. */
+ OVS_NOT_REACHED();
+ }
+ }
return 0;
}
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- ovs_mutex_lock(&dp_netdev_mutex);
+ fat_rwlock_rdlock(&dp->cls.rwlock);
stats->n_flows = hmap_count(&dp->flow_table);
- stats->n_hit = dp->n_hit;
- stats->n_missed = dp->n_missed;
- stats->n_lost = dp->n_lost;
- stats->n_masks = UINT64_MAX;
+ fat_rwlock_unlock(&dp->cls.rwlock);
+
+ stats->n_hit = ovsthread_counter_read(dp->n_hit);
+ stats->n_missed = ovsthread_counter_read(dp->n_missed);
+ stats->n_lost = ovsthread_counter_read(dp->n_lost);
+ stats->n_masks = UINT32_MAX;
stats->n_mask_hit = UINT64_MAX;
- ovs_mutex_unlock(&dp_netdev_mutex);
return 0;
}
static int
do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
odp_port_t port_no)
+ OVS_REQ_WRLOCK(dp->port_rwlock)
{
struct netdev_saved_flags *sf;
struct dp_netdev_port *port;
struct netdev_rx *rx;
enum netdev_flags flags;
const char *open_type;
- int mtu;
int error;
/* XXX reject devices already in some dp_netdev. */
port->rx = rx;
port->type = xstrdup(type);
- error = netdev_get_mtu(netdev, &mtu);
- if (!error && mtu > dp->max_mtu) {
- dp->max_mtu = mtu;
- }
-
- list_push_back(&dp->port_list, &port->node);
- dp->ports[odp_to_u32(port_no)] = port;
+ hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
seq_change(dp->port_seq);
return 0;
odp_port_t port_no;
int error;
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_rwlock_wrlock(&dp->port_rwlock);
dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
if (*port_nop != ODPP_NONE) {
- uint32_t port_idx = odp_to_u32(*port_nop);
- if (port_idx >= MAX_PORTS) {
- error = EFBIG;
- } else if (dp->ports[port_idx]) {
- error = EBUSY;
- } else {
- error = 0;
- port_no = *port_nop;
- }
+ port_no = *port_nop;
+ error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
} else {
port_no = choose_port(dp, dpif_port);
error = port_no == ODPP_NONE ? EFBIG : 0;
*port_nop = port_no;
error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_unlock(&dp->port_rwlock);
return error;
}
struct dp_netdev *dp = get_dp_netdev(dpif);
int error;
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_rwlock_wrlock(&dp->port_rwlock);
error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_unlock(&dp->port_rwlock);
return error;
}
static bool
is_valid_port_number(odp_port_t port_no)
{
- return odp_to_u32(port_no) < MAX_PORTS;
+ return port_no != ODPP_NONE;
+ }
+
+ static struct dp_netdev_port *
+ dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
+ OVS_REQ_RDLOCK(dp->port_rwlock)
+ {
+ struct dp_netdev_port *port;
+
+ HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
+ &dp->ports) {
+ if (port->port_no == port_no) {
+ return port;
+ }
+ }
+ return NULL;
}
static int
get_port_by_number(struct dp_netdev *dp,
odp_port_t port_no, struct dp_netdev_port **portp)
+ OVS_REQ_RDLOCK(dp->port_rwlock)
{
if (!is_valid_port_number(port_no)) {
*portp = NULL;
return EINVAL;
} else {
- *portp = dp->ports[odp_to_u32(port_no)];
+ *portp = dp_netdev_lookup_port(dp, port_no);
return *portp ? 0 : ENOENT;
}
}
static int
get_port_by_name(struct dp_netdev *dp,
const char *devname, struct dp_netdev_port **portp)
+ OVS_REQ_RDLOCK(dp->port_rwlock)
{
struct dp_netdev_port *port;
- LIST_FOR_EACH (port, node, &dp->port_list) {
+ HMAP_FOR_EACH (port, node, &dp->ports) {
if (!strcmp(netdev_get_name(port->netdev), devname)) {
*portp = port;
return 0;
static int
do_del_port(struct dp_netdev *dp, odp_port_t port_no)
+ OVS_REQ_WRLOCK(dp->port_rwlock)
{
struct dp_netdev_port *port;
int error;
return error;
}
- list_remove(&port->node);
- dp->ports[odp_to_u32(port_no)] = NULL;
+ hmap_remove(&dp->ports, &port->node);
seq_change(dp->port_seq);
netdev_close(port->netdev);
struct dp_netdev_port *port;
int error;
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_rwlock_rdlock(&dp->port_rwlock);
error = get_port_by_number(dp, port_no, &port);
if (!error && dpif_port) {
answer_port_query(port, dpif_port);
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_unlock(&dp->port_rwlock);
return error;
}
struct dp_netdev_port *port;
int error;
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_rwlock_rdlock(&dp->port_rwlock);
error = get_port_by_name(dp, devname, &port);
if (!error && dpif_port) {
answer_port_query(port, dpif_port);
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_unlock(&dp->port_rwlock);
return error;
}
- static uint32_t
- dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
+ static void
+ dp_netdev_remove_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
+ OVS_REQ_WRLOCK(dp->cls.rwlock)
+ OVS_REQUIRES(dp->flow_mutex)
{
- return MAX_PORTS;
+ struct cls_rule *cr = CONST_CAST(struct cls_rule *, &flow->cr);
+ struct hmap_node *node = CONST_CAST(struct hmap_node *, &flow->node);
+
+ classifier_remove(&dp->cls, cr);
+ hmap_remove(&dp->flow_table, node);
+ dp_netdev_flow_unref(flow);
}
- static void
- dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
+ static struct dp_netdev_flow *
+ dp_netdev_flow_ref(const struct dp_netdev_flow *flow_)
{
- ovs_rwlock_wrlock(&dp->cls.rwlock);
- classifier_remove(&dp->cls, &netdev_flow->cr);
- ovs_rwlock_unlock(&dp->cls.rwlock);
- cls_rule_destroy(&netdev_flow->cr);
+ struct dp_netdev_flow *flow = CONST_CAST(struct dp_netdev_flow *, flow_);
+ if (flow) {
+ ovs_refcount_ref(&flow->ref_cnt);
+ }
+ return flow;
+ }
- hmap_remove(&dp->flow_table, &netdev_flow->node);
- free(netdev_flow->actions);
- free(netdev_flow);
+ static void
+ dp_netdev_flow_unref(struct dp_netdev_flow *flow)
+ {
+ if (flow && ovs_refcount_unref(&flow->ref_cnt) == 1) {
+ cls_rule_destroy(CONST_CAST(struct cls_rule *, &flow->cr));
+ ovs_mutex_lock(&flow->mutex);
+ dp_netdev_actions_unref(flow->actions);
+ ovs_mutex_unlock(&flow->mutex);
+ ovs_mutex_destroy(&flow->mutex);
+ free(flow);
+ }
}
static void
{
struct dp_netdev_flow *netdev_flow, *next;
+ ovs_mutex_lock(&dp->flow_mutex);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
- dp_netdev_free_flow(dp, netdev_flow);
+ dp_netdev_remove_flow(dp, netdev_flow);
}
+ fat_rwlock_unlock(&dp->cls.rwlock);
+ ovs_mutex_unlock(&dp->flow_mutex);
}
static int
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- ovs_mutex_lock(&dp_netdev_mutex);
dp_netdev_flow_flush(dp);
- ovs_mutex_unlock(&dp_netdev_mutex);
-
return 0;
}
struct dp_netdev_port_state {
- odp_port_t port_no;
+ uint32_t bucket;
+ uint32_t offset;
char *name;
};
{
struct dp_netdev_port_state *state = state_;
struct dp_netdev *dp = get_dp_netdev(dpif);
- uint32_t port_idx;
+ struct hmap_node *node;
+ int retval;
- ovs_mutex_lock(&dp_netdev_mutex);
- for (port_idx = odp_to_u32(state->port_no);
- port_idx < MAX_PORTS; port_idx++) {
- struct dp_netdev_port *port = dp->ports[port_idx];
- if (port) {
- free(state->name);
- state->name = xstrdup(netdev_get_name(port->netdev));
- dpif_port->name = state->name;
- dpif_port->type = port->type;
- dpif_port->port_no = port->port_no;
- state->port_no = u32_to_odp(port_idx + 1);
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_rdlock(&dp->port_rwlock);
+ node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
+ if (node) {
+ struct dp_netdev_port *port;
- return 0;
- }
+ port = CONTAINER_OF(node, struct dp_netdev_port, node);
+
+ free(state->name);
+ state->name = xstrdup(netdev_get_name(port->netdev));
+ dpif_port->name = state->name;
+ dpif_port->type = port->type;
+ dpif_port->port_no = port->port_no;
+
+ retval = 0;
+ } else {
+ retval = EOF;
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_unlock(&dp->port_rwlock);
- return EOF;
+ return retval;
}
static int
uint64_t new_port_seq;
int error;
- ovs_mutex_lock(&dp_netdev_mutex);
new_port_seq = seq_read(dpif->dp->port_seq);
if (dpif->last_port_seq != new_port_seq) {
dpif->last_port_seq = new_port_seq;
} else {
error = EAGAIN;
}
- ovs_mutex_unlock(&dp_netdev_mutex);
return error;
}
{
struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
- ovs_mutex_lock(&dp_netdev_mutex);
seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
- ovs_mutex_unlock(&dp_netdev_mutex);
+ }
+
+ static struct dp_netdev_flow *
+ dp_netdev_flow_cast(const struct cls_rule *cr)
+ {
+ return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL;
}
static struct dp_netdev_flow *
dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
+ OVS_EXCLUDED(dp->cls.rwlock)
{
- struct cls_rule *cr;
+ struct dp_netdev_flow *netdev_flow;
- ovs_rwlock_wrlock(&dp->cls.rwlock);
- cr = classifier_lookup(&dp->cls, flow, NULL);
- ovs_rwlock_unlock(&dp->cls.rwlock);
+ fat_rwlock_rdlock(&dp->cls.rwlock);
+ netdev_flow = dp_netdev_flow_cast(classifier_lookup(&dp->cls, flow, NULL));
+ dp_netdev_flow_ref(netdev_flow);
+ fat_rwlock_unlock(&dp->cls.rwlock);
- return (cr
- ? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
- : NULL);
+ return netdev_flow;
}
static struct dp_netdev_flow *
dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
+ OVS_REQ_RDLOCK(dp->cls.rwlock)
{
struct dp_netdev_flow *netdev_flow;
HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
&dp->flow_table) {
if (flow_equal(&netdev_flow->flow, flow)) {
- return netdev_flow;
+ return dp_netdev_flow_ref(netdev_flow);
}
}
+
return NULL;
}
static void
get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
struct dpif_flow_stats *stats)
+ OVS_REQ_RDLOCK(netdev_flow->mutex)
{
stats->n_packets = netdev_flow->packet_count;
stats->n_bytes = netdev_flow->byte_count;
}
static int
- dpif_netdev_flow_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
- const struct nlattr *mask_key,
- uint32_t mask_key_len, struct flow *flow,
- struct flow *mask)
+ dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
+ const struct nlattr *mask_key,
+ uint32_t mask_key_len, const struct flow *flow,
+ struct flow *mask)
+ {
+ if (mask_key_len) {
+ if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
+ /* This should not happen: it indicates that
+ * odp_flow_key_from_mask() and odp_flow_key_to_mask()
+ * disagree on the acceptable form of a mask. Log the problem
+ * as an error, with enough details to enable debugging. */
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+ if (!VLOG_DROP_ERR(&rl)) {
+ struct ds s;
+
+ ds_init(&s);
+ odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
+ true);
+ VLOG_ERR("internal error parsing flow mask %s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+
+ return EINVAL;
+ }
+ /* Force unwildcard the in_port. */
+ mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
+ } else {
+ enum mf_field_id id;
+ /* No mask key, unwildcard everything except fields whose
+ * prerequisities are not met. */
+ memset(mask, 0x0, sizeof *mask);
+
+ for (id = 0; id < MFF_N_IDS; ++id) {
+ /* Skip registers and metadata. */
+ if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
+ && id != MFF_METADATA) {
+ const struct mf_field *mf = mf_from_id(id);
+ if (mf_are_prereqs_ok(mf, flow)) {
+ mf_mask_field(mf, mask);
+ }
+ }
+ }
+ }
+
+ return 0;
+ }
+
+ static int
+ dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
+ struct flow *flow)
{
odp_port_t in_port;
- if (odp_flow_key_to_flow(key, key_len, flow)
- || (mask_key
- && odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow))) {
+ if (odp_flow_key_to_flow(key, key_len, flow)) {
/* This should not happen: it indicates that odp_flow_key_from_flow()
- * and odp_flow_key_to_flow() disagree on the acceptable form of a flow
- * or odp_flow_key_from_mask() and odp_flow_key_to_mask() disagree on
- * the acceptable form of a mask. Log the problem as an error, with
- * enough details to enable debugging. */
+ * and odp_flow_key_to_flow() disagree on the acceptable form of a
+ * flow. Log the problem as an error, with enough details to enable
+ * debugging. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
if (!VLOG_DROP_ERR(&rl)) {
struct ds s;
ds_init(&s);
- odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
- true);
+ odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
ds_destroy(&s);
}
return EINVAL;
}
- if (mask_key) {
- /* Force unwildcard the in_port. */
- mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
- }
-
in_port = flow->in_port.odp_port;
if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
return EINVAL;
return 0;
}
- static int
- dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
- struct flow *flow)
- {
- return dpif_netdev_flow_mask_from_nlattrs(key, key_len, NULL, 0, flow,
- NULL);
- }
-
static int
dpif_netdev_flow_get(const struct dpif *dpif,
const struct nlattr *nl_key, size_t nl_key_len,
return error;
}
- ovs_mutex_lock(&dp_netdev_mutex);
+ fat_rwlock_rdlock(&dp->cls.rwlock);
netdev_flow = dp_netdev_find_flow(dp, &key);
+ fat_rwlock_unlock(&dp->cls.rwlock);
+
if (netdev_flow) {
+ struct dp_netdev_actions *actions = NULL;
+
+ ovs_mutex_lock(&netdev_flow->mutex);
if (stats) {
get_dpif_flow_stats(netdev_flow, stats);
}
if (actionsp) {
- *actionsp = ofpbuf_clone_data(netdev_flow->actions,
- netdev_flow->actions_len);
+ actions = dp_netdev_actions_ref(netdev_flow->actions);
+ }
+ ovs_mutex_unlock(&netdev_flow->mutex);
+
+ dp_netdev_flow_unref(netdev_flow);
+
+ if (actionsp) {
+ *actionsp = ofpbuf_clone_data(actions->actions, actions->size);
+ dp_netdev_actions_unref(actions);
}
} else {
error = ENOENT;
}
- ovs_mutex_unlock(&dp_netdev_mutex);
return error;
}
- static int
- set_flow_actions(struct dp_netdev_flow *netdev_flow,
- const struct nlattr *actions, size_t actions_len)
- {
- netdev_flow->actions = xrealloc(netdev_flow->actions, actions_len);
- netdev_flow->actions_len = actions_len;
- memcpy(netdev_flow->actions, actions, actions_len);
- return 0;
- }
-
static int
dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
const struct flow_wildcards *wc,
const struct nlattr *actions,
size_t actions_len)
+ OVS_REQUIRES(dp->flow_mutex)
{
struct dp_netdev_flow *netdev_flow;
struct match match;
- int error;
netdev_flow = xzalloc(sizeof *netdev_flow);
- netdev_flow->flow = *flow;
+ *CONST_CAST(struct flow *, &netdev_flow->flow) = *flow;
+ ovs_refcount_init(&netdev_flow->ref_cnt);
- match_init(&match, flow, wc);
- cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
- ovs_rwlock_wrlock(&dp->cls.rwlock);
- classifier_insert(&dp->cls, &netdev_flow->cr);
- ovs_rwlock_unlock(&dp->cls.rwlock);
+ ovs_mutex_init(&netdev_flow->mutex);
+ ovs_mutex_lock(&netdev_flow->mutex);
- error = set_flow_actions(netdev_flow, actions, actions_len);
- if (error) {
- ovs_rwlock_wrlock(&dp->cls.rwlock);
- classifier_remove(&dp->cls, &netdev_flow->cr);
- ovs_rwlock_unlock(&dp->cls.rwlock);
- cls_rule_destroy(&netdev_flow->cr);
+ netdev_flow->actions = dp_netdev_actions_create(actions, actions_len);
- free(netdev_flow);
- return error;
- }
+ match_init(&match, flow, wc);
+ cls_rule_init(CONST_CAST(struct cls_rule *, &netdev_flow->cr),
+ &match, NETDEV_RULE_PRIORITY);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
+ classifier_insert(&dp->cls,
+ CONST_CAST(struct cls_rule *, &netdev_flow->cr));
+ hmap_insert(&dp->flow_table,
+ CONST_CAST(struct hmap_node *, &netdev_flow->node),
+ flow_hash(flow, 0));
+ fat_rwlock_unlock(&dp->cls.rwlock);
+
+ ovs_mutex_unlock(&netdev_flow->mutex);
- hmap_insert(&dp->flow_table, &netdev_flow->node, flow_hash(flow, 0));
return 0;
}
static void
clear_stats(struct dp_netdev_flow *netdev_flow)
+ OVS_REQUIRES(netdev_flow->mutex)
{
netdev_flow->used = 0;
netdev_flow->packet_count = 0;
struct flow_wildcards wc;
int error;
- error = dpif_netdev_flow_mask_from_nlattrs(put->key, put->key_len,
- put->mask, put->mask_len, &flow, &wc.masks);
+ error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
+ if (error) {
+ return error;
+ }
+ error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
+ put->mask, put->mask_len,
+ &flow, &wc.masks);
if (error) {
return error;
}
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_mutex_lock(&dp->flow_mutex);
netdev_flow = dp_netdev_lookup_flow(dp, &flow);
if (!netdev_flow) {
if (put->flags & DPIF_FP_CREATE) {
} else {
if (put->flags & DPIF_FP_MODIFY
&& flow_equal(&flow, &netdev_flow->flow)) {
- error = set_flow_actions(netdev_flow, put->actions,
- put->actions_len);
- if (!error) {
- if (put->stats) {
- get_dpif_flow_stats(netdev_flow, put->stats);
- }
- if (put->flags & DPIF_FP_ZERO_STATS) {
- clear_stats(netdev_flow);
- }
+ struct dp_netdev_actions *new_actions;
+ struct dp_netdev_actions *old_actions;
+
+ new_actions = dp_netdev_actions_create(put->actions,
+ put->actions_len);
+
+ ovs_mutex_lock(&netdev_flow->mutex);
+ old_actions = netdev_flow->actions;
+ netdev_flow->actions = new_actions;
+ if (put->stats) {
+ get_dpif_flow_stats(netdev_flow, put->stats);
+ }
+ if (put->flags & DPIF_FP_ZERO_STATS) {
+ clear_stats(netdev_flow);
}
+ ovs_mutex_unlock(&netdev_flow->mutex);
+
+ dp_netdev_actions_unref(old_actions);
} else if (put->flags & DPIF_FP_CREATE) {
error = EEXIST;
} else {
/* Overlapping flow. */
error = EINVAL;
}
+ dp_netdev_flow_unref(netdev_flow);
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_mutex_unlock(&dp->flow_mutex);
return error;
}
return error;
}
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_mutex_lock(&dp->flow_mutex);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
netdev_flow = dp_netdev_find_flow(dp, &key);
if (netdev_flow) {
if (del->stats) {
+ ovs_mutex_lock(&netdev_flow->mutex);
get_dpif_flow_stats(netdev_flow, del->stats);
+ ovs_mutex_unlock(&netdev_flow->mutex);
}
- dp_netdev_free_flow(dp, netdev_flow);
+ dp_netdev_remove_flow(dp, netdev_flow);
} else {
error = ENOENT;
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ fat_rwlock_unlock(&dp->cls.rwlock);
+ ovs_mutex_unlock(&dp->flow_mutex);
return error;
}
struct dp_netdev_flow_state {
uint32_t bucket;
uint32_t offset;
- struct nlattr *actions;
+ struct dp_netdev_actions *actions;
struct odputil_keybuf keybuf;
struct odputil_keybuf maskbuf;
struct dpif_flow_stats stats;
struct dp_netdev_flow *netdev_flow;
struct hmap_node *node;
- ovs_mutex_lock(&dp_netdev_mutex);
+ fat_rwlock_rdlock(&dp->cls.rwlock);
node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
+ if (node) {
+ netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
+ dp_netdev_flow_ref(netdev_flow);
+ }
+ fat_rwlock_unlock(&dp->cls.rwlock);
if (!node) {
- ovs_mutex_unlock(&dp_netdev_mutex);
return EOF;
}
- netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
-
if (key) {
struct ofpbuf buf;
*mask_len = buf.size;
}
- if (actions) {
- free(state->actions);
- state->actions = xmemdup(netdev_flow->actions,
- netdev_flow->actions_len);
+ if (actions || stats) {
+ dp_netdev_actions_unref(state->actions);
+ state->actions = NULL;
- *actions = state->actions;
- *actions_len = netdev_flow->actions_len;
+ ovs_mutex_lock(&netdev_flow->mutex);
+ if (actions) {
+ state->actions = dp_netdev_actions_ref(netdev_flow->actions);
+ *actions = state->actions->actions;
+ *actions_len = state->actions->size;
+ }
+ if (stats) {
+ get_dpif_flow_stats(netdev_flow, &state->stats);
+ *stats = &state->stats;
+ }
+ ovs_mutex_unlock(&netdev_flow->mutex);
}
- if (stats) {
- get_dpif_flow_stats(netdev_flow, &state->stats);
- *stats = &state->stats;
- }
+ dp_netdev_flow_unref(netdev_flow);
- ovs_mutex_unlock(&dp_netdev_mutex);
return 0;
}
{
struct dp_netdev_flow_state *state = state_;
- free(state->actions);
+ dp_netdev_actions_unref(state->actions);
free(state);
return 0;
}
static int
- dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
+ dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- struct flow md;
- int error;
+ struct pkt_metadata *md = &execute->md;
+ struct flow key;
if (execute->packet->size < ETH_HEADER_LEN ||
execute->packet->size > UINT16_MAX) {
return EINVAL;
}
- /* Get packet metadata. */
- error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, &md);
- if (!error) {
- struct ofpbuf *copy;
- struct flow key;
+ /* Extract flow key. */
+ flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
+ (union flow_in_port *)&md->in_port, &key);
- /* Make a deep copy of 'packet', because we might modify its data. */
- copy = ofpbuf_clone_with_headroom(execute->packet, DP_NETDEV_HEADROOM);
+ ovs_rwlock_rdlock(&dp->port_rwlock);
+ dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
+ execute->actions_len);
+ ovs_rwlock_unlock(&dp->port_rwlock);
- /* Extract flow key. */
- flow_extract(copy, md.skb_priority, md.pkt_mark, &md.tunnel,
- &md.in_port, &key);
- ovs_mutex_lock(&dp_netdev_mutex);
- dp_netdev_execute_actions(dp, &key, copy,
- execute->actions, execute->actions_len);
- ovs_mutex_unlock(&dp_netdev_mutex);
- ofpbuf_delete(copy);
- }
- return error;
+ return 0;
}
static int
}
static struct dp_netdev_queue *
- find_nonempty_queue(struct dpif *dpif)
+ find_nonempty_queue(struct dp_netdev *dp)
+ OVS_REQUIRES(dp->queue_mutex)
{
- struct dp_netdev *dp = get_dp_netdev(dpif);
int i;
for (i = 0; i < N_QUEUES; i++) {
dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
struct ofpbuf *buf)
{
+ struct dp_netdev *dp = get_dp_netdev(dpif);
struct dp_netdev_queue *q;
int error;
- ovs_mutex_lock(&dp_netdev_mutex);
- q = find_nonempty_queue(dpif);
+ ovs_mutex_lock(&dp->queue_mutex);
+ q = find_nonempty_queue(dp);
if (q) {
struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
*upcall = u->upcall;
- upcall->packet = buf;
ofpbuf_uninit(buf);
*buf = u->buf;
} else {
error = EAGAIN;
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_mutex_unlock(&dp->queue_mutex);
return error;
}
struct dp_netdev *dp = get_dp_netdev(dpif);
uint64_t seq;
- ovs_mutex_lock(&dp_netdev_mutex);
+ ovs_mutex_lock(&dp->queue_mutex);
seq = seq_read(dp->queue_seq);
- if (find_nonempty_queue(dpif)) {
+ if (find_nonempty_queue(dp)) {
poll_immediate_wake();
} else {
seq_wait(dp->queue_seq, seq);
}
- ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_mutex_unlock(&dp->queue_mutex);
}
static void
dpif_netdev_recv_purge(struct dpif *dpif)
{
struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
- ovs_mutex_lock(&dp_netdev_mutex);
+
dp_netdev_purge_queues(dpif_netdev->dp);
- ovs_mutex_unlock(&dp_netdev_mutex);
}
\f
- static void
- dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
- const struct ofpbuf *packet)
+ /* Creates and returns a new 'struct dp_netdev_actions', with a reference count
+ * of 1, whose actions are a copy of from the 'ofpacts_len' bytes of
+ * 'ofpacts'. */
+ struct dp_netdev_actions *
+ dp_netdev_actions_create(const struct nlattr *actions, size_t size)
{
- netdev_flow->used = time_msec();
- netdev_flow->packet_count++;
- netdev_flow->byte_count += packet->size;
- netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
+ struct dp_netdev_actions *netdev_actions;
+
+ netdev_actions = xmalloc(sizeof *netdev_actions);
+ ovs_refcount_init(&netdev_actions->ref_cnt);
+ netdev_actions->actions = xmemdup(actions, size);
+ netdev_actions->size = size;
+
+ return netdev_actions;
}
- static void
- dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
- struct ofpbuf *packet, uint32_t skb_priority,
- uint32_t pkt_mark, const struct flow_tnl *tnl)
+ /* Increments 'actions''s refcount. */
+ struct dp_netdev_actions *
+ dp_netdev_actions_ref(const struct dp_netdev_actions *actions_)
{
- struct dp_netdev_flow *netdev_flow;
- struct flow key;
- union flow_in_port in_port_;
+ struct dp_netdev_actions *actions;
- if (packet->size < ETH_HEADER_LEN) {
- return;
- }
- in_port_.odp_port = port->port_no;
- flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
- netdev_flow = dp_netdev_lookup_flow(dp, &key);
- if (netdev_flow) {
- dp_netdev_flow_used(netdev_flow, packet);
- dp_netdev_execute_actions(dp, &key, packet,
- netdev_flow->actions,
- netdev_flow->actions_len);
- dp->n_hit++;
- } else {
- dp->n_missed++;
- dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
+ actions = CONST_CAST(struct dp_netdev_actions *, actions_);
+ if (actions) {
+ ovs_refcount_ref(&actions->ref_cnt);
}
+ return actions;
}
- static void
- dpif_netdev_run(struct dpif *dpif)
+ /* Decrements 'actions''s refcount and frees 'actions' if the refcount reaches
+ * 0. */
+ void
+ dp_netdev_actions_unref(struct dp_netdev_actions *actions)
{
- struct dp_netdev_port *port;
- struct dp_netdev *dp;
+ if (actions && ovs_refcount_unref(&actions->ref_cnt) == 1) {
+ free(actions->actions);
+ free(actions);
+ }
+ }
+ \f
+ static void *
+ dp_forwarder_main(void *f_)
+ {
+ struct dp_forwarder *f = f_;
+ struct dp_netdev *dp = f->dp;
struct ofpbuf packet;
- ovs_mutex_lock(&dp_netdev_mutex);
- dp = get_dp_netdev(dpif);
- ofpbuf_init(&packet,
- DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
-
- LIST_FOR_EACH (port, node, &dp->port_list) {
- int error;
+ f->name = xasprintf("forwarder_%u", ovsthread_id_self());
+ set_subprogram_name("%s", f->name);
+
+ ofpbuf_init(&packet, 0);
+ while (!latch_is_set(&dp->exit_latch)) {
+ bool received_anything;
+ int i;
+
+ ovs_rwlock_rdlock(&dp->port_rwlock);
+ for (i = 0; i < 50; i++) {
+ struct dp_netdev_port *port;
+
+ received_anything = false;
+ HMAP_FOR_EACH (port, node, &f->dp->ports) {
+ if (port->rx
+ && port->node.hash >= f->min_hash
+ && port->node.hash <= f->max_hash) {
+ int buf_size;
+ int error;
+ int mtu;
+
+ if (netdev_get_mtu(port->netdev, &mtu)) {
+ mtu = ETH_PAYLOAD_MAX;
+ }
+ buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + mtu;
+
+ ofpbuf_clear(&packet);
+ ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM,
+ buf_size);
+
+ error = netdev_rx_recv(port->rx, &packet);
+ if (!error) {
+ struct pkt_metadata md
+ = PKT_METADATA_INITIALIZER(port->port_no);
+ dp_netdev_port_input(dp, &packet, &md);
+
+ received_anything = true;
+ } else if (error != EAGAIN && error != EOPNOTSUPP) {
+ static struct vlog_rate_limit rl
+ = VLOG_RATE_LIMIT_INIT(1, 5);
+
+ VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
+ netdev_get_name(port->netdev),
+ ovs_strerror(error));
+ }
+ }
+ }
- /* Reset packet contents. */
- ofpbuf_clear(&packet);
- ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
+ if (!received_anything) {
+ break;
+ }
+ }
- error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
- if (!error) {
- dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
- } else if (error != EAGAIN && error != EOPNOTSUPP) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ if (received_anything) {
+ poll_immediate_wake();
+ } else {
+ struct dp_netdev_port *port;
- VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
- netdev_get_name(port->netdev), ovs_strerror(error));
+ HMAP_FOR_EACH (port, node, &f->dp->ports)
+ if (port->rx
+ && port->node.hash >= f->min_hash
+ && port->node.hash <= f->max_hash) {
+ netdev_rx_wait(port->rx);
+ }
+ seq_wait(dp->port_seq, seq_read(dp->port_seq));
+ latch_wait(&dp->exit_latch);
}
+ ovs_rwlock_unlock(&dp->port_rwlock);
+
+ poll_block();
}
ofpbuf_uninit(&packet);
- ovs_mutex_unlock(&dp_netdev_mutex);
+
+ free(f->name);
+
+ return NULL;
}
static void
- dpif_netdev_wait(struct dpif *dpif)
+ dp_netdev_set_threads(struct dp_netdev *dp, int n)
{
- struct dp_netdev_port *port;
+ int i;
- /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
- * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
- * A makes it to poll_block().
- *
- * But I think it doesn't matter:
- *
- * - In the dpif_port_add() case, A will not wake up when a packet
- * arrives on the new port, but this would also happen if the
- * ordering were reversed.
- *
- * - In the dpif_port_remove() case, A might wake up spuriously, but
- * that is harmless. */
+ if (n == dp->n_forwarders) {
+ return;
+ }
- ovs_mutex_lock(&dp_netdev_mutex);
- LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
- if (port->rx) {
- netdev_rx_wait(port->rx);
+ /* Stop existing threads. */
+ latch_set(&dp->exit_latch);
+ for (i = 0; i < dp->n_forwarders; i++) {
+ struct dp_forwarder *f = &dp->forwarders[i];
+
+ xpthread_join(f->thread, NULL);
+ }
+ latch_poll(&dp->exit_latch);
+ free(dp->forwarders);
+
+ /* Start new threads. */
+ dp->forwarders = xmalloc(n * sizeof *dp->forwarders);
+ dp->n_forwarders = n;
+ for (i = 0; i < n; i++) {
+ struct dp_forwarder *f = &dp->forwarders[i];
+
+ f->dp = dp;
+ f->min_hash = UINT32_MAX / n * i;
+ f->max_hash = UINT32_MAX / n * (i + 1) - 1;
+ if (i == n - 1) {
+ f->max_hash = UINT32_MAX;
}
+ xpthread_create(&f->thread, NULL, dp_forwarder_main, f);
+ }
+ }
+ \f
+ static void
+ dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
+ const struct ofpbuf *packet)
+ OVS_REQUIRES(netdev_flow->mutex)
+ {
+ netdev_flow->used = time_msec();
+ netdev_flow->packet_count++;
+ netdev_flow->byte_count += packet->size;
+ netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
+ }
+
+ static void
+ dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
+ struct pkt_metadata *md)
+ OVS_REQ_RDLOCK(dp->port_rwlock)
+ {
+ struct dp_netdev_flow *netdev_flow;
+ struct flow key;
+
+ if (packet->size < ETH_HEADER_LEN) {
+ return;
+ }
+ flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
+ (union flow_in_port *)&md->in_port, &key);
+ netdev_flow = dp_netdev_lookup_flow(dp, &key);
+ if (netdev_flow) {
+ struct dp_netdev_actions *actions;
+
+ ovs_mutex_lock(&netdev_flow->mutex);
+ dp_netdev_flow_used(netdev_flow, packet);
+ actions = dp_netdev_actions_ref(netdev_flow->actions);
+ ovs_mutex_unlock(&netdev_flow->mutex);
+
+ dp_netdev_execute_actions(dp, &key, packet, md,
+ actions->actions, actions->size);
+ dp_netdev_actions_unref(actions);
+ ovsthread_counter_inc(dp->n_hit, 1);
+ } else {
+ ovsthread_counter_inc(dp->n_missed, 1);
+ dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
}
- ovs_mutex_unlock(&dp_netdev_mutex);
}
static int
- dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
+ dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
int queue_no, const struct flow *flow,
const struct nlattr *userdata)
+ OVS_EXCLUDED(dp->queue_mutex)
{
struct dp_netdev_queue *q = &dp->queues[queue_no];
+ int error;
+
+ ovs_mutex_lock(&dp->queue_mutex);
if (q->head - q->tail < MAX_QUEUE_LEN) {
struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
struct dpif_upcall *upcall = &u->upcall;
upcall->type = queue_no;
/* Allocate buffer big enough for everything. */
- buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
+ buf_size = ODPUTIL_FLOW_KEY_BYTES;
if (userdata) {
buf_size += NLA_ALIGN(userdata->nla_len);
}
NLA_ALIGN(userdata->nla_len));
}
- /* Put packet.
- *
- * We adjust 'data' and 'size' in 'buf' so that only the packet itself
- * is visible in 'upcall->packet'. The ODP flow and (if present)
- * userdata become part of the headroom. */
- ofpbuf_put_zeros(buf, 2);
- buf->data = ofpbuf_put(buf, packet->data, packet->size);
- buf->size = packet->size;
- upcall->packet = buf;
+ /* Steal packet data. */
+ ovs_assert(packet->source == OFPBUF_MALLOC);
+ upcall->packet = *packet;
+ ofpbuf_use(packet, NULL, 0);
seq_change(dp->queue_seq);
- return 0;
+ error = 0;
} else {
- dp->n_lost++;
- return ENOBUFS;
+ ovsthread_counter_inc(dp->n_lost, 1);
+ error = ENOBUFS;
}
+ ovs_mutex_unlock(&dp->queue_mutex);
+
+ return error;
}
struct dp_netdev_execute_aux {
};
static void
- dp_netdev_action_output(void *aux_, struct ofpbuf *packet,
- const struct flow *flow OVS_UNUSED,
- odp_port_t out_port)
+ dp_execute_cb(void *aux_, struct ofpbuf *packet,
+ const struct pkt_metadata *md OVS_UNUSED,
+ const struct nlattr *a, bool may_steal)
+ OVS_NO_THREAD_SAFETY_ANALYSIS
{
struct dp_netdev_execute_aux *aux = aux_;
- struct dp_netdev_port *p = aux->dp->ports[odp_to_u32(out_port)];
- if (p) {
- netdev_send(p->netdev, packet);
- }
- }
+ int type = nl_attr_type(a);
+ struct dp_netdev_port *p;
+
+ switch ((enum ovs_action_attr)type) {
+ case OVS_ACTION_ATTR_OUTPUT:
+ p = dp_netdev_lookup_port(aux->dp, u32_to_odp(nl_attr_get_u32(a)));
+ if (p) {
+ netdev_send(p->netdev, packet);
+ }
+ break;
- static void
- dp_netdev_action_userspace(void *aux_, struct ofpbuf *packet,
- const struct flow *flow OVS_UNUSED,
- const struct nlattr *a)
- {
- struct dp_netdev_execute_aux *aux = aux_;
- const struct nlattr *userdata;
+ case OVS_ACTION_ATTR_USERSPACE: {
+ const struct nlattr *userdata;
+
+ userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
- userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
- dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
- userdata);
+ /* Make a copy if we are not allowed to steal the packet's data. */
+ if (!may_steal) {
+ packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
+ }
+ dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
+ userdata);
+ if (!may_steal) {
+ ofpbuf_uninit(packet);
+ }
+ 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_MAX:
+ OVS_NOT_REACHED();
+ }
}
static void
dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
- struct ofpbuf *packet,
+ struct ofpbuf *packet, struct pkt_metadata *md,
const struct nlattr *actions, size_t actions_len)
+ OVS_REQ_RDLOCK(dp->port_rwlock)
{
struct dp_netdev_execute_aux aux = {dp, key};
- struct flow md = *key; /* Packet metadata, may be modified by actions. */
- odp_execute_actions(&aux, packet, &md, actions, actions_len,
- dp_netdev_action_output, dp_netdev_action_userspace);
+ odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
}
- dpif_netdev_run, \
- dpif_netdev_wait, \
+#define DPIF_NETDEV_CLASS_FUNCTIONS \
+ dpif_netdev_enumerate, \
+ dpif_netdev_port_open_type, \
+ dpif_netdev_open, \
+ dpif_netdev_close, \
+ dpif_netdev_destroy, \
- dpif_netdev_get_max_ports, \
++ NULL, \
++ NULL, \
+ dpif_netdev_get_stats, \
+ dpif_netdev_port_add, \
+ dpif_netdev_port_del, \
+ dpif_netdev_port_query_by_number, \
+ dpif_netdev_port_query_by_name, \
+ NULL, /* port_get_pid */ \
+ dpif_netdev_port_dump_start, \
+ dpif_netdev_port_dump_next, \
+ dpif_netdev_port_dump_done, \
+ dpif_netdev_port_poll, \
+ dpif_netdev_port_poll_wait, \
+ dpif_netdev_flow_get, \
+ dpif_netdev_flow_put, \
+ dpif_netdev_flow_del, \
+ dpif_netdev_flow_flush, \
+ dpif_netdev_flow_dump_start, \
+ dpif_netdev_flow_dump_next, \
+ dpif_netdev_flow_dump_done, \
+ dpif_netdev_execute, \
+ NULL, /* operate */ \
+ dpif_netdev_recv_set, \
+ dpif_netdev_queue_to_priority, \
+ dpif_netdev_recv, \
+ dpif_netdev_recv_wait, \
+ dpif_netdev_recv_purge, \
+
const struct dpif_class dpif_netdev_class = {
"netdev",
- dpif_netdev_enumerate,
- dpif_netdev_port_open_type,
- dpif_netdev_open,
- dpif_netdev_close,
- dpif_netdev_destroy,
- NULL, /* run */
- NULL, /* wait */
- dpif_netdev_get_stats,
- dpif_netdev_port_add,
- dpif_netdev_port_del,
- dpif_netdev_port_query_by_number,
- dpif_netdev_port_query_by_name,
- NULL, /* port_get_pid */
- dpif_netdev_port_dump_start,
- dpif_netdev_port_dump_next,
- dpif_netdev_port_dump_done,
- dpif_netdev_port_poll,
- dpif_netdev_port_poll_wait,
- dpif_netdev_flow_get,
- dpif_netdev_flow_put,
- dpif_netdev_flow_del,
- dpif_netdev_flow_flush,
- dpif_netdev_flow_dump_start,
- dpif_netdev_flow_dump_next,
- dpif_netdev_flow_dump_done,
- dpif_netdev_execute,
- NULL, /* operate */
- dpif_netdev_recv_set,
- dpif_netdev_queue_to_priority,
- dpif_netdev_recv,
- dpif_netdev_recv_wait,
- dpif_netdev_recv_purge,
+ DPIF_NETDEV_CLASS_FUNCTIONS
+};
+
+const struct dpif_class dpif_planetlab_class = {
+ "planetlab",
+ DPIF_NETDEV_CLASS_FUNCTIONS
};
static void
{
struct dp_netdev_port *port;
struct dp_netdev *dp;
- int port_no;
+ odp_port_t port_no;
+ ovs_mutex_lock(&dp_netdev_mutex);
dp = shash_find_data(&dp_netdevs, argv[1]);
if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
+ ovs_mutex_unlock(&dp_netdev_mutex);
unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
return;
}
+ ovs_refcount_ref(&dp->ref_cnt);
+ ovs_mutex_unlock(&dp_netdev_mutex);
+ ovs_rwlock_wrlock(&dp->port_rwlock);
if (get_port_by_name(dp, argv[2], &port)) {
unixctl_command_reply_error(conn, "unknown port");
- return;
+ goto exit;
}
- port_no = atoi(argv[3]);
- if (port_no <= 0 || port_no >= MAX_PORTS) {
+ port_no = u32_to_odp(atoi(argv[3]));
+ if (!port_no || port_no == ODPP_NONE) {
unixctl_command_reply_error(conn, "bad port number");
- return;
+ goto exit;
}
- if (dp->ports[port_no]) {
+ if (dp_netdev_lookup_port(dp, port_no)) {
unixctl_command_reply_error(conn, "port number already in use");
- return;
+ goto exit;
}
- dp->ports[odp_to_u32(port->port_no)] = NULL;
- dp->ports[port_no] = port;
- port->port_no = u32_to_odp(port_no);
+ hmap_remove(&dp->ports, &port->node);
+ port->port_no = port_no;
+ hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
seq_change(dp->port_seq);
unixctl_command_reply(conn, NULL);
+
+ exit:
+ ovs_rwlock_unlock(&dp->port_rwlock);
+ dp_netdev_unref(dp);
}
static void
"DP PORT NEW-NUMBER",
3, 3, dpif_dummy_change_port_number, NULL);
}
+
int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
struct dpif_port *port);
- /* Returns one greater than the largest port number accepted in flow
- * actions. */
- uint32_t (*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'.
int (*flow_dump_done)(const struct dpif *dpif, void *state);
/* Performs the 'execute->actions_len' bytes of actions in
- * 'execute->actions' on the Ethernet frame specified in 'execute->packet'
- * taken from the flow specified in the 'execute->key_len' bytes of
- * 'execute->key'. ('execute->key' is mostly redundant with
- * 'execute->packet', but it contains some metadata that cannot be
- * recovered from 'execute->packet', such as tunnel and in_port.) */
- int (*execute)(struct dpif *dpif, const struct dpif_execute *execute);
+ * 'execute->actions' on the Ethernet frame in 'execute->packet'
+ * and on the packet metadata in 'execute->md'.
+ * May modify both packet and metadata. */
+ int (*execute)(struct dpif *dpif, struct dpif_execute *execute);
/* 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
* '*upcall', using 'buf' for storage. Should only be called if 'recv_set'
* has been used to enable receiving packets from 'dpif'.
*
- * The implementation should point 'upcall->packet' and 'upcall->key' into
- * data in the caller-provided 'buf'. If necessary to make room, the
- * implementation may expand the data in 'buf'. (This is hardly a great
- * way to do things but it works out OK for the dpif providers that exist
- * so far.)
+ * The implementation should point 'upcall->key' and 'upcall->userdata'
+ * (if any) into data in the caller-provided 'buf'. The implementation may
+ * also use 'buf' for storing the data of 'upcall->packet'. If necessary
+ * to make room, the implementation may reallocate the data in 'buf'.
+ *
+ * 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.
*
* This function must not block. If no upcall is pending when it is
* called, it should return EAGAIN without blocking. */
extern const struct dpif_class dpif_linux_class;
extern const struct dpif_class dpif_netdev_class;
+extern const struct dpif_class dpif_planetlab_class;
#ifdef __cplusplus
}
&dpif_linux_class,
#endif
&dpif_netdev_class,
+ &dpif_planetlab_class,
};
struct registered_dpif_class {
int error;
};
+ /* This is called for actions that need the context of the datapath to be
+ * meaningful. */
static void
- dpif_execute_helper_execute__(void *aux_, struct ofpbuf *packet,
- const struct flow *flow,
- const struct nlattr *actions, size_t actions_len)
+ dpif_execute_helper_cb(void *aux_, struct ofpbuf *packet,
+ const struct pkt_metadata *md,
+ const struct nlattr *action, bool may_steal OVS_UNUSED)
{
struct dpif_execute_helper_aux *aux = aux_;
struct dpif_execute execute;
- struct odputil_keybuf key_stub;
- struct ofpbuf key;
- int error;
-
- ofpbuf_use_stub(&key, &key_stub, sizeof key_stub);
- odp_flow_key_from_flow(&key, flow, flow->in_port.odp_port);
-
- execute.key = key.data;
- execute.key_len = key.size;
- execute.actions = actions;
- execute.actions_len = actions_len;
- execute.packet = packet;
- execute.needs_help = false;
-
- error = aux->dpif->dpif_class->execute(aux->dpif, &execute);
- if (error) {
- aux->error = error;
+ 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_MAX:
+ OVS_NOT_REACHED();
}
}
- static void
- dpif_execute_helper_output_cb(void *aux, struct ofpbuf *packet,
- const struct flow *flow, odp_port_t out_port)
- {
- uint64_t actions_stub[DIV_ROUND_UP(NL_A_U32_SIZE, 8)];
- struct ofpbuf actions;
-
- ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
- nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, odp_to_u32(out_port));
-
- dpif_execute_helper_execute__(aux, packet, flow,
- actions.data, actions.size);
- }
-
- static void
- dpif_execute_helper_userspace_cb(void *aux, struct ofpbuf *packet,
- const struct flow *flow,
- const struct nlattr *action)
- {
- dpif_execute_helper_execute__(aux, packet, flow,
- action, NLA_ALIGN(action->nla_len));
- }
-
/* 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, const struct dpif_execute *execute)
+ dpif_execute_with_help(struct dpif *dpif, struct dpif_execute *execute)
{
- struct dpif_execute_helper_aux aux;
- enum odp_key_fitness fit;
- struct ofpbuf *packet;
- struct flow flow;
+ struct dpif_execute_helper_aux aux = {dpif, 0};
COVERAGE_INC(dpif_execute_with_help);
- fit = odp_flow_key_to_flow(execute->key, execute->key_len, &flow);
- if (fit == ODP_FIT_ERROR) {
- return EINVAL;
- }
-
- aux.dpif = dpif;
- aux.error = 0;
-
- packet = ofpbuf_clone_with_headroom(execute->packet, VLAN_HEADER_LEN);
- odp_execute_actions(&aux, packet, &flow,
+ odp_execute_actions(&aux, execute->packet, &execute->md,
execute->actions, execute->actions_len,
- dpif_execute_helper_output_cb,
- dpif_execute_helper_userspace_cb);
- ofpbuf_delete(packet);
-
+ dpif_execute_helper_cb);
return aux.error;
}
- static int
- dpif_execute__(struct dpif *dpif, const struct dpif_execute *execute)
+ /* 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, struct dpif_execute *execute)
{
int error;
COVERAGE_INC(dpif_execute);
if (execute->actions_len > 0) {
- error = (execute->needs_help
+ error = (execute->needs_help || nl_attr_oversized(execute->actions_len)
? dpif_execute_with_help(dpif, execute)
: dpif->dpif_class->execute(dpif, execute));
} else {
return error;
}
- /* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on
- * the Ethernet frame specified in 'packet' taken from the flow specified in
- * the 'key_len' bytes of 'key'. ('key' is mostly redundant with 'packet', but
- * it contains some metadata that cannot be recovered from 'packet', such as
- * tunnel and in_port.)
- *
- * 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 '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 *key, size_t key_len,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *buf,
- bool needs_help)
- {
- struct dpif_execute execute;
-
- execute.key = key;
- execute.key_len = key_len;
- execute.actions = actions;
- execute.actions_len = actions_len;
- execute.packet = buf;
- execute.needs_help = needs_help || nl_attr_oversized(actions_len);
- return dpif_execute__(dpif, &execute);
- }
-
/* 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.
/* Help the dpif provider to execute one op. */
struct dpif_op *op = ops[0];
- op->error = dpif_execute__(dpif, &op->u.execute);
+ op->error = dpif_execute(dpif, &op->u.execute);
ops++;
n_ops--;
}
break;
case DPIF_OP_EXECUTE:
- op->error = dpif_execute__(dpif, &op->u.execute);
+ op->error = dpif_execute(dpif, &op->u.execute);
break;
default:
- NOT_REACHED();
+ OVS_NOT_REACHED();
}
}
}
* '*upcall', using 'buf' for storage. Should only be called if
* dpif_recv_set() has been used to enable receiving packets on 'dpif'.
*
- * 'upcall->packet' and 'upcall->key' point into data in the caller-provided
- * 'buf', so their memory cannot be freed separately from 'buf'. (This is
- * hardly a great way to do things but it works out OK for the dpif providers
- * and clients that exist so far.)
+ * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
+ * 'buf', so their memory cannot be freed separately from 'buf'.
+ *
+ * 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. */
struct ds flow;
char *packet;
- packet = ofp_packet_to_string(upcall->packet->data,
- upcall->packet->size);
+ packet = ofp_packet_to_string(upcall->packet.data,
+ upcall->packet.size);
ds_init(&flow);
odp_flow_key_format(upcall->key, upcall->key_len, &flow);
--- /dev/null
- netdev_pltap_rx_recv(struct netdev_rx *rx_, void *buffer, size_t size)
+/*
+ * Copyright (c) 2012 Giuseppe Lettieri
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+
+#include <sys/types.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <arpa/inet.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <net/if.h>
+#include <net/if_arp.h>
+#include <linux/if_tun.h>
+#include <netinet/in.h>
+#include <errno.h>
+
+#include "flow.h"
+#include "list.h"
+#include "netdev-provider.h"
+#include "odp-util.h"
+#include "ofp-print.h"
+#include "ofpbuf.h"
+#include "packets.h"
+#include "poll-loop.h"
+#include "shash.h"
+#include "sset.h"
+#include "unixctl.h"
+#include "socket-util.h"
+#include "vlog.h"
+#include "tunalloc.h"
+
+VLOG_DEFINE_THIS_MODULE(netdev_pltap);
+
+/* Protects 'sync_list'. */
+static struct ovs_mutex sync_list_mutex = OVS_MUTEX_INITIALIZER;
+
+static struct list sync_list OVS_GUARDED_BY(sync_list_mutex)
+ = LIST_INITIALIZER(&sync_list);
+
+struct netdev_pltap {
+ struct netdev up;
+
+ /* In sync_list. */
+ struct list sync_list OVS_GUARDED_BY(sync_list_mutex);
+
+ /* Protects all members below. */
+ struct ovs_mutex mutex OVS_ACQ_AFTER(sync_list_mutex);
+
+ char *real_name;
+ struct netdev_stats stats;
+ enum netdev_flags new_flags;
+ enum netdev_flags flags;
+ int fd;
+ struct sockaddr_in local_addr;
+ int local_netmask;
+ bool valid_local_ip;
+ bool valid_local_netmask;
+ bool sync_flags_needed;
+ unsigned int change_seq;
+};
+
+
+struct netdev_rx_pltap {
+ struct netdev_rx up;
+ int fd;
+};
+
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+
+/* Protects 'pltap_netdevs' */
+static struct ovs_mutex pltap_netdevs_mutex = OVS_MUTEX_INITIALIZER;
+static struct shash pltap_netdevs OVS_GUARDED_BY(pltap_netdevs_mutex)
+ = SHASH_INITIALIZER(&pltap_netdevs);
+
+static int netdev_pltap_construct(struct netdev *netdev_);
+
+static void netdev_pltap_update_seq(struct netdev_pltap *)
+ OVS_REQUIRES(dev->mutex);
+static int get_flags(struct netdev_pltap *dev, enum netdev_flags *flags)
+ OVS_REQUIRES(dev->mutex);
+
+static bool
+netdev_pltap_finalized(struct netdev_pltap *dev)
+ OVS_REQUIRES(dev->mutex)
+{
+ return dev->valid_local_ip && dev->valid_local_netmask;
+}
+
+static bool
+is_netdev_pltap_class(const struct netdev_class *class)
+{
+ return class->construct == netdev_pltap_construct;
+}
+
+static struct netdev_pltap *
+netdev_pltap_cast(const struct netdev *netdev)
+{
+ ovs_assert(is_netdev_pltap_class(netdev_get_class(netdev)));
+ return CONTAINER_OF(netdev, struct netdev_pltap, up);
+}
+
+static struct netdev_rx_pltap*
+netdev_rx_pltap_cast(const struct netdev_rx *rx)
+{
+ ovs_assert(is_netdev_pltap_class(netdev_get_class(rx->netdev)));
+ return CONTAINER_OF(rx, struct netdev_rx_pltap, up);
+}
+
+static void sync_needed(struct netdev_pltap *dev)
+ OVS_REQUIRES(dev->mutex, sync_list_mutex)
+{
+ if (dev->sync_flags_needed)
+ return;
+
+ dev->sync_flags_needed = true;
+ list_insert(&sync_list, &dev->sync_list);
+}
+
+static void sync_done(struct netdev_pltap *dev)
+ OVS_REQUIRES(dev->mutex, sync_list_mutex)
+{
+ if (!dev->sync_flags_needed)
+ return;
+
+ (void) list_remove(&dev->sync_list);
+ dev->sync_flags_needed = false;
+}
+
+static struct netdev *
+netdev_pltap_alloc(void)
+{
+ struct netdev_pltap *netdev = xzalloc(sizeof *netdev);
+ return &netdev->up;
+}
+
+static int
+netdev_pltap_construct(struct netdev *netdev_)
+{
+ struct netdev_pltap *netdev = netdev_pltap_cast(netdev_);
+ int error;
+
+ ovs_mutex_init(&netdev->mutex);
+ netdev->real_name = xzalloc(IFNAMSIZ + 1);
+ memset(&netdev->local_addr, 0, sizeof(netdev->local_addr));
+ netdev->valid_local_ip = false;
+ netdev->valid_local_netmask = false;
+ netdev->flags = 0;
+ netdev->sync_flags_needed = false;
+ netdev->change_seq = 1;
+
+
+ /* Open tap device. */
+ netdev->fd = tun_alloc(IFF_TAP, netdev->real_name);
+ if (netdev->fd < 0) {
+ error = errno;
+ VLOG_WARN("tun_alloc(IFF_TAP, %s) failed: %s",
+ netdev_get_name(netdev_), ovs_strerror(error));
+ return error;
+ }
+ VLOG_DBG("real_name = %s", netdev->real_name);
+
+ /* Make non-blocking. */
+ error = set_nonblocking(netdev->fd);
+ if (error) {
+ return error;
+ }
+
+ ovs_mutex_lock(&pltap_netdevs_mutex);
+ shash_add(&pltap_netdevs, netdev_get_name(netdev_), netdev);
+ ovs_mutex_unlock(&pltap_netdevs_mutex);
+ return 0;
+}
+
+static void
+netdev_pltap_destruct(struct netdev *netdev_)
+{
+ struct netdev_pltap *netdev = netdev_pltap_cast(netdev_);
+
+ ovs_mutex_lock(&pltap_netdevs_mutex);
+ if (netdev->fd != -1)
+ close(netdev->fd);
+
+ if (netdev->sync_flags_needed) {
+ ovs_mutex_lock(&sync_list_mutex);
+ (void) list_remove(&netdev->sync_list);
+ ovs_mutex_unlock(&sync_list_mutex);
+ }
+
+ shash_find_and_delete(&pltap_netdevs,
+ netdev_get_name(netdev_));
+ ovs_mutex_unlock(&pltap_netdevs_mutex);
+ ovs_mutex_destroy(&netdev->mutex);
+}
+
+static void
+netdev_pltap_dealloc(struct netdev *netdev_)
+{
+ struct netdev_pltap *netdev = netdev_pltap_cast(netdev_);
+ free(netdev);
+}
+
+static int netdev_pltap_up(struct netdev_pltap *dev) OVS_REQUIRES(dev->mutex);
+
+static struct netdev_rx *
+netdev_pltap_rx_alloc(void)
+{
+ struct netdev_rx_pltap *rx = xzalloc(sizeof *rx);
+ return &rx->up;
+}
+
+static int
+netdev_pltap_rx_construct(struct netdev_rx *rx_)
+{
+ struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
+ struct netdev *netdev_ = rx->up.netdev;
+ struct netdev_pltap *netdev =
+ netdev_pltap_cast(netdev_);
+ int error = 0;
+
+ ovs_mutex_lock(&netdev->mutex);
+ rx->fd = netdev->fd;
+ if (!netdev_pltap_finalized(netdev))
+ goto out;
+ error = netdev_pltap_up(netdev);
+ if (error) {
+ goto out;
+ }
+out:
+ ovs_mutex_unlock(&netdev->mutex);
+ return error;
+}
+
+static void
+netdev_pltap_rx_destruct(struct netdev_rx *rx_ OVS_UNUSED)
+{
+}
+
+static void
+netdev_pltap_rx_dealloc(struct netdev_rx *rx_)
+{
+ struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
+
+ free(rx);
+}
+
+static int vsys_transaction(const char *script,
+ const char **preply, char *format, ...)
+{
+ char *msg = NULL, *reply = NULL;
+ const size_t reply_size = 1024;
+ int ifd = -1, ofd = -1, maxfd;
+ size_t bytes_to_write, bytes_to_read,
+ bytes_written = 0, bytes_read = 0;
+ int error = 0;
+ char *ofname = NULL, *ifname = NULL;
+ va_list args;
+
+ va_start(args, format);
+ msg = xvasprintf(format, args);
+ va_end(args);
+ reply = (char*)xmalloc(reply_size);
+ if (!msg || !reply) {
+ VLOG_ERR("Out of memory");
+ error = ENOMEM;
+ goto cleanup;
+ }
+
+ ofname = xasprintf("/vsys/%s.out", script);
+ ifname = xasprintf("/vsys/%s.in", script);
+ if (!ofname || !ifname) {
+ VLOG_ERR("Out of memory");
+ error = ENOMEM;
+ goto cleanup;
+ }
+
+ ofd = open(ofname, O_RDONLY | O_NONBLOCK);
+ if (ofd < 0) {
+ VLOG_ERR("Cannot open %s: %s", ofname, ovs_strerror(errno));
+ error = errno;
+ goto cleanup;
+ }
+ ifd = open(ifname, O_WRONLY | O_NONBLOCK);
+ if (ifd < 0) {
+ VLOG_ERR("Cannot open %s: %s", ifname, ovs_strerror(errno));
+ error = errno;
+ goto cleanup;
+ }
+ maxfd = (ifd < ofd) ? ofd : ifd;
+
+ bytes_to_write = strlen(msg);
+ bytes_to_read = reply_size;
+ while (bytes_to_write || bytes_to_read) {
+ fd_set readset, writeset, errorset;
+
+ FD_ZERO(&readset);
+ FD_ZERO(&writeset);
+ FD_ZERO(&errorset);
+ if (bytes_to_write) {
+ FD_SET(ifd, &writeset);
+ FD_SET(ifd, &errorset);
+ }
+ FD_SET(ofd, &readset);
+ FD_SET(ofd, &errorset);
+ if (select(maxfd + 1, &readset, &writeset, &errorset, NULL) < 0) {
+ if (errno == EINTR)
+ continue;
+ VLOG_ERR("selec error: %s", ovs_strerror(errno));
+ error = errno;
+ goto cleanup;
+ }
+ if (FD_ISSET(ifd, &errorset) || FD_ISSET(ofd, &errorset)) {
+ VLOG_ERR("error condition on ifd or ofd");
+ goto cleanup;
+ }
+ if (FD_ISSET(ifd, &writeset)) {
+ ssize_t n = write(ifd, msg + bytes_written, bytes_to_write);
+ if (n < 0) {
+ if (errno != EAGAIN && errno != EINTR) {
+ VLOG_ERR("write on %s: %s", ifname, ovs_strerror(errno));
+ error = errno;
+ goto cleanup;
+ }
+ } else {
+ bytes_written += n;
+ bytes_to_write -= n;
+ if (bytes_to_write == 0)
+ close(ifd);
+ }
+ }
+ if (FD_ISSET(ofd, &readset)) {
+ ssize_t n = read(ofd, reply + bytes_read, bytes_to_read);
+ if (n < 0) {
+ if (errno != EAGAIN && errno != EINTR) {
+ VLOG_ERR("read on %s: %s", ofname, ovs_strerror(errno));
+ error = errno;
+ goto cleanup;
+ }
+ } else if (n == 0) {
+ bytes_to_read = 0;
+ } else {
+ bytes_read += n;
+ bytes_to_read -= n;
+ }
+ }
+ }
+ if (bytes_read) {
+ reply[bytes_read] = '\0';
+ if (preply) {
+ *preply = reply;
+ reply = NULL; /* prevent freeing the reply msg */
+ } else {
+ VLOG_ERR("%s returned: %s", script, reply);
+ }
+ error = EAGAIN;
+ goto cleanup;
+ }
+
+cleanup:
+ free(msg);
+ free(reply);
+ free(ofname);
+ free(ifname);
+ close(ifd);
+ close(ofd);
+ return error;
+}
+
+static int
+netdev_pltap_up(struct netdev_pltap *dev)
+ OVS_REQUIRES(dev->mutex)
+{
+ if (!netdev_pltap_finalized(dev)) {
+ return 0;
+ }
+
+ return vsys_transaction("vif_up", NULL, "%s\n"IP_FMT"\n%d\n",
+ dev->real_name,
+ IP_ARGS(dev->local_addr.sin_addr.s_addr),
+ dev->local_netmask);
+}
+
+static int
+netdev_pltap_down(struct netdev_pltap *dev)
+ OVS_REQUIRES(dev->mutex)
+{
+ if (!netdev_pltap_finalized(dev)) {
+ return 0;
+ }
+
+ return vsys_transaction("vif_down", NULL, "%s\n", dev->real_name);
+}
+
+static int
+netdev_pltap_promisc(struct netdev_pltap *dev, bool promisc)
+ OVS_REQUIRES(dev-mutex)
+{
+ if (!netdev_pltap_finalized(dev)) {
+ return 0;
+ }
+
+ return vsys_transaction("promisc", NULL, "%s\n%s",
+ dev->real_name,
+ (promisc ? "" : "-\n"));
+}
+
+static void
+netdev_pltap_sync_flags(struct netdev_pltap *dev)
+ OVS_REQUIRES(sync_list_mutex)
+{
+
+ ovs_mutex_lock(&dev->mutex);
+
+ if (dev->fd < 0 || !netdev_pltap_finalized(dev)) {
+ goto out;
+ }
+
+ VLOG_DBG("sync_flags(%s): current: %s %s target: %s %s",
+ dev->real_name,
+ (dev->flags & NETDEV_UP ? "UP" : "-"),
+ (dev->flags & NETDEV_PROMISC ? "PROMISC" : "-"),
+ (dev->new_flags & NETDEV_UP ? "UP" : "-"),
+ (dev->new_flags & NETDEV_PROMISC ? "PROMISC" : "-"));
+
+ if ((dev->new_flags & NETDEV_UP) && !(dev->flags & NETDEV_UP)) {
+ (void) netdev_pltap_up(dev);
+ } else if (!(dev->new_flags & NETDEV_UP) && (dev->flags & NETDEV_UP)) {
+ (void) netdev_pltap_down(dev);
+ }
+
+ if ((dev->new_flags & NETDEV_PROMISC) ^ (dev->flags & NETDEV_PROMISC)) {
+ (void) netdev_pltap_promisc(dev, dev->new_flags & NETDEV_PROMISC);
+ }
+
+ netdev_pltap_update_seq(dev);
+
+out:
+ sync_done(dev);
+ ovs_mutex_unlock(&dev->mutex);
+}
+
+
+static int
+netdev_pltap_get_config(const struct netdev *dev_, struct smap *args)
+{
+ struct netdev_pltap *netdev = netdev_pltap_cast(dev_);
+
+ ovs_mutex_lock(&netdev->mutex);
+ if (netdev->valid_local_ip)
+ smap_add_format(args, "local_ip", IP_FMT,
+ IP_ARGS(netdev->local_addr.sin_addr.s_addr));
+ if (netdev->valid_local_netmask)
+ smap_add_format(args, "local_netmask", "%"PRIu32,
+ ntohs(netdev->local_netmask));
+ ovs_mutex_unlock(&netdev->mutex);
+ return 0;
+}
+
+static int
+netdev_pltap_set_config(struct netdev *dev_, const struct smap *args)
+{
+ struct netdev_pltap *netdev = netdev_pltap_cast(dev_);
+ struct shash_node *node;
+
+ ovs_mutex_lock(&sync_list_mutex);
+ ovs_mutex_lock(&netdev->mutex);
+ VLOG_DBG("pltap_set_config(%s)", netdev_get_name(dev_));
+ SMAP_FOR_EACH(node, args) {
+ VLOG_DBG("arg: %s->%s", node->name, (char*)node->data);
+ if (!strcmp(node->name, "local_ip")) {
+ struct in_addr addr;
+ if (lookup_ip(node->data, &addr)) {
+ VLOG_WARN("%s: bad 'local_ip'", node->name);
+ } else {
+ netdev->local_addr.sin_addr = addr;
+ netdev->valid_local_ip = true;
+ }
+ } else if (!strcmp(node->name, "local_netmask")) {
+ netdev->local_netmask = atoi(node->data);
+ // XXX check valididy
+ netdev->valid_local_netmask = true;
+ } else {
+ VLOG_WARN("%s: unknown argument '%s'",
+ netdev_get_name(dev_), node->name);
+ }
+ }
+ if (netdev_pltap_finalized(netdev)) {
+ netdev->new_flags |= NETDEV_UP;
+ sync_needed(netdev);
+ }
+ ovs_mutex_unlock(&netdev->mutex);
+ ovs_mutex_unlock(&sync_list_mutex);
+ return 0;
+}
+
+static int
- { .iov_base = buffer, .iov_len = size }
++netdev_pltap_rx_recv(struct netdev_rx *rx_, struct ofpbuf *buffer)
+{
++ size_t size = ofpbuf_tailroom(buffer);
+ struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
+ struct tun_pi pi;
+ struct iovec iov[2] = {
+ { .iov_base = &pi, .iov_len = sizeof(pi) },
- netdev_pltap_change_seq,
-
++ { .iov_base = buffer->data, .iov_len = size }
+ };
+ for (;;) {
+ ssize_t retval;
+ retval = readv(rx->fd, iov, 2);
+ if (retval >= 0) {
+ if (retval <= size) {
+ return retval;
+ } else {
+ return -EMSGSIZE;
+ }
+ } else if (errno != EINTR) {
+ if (errno != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error receiveing Ethernet packet on %s: %s",
+ netdev_rx_get_name(rx_), ovs_strerror(errno));
+ }
+ return -errno;
+ }
+ }
+}
+
+static void
+netdev_pltap_rx_wait(struct netdev_rx *rx_)
+{
+ struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
+ struct netdev_pltap *netdev =
+ netdev_pltap_cast(rx->up.netdev);
+ if (rx->fd >= 0 && netdev_pltap_finalized(netdev)) {
+ poll_fd_wait(rx->fd, POLLIN);
+ }
+}
+
+static int
+netdev_pltap_send(struct netdev *netdev_, const void *buffer, size_t size)
+{
+ struct netdev_pltap *dev =
+ netdev_pltap_cast(netdev_);
+ struct tun_pi pi = { 0, 0x86 };
+ struct iovec iov[2] = {
+ { .iov_base = &pi, .iov_len = sizeof(pi) },
+ { .iov_base = (char*) buffer, .iov_len = size }
+ };
+ if (dev->fd < 0)
+ return EAGAIN;
+ for (;;) {
+ ssize_t retval;
+ retval = writev(dev->fd, iov, 2);
+ if (retval >= 0) {
+ if (retval != size + 4) {
+ VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIdSIZE" bytes of %"PRIuSIZE") on %s",
+ retval, size + 4, netdev_get_name(netdev_));
+ }
+ return 0;
+ } else if (errno != EINTR) {
+ if (errno != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
+ netdev_get_name(netdev_), ovs_strerror(errno));
+ }
+ return errno;
+ }
+ }
+}
+
+static void
+netdev_pltap_send_wait(struct netdev *netdev_)
+{
+ struct netdev_pltap *dev =
+ netdev_pltap_cast(netdev_);
+ if (dev->fd >= 0 && netdev_pltap_finalized(dev)) {
+ poll_fd_wait(dev->fd, POLLOUT);
+ }
+}
+
+static int
+netdev_pltap_rx_drain(struct netdev_rx *rx_)
+{
+ struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
+ char buffer[128];
+ int error;
+
+ if (rx->fd < 0)
+ return EAGAIN;
+ for (;;) {
+ error = recv(rx->fd, buffer, 128, MSG_TRUNC);
+ if (error) {
+ if (error == -EAGAIN)
+ break;
+ else if (error != -EMSGSIZE)
+ return error;
+ }
+ }
+ return 0;
+}
+
+static int
+netdev_pltap_set_etheraddr(struct netdev *netdevi OVS_UNUSED,
+ const uint8_t mac[ETH_ADDR_LEN] OVS_UNUSED)
+{
+ return ENOTSUP;
+}
+
+
+// XXX from netdev-linux.c
+static int
+get_etheraddr(struct netdev_pltap *dev, uint8_t ea[ETH_ADDR_LEN])
+ OVS_REQUIRES(dev->mutex)
+{
+ struct ifreq ifr;
+ int hwaddr_family;
+ int error;
+
+ memset(&ifr, 0, sizeof ifr);
+ ovs_strzcpy(ifr.ifr_name, dev->real_name, sizeof ifr.ifr_name);
+ error = af_inet_ifreq_ioctl(dev->real_name, &ifr,
+ SIOCGIFHWADDR, "SIOCGIFHWADDR");
+ if (error) {
+ return error;
+ }
+ hwaddr_family = ifr.ifr_hwaddr.sa_family;
+ if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
+ VLOG_WARN("%s device has unknown hardware address family %d",
+ dev->real_name, hwaddr_family);
+ }
+ memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
+ return 0;
+}
+
+static int
+get_flags(struct netdev_pltap *dev, enum netdev_flags *flags)
+ OVS_REQUIRES(dev->mutex)
+{
+ struct ifreq ifr;
+ int error;
+
+ error = af_inet_ifreq_ioctl(dev->real_name, &ifr,
+ SIOCGIFFLAGS, "SIOCGIFFLAGS");
+ if (error) {
+ return error;
+ }
+ *flags = 0;
+ if (ifr.ifr_flags & IFF_UP)
+ *flags |= NETDEV_UP;
+ if (ifr.ifr_flags & IFF_PROMISC)
+ *flags |= NETDEV_PROMISC;
+ return 0;
+}
+
+static int
+netdev_pltap_get_etheraddr(const struct netdev *netdev,
+ uint8_t mac[ETH_ADDR_LEN])
+{
+ struct netdev_pltap *dev =
+ netdev_pltap_cast(netdev);
+ int error = 0;
+
+ ovs_mutex_lock(&dev->mutex);
+ if (dev->fd < 0) {
+ error = EAGAIN;
+ goto out;
+ }
+ error = get_etheraddr(dev, mac);
+
+out:
+ ovs_mutex_unlock(&dev->mutex);
+ return error;
+}
+
+
+// XXX can we read stats in planetlab?
+static int
+netdev_pltap_get_stats(const struct netdev *netdev OVS_UNUSED, struct netdev_stats *stats OVS_UNUSED)
+{
+ return ENOTSUP;
+}
+
+static int
+netdev_pltap_set_stats(struct netdev *netdev OVS_UNUSED, const struct netdev_stats *stats OVS_UNUSED)
+{
+ return ENOTSUP;
+}
+
+
+static int
+netdev_pltap_update_flags(struct netdev *dev_,
+ enum netdev_flags off, enum netdev_flags on,
+ enum netdev_flags *old_flagsp)
+{
+ struct netdev_pltap *netdev =
+ netdev_pltap_cast(dev_);
+ int error = 0;
+
+ ovs_mutex_lock(&sync_list_mutex);
+ ovs_mutex_lock(&netdev->mutex);
+ if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) {
+ error = EINVAL;
+ goto out;
+ }
+
+ if (netdev_pltap_finalized(netdev)) {
+ error = get_flags(netdev, &netdev->flags);
+ }
+ *old_flagsp = netdev->flags;
+ netdev->new_flags |= on;
+ netdev->new_flags &= ~off;
+ if (netdev->flags != netdev->new_flags) {
+ /* we cannot sync here, since we may be in a signal handler */
+ sync_needed(netdev);
+ }
+
+out:
+ ovs_mutex_unlock(&netdev->mutex);
+ ovs_mutex_unlock(&sync_list_mutex);
+ return error;
+}
+
+static unsigned int
+netdev_pltap_change_seq(const struct netdev *netdev)
+{
+ struct netdev_pltap *dev =
+ netdev_pltap_cast(netdev);
+ unsigned int change_seq;
+
+ ovs_mutex_lock(&dev->mutex);
+ change_seq = dev->change_seq;
+ ovs_mutex_unlock(&dev->mutex);
+
+ return change_seq;
+}
+\f
+/* Helper functions. */
+
+static void
+netdev_pltap_update_seq(struct netdev_pltap *dev)
+ OVS_REQUIRES(dev->mutex)
+{
+ dev->change_seq++;
+ if (!dev->change_seq) {
+ dev->change_seq++;
+ }
+}
+
+static void
+netdev_pltap_get_real_name(struct unixctl_conn *conn,
+ int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED)
+{
+ struct netdev_pltap *pltap_dev;
+
+ ovs_mutex_lock(&pltap_netdevs_mutex);
+ pltap_dev = shash_find_data(&pltap_netdevs, argv[1]);
+ if (!pltap_dev) {
+ unixctl_command_reply_error(conn, "no such pltap netdev");
+ goto out;
+ }
+ if (pltap_dev->fd < 0) {
+ unixctl_command_reply_error(conn, "no real device attached");
+ goto out;
+ }
+
+ unixctl_command_reply(conn, pltap_dev->real_name);
+
+out:
+ ovs_mutex_unlock(&pltap_netdevs_mutex);
+}
+
+static int
+netdev_pltap_init(void)
+{
+ unixctl_command_register("netdev-pltap/get-tapname", "port",
+ 1, 1, netdev_pltap_get_real_name, NULL);
+ return 0;
+}
+
+static void
+netdev_pltap_run(void)
+{
+ struct netdev_pltap *iter, *next;
+ ovs_mutex_lock(&sync_list_mutex);
+ LIST_FOR_EACH_SAFE(iter, next, sync_list, &sync_list) {
+ netdev_pltap_sync_flags(iter);
+ }
+ ovs_mutex_unlock(&sync_list_mutex);
+}
+
+static void
+netdev_pltap_wait(void)
+{
+ ovs_mutex_lock(&sync_list_mutex);
+ if (!list_is_empty(&sync_list)) {
+ VLOG_DBG("netdev_pltap: scheduling sync");
+ poll_immediate_wake();
+ }
+ ovs_mutex_unlock(&sync_list_mutex);
+}
+
+const struct netdev_class netdev_pltap_class = {
+ "pltap",
+ netdev_pltap_init,
+ netdev_pltap_run,
+ netdev_pltap_wait,
+
+ netdev_pltap_alloc,
+ netdev_pltap_construct,
+ netdev_pltap_destruct,
+ netdev_pltap_dealloc,
+ netdev_pltap_get_config,
+ netdev_pltap_set_config,
+ NULL, /* get_tunnel_config */
+
+ netdev_pltap_send,
+ netdev_pltap_send_wait,
+
+ netdev_pltap_set_etheraddr,
+ netdev_pltap_get_etheraddr,
+ NULL, /* get_mtu */
+ NULL, /* set_mtu */
+ NULL, /* get_ifindex */
+ NULL, /* get_carrier */
+ NULL, /* get_carrier_resets */
+ NULL, /* get_miimon */
+ netdev_pltap_get_stats,
+ netdev_pltap_set_stats,
+
+ NULL, /* get_features */
+ NULL, /* set_advertisements */
+
+ NULL, /* set_policing */
+ NULL, /* get_qos_types */
+ NULL, /* get_qos_capabilities */
+ NULL, /* get_qos */
+ NULL, /* set_qos */
+ NULL, /* get_queue */
+ NULL, /* set_queue */
+ NULL, /* delete_queue */
+ NULL, /* get_queue_stats */
+ NULL, /* queue_dump_start */
+ NULL, /* queue_dump_next */
+ NULL, /* queue_dump_done */
+ NULL, /* dump_queue_stats */
+
+ NULL, /* get_in4 */
+ NULL, /* set_in4 */
+ NULL, /* get_in6 */
+ NULL, /* add_router */
+ NULL, /* get_next_hop */
+ NULL, /* get_drv_info */
+ NULL, /* arp_lookup */
+
+ netdev_pltap_update_flags,
+
+ netdev_pltap_rx_alloc,
+ netdev_pltap_rx_construct,
+ netdev_pltap_rx_destruct,
+ netdev_pltap_rx_dealloc,
+ netdev_pltap_rx_recv,
+ netdev_pltap_rx_wait,
+ netdev_pltap_rx_drain,
+};
* Each "dealloc" function frees raw memory that was allocated by the the
* "alloc" function. The memory's base and derived members might not have ever
* been initialized (but if "construct" returned successfully, then it has been
- * "destruct"ed already). The "dealloc" function is not allowed to fail. */
+ * "destruct"ed already). The "dealloc" function is not allowed to fail.
+ *
+ *
+ * Device Change Notification
+ * ==========================
+ *
+ * Minimally, implementations are required to report changes to netdev flags,
+ * features, ethernet address or carrier through connectivity_seq. Changes to
+ * other properties are allowed to cause notification through this interface,
+ * although implementations should try to avoid this. connectivity_seq_get()
+ * can be used to acquire a reference to the struct seq. The interface is
+ * described in detail in seq.h. */
struct netdev_class {
/* Type of netdevs in this class, e.g. "system", "tap", "gre", etc.
*
int (*update_flags)(struct netdev *netdev, enum netdev_flags off,
enum netdev_flags on, enum netdev_flags *old_flags);
- /* Returns a sequence number which indicates changes in one of 'netdev''s
- * properties. The returned sequence number must be nonzero so that
- * callers have a value which they may use as a reset when tracking
- * 'netdev'.
- *
- * Minimally, the returned sequence number is required to change whenever
- * 'netdev''s flags, features, ethernet address, or carrier changes. The
- * returned sequence number is allowed to change even when 'netdev' doesn't
- * change, although implementations should try to avoid this. */
- unsigned int (*change_seq)(const struct netdev *netdev);
-
/* ## ------------------- ## */
/* ## netdev_rx Functions ## */
/* ## ------------------- ## */
void (*rx_destruct)(struct netdev_rx *);
void (*rx_dealloc)(struct netdev_rx *);
- /* Attempts to receive a packet from 'rx' into the 'size' bytes in
- * 'buffer'. If successful, returns the number of bytes in the received
- * packet, otherwise a negative errno value. Returns -EAGAIN immediately
- * if no packet is ready to be received.
+ /* Attempts to receive a packet from 'rx' into the tailroom of 'buffer',
+ * which should initially be empty. If successful, returns 0 and
+ * increments 'buffer->size' by the number of bytes in the received packet,
+ * otherwise a positive errno value. Returns EAGAIN immediately if no
+ * packet is ready to be received.
+ *
+ * Must return EMSGSIZE, and discard the packet, if the received packet
+ * is longer than 'ofpbuf_tailroom(buffer)'.
+ *
+ * Implementations may make use of VLAN_HEADER_LEN bytes of tailroom to
+ * add a VLAN header which is obtained out-of-band to the packet. If
+ * this occurs then VLAN_HEADER_LEN bytes of tailroom will no longer be
+ * available for the packet, otherwise it may be used for the packet
+ * itself.
*
- * Must return -EMSGSIZE, and discard the packet, if the received packet
- * is longer than 'size' bytes.
+ * It is advised that the tailroom of 'buffer' should be
+ * VLAN_HEADER_LEN bytes longer than the MTU to allow space for an
+ * out-of-band VLAN header to be added to the packet.
*
- * Specify NULL if this */
- int (*rx_recv)(struct netdev_rx *rx, void *buffer, size_t size);
+ * This function may be set to null if it would always return EOPNOTSUPP
+ * anyhow. */
+ int (*rx_recv)(struct netdev_rx *rx, struct ofpbuf *buffer);
/* Registers with the poll loop to wake up from the next call to
* poll_block() when a packet is ready to be received with netdev_rx_recv()
extern const struct netdev_class netdev_bsd_class;
#endif
+extern const struct netdev_class netdev_tunnel_class;
+extern const struct netdev_class netdev_pltap_class;
+
#ifdef __cplusplus
}
#endif
--- /dev/null
- netdev_tunnel_rx_recv(struct netdev_rx *rx_, void *buffer, size_t size)
+/*
+ * Copyright (c) 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.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+
+#include <unistd.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+#include <errno.h>
+
+#include "flow.h"
+#include "list.h"
+#include "netdev-provider.h"
+#include "odp-util.h"
+#include "ofp-print.h"
+#include "ofpbuf.h"
+#include "packets.h"
+#include "poll-loop.h"
+#include "shash.h"
+#include "sset.h"
+#include "unixctl.h"
+#include "socket-util.h"
+#include "vlog.h"
+
+VLOG_DEFINE_THIS_MODULE(netdev_tunnel);
+
+struct netdev_tunnel {
+ struct netdev up;
+
+ /* Protects all members below. */
+ struct ovs_mutex mutex;
+
+ uint8_t hwaddr[ETH_ADDR_LEN];
+ struct netdev_stats stats;
+ enum netdev_flags flags;
+ int sockfd;
+ struct sockaddr_in local_addr;
+ struct sockaddr_in remote_addr;
+ bool valid_remote_ip;
+ bool valid_remote_port;
+ bool connected;
+ unsigned int change_seq;
+};
+
+struct netdev_rx_tunnel {
+ struct netdev_rx up;
+ int fd;
+};
+
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+
+static struct ovs_mutex tunnel_netdevs_mutex = OVS_MUTEX_INITIALIZER;
+static struct shash tunnel_netdevs OVS_GUARDED_BY(tunnel_netdevs_mutex)
+ = SHASH_INITIALIZER(&tunnel_netdevs);
+
+static int netdev_tunnel_construct(struct netdev *netdevp_);
+static void netdev_tunnel_update_seq(struct netdev_tunnel *);
+
+static bool
+is_netdev_tunnel_class(const struct netdev_class *class)
+{
+ return class->construct == netdev_tunnel_construct;
+}
+
+static struct netdev_tunnel *
+netdev_tunnel_cast(const struct netdev *netdev)
+{
+ ovs_assert(is_netdev_tunnel_class(netdev_get_class(netdev)));
+ return CONTAINER_OF(netdev, struct netdev_tunnel, up);
+}
+
+static struct netdev_rx_tunnel *
+netdev_rx_tunnel_cast(const struct netdev_rx *rx)
+{
+ ovs_assert(is_netdev_tunnel_class(netdev_get_class(rx->netdev)));
+ return CONTAINER_OF(rx, struct netdev_rx_tunnel, up);
+}
+
+static struct netdev *
+netdev_tunnel_alloc(void)
+{
+ struct netdev_tunnel *netdev = xzalloc(sizeof *netdev);
+ return &netdev->up;
+}
+
+static int
+netdev_tunnel_construct(struct netdev *netdev_)
+{
+ static atomic_uint next_n = ATOMIC_VAR_INIT(0);
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(netdev_);
+ unsigned int n;
+
+ atomic_add(&next_n, 1, &n);
+
+ ovs_mutex_init(&netdev->mutex);
+ netdev->hwaddr[0] = 0xfe;
+ netdev->hwaddr[1] = 0xff;
+ netdev->hwaddr[2] = 0xff;
+ netdev->hwaddr[3] = n >> 16;
+ netdev->hwaddr[4] = n >> 8;
+ netdev->hwaddr[5] = n;
+ netdev->flags = 0;
+ netdev->change_seq = 1;
+ memset(&netdev->remote_addr, 0, sizeof(netdev->remote_addr));
+ netdev->valid_remote_ip = false;
+ netdev->valid_remote_port = false;
+ netdev->connected = false;
+
+
+ netdev->sockfd = inet_open_passive(SOCK_DGRAM, "", 0, &netdev->local_addr, 0);
+ if (netdev->sockfd < 0) {
+ return netdev->sockfd;
+ }
+
+
+ shash_add(&tunnel_netdevs, netdev_get_name(netdev_), netdev);
+
+ n++;
+
+ VLOG_DBG("tunnel_create: name=%s, fd=%d, port=%d",
+ netdev_get_name(netdev_), netdev->sockfd, netdev->local_addr.sin_port);
+
+ return 0;
+
+}
+
+static void
+netdev_tunnel_destruct(struct netdev *netdev_)
+{
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(netdev_);
+
+ ovs_mutex_lock(&tunnel_netdevs_mutex);
+
+ if (netdev->sockfd != -1)
+ close(netdev->sockfd);
+
+ shash_find_and_delete(&tunnel_netdevs,
+ netdev_get_name(netdev_));
+
+ ovs_mutex_destroy(&netdev->mutex);
+ ovs_mutex_unlock(&tunnel_netdevs_mutex);
+}
+
+static void
+netdev_tunnel_dealloc(struct netdev *netdev_)
+{
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(netdev_);
+ free(netdev);
+}
+
+static int
+netdev_tunnel_get_config(const struct netdev *dev_, struct smap *args)
+{
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(dev_);
+
+ ovs_mutex_lock(&netdev->mutex);
+ if (netdev->valid_remote_ip)
+ smap_add_format(args, "remote_ip", IP_FMT,
+ IP_ARGS(netdev->remote_addr.sin_addr.s_addr));
+ if (netdev->valid_remote_port)
+ smap_add_format(args, "remote_port", "%"PRIu16,
+ ntohs(netdev->remote_addr.sin_port));
+ ovs_mutex_unlock(&netdev->mutex);
+ return 0;
+}
+
+static int
+netdev_tunnel_connect(struct netdev_tunnel *dev)
+ OVS_REQUIRES(dev->mutex)
+{
+ char buf[1024];
+ if (dev->sockfd < 0)
+ return EBADF;
+ if (!dev->valid_remote_ip || !dev->valid_remote_port)
+ return 0;
+ dev->remote_addr.sin_family = AF_INET;
+ if (connect(dev->sockfd, (struct sockaddr*) &dev->remote_addr, sizeof(dev->remote_addr)) < 0) {
+ return errno;
+ }
+ dev->connected = true;
+ netdev_tunnel_update_seq(dev);
+ VLOG_DBG("%s: connected to (%s, %d)", netdev_get_name(&dev->up),
+ inet_ntop(AF_INET, &dev->remote_addr.sin_addr, buf, 1024), ntohs(dev->remote_addr.sin_port));
+ return 0;
+}
+
+static int
+netdev_tunnel_set_config(struct netdev *dev_, const struct smap *args)
+{
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(dev_);
+ struct shash_node *node;
+ int error;
+
+ ovs_mutex_lock(&netdev->mutex);
+ VLOG_DBG("tunnel_set_config(%s)", netdev_get_name(dev_));
+ SMAP_FOR_EACH(node, args) {
+ VLOG_DBG("arg: %s->%s", node->name, (char*)node->data);
+ if (!strcmp(node->name, "remote_ip")) {
+ struct in_addr addr;
+ if (lookup_ip(node->data, &addr)) {
+ VLOG_WARN("%s: bad 'remote_ip'", node->name);
+ } else {
+ netdev->remote_addr.sin_addr = addr;
+ netdev->valid_remote_ip = true;
+ }
+ } else if (!strcmp(node->name, "remote_port")) {
+ netdev->remote_addr.sin_port = htons(atoi(node->data));
+ netdev->valid_remote_port = true;
+ } else {
+ VLOG_WARN("%s: unknown argument '%s'",
+ netdev_get_name(dev_), node->name);
+ }
+ }
+ error = netdev_tunnel_connect(netdev);
+ ovs_mutex_unlock(&netdev->mutex);
+ return error;
+}
+
+static struct netdev_rx *
+netdev_tunnel_rx_alloc(void)
+{
+ struct netdev_rx_tunnel *rx = xzalloc(sizeof *rx);
+ return &rx->up;
+}
+
+static int
+netdev_tunnel_rx_construct(struct netdev_rx *rx_)
+{
+ struct netdev_rx_tunnel *rx = netdev_rx_tunnel_cast(rx_);
+ struct netdev *netdev_ = rx->up.netdev;
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(netdev_);
+
+ ovs_mutex_lock(&netdev->mutex);
+ rx->fd = netdev->sockfd;
+ ovs_mutex_unlock(&netdev->mutex);
+ return 0;
+}
+
+static void
+netdev_tunnel_rx_destruct(struct netdev_rx *rx_ OVS_UNUSED)
+{
+}
+
+static void
+netdev_tunnel_rx_dealloc(struct netdev_rx *rx_)
+{
+ struct netdev_rx_tunnel *rx = netdev_rx_tunnel_cast(rx_);
+
+ free(rx);
+}
+
+static int
- retval = recv(rx->fd, buffer, size, MSG_TRUNC);
++netdev_tunnel_rx_recv(struct netdev_rx *rx_, struct ofpbuf *buffer)
+{
++ size_t size = ofpbuf_tailroom(buffer);
+ struct netdev_rx_tunnel *rx = netdev_rx_tunnel_cast(rx_);
+ struct netdev_tunnel *netdev =
+ netdev_tunnel_cast(rx_->netdev);
+ if (!netdev->connected)
+ return -EAGAIN;
+ for (;;) {
+ ssize_t retval;
- netdev_rx_get_name(rx_), (uintptr_t)buffer, size, retval);
++ retval = recv(rx->fd, buffer->data, size, MSG_TRUNC);
+ VLOG_DBG("%s: recv(%"PRIxPTR", %"PRIuSIZE", MSG_TRUNC) = %"PRIdSIZE,
- netdev->stats.rx_packets++;
- netdev->stats.rx_bytes += retval;
++ netdev_rx_get_name(rx_), (uintptr_t)buffer->data, size, retval);
+ if (retval >= 0) {
- NULL, /* get_drv_info */
++ netdev->stats.rx_packets++;
++ netdev->stats.rx_bytes += retval;
+ if (retval <= size) {
+ return retval;
+ } else {
+ netdev->stats.rx_errors++;
+ netdev->stats.rx_length_errors++;
+ return -EMSGSIZE;
+ }
+ } else if (errno != EINTR) {
+ if (errno != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error receiveing Ethernet packet on %s: %s",
+ netdev_rx_get_name(rx_), ovs_strerror(errno));
+ netdev->stats.rx_errors++;
+ }
+ return -errno;
+ }
+ }
+}
+
+static void
+netdev_tunnel_rx_wait(struct netdev_rx *rx_)
+{
+ struct netdev_rx_tunnel *rx =
+ netdev_rx_tunnel_cast(rx_);
+ if (rx->fd >= 0) {
+ poll_fd_wait(rx->fd, POLLIN);
+ }
+}
+
+static int
+netdev_tunnel_send(struct netdev *netdev_, const void *buffer, size_t size)
+{
+ struct netdev_tunnel *dev =
+ netdev_tunnel_cast(netdev_);
+ if (!dev->connected)
+ return EAGAIN;
+ for (;;) {
+ ssize_t retval;
+ retval = send(dev->sockfd, buffer, size, 0);
+ VLOG_DBG("%s: send(%"PRIxPTR", %"PRIuSIZE") = %"PRIdSIZE,
+ netdev_get_name(netdev_), (uintptr_t)buffer, size, retval);
+ if (retval >= 0) {
+ dev->stats.tx_packets++;
+ dev->stats.tx_bytes += retval;
+ if (retval != size) {
+ VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIdSIZE" bytes of "
+ "%"PRIuSIZE") on %s", retval, size, netdev_get_name(netdev_));
+ dev->stats.tx_errors++;
+ }
+ return 0;
+ } else if (errno != EINTR) {
+ if (errno != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
+ netdev_get_name(netdev_), ovs_strerror(errno));
+ dev->stats.tx_errors++;
+ }
+ return errno;
+ }
+ }
+}
+
+static void
+netdev_tunnel_send_wait(struct netdev *netdev_)
+{
+ struct netdev_tunnel *dev = netdev_tunnel_cast(netdev_);
+ if (dev->sockfd >= 0) {
+ poll_fd_wait(dev->sockfd, POLLOUT);
+ }
+}
+
+static int
+netdev_tunnel_rx_drain(struct netdev_rx *rx_)
+{
+ struct netdev_tunnel *netdev =
+ netdev_tunnel_cast(rx_->netdev);
+ struct netdev_rx_tunnel *rx =
+ netdev_rx_tunnel_cast(rx_);
+ char buffer[128];
+ int error;
+
+ if (!netdev->connected)
+ return 0;
+ for (;;) {
+ error = recv(rx->fd, buffer, 128, MSG_TRUNC);
+ if (error) {
+ if (error == -EAGAIN)
+ break;
+ else if (error != -EMSGSIZE)
+ return error;
+ }
+ }
+ return 0;
+}
+
+static int
+netdev_tunnel_set_etheraddr(struct netdev *netdev,
+ const uint8_t mac[ETH_ADDR_LEN])
+{
+ struct netdev_tunnel *dev = netdev_tunnel_cast(netdev);
+
+ ovs_mutex_lock(&dev->mutex);
+ if (!eth_addr_equals(dev->hwaddr, mac)) {
+ memcpy(dev->hwaddr, mac, ETH_ADDR_LEN);
+ netdev_tunnel_update_seq(dev);
+ }
+ ovs_mutex_unlock(&dev->mutex);
+
+ return 0;
+}
+
+static int
+netdev_tunnel_get_etheraddr(const struct netdev *netdev,
+ uint8_t mac[ETH_ADDR_LEN])
+{
+ const struct netdev_tunnel *dev = netdev_tunnel_cast(netdev);
+
+ ovs_mutex_lock(&dev->mutex);
+ memcpy(mac, dev->hwaddr, ETH_ADDR_LEN);
+ ovs_mutex_unlock(&dev->mutex);
+ return 0;
+}
+
+
+static int
+netdev_tunnel_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
+{
+ const struct netdev_tunnel *dev = netdev_tunnel_cast(netdev);
+
+ ovs_mutex_lock(&dev->mutex);
+ *stats = dev->stats;
+ ovs_mutex_unlock(&dev->mutex);
+ return 0;
+}
+
+static int
+netdev_tunnel_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
+{
+ struct netdev_tunnel *dev = netdev_tunnel_cast(netdev);
+
+ ovs_mutex_lock(&dev->mutex);
+ dev->stats = *stats;
+ ovs_mutex_unlock(&dev->mutex);
+ return 0;
+}
+
+static int
+netdev_tunnel_update_flags(struct netdev *dev_,
+ enum netdev_flags off, enum netdev_flags on,
+ enum netdev_flags *old_flagsp)
+{
+ struct netdev_tunnel *netdev =
+ netdev_tunnel_cast(dev_);
+ int error = 0;
+
+ ovs_mutex_lock(&netdev->mutex);
+ if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) {
+ error = EINVAL;
+ goto out;
+ }
+
+ // XXX should we actually do something with these flags?
+ *old_flagsp = netdev->flags;
+ netdev->flags |= on;
+ netdev->flags &= ~off;
+ if (*old_flagsp != netdev->flags) {
+ netdev_tunnel_update_seq(netdev);
+ }
+
+out:
+ ovs_mutex_unlock(&netdev->mutex);
+ return error;
+}
+
+static unsigned int
+netdev_tunnel_change_seq(const struct netdev *netdev_)
+{
+ struct netdev_tunnel *netdev = netdev_tunnel_cast(netdev_);
+ unsigned int change_seq;
+
+
+ ovs_mutex_lock(&netdev->mutex);
+ change_seq = netdev->change_seq;
+ ovs_mutex_unlock(&netdev->mutex);
+ return change_seq;
+}
+\f
+/* Helper functions. */
+
+static void
+netdev_tunnel_update_seq(struct netdev_tunnel *dev)
+ OVS_REQUIRES(dev->mutex)
+{
+ dev->change_seq++;
+ if (!dev->change_seq) {
+ dev->change_seq++;
+ }
+}
+
+static void
+netdev_tunnel_get_port(struct unixctl_conn *conn,
+ int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED)
+{
+ struct netdev_tunnel *tunnel_dev;
+ char buf[6];
+
+ ovs_mutex_lock(&tunnel_netdevs_mutex);
+ tunnel_dev = shash_find_data(&tunnel_netdevs, argv[1]);
+ if (!tunnel_dev) {
+ unixctl_command_reply_error(conn, "no such tunnel netdev");
+ goto out;
+ }
+
+ ovs_mutex_lock(&tunnel_dev->mutex);
+ sprintf(buf, "%d", ntohs(tunnel_dev->local_addr.sin_port));
+ ovs_mutex_unlock(&tunnel_dev->mutex);
+
+ unixctl_command_reply(conn, buf);
+out:
+ ovs_mutex_unlock(&tunnel_netdevs_mutex);
+}
+
+static void
+netdev_tunnel_get_tx_bytes(struct unixctl_conn *conn,
+ int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED)
+{
+ struct netdev_tunnel *tunnel_dev;
+ char buf[128];
+
+ ovs_mutex_lock(&tunnel_netdevs_mutex);
+ tunnel_dev = shash_find_data(&tunnel_netdevs, argv[1]);
+ if (!tunnel_dev) {
+ unixctl_command_reply_error(conn, "no such tunnel netdev");
+ goto out;
+ }
+
+ ovs_mutex_lock(&tunnel_dev->mutex);
+ sprintf(buf, "%"PRIu64, tunnel_dev->stats.tx_bytes);
+ ovs_mutex_unlock(&tunnel_dev->mutex);
+ unixctl_command_reply(conn, buf);
+out:
+ ovs_mutex_unlock(&tunnel_netdevs_mutex);
+}
+
+static void
+netdev_tunnel_get_rx_bytes(struct unixctl_conn *conn,
+ int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED)
+{
+ struct netdev_tunnel *tunnel_dev;
+ char buf[128];
+
+ ovs_mutex_lock(&tunnel_netdevs_mutex);
+ tunnel_dev = shash_find_data(&tunnel_netdevs, argv[1]);
+ if (!tunnel_dev) {
+ unixctl_command_reply_error(conn, "no such tunnel netdev");
+ goto out;
+ }
+
+ sprintf(buf, "%"PRIu64, tunnel_dev->stats.rx_bytes);
+ unixctl_command_reply(conn, buf);
+out:
+ ovs_mutex_unlock(&tunnel_netdevs_mutex);
+}
+
+
+static int
+netdev_tunnel_init(void)
+{
+ unixctl_command_register("netdev-tunnel/get-port", "NAME",
+ 1, 1, netdev_tunnel_get_port, NULL);
+ unixctl_command_register("netdev-tunnel/get-tx-bytes", "NAME",
+ 1, 1, netdev_tunnel_get_tx_bytes, NULL);
+ unixctl_command_register("netdev-tunnel/get-rx-bytes", "NAME",
+ 1, 1, netdev_tunnel_get_rx_bytes, NULL);
+ return 0;
+}
+
+static void
+netdev_tunnel_run(void)
+{
+}
+
+static void
+netdev_tunnel_wait(void)
+{
+}
+
+const struct netdev_class netdev_tunnel_class = {
+ "tunnel",
+ netdev_tunnel_init,
+ netdev_tunnel_run,
+ netdev_tunnel_wait,
+
+ netdev_tunnel_alloc,
+ netdev_tunnel_construct,
+ netdev_tunnel_destruct,
+ netdev_tunnel_dealloc,
+ netdev_tunnel_get_config,
+ netdev_tunnel_set_config,
+ NULL, /* get_tunnel_config */
+
+ netdev_tunnel_send,
+ netdev_tunnel_send_wait,
+
+ netdev_tunnel_set_etheraddr,
+ netdev_tunnel_get_etheraddr,
+ NULL, /* get_mtu */
+ NULL, /* set_mtu */
+ NULL, /* get_ifindex */
+ NULL, /* get_carrier */
+ NULL, /* get_carrier_resets */
+ NULL, /* get_miimon */
+ netdev_tunnel_get_stats,
+ netdev_tunnel_set_stats,
+
+ NULL, /* get_features */
+ NULL, /* set_advertisements */
+
+ NULL, /* set_policing */
+ NULL, /* get_qos_types */
+ NULL, /* get_qos_capabilities */
+ NULL, /* get_qos */
+ NULL, /* set_qos */
+ NULL, /* get_queue */
+ NULL, /* set_queue */
+ NULL, /* delete_queue */
+ NULL, /* get_queue_stats */
+ NULL, /* queue_dump_start */
+ NULL, /* queue_dump_next */
+ NULL, /* queue_dump_done */
+ NULL, /* dump_queue_stats */
+
+ NULL, /* get_in4 */
+ NULL, /* set_in4 */
+ NULL, /* get_in6 */
+ NULL, /* add_router */
+ NULL, /* get_next_hop */
- netdev_tunnel_change_seq,
-
++ NULL, /* get_status */
+ NULL, /* arp_lookup */
+
+ netdev_tunnel_update_flags,
+
+ netdev_tunnel_rx_alloc,
+ netdev_tunnel_rx_construct,
+ netdev_tunnel_rx_destruct,
+ netdev_tunnel_rx_dealloc,
+ netdev_tunnel_rx_recv,
+ netdev_tunnel_rx_wait,
+ netdev_tunnel_rx_drain,
+};
#include <string.h>
#include <unistd.h>
+ #include "connectivity.h"
#include "coverage.h"
#include "dpif.h"
#include "dynamic-string.h"
#include "openflow/openflow.h"
#include "packets.h"
#include "poll-loop.h"
+ #include "seq.h"
#include "shash.h"
#include "smap.h"
#include "sset.h"
netdev_register_provider(&netdev_tap_class);
netdev_register_provider(&netdev_bsd_class);
#endif
+ netdev_register_provider(&netdev_tunnel_class);
+ netdev_register_provider(&netdev_pltap_class);
ovsthread_once_done(&once);
}
atomic_read(&rc->ref_cnt, &ref_cnt);
if (!ref_cnt) {
hmap_remove(&netdev_classes, &rc->hmap_node);
+ atomic_destroy(&rc->ref_cnt);
free(rc);
error = 0;
} else {
int old_ref_cnt;
atomic_add(&rc->ref_cnt, 1, &old_ref_cnt);
+ seq_change(connectivity_seq_get());
} else {
free(netdev->name);
ovs_assert(list_is_empty(&netdev->saved_flags_list));
{
if (netdev->netdev_class->set_config) {
const struct smap no_args = SMAP_INITIALIZER(&no_args);
- return netdev->netdev_class->set_config(netdev,
- args ? args : &no_args);
+ int error;
+
+ error = netdev->netdev_class->set_config(netdev,
+ args ? args : &no_args);
+ if (error) {
+ VLOG_WARN("%s: could not set configuration (%s)",
+ netdev_get_name(netdev), ovs_strerror(error));
+ }
+ return error;
} else if (args && !smap_is_empty(args)) {
VLOG_WARN("%s: arguments provided to device that is not configurable",
netdev_get_name(netdev));
}
-
return 0;
}
}
}
+ /* Attempts to open a netdev_rx handle for obtaining packets received on
+ * 'netdev'. On success, returns 0 and stores a nonnull 'netdev_rx *' into
+ * '*rxp'. On failure, returns a positive errno value and stores NULL into
+ * '*rxp'.
+ *
+ * Some kinds of network devices might not support receiving packets. This
+ * function returns EOPNOTSUPP in that case.*/
int
netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp)
OVS_EXCLUDED(netdev_mutex)
return error;
}
+ /* Closes 'rx'. */
void
netdev_rx_close(struct netdev_rx *rx)
OVS_EXCLUDED(netdev_mutex)
}
}
+ /* Attempts to receive a packet from 'rx' into the tailroom of 'buffer', which
+ * must initially be empty. If successful, returns 0 and increments
+ * 'buffer->size' by the number of bytes in the received packet, otherwise a
+ * positive errno value.
+ *
+ * Returns EAGAIN immediately if no packet is ready to be received.
+ *
+ * Returns EMSGSIZE, and discards the packet, if the received packet is longer
+ * than 'ofpbuf_tailroom(buffer)'.
+ *
+ * Implementations may make use of VLAN_HEADER_LEN bytes of tailroom to
+ * add a VLAN header which is obtained out-of-band to the packet. If
+ * this occurs then VLAN_HEADER_LEN bytes of tailroom will no longer be
+ * available for the packet, otherwise it may be used for the packet
+ * itself.
+ *
+ * It is advised that the tailroom of 'buffer' should be
+ * VLAN_HEADER_LEN bytes longer than the MTU to allow space for an
+ * out-of-band VLAN header to be added to the packet. At the very least,
+ * 'buffer' must have at least ETH_TOTAL_MIN bytes of tailroom.
+ *
+ * This function may be set to null if it would always return EOPNOTSUPP
+ * anyhow. */
int
netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer)
{
ovs_assert(buffer->size == 0);
ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN);
- retval = rx->netdev->netdev_class->rx_recv(rx, buffer->data,
- ofpbuf_tailroom(buffer));
- if (retval >= 0) {
+ retval = rx->netdev->netdev_class->rx_recv(rx, buffer);
+ if (!retval) {
COVERAGE_INC(netdev_received);
- buffer->size += retval;
if (buffer->size < ETH_TOTAL_MIN) {
ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size);
}
return 0;
} else {
- return -retval;
+ return retval;
}
}
+ /* Arranges for poll_block() to wake up when a packet is ready to be received
+ * on 'rx'. */
void
netdev_rx_wait(struct netdev_rx *rx)
{
rx->netdev->netdev_class->rx_wait(rx);
}
+ /* Discards any packets ready to be received on 'rx'. */
int
netdev_rx_drain(struct netdev_rx *rx)
{
: EOPNOTSUPP);
}
- /* Returns a sequence number which indicates changes in one of 'netdev''s
- * properties. The returned sequence will be nonzero so that callers have a
- * value which they may use as a reset when tracking 'netdev'.
- *
- * The returned sequence number will change whenever 'netdev''s flags,
- * features, ethernet address, or carrier changes. It may change for other
- * reasons as well, or no reason at all. */
- unsigned int
- netdev_change_seq(const struct netdev *netdev)
- {
- return netdev->netdev_class->change_seq(netdev);
- }
\f
/* Returns the class type of 'netdev'.
*
--- /dev/null
- %define version 2.0.90
+%define name sliver-openvswitch
+# to check for any change:
+# grep AC_INIT configure.ac
++%define version 2.1.90
+%define taglevel 0
+
+%define debug_package %{nil}
+
+%define release %{taglevel}%{?pldistro:.%{pldistro}}%{?date:.%{date}}
+
+Vendor: OneLab
+Packager: OneLab <support@planet-lab.eu>
+Distribution: PlanetLab %{plrelease}
+URL: %{SCMURL}
+# Dependencies
+# mar 2013 - because of the move to f18 I have to turn off auto requires
+# this is because rpm would otherwise find deps to /bin/python and /bin/perl
+# In other modules I was able to solve this by referring to /usr/bin/python
+# instead of just python in the builds scripts, but here it looks too complex
+AutoReq: no
+
+Summary: Openvswitch modified for running from a PlanetLab sliver
+Name: %{name}
+Version: %{version}
+Release: %{release}
+License: GPL
+Group: System Environment/Applications
+BuildRoot: %{_tmppath}/%{name}-%{version}-%{release}-buildroot
+Source0: sliver-openvswitch-%{version}.tar.gz
+
+%description
+Openvswitch tuned for running within a PlanetLab sliver
+
+%prep
+%setup -q
+
+%build
+./boot.sh
+# let's be as close as the regular linux/fedora layout
+./configure --prefix=/usr --sysconfdir=/etc --localstatedir=/var --with-logdir=/var/log
+make
+
+%install
+make install DESTDIR=$RPM_BUILD_ROOT
+
+%clean
+rm -rf $RPM_BUILD_ROOT
+
+%files
+/usr/bin/*
+/usr/sbin/*
+/usr/share/openvswitch
+/usr/share/man
+
+%post
+
+%postun
+
+%changelog
+* Sat Jul 06 2013 Giuseppe Lettieri <g.lettieri@iet.unipi.it> - sliver-openvswitch-1.11.90-1
+- merge with mainstream
+
+* Sat Jul 06 2013 Giuseppe Lettieri <g.lettieri@iet.unipi.it> - sliver-openvswitch-1.10.90-3
+- merge with mainstream
+
+* Wed May 01 2013 Giuseppe Lettieri <g.lettieri@iet.unipi.it> - sliver-openvswitch-1.10.90-2
+- - fixed several bugs in the external-nodes support in exp-tool/Makefile
+- - let sliver-ovs return an error if tap device configuration failed
+
+* Mon Apr 22 2013 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.10.90-1
+- merged with upstream (develoment version 1.10.90)
+- integrated ALLEGRA contributions for the termination of virtual cables in external nodes.
+
+* Fri Feb 22 2013 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.9.90-3
+- pulled mainstream - amazingly this is still known as 1.9.90 despite the size of changes
+
+* Fri Dec 21 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.9.90-2
+- merged with upstream
+- handling of promisc &up/down flags for tap devices
+- small improvements to the Makefile
+
+* Fri Nov 23 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.8.90-6
+- fixes in the exp-tool makefile (bash redirections, scp with key..)
+
+* Tue Oct 16 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.8.90-5
+- numerous additional make targets for finer control (use make help)
+- including gprobe for reporting traffic to an ndnmap instance
+- related, more functions in sliver-ovs as well, like exposing
+- detailed info (mac, dpids..) relevant to the OF controller
+- retrieving rx_bytes/tx_bytes (fixed) accessible through ovs-appctl
+
+* Fri Sep 28 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.8.90-4
+- fix file descriptor leaks
+
+* Fri Sep 28 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.8.90-3
+- can specify OpenFlow controller ip/port for each ovs instance
+- through $(CONTROLLER_<id>), or $(CONTROLLER) by default
+
+* Thu Sep 27 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.8.90-2
+- add/skip packet information on tap send/recv
+
+* Wed Sep 26 2012 Thierry Parmentelat <thierry.parmentelat@sophia.inria.fr> - sliver-openvswitch-1.8.90-1
+- merged mainstream 1.8.90
+- planetlab extensions to the openvswitch: single helper command tool 'sliver-ovs' in /usr/sbin
+- planetlab exp-tool : single config file (conf.mk)
+- planetlab exp-tool : can retrieve and save current topology