/*
- * Copyright (c) 2009, 2010 Nicira Networks.
+ * Copyright (c) 2009, 2010, 2011 Nicira Networks.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <linux/ip.h>
#include <linux/types.h>
#include <linux/ethtool.h>
+#include <linux/mii.h>
#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
#include <unistd.h>
#include "coverage.h"
+#include "dpif-linux.h"
#include "dynamic-string.h"
#include "fatal-signal.h"
+#include "hash.h"
+#include "hmap.h"
#include "netdev-provider.h"
#include "netdev-vport.h"
#include "netlink.h"
+#include "netlink-socket.h"
#include "ofpbuf.h"
#include "openflow/openflow.h"
#include "packets.h"
#include "poll-loop.h"
-#include "port-array.h"
#include "rtnetlink.h"
+#include "rtnetlink-link.h"
#include "socket-util.h"
#include "shash.h"
-#include "svec.h"
+#include "sset.h"
#include "vlog.h"
-VLOG_DEFINE_THIS_MODULE(netdev_linux)
+VLOG_DEFINE_THIS_MODULE(netdev_linux);
+
+COVERAGE_DEFINE(netdev_get_vlan_vid);
+COVERAGE_DEFINE(netdev_set_policing);
+COVERAGE_DEFINE(netdev_arp_lookup);
+COVERAGE_DEFINE(netdev_get_ifindex);
+COVERAGE_DEFINE(netdev_get_hwaddr);
+COVERAGE_DEFINE(netdev_set_hwaddr);
+COVERAGE_DEFINE(netdev_ethtool);
\f
/* These were introduced in Linux 2.6.14, so they might be missing if we have
* old headers. */
/* Traffic control. */
/* An instance of a traffic control class. Always associated with a particular
- * network device. */
+ * network device.
+ *
+ * Each TC implementation subclasses this with whatever additional data it
+ * needs. */
struct tc {
const struct tc_ops *ops;
+ struct hmap queues; /* Contains "struct tc_queue"s.
+ * Read by generic TC layer.
+ * Written only by TC implementation. */
+};
- /* Maps from queue ID to tc-specific data.
- *
- * The generic netdev TC layer uses this to the following extent: if an
- * entry is nonnull, then the queue whose ID is the index is assumed to
- * exist; if an entry is null, then that queue is assumed not to exist.
- * Implementations must adhere to this scheme, although they may store
- * whatever they like as data.
- */
- struct port_array queues;
+/* One traffic control queue.
+ *
+ * Each TC implementation subclasses this with whatever additional data it
+ * needs. */
+struct tc_queue {
+ struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */
+ unsigned int queue_id; /* OpenFlow queue ID. */
};
/* A particular kind of traffic control. Each implementation generally maps to
*/
int (*qdisc_set)(struct netdev *, const struct shash *details);
- /* Retrieves details of 'queue_id' on 'netdev->tc' into 'details'. The
- * caller ensures that 'queues' has a nonnull value for index 'queue_id.
+ /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is
+ * one of the 'struct tc_queue's within 'netdev->tc->queues'.
*
* The contents of 'details' should be documented as valid for 'ovs_name'
* in the "other_config" column in the "Queue" table in
*
* This function may be null if 'tc' does not have queues ('n_queues' is
* 0). */
- int (*class_get)(const struct netdev *netdev, unsigned int queue_id,
+ int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue,
struct shash *details);
/* Configures or reconfigures 'queue_id' on 'netdev->tc' according to
int (*class_set)(struct netdev *, unsigned int queue_id,
const struct shash *details);
- /* Deletes 'queue_id' from 'netdev->tc'. The caller ensures that 'queues'
- * has a nonnull value for index 'queue_id.
+ /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct
+ * tc_queue's within 'netdev->tc->queues'.
*
* This function may be null if 'tc' does not have queues or its queues
* cannot be deleted. */
- int (*class_delete)(struct netdev *, unsigned int queue_id);
+ int (*class_delete)(struct netdev *, struct tc_queue *queue);
- /* Obtains stats for 'queue' from 'netdev->tc'. The caller ensures that
- * 'queues' has a nonnull value for index 'queue_id.
+ /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the
+ * 'struct tc_queue's within 'netdev->tc->queues'.
*
* On success, initializes '*stats'.
*
* This function may be null if 'tc' does not have queues or if it cannot
* report queue statistics. */
- int (*class_get_stats)(const struct netdev *netdev, unsigned int queue_id,
+ int (*class_get_stats)(const struct netdev *netdev,
+ const struct tc_queue *queue,
struct netdev_queue_stats *stats);
/* Extracts queue stats from 'nlmsg', which is a response to a
tc_init(struct tc *tc, const struct tc_ops *ops)
{
tc->ops = ops;
- port_array_init(&tc->queues);
+ hmap_init(&tc->queues);
}
static void
tc_destroy(struct tc *tc)
{
- port_array_destroy(&tc->queues);
+ hmap_destroy(&tc->queues);
}
static const struct tc_ops tc_ops_htb;
+static const struct tc_ops tc_ops_hfsc;
static const struct tc_ops tc_ops_default;
static const struct tc_ops tc_ops_other;
static const struct tc_ops *tcs[] = {
&tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
+ &tc_ops_hfsc, /* Hierarchical fair service curve. */
&tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
&tc_ops_other, /* Some other qdisc. */
NULL
int fd;
};
-/* An AF_INET socket (used for ioctl operations). */
-static int af_inet_sock = -1;
+/* Sockets used for ioctl operations. */
+static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */
+static int af_packet_sock = -1; /* AF_PACKET, SOCK_RAW. */
/* A Netlink routing socket that is not subscribed to any multicast groups. */
static struct nl_sock *rtnl_sock;
status = af_inet_sock >= 0 ? 0 : errno;
if (status) {
VLOG_ERR("failed to create inet socket: %s", strerror(status));
+ } else {
+ /* Create AF_PACKET socket. */
+ af_packet_sock = socket(AF_PACKET, SOCK_RAW, 0);
+ status = af_packet_sock >= 0 ? 0 : errno;
+ if (status) {
+ VLOG_ERR("failed to create packet socket: %s",
+ strerror(status));
+ }
}
/* Create rtnetlink socket. */
if (!status) {
- status = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
+ status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
if (status) {
VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
strerror(status));
static void
netdev_linux_run(void)
{
- rtnetlink_notifier_run();
+ rtnetlink_link_notifier_run();
}
static void
netdev_linux_wait(void)
{
- rtnetlink_notifier_wait();
+ rtnetlink_link_notifier_wait();
}
static void
-netdev_linux_cache_cb(const struct rtnetlink_change *change,
+netdev_linux_cache_cb(const struct rtnetlink_link_change *change,
void *aux OVS_UNUSED)
{
struct netdev_dev_linux *dev;
}
}
-/* Creates the netdev device of 'type' with 'name'. */
+/* Creates system and internal devices. */
static int
-netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
- const struct shash *args, struct netdev_dev **netdev_devp)
+netdev_linux_create(const struct netdev_class *class,
+ const char *name, const struct shash *args,
+ struct netdev_dev **netdev_devp)
{
struct netdev_dev_linux *netdev_dev;
int error;
if (!shash_is_empty(args)) {
- VLOG_WARN("%s: arguments for system devices should be empty", name);
+ VLOG_WARN("%s: arguments for %s devices should be empty",
+ name, class->type);
}
if (!cache_notifier_refcount) {
- error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
- netdev_linux_cache_cb, NULL);
+ error = rtnetlink_link_notifier_register(&netdev_linux_cache_notifier,
+ netdev_linux_cache_cb, NULL);
if (error) {
return error;
}
cache_notifier_refcount++;
netdev_dev = xzalloc(sizeof *netdev_dev);
- netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
+ netdev_dev_init(&netdev_dev->netdev_dev, name, args, class);
*netdev_devp = &netdev_dev->netdev_dev;
return 0;
* buffers, across all readers. Therefore once data is read it will
* be unavailable to other reads for tap devices. */
static int
-netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
- const struct shash *args, struct netdev_dev **netdev_devp)
+netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED,
+ const char *name, const struct shash *args,
+ struct netdev_dev **netdev_devp)
{
struct netdev_dev_linux *netdev_dev;
struct tap_state *state;
/* Create tap device. */
ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
- strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
+ ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
VLOG_WARN("%s: creating tap device failed: %s", name,
strerror(errno));
goto error;
}
- netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
+ netdev_dev_init(&netdev_dev->netdev_dev, name, args, &netdev_tap_class);
*netdev_devp = &netdev_dev->netdev_dev;
return 0;
netdev_linux_destroy(struct netdev_dev *netdev_dev_)
{
struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
- const char *type = netdev_dev_get_type(netdev_dev_);
+ const struct netdev_class *class = netdev_dev_get_class(netdev_dev_);
if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) {
netdev_dev->tc->ops->tc_destroy(netdev_dev->tc);
}
- if (!strcmp(type, "system")) {
+ if (class == &netdev_linux_class || class == &netdev_internal_class) {
cache_notifier_refcount--;
if (!cache_notifier_refcount) {
- rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
+ rtnetlink_link_notifier_unregister(&netdev_linux_cache_notifier);
}
- } else if (!strcmp(type, "tap")) {
+ } else if (class == &netdev_tap_class) {
destroy_tap(netdev_dev);
+ } else {
+ NOT_REACHED();
}
free(netdev_dev);
netdev->fd = -1;
netdev_init(&netdev->netdev, netdev_dev_);
- error = netdev_get_flags(&netdev->netdev, &flags);
- if (error == ENODEV) {
- goto error;
+ /* Verify that the device really exists, by attempting to read its flags.
+ * (The flags might be cached, in which case this won't actually do an
+ * ioctl.)
+ *
+ * Don't do this for "internal" netdevs, though, because those have to be
+ * created as netdev objects before they exist in the kernel, because
+ * creating them in the kernel happens by passing a netdev object to
+ * dpif_port_add(). */
+ if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) {
+ error = netdev_get_flags(&netdev->netdev, &flags);
+ if (error == ENODEV) {
+ goto error;
+ }
}
if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") &&
free(netdev);
}
-/* Initializes 'svec' with a list of the names of all known network devices. */
+/* Initializes 'sset' with a list of the names of all known network devices. */
static int
-netdev_linux_enumerate(struct svec *svec)
+netdev_linux_enumerate(struct sset *sset)
{
struct if_nameindex *names;
size_t i;
for (i = 0; names[i].if_name != NULL; i++) {
- svec_add(svec, names[i].if_name);
+ sset_add(sset, names[i].if_name);
}
if_freenameindex(names);
return 0;
static int
netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct sockaddr_ll sll;
+ struct msghdr msg;
+ struct iovec iov;
+ int ifindex;
+ int error;
- /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
- */
- if (netdev->fd < 0) {
- return EPIPE;
+ error = get_ifindex(netdev_, &ifindex);
+ if (error) {
+ return error;
}
+ /* We don't bother setting most fields in sockaddr_ll because the kernel
+ * ignores them for SOCK_RAW. */
+ memset(&sll, 0, sizeof sll);
+ sll.sll_family = AF_PACKET;
+ sll.sll_ifindex = ifindex;
+
+ iov.iov_base = (void *) data;
+ iov.iov_len = size;
+
+ msg.msg_name = &sll;
+ msg.msg_namelen = sizeof sll;
+ msg.msg_iov = &iov;
+ msg.msg_iovlen = 1;
+ msg.msg_control = NULL;
+ msg.msg_controllen = 0;
+ msg.msg_flags = 0;
+
for (;;) {
- ssize_t retval = write(netdev->fd, data, size);
+ ssize_t retval = sendmsg(af_packet_sock, &msg, 0);
if (retval < 0) {
/* The Linux AF_PACKET implementation never blocks waiting for room
* for packets, instead returning ENOBUFS. Translate this into
return error;
}
+static int
+netdev_linux_do_miimon(const struct netdev *netdev, int cmd,
+ const char *cmd_name, struct mii_ioctl_data *data)
+{
+ struct ifreq ifr;
+ int error;
+
+ memset(&ifr, 0, sizeof ifr);
+ memcpy(&ifr.ifr_data, data, sizeof *data);
+ error = netdev_linux_do_ioctl(netdev_get_name(netdev),
+ &ifr, cmd, cmd_name);
+ memcpy(data, &ifr.ifr_data, sizeof *data);
+
+ return error;
+}
+
+static int
+netdev_linux_get_miimon(const struct netdev *netdev, bool *miimon)
+{
+ const char *name = netdev_get_name(netdev);
+ struct mii_ioctl_data data;
+ int error;
+
+ *miimon = false;
+
+ memset(&data, 0, sizeof data);
+ error = netdev_linux_do_miimon(netdev, SIOCGMIIPHY, "SIOCGMIIPHY", &data);
+ if (!error) {
+ /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */
+ data.reg_num = MII_BMSR;
+ error = netdev_linux_do_miimon(netdev, SIOCGMIIREG, "SIOCGMIIREG",
+ &data);
+
+ if (!error) {
+ *miimon = !!(data.val_out & BMSR_LSTATUS);
+ } else {
+ VLOG_WARN_RL(&rl, "%s: failed to query MII", name);
+ }
+ } else {
+ struct ethtool_cmd ecmd;
+
+ VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool",
+ name);
+
+ memset(&ecmd, 0, sizeof ecmd);
+ error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK,
+ "ETHTOOL_GLINK");
+ if (!error) {
+ struct ethtool_value eval;
+
+ memcpy(&eval, &ecmd, sizeof eval);
+ *miimon = !!eval.data;
+ } else {
+ VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name);
+ }
+ }
+
+ return error;
+}
+
/* Check whether we can we use RTM_GETLINK to get network device statistics.
* In pre-2.6.19 kernels, this was only available if wireless extensions were
* enabled. */
if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
-
- netdev_dev->is_tap = !strcmp(type, "tap");
- netdev_dev->is_internal = false;
- if (!netdev_dev->is_tap) {
- struct ethtool_drvinfo drvinfo;
- int error;
-
- memset(&drvinfo, 0, sizeof drvinfo);
- error = netdev_linux_do_ethtool(name,
- (struct ethtool_cmd *)&drvinfo,
- ETHTOOL_GDRVINFO,
- "ETHTOOL_GDRVINFO");
-
- if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
- netdev_dev->is_internal = true;
- }
- }
+ netdev_dev->is_tap = !strcmp(type, "tap");
+ netdev_dev->is_internal = (!netdev_dev->is_tap
+ && dpif_linux_is_internal_device(name));
netdev_dev->cache_valid |= VALID_IS_PSEUDO;
}
}
static int use_netlink_stats = -1;
int error;
- COVERAGE_INC(netdev_get_stats);
-
if (netdev_dev->have_vport_stats ||
!(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) {
* bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
* successful, otherwise a positive errno value. */
static int
-netdev_linux_get_features(struct netdev *netdev,
+netdev_linux_get_features(const struct netdev *netdev,
uint32_t *current, uint32_t *advertised,
uint32_t *supported, uint32_t *peer)
{
int error;
tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = tc_make_handle(0xffff, 0);
tcmsg->tcm_parent = TC_H_INGRESS;
nl_msg_put_string(&request, TCA_KIND, "ingress");
static int
netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED,
- struct svec *types)
+ struct sset *types)
{
const struct tc_ops **opsp;
for (opsp = tcs; *opsp != NULL; opsp++) {
const struct tc_ops *ops = *opsp;
if (ops->tc_install && ops->ovs_name[0] != '\0') {
- svec_add(types, ops->ovs_name);
+ sset_add(types, ops->ovs_name);
}
}
return 0;
return NULL;
}
+static struct tc_queue *
+tc_find_queue__(const struct netdev *netdev, unsigned int queue_id,
+ size_t hash)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ struct tc_queue *queue;
+
+ HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) {
+ if (queue->queue_id == queue_id) {
+ return queue;
+ }
+ }
+ return NULL;
+}
+
+static struct tc_queue *
+tc_find_queue(const struct netdev *netdev, unsigned int queue_id)
+{
+ return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0));
+}
+
static int
netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED,
const char *type,
error = tc_query_qdisc(netdev);
if (error) {
return error;
- } else if (queue_id > UINT16_MAX
- || !port_array_get(&netdev_dev->tc->queues, queue_id)) {
- return ENOENT;
+ } else {
+ struct tc_queue *queue = tc_find_queue(netdev, queue_id);
+ return (queue
+ ? netdev_dev->tc->ops->class_get(netdev, queue, details)
+ : ENOENT);
}
-
- return netdev_dev->tc->ops->class_get(netdev, queue_id, details);
}
static int
return error;
} else if (!netdev_dev->tc->ops->class_delete) {
return EINVAL;
- } else if (queue_id > UINT16_MAX
- || !port_array_get(&netdev_dev->tc->queues, queue_id)) {
- return ENOENT;
+ } else {
+ struct tc_queue *queue = tc_find_queue(netdev, queue_id);
+ return (queue
+ ? netdev_dev->tc->ops->class_delete(netdev, queue)
+ : ENOENT);
}
-
- return netdev_dev->tc->ops->class_delete(netdev, queue_id);
}
static int
error = tc_query_qdisc(netdev);
if (error) {
return error;
- } else if (queue_id > UINT16_MAX
- || !port_array_get(&netdev_dev->tc->queues, queue_id)) {
- return ENOENT;
} else if (!netdev_dev->tc->ops->class_get_stats) {
return EOPNOTSUPP;
+ } else {
+ const struct tc_queue *queue = tc_find_queue(netdev, queue_id);
+ return (queue
+ ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats)
+ : ENOENT);
}
-
- return netdev_dev->tc->ops->class_get_stats(netdev, queue_id, stats);
}
-static void
+static bool
start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
{
struct ofpbuf request;
struct tcmsg *tcmsg;
tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request);
+ if (!tcmsg) {
+ return false;
+ }
tcmsg->tcm_parent = 0;
nl_dump_start(dump, rtnl_sock, &request);
ofpbuf_uninit(&request);
+ return true;
}
static int
{
struct netdev_dev_linux *netdev_dev =
netdev_dev_linux_cast(netdev_get_dev(netdev));
- unsigned int queue_id;
+ struct tc_queue *queue;
struct shash details;
int last_error;
- void *queue;
int error;
error = tc_query_qdisc(netdev);
last_error = 0;
shash_init(&details);
- PORT_ARRAY_FOR_EACH (queue, &netdev_dev->tc->queues, queue_id) {
+ HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) {
shash_clear(&details);
- error = netdev_dev->tc->ops->class_get(netdev, queue_id, &details);
+ error = netdev_dev->tc->ops->class_get(netdev, queue, &details);
if (!error) {
- (*cb)(queue_id, &details, aux);
+ (*cb)(queue->queue_id, &details, aux);
} else {
last_error = error;
}
}
last_error = 0;
- start_queue_dump(netdev, &dump);
+ if (!start_queue_dump(netdev, &dump)) {
+ return ENODEV;
+ }
while (nl_dump_next(&dump, &msg)) {
error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux);
if (error) {
if (file != NULL) {
const char *name = netdev_get_name(netdev_);
while (fgets(line, sizeof line, file)) {
- struct in6_addr in6;
+ struct in6_addr in6_tmp;
char ifname[16 + 1];
- if (parse_if_inet6_line(line, &in6, ifname)
+ if (parse_if_inet6_line(line, &in6_tmp, ifname)
&& !strcmp(name, ifname))
{
- netdev_dev->in6 = in6;
+ netdev_dev->in6 = in6_tmp;
break;
}
}
int ioctl_nr, const char *ioctl_name, struct in_addr addr)
{
struct ifreq ifr;
- strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
+ ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
make_in4_sockaddr(&ifr.ifr_addr, addr);
return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
make_in4_sockaddr(&rt.rt_gateway, router);
make_in4_sockaddr(&rt.rt_genmask, any);
rt.rt_flags = RTF_UP | RTF_GATEWAY;
- COVERAGE_INC(netdev_add_router);
error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
if (error) {
VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
iface, &dest, &gateway, &flags, &refcnt,
&use, &metric, &mask, &mtu, &window, &irtt) != 11) {
- VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
+ VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
fn, ln, line);
continue;
}
}
/* The output of 'dest', 'mask', and 'gateway' were given in
- * network byte order, so we don't need need any endian
+ * network byte order, so we don't need need any endian
* conversions here. */
if ((dest & mask) == (host->s_addr & mask)) {
if (!gateway) {
return ENXIO;
}
+static int
+netdev_linux_get_status(const struct netdev *netdev, struct shash *sh)
+{
+ struct ethtool_drvinfo drvinfo;
+ int error;
+
+ memset(&drvinfo, 0, sizeof drvinfo);
+ error = netdev_linux_do_ethtool(netdev_get_name(netdev),
+ (struct ethtool_cmd *)&drvinfo,
+ ETHTOOL_GDRVINFO,
+ "ETHTOOL_GDRVINFO");
+ if (!error) {
+ shash_add(sh, "driver_name", xstrdup(drvinfo.driver));
+ shash_add(sh, "driver_version", xstrdup(drvinfo.version));
+ shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version));
+ }
+
+ return error;
+}
+
/* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be
* successfully retrieved, it stores the corresponding MAC address in 'mac' and
* returns 0. Otherwise, it returns a positive errno value; in particular,
int retval;
memset(&r, 0, sizeof r);
+ memset(&sin, 0, sizeof sin);
sin.sin_family = AF_INET;
sin.sin_addr.s_addr = ip;
sin.sin_port = 0;
memcpy(&r.arp_pa, &sin, sizeof sin);
r.arp_ha.sa_family = ARPHRD_ETHER;
r.arp_flags = 0;
- strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
+ ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
COVERAGE_INC(netdev_arp_lookup);
retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
if (!retval) {
poll_notify(struct list *list)
{
struct netdev_linux_notifier *notifier;
- LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
+ LIST_FOR_EACH (notifier, node, list) {
struct netdev_notifier *n = ¬ifier->notifier;
n->cb(n);
}
}
static void
-netdev_linux_poll_cb(const struct rtnetlink_change *change,
+netdev_linux_poll_cb(const struct rtnetlink_link_change *change,
void *aux OVS_UNUSED)
{
if (change) {
struct list *list;
if (shash_is_empty(&netdev_linux_notifiers)) {
- int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
- netdev_linux_poll_cb, NULL);
+ int error;
+ error = rtnetlink_link_notifier_register(&netdev_linux_poll_notifier,
+ netdev_linux_poll_cb, NULL);
if (error) {
return error;
}
/* If that was the last notifier, unregister. */
if (shash_is_empty(&netdev_linux_notifiers)) {
- rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
+ rtnetlink_link_notifier_unregister(&netdev_linux_poll_notifier);
}
}
-const struct netdev_class netdev_linux_class = {
- "system",
-
- netdev_linux_init,
- netdev_linux_run,
- netdev_linux_wait,
-
- netdev_linux_create_system,
- netdev_linux_destroy,
- NULL, /* reconfigure */
-
- netdev_linux_open,
- netdev_linux_close,
-
- netdev_linux_enumerate,
-
- netdev_linux_recv,
- netdev_linux_recv_wait,
- netdev_linux_drain,
-
- netdev_linux_send,
- netdev_linux_send_wait,
-
- netdev_linux_set_etheraddr,
- netdev_linux_get_etheraddr,
- netdev_linux_get_mtu,
- netdev_linux_get_ifindex,
- netdev_linux_get_carrier,
- netdev_linux_get_stats,
- netdev_vport_set_stats,
-
- netdev_linux_get_features,
- netdev_linux_set_advertisements,
- netdev_linux_get_vlan_vid,
-
- netdev_linux_set_policing,
- netdev_linux_get_qos_types,
- netdev_linux_get_qos_capabilities,
- netdev_linux_get_qos,
- netdev_linux_set_qos,
- netdev_linux_get_queue,
- netdev_linux_set_queue,
- netdev_linux_delete_queue,
- netdev_linux_get_queue_stats,
- netdev_linux_dump_queues,
- netdev_linux_dump_queue_stats,
-
- netdev_linux_get_in4,
- netdev_linux_set_in4,
- netdev_linux_get_in6,
- netdev_linux_add_router,
- netdev_linux_get_next_hop,
- netdev_linux_arp_lookup,
-
- netdev_linux_update_flags,
-
- netdev_linux_poll_add,
- netdev_linux_poll_remove,
-};
+#define NETDEV_LINUX_CLASS(NAME, CREATE, ENUMERATE, SET_STATS) \
+{ \
+ NAME, \
+ \
+ netdev_linux_init, \
+ netdev_linux_run, \
+ netdev_linux_wait, \
+ \
+ CREATE, \
+ netdev_linux_destroy, \
+ NULL, /* set_config */ \
+ \
+ netdev_linux_open, \
+ netdev_linux_close, \
+ \
+ ENUMERATE, \
+ \
+ netdev_linux_recv, \
+ netdev_linux_recv_wait, \
+ netdev_linux_drain, \
+ \
+ netdev_linux_send, \
+ netdev_linux_send_wait, \
+ \
+ netdev_linux_set_etheraddr, \
+ netdev_linux_get_etheraddr, \
+ netdev_linux_get_mtu, \
+ netdev_linux_get_ifindex, \
+ netdev_linux_get_carrier, \
+ netdev_linux_get_miimon, \
+ netdev_linux_get_stats, \
+ SET_STATS, \
+ \
+ netdev_linux_get_features, \
+ netdev_linux_set_advertisements, \
+ netdev_linux_get_vlan_vid, \
+ \
+ netdev_linux_set_policing, \
+ netdev_linux_get_qos_types, \
+ netdev_linux_get_qos_capabilities, \
+ netdev_linux_get_qos, \
+ netdev_linux_set_qos, \
+ netdev_linux_get_queue, \
+ netdev_linux_set_queue, \
+ netdev_linux_delete_queue, \
+ netdev_linux_get_queue_stats, \
+ netdev_linux_dump_queues, \
+ netdev_linux_dump_queue_stats, \
+ \
+ netdev_linux_get_in4, \
+ netdev_linux_set_in4, \
+ netdev_linux_get_in6, \
+ netdev_linux_add_router, \
+ netdev_linux_get_next_hop, \
+ netdev_linux_get_status, \
+ netdev_linux_arp_lookup, \
+ \
+ netdev_linux_update_flags, \
+ \
+ netdev_linux_poll_add, \
+ netdev_linux_poll_remove \
+}
-const struct netdev_class netdev_tap_class = {
- "tap",
-
- netdev_linux_init,
- netdev_linux_run,
- netdev_linux_wait,
-
- netdev_linux_create_tap,
- netdev_linux_destroy,
- NULL, /* reconfigure */
-
- netdev_linux_open,
- netdev_linux_close,
-
- NULL, /* enumerate */
-
- netdev_linux_recv,
- netdev_linux_recv_wait,
- netdev_linux_drain,
-
- netdev_linux_send,
- netdev_linux_send_wait,
-
- netdev_linux_set_etheraddr,
- netdev_linux_get_etheraddr,
- netdev_linux_get_mtu,
- netdev_linux_get_ifindex,
- netdev_linux_get_carrier,
- netdev_linux_get_stats,
- NULL, /* set_stats */
-
- netdev_linux_get_features,
- netdev_linux_set_advertisements,
- netdev_linux_get_vlan_vid,
-
- netdev_linux_set_policing,
- netdev_linux_get_qos_types,
- netdev_linux_get_qos_capabilities,
- netdev_linux_get_qos,
- netdev_linux_set_qos,
- netdev_linux_get_queue,
- netdev_linux_set_queue,
- netdev_linux_delete_queue,
- netdev_linux_get_queue_stats,
- netdev_linux_dump_queues,
- netdev_linux_dump_queue_stats,
-
- netdev_linux_get_in4,
- netdev_linux_set_in4,
- netdev_linux_get_in6,
- netdev_linux_add_router,
- netdev_linux_get_next_hop,
- netdev_linux_arp_lookup,
-
- netdev_linux_update_flags,
-
- netdev_linux_poll_add,
- netdev_linux_poll_remove,
-};
+const struct netdev_class netdev_linux_class =
+ NETDEV_LINUX_CLASS(
+ "system",
+ netdev_linux_create,
+ netdev_linux_enumerate,
+ NULL); /* set_stats */
+
+const struct netdev_class netdev_tap_class =
+ NETDEV_LINUX_CLASS(
+ "tap",
+ netdev_linux_create_tap,
+ NULL, /* enumerate */
+ NULL); /* set_stats */
+
+const struct netdev_class netdev_internal_class =
+ NETDEV_LINUX_CLASS(
+ "internal",
+ netdev_linux_create,
+ NULL, /* enumerate */
+ netdev_vport_set_stats);
\f
/* HTB traffic control class. */
};
struct htb_class {
+ struct tc_queue tc_queue;
unsigned int min_rate; /* In bytes/s. */
unsigned int max_rate; /* In bytes/s. */
unsigned int burst; /* In bytes. */
return CONTAINER_OF(netdev_dev->tc, struct htb, tc);
}
-static struct htb *
+static void
htb_install__(struct netdev *netdev, uint64_t max_rate)
{
struct netdev_dev_linux *netdev_dev =
htb->max_rate = max_rate;
netdev_dev->tc = &htb->tc;
-
- return htb;
}
/* Create an HTB qdisc.
*
- * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default
- * 0". */
+ * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
static int
htb_setup_qdisc__(struct netdev *netdev)
{
tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
NLM_F_EXCL | NLM_F_CREATE, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = tc_make_handle(1, 0);
tcmsg->tcm_parent = TC_H_ROOT;
memset(&opt, 0, sizeof opt);
opt.rate2quantum = 10;
opt.version = 3;
- opt.defcls = 0;
+ opt.defcls = 1;
opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
int mtu;
netdev_get_mtu(netdev, &mtu);
+ if (mtu == INT_MAX) {
+ VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
+ netdev_get_name(netdev));
+ return EINVAL;
+ }
memset(&opt, 0, sizeof opt);
tc_fill_rate(&opt.rate, class->min_rate, mtu);
opt.prio = class->priority;
tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = handle;
tcmsg->tcm_parent = parent;
const char *priority_s = shash_find_data(details, "priority");
int mtu;
- /* min-rate */
- if (!min_rate_s) {
- /* min-rate is required. */
+ netdev_get_mtu(netdev, &mtu);
+ if (mtu == INT_MAX) {
+ VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU",
+ netdev_get_name(netdev));
return EINVAL;
}
- hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8;
- hc->min_rate = MAX(hc->min_rate, 0);
+
+ /* HTB requires at least an mtu sized min-rate to send any traffic even
+ * on uncongested links. */
+ hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
+ hc->min_rate = MAX(hc->min_rate, mtu);
hc->min_rate = MIN(hc->min_rate, htb->max_rate);
/* max-rate */
* doesn't include the Ethernet header, we need to add at least 14 (18?) to
* the MTU. We actually add 64, instead of 14, as a guard against
* additional headers get tacked on somewhere that we're not aware of. */
- netdev_get_mtu(netdev, &mtu);
hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0;
hc->burst = MAX(hc->burst, mtu + 64);
return error;
}
+static struct htb_class *
+htb_class_cast__(const struct tc_queue *queue)
+{
+ return CONTAINER_OF(queue, struct htb_class, tc_queue);
+}
+
static void
htb_update_queue__(struct netdev *netdev, unsigned int queue_id,
const struct htb_class *hc)
{
struct htb *htb = htb_get__(netdev);
+ size_t hash = hash_int(queue_id, 0);
+ struct tc_queue *queue;
struct htb_class *hcp;
- hcp = port_array_get(&htb->tc.queues, queue_id);
- if (!hcp) {
+ queue = tc_find_queue__(netdev, queue_id, hash);
+ if (queue) {
+ hcp = htb_class_cast__(queue);
+ } else {
hcp = xmalloc(sizeof *hcp);
- port_array_set(&htb->tc.queues, queue_id, hcp);
+ queue = &hcp->tc_queue;
+ queue->queue_id = queue_id;
+ hmap_insert(&htb->tc.queues, &queue->hmap_node, hash);
}
- *hcp = *hc;
+
+ hcp->min_rate = hc->min_rate;
+ hcp->max_rate = hc->max_rate;
+ hcp->burst = hc->burst;
+ hcp->priority = hc->priority;
}
static int
htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
- struct shash details = SHASH_INITIALIZER(&details);
struct ofpbuf msg;
struct nl_dump dump;
struct htb_class hc;
- struct htb *htb;
/* Get qdisc options. */
hc.max_rate = 0;
htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
- htb = htb_install__(netdev, hc.max_rate);
+ htb_install__(netdev, hc.max_rate);
/* Get queues. */
- start_queue_dump(netdev, &dump);
- shash_init(&details);
+ if (!start_queue_dump(netdev, &dump)) {
+ return ENODEV;
+ }
while (nl_dump_next(&dump, &msg)) {
unsigned int queue_id;
htb_tc_destroy(struct tc *tc)
{
struct htb *htb = CONTAINER_OF(tc, struct htb, tc);
- unsigned int queue_id;
- struct htb_class *hc;
+ struct htb_class *hc, *next;
- PORT_ARRAY_FOR_EACH (hc, &htb->tc.queues, queue_id) {
+ HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) {
+ hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
free(hc);
}
tc_destroy(tc);
}
static int
-htb_class_get(const struct netdev *netdev, unsigned int queue_id,
- struct shash *details)
+htb_class_get(const struct netdev *netdev OVS_UNUSED,
+ const struct tc_queue *queue, struct shash *details)
{
- const struct htb *htb = htb_get__(netdev);
- const struct htb_class *hc;
-
- hc = port_array_get(&htb->tc.queues, queue_id);
- assert(hc != NULL);
+ const struct htb_class *hc = htb_class_cast__(queue);
shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
if (hc->min_rate != hc->max_rate) {
}
static int
-htb_class_delete(struct netdev *netdev, unsigned int queue_id)
+htb_class_delete(struct netdev *netdev, struct tc_queue *queue)
{
+ struct htb_class *hc = htb_class_cast__(queue);
struct htb *htb = htb_get__(netdev);
- struct htb_class *hc;
int error;
- hc = port_array_get(&htb->tc.queues, queue_id);
- assert(hc != NULL);
-
- error = tc_delete_class(netdev, tc_make_handle(1, queue_id + 1));
+ error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
if (!error) {
+ hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node);
free(hc);
- port_array_delete(&htb->tc.queues, queue_id);
}
return error;
}
static int
-htb_class_get_stats(const struct netdev *netdev, unsigned int queue_id,
+htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
struct netdev_queue_stats *stats)
{
- return htb_query_class__(netdev, tc_make_handle(1, queue_id + 1),
+ return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
tc_make_handle(1, 0xfffe), NULL, stats);
}
major = tc_get_major(handle);
minor = tc_get_minor(handle);
if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) {
- (*cb)(tc_get_minor(handle), &stats, aux);
+ (*cb)(minor - 1, &stats, aux);
}
return 0;
}
htb_class_dump_stats
};
\f
+/* "linux-hfsc" traffic control class. */
+
+#define HFSC_N_QUEUES 0xf000
+
+struct hfsc {
+ struct tc tc;
+ uint32_t max_rate;
+};
+
+struct hfsc_class {
+ struct tc_queue tc_queue;
+ uint32_t min_rate;
+ uint32_t max_rate;
+};
+
+static struct hfsc *
+hfsc_get__(const struct netdev *netdev)
+{
+ struct netdev_dev_linux *netdev_dev;
+ netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
+ return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc);
+}
+
+static struct hfsc_class *
+hfsc_class_cast__(const struct tc_queue *queue)
+{
+ return CONTAINER_OF(queue, struct hfsc_class, tc_queue);
+}
+
+static void
+hfsc_install__(struct netdev *netdev, uint32_t max_rate)
+{
+ struct netdev_dev_linux * netdev_dev;
+ struct hfsc *hfsc;
+
+ netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev));
+ hfsc = xmalloc(sizeof *hfsc);
+ tc_init(&hfsc->tc, &tc_ops_hfsc);
+ hfsc->max_rate = max_rate;
+ netdev_dev->tc = &hfsc->tc;
+}
+
+static void
+hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id,
+ const struct hfsc_class *hc)
+{
+ size_t hash;
+ struct hfsc *hfsc;
+ struct hfsc_class *hcp;
+ struct tc_queue *queue;
+
+ hfsc = hfsc_get__(netdev);
+ hash = hash_int(queue_id, 0);
+
+ queue = tc_find_queue__(netdev, queue_id, hash);
+ if (queue) {
+ hcp = hfsc_class_cast__(queue);
+ } else {
+ hcp = xmalloc(sizeof *hcp);
+ queue = &hcp->tc_queue;
+ queue->queue_id = queue_id;
+ hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash);
+ }
+
+ hcp->min_rate = hc->min_rate;
+ hcp->max_rate = hc->max_rate;
+}
+
+static int
+hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class)
+{
+ const struct tc_service_curve *rsc, *fsc, *usc;
+ static const struct nl_policy tca_hfsc_policy[] = {
+ [TCA_HFSC_RSC] = {
+ .type = NL_A_UNSPEC,
+ .optional = false,
+ .min_len = sizeof(struct tc_service_curve),
+ },
+ [TCA_HFSC_FSC] = {
+ .type = NL_A_UNSPEC,
+ .optional = false,
+ .min_len = sizeof(struct tc_service_curve),
+ },
+ [TCA_HFSC_USC] = {
+ .type = NL_A_UNSPEC,
+ .optional = false,
+ .min_len = sizeof(struct tc_service_curve),
+ },
+ };
+ struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)];
+
+ if (!nl_parse_nested(nl_options, tca_hfsc_policy,
+ attrs, ARRAY_SIZE(tca_hfsc_policy))) {
+ VLOG_WARN_RL(&rl, "failed to parse HFSC class options");
+ return EPROTO;
+ }
+
+ rsc = nl_attr_get(attrs[TCA_HFSC_RSC]);
+ fsc = nl_attr_get(attrs[TCA_HFSC_FSC]);
+ usc = nl_attr_get(attrs[TCA_HFSC_USC]);
+
+ if (rsc->m1 != 0 || rsc->d != 0 ||
+ fsc->m1 != 0 || fsc->d != 0 ||
+ usc->m1 != 0 || usc->d != 0) {
+ VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
+ "Non-linear service curves are not supported.");
+ return EPROTO;
+ }
+
+ if (rsc->m2 != fsc->m2) {
+ VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
+ "Real-time service curves are not supported ");
+ return EPROTO;
+ }
+
+ if (rsc->m2 > usc->m2) {
+ VLOG_WARN_RL(&rl, "failed to parse HFSC class options. "
+ "Min-rate service curve is greater than "
+ "the max-rate service curve.");
+ return EPROTO;
+ }
+
+ class->min_rate = fsc->m2;
+ class->max_rate = usc->m2;
+ return 0;
+}
+
+static int
+hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id,
+ struct hfsc_class *options,
+ struct netdev_queue_stats *stats)
+{
+ int error;
+ unsigned int handle;
+ struct nlattr *nl_options;
+
+ error = tc_parse_class(tcmsg, &handle, &nl_options, stats);
+ if (error) {
+ return error;
+ }
+
+ if (queue_id) {
+ unsigned int major, minor;
+
+ major = tc_get_major(handle);
+ minor = tc_get_minor(handle);
+ if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
+ *queue_id = minor - 1;
+ } else {
+ return EPROTO;
+ }
+ }
+
+ if (options) {
+ error = hfsc_parse_tca_options__(nl_options, options);
+ }
+
+ return error;
+}
+
+static int
+hfsc_query_class__(const struct netdev *netdev, unsigned int handle,
+ unsigned int parent, struct hfsc_class *options,
+ struct netdev_queue_stats *stats)
+{
+ int error;
+ struct ofpbuf *reply;
+
+ error = tc_query_class(netdev, handle, parent, &reply);
+ if (error) {
+ return error;
+ }
+
+ error = hfsc_parse_tcmsg__(reply, NULL, options, stats);
+ ofpbuf_delete(reply);
+ return error;
+}
+
+static void
+hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details,
+ struct hfsc_class *class)
+{
+ uint32_t max_rate;
+ const char *max_rate_s;
+
+ max_rate_s = shash_find_data(details, "max-rate");
+ max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0;
+
+ if (!max_rate) {
+ uint32_t current;
+
+ netdev_get_features(netdev, ¤t, NULL, NULL, NULL);
+ max_rate = netdev_features_to_bps(current) / 8;
+ }
+
+ class->min_rate = max_rate;
+ class->max_rate = max_rate;
+}
+
+static int
+hfsc_parse_class_details__(struct netdev *netdev,
+ const struct shash *details,
+ struct hfsc_class * class)
+{
+ const struct hfsc *hfsc;
+ uint32_t min_rate, max_rate;
+ const char *min_rate_s, *max_rate_s;
+
+ hfsc = hfsc_get__(netdev);
+ min_rate_s = shash_find_data(details, "min-rate");
+ max_rate_s = shash_find_data(details, "max-rate");
+
+ min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0;
+ min_rate = MAX(min_rate, 1);
+ min_rate = MIN(min_rate, hfsc->max_rate);
+
+ max_rate = (max_rate_s
+ ? strtoull(max_rate_s, NULL, 10) / 8
+ : hfsc->max_rate);
+ max_rate = MAX(max_rate, min_rate);
+ max_rate = MIN(max_rate, hfsc->max_rate);
+
+ class->min_rate = min_rate;
+ class->max_rate = max_rate;
+
+ return 0;
+}
+
+/* Create an HFSC qdisc.
+ *
+ * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */
+static int
+hfsc_setup_qdisc__(struct netdev * netdev)
+{
+ struct tcmsg *tcmsg;
+ struct ofpbuf request;
+ struct tc_hfsc_qopt opt;
+
+ tc_del_qdisc(netdev);
+
+ tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
+ NLM_F_EXCL | NLM_F_CREATE, &request);
+
+ if (!tcmsg) {
+ return ENODEV;
+ }
+
+ tcmsg->tcm_handle = tc_make_handle(1, 0);
+ tcmsg->tcm_parent = TC_H_ROOT;
+
+ memset(&opt, 0, sizeof opt);
+ opt.defcls = 1;
+
+ nl_msg_put_string(&request, TCA_KIND, "hfsc");
+ nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
+
+ return tc_transact(&request, NULL);
+}
+
+/* Create an HFSC class.
+ *
+ * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc
+ * sc rate <min_rate> ul rate <max_rate>" */
+static int
+hfsc_setup_class__(struct netdev *netdev, unsigned int handle,
+ unsigned int parent, struct hfsc_class *class)
+{
+ int error;
+ size_t opt_offset;
+ struct tcmsg *tcmsg;
+ struct ofpbuf request;
+ struct tc_service_curve min, max;
+
+ tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request);
+
+ if (!tcmsg) {
+ return ENODEV;
+ }
+
+ tcmsg->tcm_handle = handle;
+ tcmsg->tcm_parent = parent;
+
+ min.m1 = 0;
+ min.d = 0;
+ min.m2 = class->min_rate;
+
+ max.m1 = 0;
+ max.d = 0;
+ max.m2 = class->max_rate;
+
+ nl_msg_put_string(&request, TCA_KIND, "hfsc");
+ opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
+ nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min);
+ nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min);
+ nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max);
+ nl_msg_end_nested(&request, opt_offset);
+
+ error = tc_transact(&request, NULL);
+ if (error) {
+ VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, "
+ "min-rate %ubps, max-rate %ubps (%s)",
+ netdev_get_name(netdev),
+ tc_get_major(handle), tc_get_minor(handle),
+ tc_get_major(parent), tc_get_minor(parent),
+ class->min_rate, class->max_rate, strerror(error));
+ }
+
+ return error;
+}
+
+static int
+hfsc_tc_install(struct netdev *netdev, const struct shash *details)
+{
+ int error;
+ struct hfsc_class class;
+
+ error = hfsc_setup_qdisc__(netdev);
+
+ if (error) {
+ return error;
+ }
+
+ hfsc_parse_qdisc_details__(netdev, details, &class);
+ error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
+ tc_make_handle(1, 0), &class);
+
+ if (error) {
+ return error;
+ }
+
+ hfsc_install__(netdev, class.max_rate);
+ return 0;
+}
+
+static int
+hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
+{
+ struct ofpbuf msg;
+ struct nl_dump dump;
+ struct hfsc_class hc;
+
+ hc.max_rate = 0;
+ hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
+ hfsc_install__(netdev, hc.max_rate);
+
+ if (!start_queue_dump(netdev, &dump)) {
+ return ENODEV;
+ }
+
+ while (nl_dump_next(&dump, &msg)) {
+ unsigned int queue_id;
+
+ if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
+ hfsc_update_queue__(netdev, queue_id, &hc);
+ }
+ }
+
+ nl_dump_done(&dump);
+ return 0;
+}
+
+static void
+hfsc_tc_destroy(struct tc *tc)
+{
+ struct hfsc *hfsc;
+ struct hfsc_class *hc, *next;
+
+ hfsc = CONTAINER_OF(tc, struct hfsc, tc);
+
+ HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) {
+ hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
+ free(hc);
+ }
+
+ tc_destroy(tc);
+ free(hfsc);
+}
+
+static int
+hfsc_qdisc_get(const struct netdev *netdev, struct shash *details)
+{
+ const struct hfsc *hfsc;
+ hfsc = hfsc_get__(netdev);
+ shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate));
+ return 0;
+}
+
+static int
+hfsc_qdisc_set(struct netdev *netdev, const struct shash *details)
+{
+ int error;
+ struct hfsc_class class;
+
+ hfsc_parse_qdisc_details__(netdev, details, &class);
+ error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe),
+ tc_make_handle(1, 0), &class);
+
+ if (!error) {
+ hfsc_get__(netdev)->max_rate = class.max_rate;
+ }
+
+ return error;
+}
+
+static int
+hfsc_class_get(const struct netdev *netdev OVS_UNUSED,
+ const struct tc_queue *queue, struct shash *details)
+{
+ const struct hfsc_class *hc;
+
+ hc = hfsc_class_cast__(queue);
+ shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate));
+ if (hc->min_rate != hc->max_rate) {
+ shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate));
+ }
+ return 0;
+}
+
+static int
+hfsc_class_set(struct netdev *netdev, unsigned int queue_id,
+ const struct shash *details)
+{
+ int error;
+ struct hfsc_class class;
+
+ error = hfsc_parse_class_details__(netdev, details, &class);
+ if (error) {
+ return error;
+ }
+
+ error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1),
+ tc_make_handle(1, 0xfffe), &class);
+ if (error) {
+ return error;
+ }
+
+ hfsc_update_queue__(netdev, queue_id, &class);
+ return 0;
+}
+
+static int
+hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue)
+{
+ int error;
+ struct hfsc *hfsc;
+ struct hfsc_class *hc;
+
+ hc = hfsc_class_cast__(queue);
+ hfsc = hfsc_get__(netdev);
+
+ error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1));
+ if (!error) {
+ hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node);
+ free(hc);
+ }
+ return error;
+}
+
+static int
+hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue,
+ struct netdev_queue_stats *stats)
+{
+ return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1),
+ tc_make_handle(1, 0xfffe), NULL, stats);
+}
+
+static int
+hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED,
+ const struct ofpbuf *nlmsg,
+ netdev_dump_queue_stats_cb *cb, void *aux)
+{
+ struct netdev_queue_stats stats;
+ unsigned int handle, major, minor;
+ int error;
+
+ error = tc_parse_class(nlmsg, &handle, NULL, &stats);
+ if (error) {
+ return error;
+ }
+
+ major = tc_get_major(handle);
+ minor = tc_get_minor(handle);
+ if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) {
+ (*cb)(minor - 1, &stats, aux);
+ }
+ return 0;
+}
+
+static const struct tc_ops tc_ops_hfsc = {
+ "hfsc", /* linux_name */
+ "linux-hfsc", /* ovs_name */
+ HFSC_N_QUEUES, /* n_queues */
+ hfsc_tc_install, /* tc_install */
+ hfsc_tc_load, /* tc_load */
+ hfsc_tc_destroy, /* tc_destroy */
+ hfsc_qdisc_get, /* qdisc_get */
+ hfsc_qdisc_set, /* qdisc_set */
+ hfsc_class_get, /* class_get */
+ hfsc_class_set, /* class_set */
+ hfsc_class_delete, /* class_delete */
+ hfsc_class_get_stats, /* class_get_stats */
+ hfsc_class_dump_stats /* class_dump_stats */
+};
+\f
/* "linux-default" traffic control class.
*
* This class represents the default, unnamed Linux qdisc. It corresponds to
if (!buffer_hz) {
read_psched();
}
- return ((unsigned long long int) ticks_per_s * size) / rate;
+ return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0;
}
/* Returns the number of bytes that need to be reserved for qdisc buffering at
int error;
tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = handle;
tcmsg->tcm_parent = parent;
int error;
tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = handle;
tcmsg->tcm_parent = 0;
int error;
tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = tc_make_handle(1, 0);
tcmsg->tcm_parent = TC_H_ROOT;
* We could check for Linux 2.6.35+ and use a more straightforward method
* there. */
tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
tcmsg->tcm_handle = tc_make_handle(1, 0);
tcmsg->tcm_parent = 0;
/* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a
* rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested
* burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of
- * 0 is fine.)
- *
- * This */
+ * 0 is fine.) */
static int
tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes)
{
{
struct ifreq ifr;
- strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
+ ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
COVERAGE_INC(netdev_get_ifindex);
if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
int hwaddr_family;
memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
+ ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
COVERAGE_INC(netdev_get_hwaddr);
if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
struct ifreq ifr;
memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
+ ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
ifr.ifr_hwaddr.sa_family = hwaddr_family;
memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
COVERAGE_INC(netdev_set_hwaddr);
struct ifreq ifr;
memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
+ ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
ifr.ifr_data = (caddr_t) ecmd;
ecmd->cmd = cmd;
netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
const char *cmd_name)
{
- strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
+ ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
if (ioctl(af_inet_sock, cmd, ifr) == -1) {
VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
strerror(errno));