X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=5c7065fe1ab31309be55567ef9d10c5372b683ec;hb=003ce655b7116d18c86a74c50391e54990346931;hp=2fa05b66e3e2162f346a0369d1e2c641a3a15c76;hpb=61b999dd6fa3a8942b5002f9392a0068324a80b6;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 2fa05b66e..5c7065fe1 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010 Nicira Networks. + * 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. @@ -15,26 +15,29 @@ */ #include -#include + +#include "netdev-linux.h" + #include #include #include #include +#include +#include +#include #include -#include #include #include +#include +#include #include #include #include -#include #include #include #include #include -#include #include -#include #include #include #include @@ -45,23 +48,40 @@ #include #include "coverage.h" +#include "dpif-linux.h" +#include "dpif-netdev.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-notifier.h" +#include "netlink-socket.h" #include "netlink.h" #include "ofpbuf.h" #include "openflow/openflow.h" -#include "openvswitch/internal_dev.h" -#include "openvswitch/gre.h" +#include "ovs-atomic.h" #include "packets.h" #include "poll-loop.h" -#include "rtnetlink.h" -#include "socket-util.h" +#include "rtnetlink-link.h" #include "shash.h" -#include "svec.h" - -#define THIS_MODULE VLM_netdev_linux +#include "socket-util.h" +#include "sset.h" +#include "timer.h" +#include "unaligned.h" #include "vlog.h" + +VLOG_DEFINE_THIS_MODULE(netdev_linux); + +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_get_ethtool); +COVERAGE_DEFINE(netdev_set_ethtool); + /* These were introduced in Linux 2.6.14, so they might be missing if we have * old headers. */ @@ -72,31 +92,306 @@ #define ADVERTISED_Asym_Pause (1 << 14) #endif -static struct rtnetlink_notifier netdev_linux_cache_notifier; -static int cache_notifier_refcount; +/* These were introduced in Linux 2.6.24, so they might be missing if we + * have old headers. */ +#ifndef ETHTOOL_GFLAGS +#define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ +#endif +#ifndef ETHTOOL_SFLAGS +#define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ +#endif + +/* This was introduced in Linux 2.6.25, so it might be missing if we have old + * headers. */ +#ifndef TC_RTAB_SIZE +#define TC_RTAB_SIZE 1024 +#endif + +/* Linux 2.6.21 introduced struct tpacket_auxdata. + * Linux 2.6.27 added the tp_vlan_tci member. + * Linux 3.0 defined TP_STATUS_VLAN_VALID. + * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined + * TP_STATUS_VLAN_TPID_VALID. + * + * With all this churn it's easiest to unconditionally define a replacement + * structure that has everything we want. + */ +#ifndef PACKET_AUXDATA +#define PACKET_AUXDATA 8 +#endif +#ifndef TP_STATUS_VLAN_VALID +#define TP_STATUS_VLAN_VALID (1 << 4) +#endif +#ifndef TP_STATUS_VLAN_TPID_VALID +#define TP_STATUS_VLAN_TPID_VALID (1 << 6) +#endif +#undef tpacket_auxdata +#define tpacket_auxdata rpl_tpacket_auxdata +struct tpacket_auxdata { + uint32_t tp_status; + uint32_t tp_len; + uint32_t tp_snaplen; + uint16_t tp_mac; + uint16_t tp_net; + uint16_t tp_vlan_tci; + uint16_t tp_vlan_tpid; +}; enum { - VALID_IFINDEX = 1 << 0, - VALID_ETHERADDR = 1 << 1, - VALID_IN4 = 1 << 2, - VALID_IN6 = 1 << 3, - VALID_MTU = 1 << 4, - VALID_CARRIER = 1 << 5, - VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */ - VALID_POLICING = 1 << 7 + VALID_IFINDEX = 1 << 0, + VALID_ETHERADDR = 1 << 1, + VALID_IN4 = 1 << 2, + VALID_IN6 = 1 << 3, + VALID_MTU = 1 << 4, + VALID_POLICING = 1 << 5, + VALID_VPORT_STAT_ERROR = 1 << 6, + VALID_DRVINFO = 1 << 7, + VALID_FEATURES = 1 << 8, +}; + +/* Traffic control. */ + +/* An instance of a traffic control class. Always associated with a particular + * 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. */ }; -struct tap_state { - int fd; - bool opened; +#define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->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. */ + long long int created; /* Time queue was created, in msecs. */ +}; + +/* A particular kind of traffic control. Each implementation generally maps to + * one particular Linux qdisc class. + * + * The functions below return 0 if successful or a positive errno value on + * failure, except where otherwise noted. All of them must be provided, except + * where otherwise noted. */ +struct tc_ops { + /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". + * This is null for tc_ops_default and tc_ops_other, for which there are no + * appropriate values. */ + const char *linux_name; + + /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ + const char *ovs_name; + + /* Number of supported OpenFlow queues, 0 for qdiscs that have no + * queues. The queues are numbered 0 through n_queues - 1. */ + unsigned int n_queues; + + /* Called to install this TC class on 'netdev'. The implementation should + * make the Netlink calls required to set up 'netdev' with the right qdisc + * and configure it according to 'details'. The implementation may assume + * that the current qdisc is the default; that is, there is no need for it + * to delete the current qdisc before installing itself. + * + * The contents of 'details' should be documented as valid for 'ovs_name' + * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml + * (which is built as ovs-vswitchd.conf.db(8)). + * + * This function must return 0 if and only if it sets 'netdev->tc' to an + * initialized 'struct tc'. + * + * (This function is null for tc_ops_other, which cannot be installed. For + * other TC classes it should always be nonnull.) */ + int (*tc_install)(struct netdev *netdev, const struct smap *details); + + /* Called when the netdev code determines (through a Netlink query) that + * this TC class's qdisc is installed on 'netdev', but we didn't install + * it ourselves and so don't know any of the details. + * + * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for + * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The + * implementation should parse the other attributes of 'nlmsg' as + * necessary to determine its configuration. If necessary it should also + * use Netlink queries to determine the configuration of queues on + * 'netdev'. + * + * This function must return 0 if and only if it sets 'netdev->tc' to an + * initialized 'struct tc'. */ + int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); + + /* Destroys the data structures allocated by the implementation as part of + * 'tc'. (This includes destroying 'tc->queues' by calling + * tc_destroy(tc). + * + * The implementation should not need to perform any Netlink calls. If + * desirable, the caller is responsible for deconfiguring the kernel qdisc. + * (But it may not be desirable.) + * + * This function may be null if 'tc' is trivial. */ + void (*tc_destroy)(struct tc *tc); + + /* Retrieves details of 'netdev->tc' configuration into 'details'. + * + * The implementation should not need to perform any Netlink calls, because + * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have + * cached the configuration. + * + * The contents of 'details' should be documented as valid for 'ovs_name' + * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml + * (which is built as ovs-vswitchd.conf.db(8)). + * + * This function may be null if 'tc' is not configurable. + */ + int (*qdisc_get)(const struct netdev *netdev, struct smap *details); + + /* Reconfigures 'netdev->tc' according to 'details', performing any + * required Netlink calls to complete the reconfiguration. + * + * The contents of 'details' should be documented as valid for 'ovs_name' + * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml + * (which is built as ovs-vswitchd.conf.db(8)). + * + * This function may be null if 'tc' is not configurable. + */ + int (*qdisc_set)(struct netdev *, const struct smap *details); + + /* 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 + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * The implementation should not need to perform any Netlink calls, because + * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have + * cached the queue configuration. + * + * This function may be null if 'tc' does not have queues ('n_queues' is + * 0). */ + int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, + struct smap *details); + + /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to + * 'details', perfoming any required Netlink calls to complete the + * reconfiguration. The caller ensures that 'queue_id' is less than + * 'n_queues'. + * + * The contents of 'details' should be documented as valid for 'ovs_name' + * in the "other_config" column in the "Queue" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * This function may be null if 'tc' does not have queues or its queues are + * not configurable. */ + int (*class_set)(struct netdev *, unsigned int queue_id, + const struct smap *details); + + /* 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 *, struct tc_queue *queue); + + /* 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, + const struct tc_queue *queue, + struct netdev_queue_stats *stats); + + /* Extracts queue stats from 'nlmsg', which is a response to a + * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. + * + * This function may be null if 'tc' does not have queues or if it cannot + * report queue statistics. */ + int (*class_dump_stats)(const struct netdev *netdev, + const struct ofpbuf *nlmsg, + netdev_dump_queue_stats_cb *cb, void *aux); +}; + +static void +tc_init(struct tc *tc, const struct tc_ops *ops) +{ + tc->ops = ops; + hmap_init(&tc->queues); +} + +static void +tc_destroy(struct tc *tc) +{ + 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 *const 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 }; -struct netdev_dev_linux { - struct netdev_dev netdev_dev; +static unsigned int tc_make_handle(unsigned int major, unsigned int minor); +static unsigned int tc_get_major(unsigned int handle); +static unsigned int tc_get_minor(unsigned int handle); + +static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); +static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); +static unsigned int tc_buffer_per_jiffy(unsigned int rate); + +static struct tcmsg *tc_make_request(const struct netdev *, int type, + unsigned int flags, struct ofpbuf *); +static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp); +static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add); +static int tc_add_policer(struct netdev *netdev, int kbits_rate, + int kbits_burst); + +static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, + struct nlattr **options); +static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, + struct nlattr **options, + struct netdev_queue_stats *); +static int tc_query_class(const struct netdev *, + unsigned int handle, unsigned int parent, + struct ofpbuf **replyp); +static int tc_delete_class(const struct netdev *, unsigned int handle); + +static int tc_del_qdisc(struct netdev *netdev); +static int tc_query_qdisc(const struct netdev *netdev); + +static int tc_calc_cell_log(unsigned int mtu); +static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); +static void tc_put_rtab(struct ofpbuf *, uint16_t type, + const struct tc_ratespec *rate); +static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); + +struct netdev_linux { + struct netdev up; + + /* Protects all members below. */ + struct ovs_mutex mutex; - struct shash_node *shash_node; unsigned int cache_valid; + bool miimon; /* Link status of last poll. */ + long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ + struct timer miimon_timer; + /* The following are figured out "on demand" only. They are only valid * when the corresponding VALID_* bit in 'cache_valid' is set. */ int ifindex; @@ -104,166 +399,297 @@ struct netdev_dev_linux { struct in_addr address, netmask; struct in6_addr in6; int mtu; - int carrier; - bool is_internal; /* Is this an openvswitch internal device? */ - bool is_tap; /* Is this a tuntap device? */ + unsigned int ifi_flags; + long long int carrier_resets; uint32_t kbits_rate; /* Policing data. */ uint32_t kbits_burst; - - union { - struct tap_state tap; - } state; + int vport_stats_error; /* Cached error code from vport_get_stats(). + 0 or an errno value. */ + int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */ + int ether_addr_error; /* Cached error code from set/get etheraddr. */ + int netdev_policing_error; /* Cached error code from set policing. */ + int get_features_error; /* Cached error code from ETHTOOL_GSET. */ + int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */ + + enum netdev_features current; /* Cached from ETHTOOL_GSET. */ + enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ + enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ + + struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ + struct tc *tc; + + /* For devices of class netdev_tap_class only. */ + int tap_fd; }; -struct netdev_linux { - struct netdev netdev; +struct netdev_rxq_linux { + struct netdev_rxq up; + bool is_tap; int fd; }; -/* An AF_INET socket (used for ioctl operations). */ -static int af_inet_sock = -1; - -struct netdev_linux_notifier { - struct netdev_notifier notifier; - struct list node; -}; - -static struct shash netdev_linux_notifiers = - SHASH_INITIALIZER(&netdev_linux_notifiers); -static struct rtnetlink_notifier netdev_linux_poll_notifier; - /* This is set pretty low because we probably won't learn anything from the * additional log messages. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); -static int netdev_linux_init(void); +/* Polling miimon status for all ports causes performance degradation when + * handling a large number of ports. If there are no devices using miimon, then + * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). */ +static atomic_int miimon_cnt = ATOMIC_VAR_INIT(0); + +static void netdev_linux_run(void); static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, int cmd, const char *cmd_name); -static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, - const char *cmd_name); static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, int cmd, const char *cmd_name); -static int get_flags(const struct netdev *, int *flagsp); -static int set_flags(struct netdev *, int flags); +static int get_flags(const struct netdev *, unsigned int *flags); +static int set_flags(const char *, unsigned int flags); +static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) + OVS_REQUIRES(netdev->mutex); static int do_get_ifindex(const char *netdev_name); static int get_ifindex(const struct netdev *, int *ifindexp); static int do_set_addr(struct netdev *netdev, int ioctl_nr, const char *ioctl_name, struct in_addr addr); static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); -static int set_etheraddr(const char *netdev_name, int hwaddr_family, - const uint8_t[ETH_ADDR_LEN]); -static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); -static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); -static int get_rtnl_sock(struct nl_sock **); +static int set_etheraddr(const char *netdev_name, const uint8_t[ETH_ADDR_LEN]); +static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *); +static int af_packet_sock(void); +static bool netdev_linux_miimon_enabled(void); +static void netdev_linux_miimon_run(void); +static void netdev_linux_miimon_wait(void); +static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup); static bool is_netdev_linux_class(const struct netdev_class *netdev_class) { - return netdev_class->init == netdev_linux_init; + return netdev_class->run == netdev_linux_run; } -static struct netdev_dev_linux * -netdev_dev_linux_cast(const struct netdev_dev *netdev_dev) +static bool +is_tap_netdev(const struct netdev *netdev) { - const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); - assert(is_netdev_linux_class(netdev_class)); - - return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev); + return netdev_get_class(netdev) == &netdev_tap_class; } static struct netdev_linux * netdev_linux_cast(const struct netdev *netdev) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); - const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); - assert(is_netdev_linux_class(netdev_class)); + ovs_assert(is_netdev_linux_class(netdev_get_class(netdev))); - return CONTAINER_OF(netdev, struct netdev_linux, netdev); + return CONTAINER_OF(netdev, struct netdev_linux, up); } -static int -netdev_linux_init(void) +static struct netdev_rxq_linux * +netdev_rxq_linux_cast(const struct netdev_rxq *rx) { - static int status = -1; - if (status < 0) { - af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); - status = af_inet_sock >= 0 ? 0 : errno; - if (status) { - VLOG_ERR("failed to create inet socket: %s", strerror(status)); + ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev))); + return CONTAINER_OF(rx, struct netdev_rxq_linux, up); +} + +static void netdev_linux_update(struct netdev_linux *netdev, + const struct rtnetlink_link_change *) + OVS_REQUIRES(netdev->mutex); +static void netdev_linux_changed(struct netdev_linux *netdev, + unsigned int ifi_flags, unsigned int mask) + OVS_REQUIRES(netdev->mutex); + +/* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK changes, or NULL + * if no such socket could be created. */ +static struct nl_sock * +netdev_linux_notify_sock(void) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static struct nl_sock *sock; + + if (ovsthread_once_start(&once)) { + int error; + + error = nl_sock_create(NETLINK_ROUTE, &sock); + if (!error) { + error = nl_sock_join_mcgroup(sock, RTNLGRP_LINK); + if (error) { + nl_sock_destroy(sock); + sock = NULL; + } } + ovsthread_once_done(&once); } - return status; + + return sock; +} + +static bool +netdev_linux_miimon_enabled(void) +{ + int miimon; + + atomic_read(&miimon_cnt, &miimon); + return miimon > 0; } static void netdev_linux_run(void) { - rtnetlink_notifier_run(); + struct nl_sock *sock; + int error; + + if (netdev_linux_miimon_enabled()) { + netdev_linux_miimon_run(); + } + + sock = netdev_linux_notify_sock(); + if (!sock) { + return; + } + + do { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + uint64_t buf_stub[4096 / 8]; + struct ofpbuf buf; + + ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); + error = nl_sock_recv(sock, &buf, false); + if (!error) { + struct rtnetlink_link_change change; + + if (rtnetlink_link_parse(&buf, &change)) { + struct netdev *netdev_ = netdev_from_name(change.ifname); + if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); + netdev_linux_update(netdev, &change); + ovs_mutex_unlock(&netdev->mutex); + } + netdev_close(netdev_); + } + } else if (error == ENOBUFS) { + struct shash device_shash; + struct shash_node *node; + + nl_sock_drain(sock); + + shash_init(&device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev *netdev_ = node->data; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + unsigned int flags; + + ovs_mutex_lock(&netdev->mutex); + get_flags(netdev_, &flags); + netdev_linux_changed(netdev, flags, 0); + ovs_mutex_unlock(&netdev->mutex); + + netdev_close(netdev_); + } + shash_destroy(&device_shash); + } else if (error != EAGAIN) { + VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)", + ovs_strerror(error)); + } + ofpbuf_uninit(&buf); + } while (!error); } static void netdev_linux_wait(void) { - rtnetlink_notifier_wait(); + struct nl_sock *sock; + + if (netdev_linux_miimon_enabled()) { + netdev_linux_miimon_wait(); + } + sock = netdev_linux_notify_sock(); + if (sock) { + nl_sock_wait(sock, POLLIN); + } } static void -netdev_linux_cache_cb(const struct rtnetlink_change *change, - void *aux OVS_UNUSED) -{ - struct netdev_dev_linux *dev; - if (change) { - struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname); - if (base_dev) { - const struct netdev_class *netdev_class = - netdev_dev_get_class(base_dev); - - if (is_netdev_linux_class(netdev_class)) { - dev = netdev_dev_linux_cast(base_dev); - dev->cache_valid = 0; - } +netdev_linux_changed(struct netdev_linux *dev, + unsigned int ifi_flags, unsigned int mask) + OVS_REQUIRES(dev->mutex) +{ + netdev_change_seq_changed(&dev->up); + + if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { + dev->carrier_resets++; + } + dev->ifi_flags = ifi_flags; + + dev->cache_valid &= mask; +} + +static void +netdev_linux_update(struct netdev_linux *dev, + const struct rtnetlink_link_change *change) + OVS_REQUIRES(dev->mutex) +{ + if (change->nlmsg_type == RTM_NEWLINK) { + /* Keep drv-info */ + netdev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO); + + /* Update netdev from rtnl-change msg. */ + if (change->mtu) { + dev->mtu = change->mtu; + dev->cache_valid |= VALID_MTU; + dev->netdev_mtu_error = 0; } - } else { - struct shash device_shash; - struct shash_node *node; - - shash_init(&device_shash); - netdev_dev_get_devices(&netdev_linux_class, &device_shash); - SHASH_FOR_EACH (node, &device_shash) { - dev = node->data; - dev->cache_valid = 0; + + if (!eth_addr_is_zero(change->addr)) { + memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN); + dev->cache_valid |= VALID_ETHERADDR; + dev->ether_addr_error = 0; } - shash_destroy(&device_shash); + + dev->ifindex = change->ifi_index; + dev->cache_valid |= VALID_IFINDEX; + dev->get_ifindex_error = 0; + + } else { + netdev_linux_changed(dev, change->ifi_flags, 0); } } -/* Creates the netdev device of 'type' with 'name'. */ +static struct netdev * +netdev_linux_alloc(void) +{ + struct netdev_linux *netdev = xzalloc(sizeof *netdev); + return &netdev->up; +} + +static void +netdev_linux_common_construct(struct netdev_linux *netdev) +{ + ovs_mutex_init(&netdev->mutex); +} + +/* 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_construct(struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for system devices should be empty", name); - } + netdev_linux_common_construct(netdev); - if (!cache_notifier_refcount) { - error = rtnetlink_notifier_register(&netdev_linux_cache_notifier, - netdev_linux_cache_cb, NULL); - if (error) { - return error; + error = get_flags(&netdev->up, &netdev->ifi_flags); + if (error == ENODEV) { + if (netdev->up.netdev_class != &netdev_internal_class) { + /* The device does not exist, so don't allow it to be opened. */ + return ENODEV; + } else { + /* "Internal" netdevs 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(). + * Therefore, ignore the error. */ } } - cache_notifier_refcount++; - netdev_dev = xzalloc(sizeof *netdev_dev); - netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class); - - *netdev_devp = &netdev_dev->netdev_dev; return 0; } @@ -274,136 +700,134 @@ netdev_linux_create_system(const char *name, const char *type OVS_UNUSED, * 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_construct_tap(struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev; - struct tap_state *state; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); static const char tap_dev[] = "/dev/net/tun"; + const char *name = netdev_->name; struct ifreq ifr; int error; - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for TAP devices should be empty", name); - } - - netdev_dev = xzalloc(sizeof *netdev_dev); - state = &netdev_dev->state.tap; + netdev_linux_common_construct(netdev); /* Open tap device. */ - state->fd = open(tap_dev, O_RDWR); - if (state->fd < 0) { + netdev->tap_fd = open(tap_dev, O_RDWR); + if (netdev->tap_fd < 0) { error = errno; - VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); - goto error; + VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); + return error; } /* Create tap device. */ ifr.ifr_flags = IFF_TAP | IFF_NO_PI; - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { + ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); + if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { VLOG_WARN("%s: creating tap device failed: %s", name, - strerror(errno)); + ovs_strerror(errno)); error = errno; - goto error; + goto error_close; } /* Make non-blocking. */ - error = set_nonblocking(state->fd); + error = set_nonblocking(netdev->tap_fd); if (error) { - goto error; + goto error_close; } - netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class); - *netdev_devp = &netdev_dev->netdev_dev; return 0; -error: - free(netdev_dev); +error_close: + close(netdev->tap_fd); return error; } static void -destroy_tap(struct netdev_dev_linux *netdev_dev) +netdev_linux_destruct(struct netdev *netdev_) { - struct tap_state *state = &netdev_dev->state.tap; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (netdev->tc && netdev->tc->ops->tc_destroy) { + netdev->tc->ops->tc_destroy(netdev->tc); + } - if (state->fd >= 0) { - close(state->fd); + if (netdev_get_class(netdev_) == &netdev_tap_class + && netdev->tap_fd >= 0) + { + close(netdev->tap_fd); } + + if (netdev->miimon_interval > 0) { + int junk; + atomic_sub(&miimon_cnt, 1, &junk); + } + + ovs_mutex_destroy(&netdev->mutex); } -/* Destroys the netdev device 'netdev_dev_'. */ static void -netdev_linux_destroy(struct netdev_dev *netdev_dev_) +netdev_linux_dealloc(struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); - const char *type = netdev_dev_get_type(netdev_dev_); - - if (!strcmp(type, "system")) { - cache_notifier_refcount--; - - if (!cache_notifier_refcount) { - rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); - } - } else if (!strcmp(type, "tap")) { - destroy_tap(netdev_dev); - } + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + free(netdev); +} - free(netdev_dev); +static struct netdev_rxq * +netdev_linux_rxq_alloc(void) +{ + struct netdev_rxq_linux *rx = xzalloc(sizeof *rx); + return &rx->up; } static int -netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, - struct netdev **netdevp) +netdev_linux_rxq_construct(struct netdev_rxq *rxq_) { - struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); - struct netdev_linux *netdev; - enum netdev_flags flags; + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + struct netdev *netdev_ = rx->up.netdev; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - /* Allocate network device. */ - netdev = xzalloc(sizeof *netdev); - netdev->fd = -1; - netdev_init(&netdev->netdev, netdev_dev_); - - error = netdev_get_flags(&netdev->netdev, &flags); - if (error == ENODEV) { - goto error; - } - - if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") && - !netdev_dev->state.tap.opened) { - - /* We assume that the first user of the tap device is the primary user - * and give them the tap FD. Subsequent users probably just expect - * this to be a system device so open it normally to avoid send/receive - * directions appearing to be reversed. */ - netdev->fd = netdev_dev->state.tap.fd; - netdev_dev->state.tap.opened = true; - } else if (ethertype != NETDEV_ETH_TYPE_NONE) { + ovs_mutex_lock(&netdev->mutex); + rx->is_tap = is_tap_netdev(netdev_); + if (rx->is_tap) { + rx->fd = netdev->tap_fd; + } else { struct sockaddr_ll sll; - int protocol; - int ifindex; + int ifindex, val; + /* Result of tcpdump -dd inbound */ + static const struct sock_filter filt[] = { + { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ + { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ + { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ + { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ + }; + static const struct sock_fprog fprog = { + ARRAY_SIZE(filt), (struct sock_filter *) filt + }; /* Create file descriptor. */ - protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL - : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 - : ethertype); - netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); - if (netdev->fd < 0) { + rx->fd = socket(PF_PACKET, SOCK_RAW, 0); + if (rx->fd < 0) { + error = errno; + VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); + goto error; + } + + val = 1; + if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) { error = errno; + VLOG_ERR("%s: failed to mark socket for auxdata (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); goto error; } /* Set non-blocking mode. */ - error = set_nonblocking(netdev->fd); + error = set_nonblocking(rx->fd); if (error) { goto error; } /* Get ethernet device index. */ - error = get_ifindex(&netdev->netdev, &ifindex); + error = get_ifindex(&netdev->up, &ifindex); if (error) { goto error; } @@ -412,119 +836,209 @@ netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, memset(&sll, 0, sizeof sll); sll.sll_family = AF_PACKET; sll.sll_ifindex = ifindex; - if (bind(netdev->fd, - (struct sockaddr *) &sll, sizeof sll) < 0) { + sll.sll_protocol = htons(ETH_P_ALL); + if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { error = errno; - VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), - strerror(error)); + VLOG_ERR("%s: failed to bind raw socket (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); goto error; } - /* Between the socket() and bind() calls above, the socket receives all - * packets of the requested type on all system interfaces. We do not - * want to receive that data, but there is no way to avoid it. So we - * must now drain out the receive queue. */ - error = drain_rcvbuf(netdev->fd); + /* Filter for only inbound packets. */ + error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, + sizeof fprog); if (error) { + error = errno; + VLOG_ERR("%s: failed to attach filter (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); goto error; } } + ovs_mutex_unlock(&netdev->mutex); - *netdevp = &netdev->netdev; return 0; error: - netdev_uninit(&netdev->netdev, true); + if (rx->fd >= 0) { + close(rx->fd); + } + ovs_mutex_unlock(&netdev->mutex); return error; } -/* Closes and destroys 'netdev'. */ static void -netdev_linux_close(struct netdev *netdev_) +netdev_linux_rxq_destruct(struct netdev_rxq *rxq_) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); - if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) { - close(netdev->fd); + if (!rx->is_tap) { + close(rx->fd); } - free(netdev); } -/* Initializes 'svec' with a list of the names of all known network devices. */ +static void +netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_) +{ + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + + free(rx); +} + +static ovs_be16 +auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux) +{ + if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) { + return htons(aux->tp_vlan_tpid); + } else { + return htons(ETH_TYPE_VLAN); + } +} + +static bool +auxdata_has_vlan_tci(const struct tpacket_auxdata *aux) +{ + return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID; +} + static int -netdev_linux_enumerate(struct svec *svec) +netdev_linux_rxq_recv_sock(int fd, struct ofpbuf *buffer) { - struct if_nameindex *names; + size_t size; + ssize_t retval; + struct iovec iov; + struct cmsghdr *cmsg; + union { + struct cmsghdr cmsg; + char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; + } cmsg_buffer; + struct msghdr msgh; + + /* Reserve headroom for a single VLAN tag */ + ofpbuf_reserve(buffer, VLAN_HEADER_LEN); + size = ofpbuf_tailroom(buffer); + + iov.iov_base = ofpbuf_data(buffer); + iov.iov_len = size; + msgh.msg_name = NULL; + msgh.msg_namelen = 0; + msgh.msg_iov = &iov; + msgh.msg_iovlen = 1; + msgh.msg_control = &cmsg_buffer; + msgh.msg_controllen = sizeof cmsg_buffer; + msgh.msg_flags = 0; + + do { + retval = recvmsg(fd, &msgh, MSG_TRUNC); + } while (retval < 0 && errno == EINTR); + + if (retval < 0) { + return errno; + } else if (retval > size) { + return EMSGSIZE; + } - names = if_nameindex(); - if (names) { - size_t i; + ofpbuf_set_size(buffer, ofpbuf_size(buffer) + retval); - for (i = 0; names[i].if_name != NULL; i++) { - svec_add(svec, names[i].if_name); + for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) { + const struct tpacket_auxdata *aux; + + if (cmsg->cmsg_level != SOL_PACKET + || cmsg->cmsg_type != PACKET_AUXDATA + || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) { + continue; } - if_freenameindex(names); - return 0; - } else { - VLOG_WARN("could not obtain list of network device names: %s", - strerror(errno)); + + aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg)); + if (auxdata_has_vlan_tci(aux)) { + if (retval < ETH_HEADER_LEN) { + return EINVAL; + } + + eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux), + htons(aux->tp_vlan_tci)); + break; + } + } + + return 0; +} + +static int +netdev_linux_rxq_recv_tap(int fd, struct ofpbuf *buffer) +{ + ssize_t retval; + size_t size = ofpbuf_tailroom(buffer); + + do { + retval = read(fd, ofpbuf_data(buffer), size); + } while (retval < 0 && errno == EINTR); + + if (retval < 0) { return errno; + } else if (retval > size) { + return EMSGSIZE; } + + ofpbuf_set_size(buffer, ofpbuf_size(buffer) + retval); + return 0; } static int -netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) +netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct ofpbuf **packet, int *c) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + struct netdev *netdev = rx->up.netdev; + struct ofpbuf *buffer; + ssize_t retval; + int mtu; - if (netdev->fd < 0) { - /* Device was opened with NETDEV_ETH_TYPE_NONE. */ - return -EAGAIN; + if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { + mtu = ETH_PAYLOAD_MAX; } - for (;;) { - ssize_t retval = read(netdev->fd, data, size); - if (retval >= 0) { - return retval; - } else if (errno != EINTR) { - if (errno != EAGAIN) { - VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", - strerror(errno), netdev_get_name(netdev_)); - } - return -errno; + buffer = ofpbuf_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu, DP_NETDEV_HEADROOM); + + retval = (rx->is_tap + ? netdev_linux_rxq_recv_tap(rx->fd, buffer) + : netdev_linux_rxq_recv_sock(rx->fd, buffer)); + + if (retval) { + if (retval != EAGAIN && retval != EMSGSIZE) { + VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", + ovs_strerror(errno), netdev_rxq_get_name(rxq_)); } + ofpbuf_delete(buffer); + } else { + dp_packet_pad(buffer); + packet[0] = buffer; + *c = 1; } + + return retval; } -/* 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_recv() on 'netdev'. */ static void -netdev_linux_recv_wait(struct netdev *netdev_) +netdev_linux_rxq_wait(struct netdev_rxq *rxq_) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->fd >= 0) { - poll_fd_wait(netdev->fd, POLLIN); - } + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + poll_fd_wait(rx->fd, POLLIN); } -/* Discards all packets waiting to be received from 'netdev'. */ static int -netdev_linux_drain(struct netdev *netdev_) +netdev_linux_rxq_drain(struct netdev_rxq *rxq_) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->fd < 0) { - return 0; - } else if (!strcmp(netdev_get_type(netdev_), "tap")) { + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + if (rx->is_tap) { struct ifreq ifr; - int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, - SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); + int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr, + SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); if (error) { return error; } - drain_fd(netdev->fd, ifr.ifr_qlen); + drain_fd(rx->fd, ifr.ifr_qlen); return 0; } else { - return drain_rcvbuf(netdev->fd); + return drain_rcvbuf(rx->fd); } } @@ -538,18 +1052,65 @@ netdev_linux_drain(struct netdev *netdev_) * The kernel maintains a packet transmission queue, so the caller is not * expected to do additional queuing of packets. */ static int -netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) +netdev_linux_send(struct netdev *netdev_, struct ofpbuf *pkt, bool may_steal) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - - /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. - */ - if (netdev->fd < 0) { - return EPIPE; - } + const void *data = ofpbuf_data(pkt); + size_t size = ofpbuf_size(pkt); for (;;) { - ssize_t retval = write(netdev->fd, data, size); + ssize_t retval; + + if (!is_tap_netdev(netdev_)) { + /* Use our AF_PACKET socket to send to this device. */ + struct sockaddr_ll sll; + struct msghdr msg; + struct iovec iov; + int ifindex; + int sock; + + sock = af_packet_sock(); + if (sock < 0) { + return -sock; + } + + ifindex = netdev_get_ifindex(netdev_); + if (ifindex < 0) { + return -ifindex; + } + + /* 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 = CONST_CAST(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; + + retval = sendmsg(sock, &msg, 0); + } else { + /* Use the tap fd to send to this device. This is essential for + * tap devices, because packets sent to a tap device with an + * AF_PACKET socket will loop back to be *received* again on the + * tap device. This doesn't occur on other interface types + * because we attach a socket filter to the rx socket. */ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + retval = write(netdev->tap_fd, data, size); + } + + if (may_steal) { + ofpbuf_delete(pkt); + } + if (retval < 0) { /* The Linux AF_PACKET implementation never blocks waiting for room * for packets, instead returning ENOBUFS. Translate this into @@ -560,12 +1121,12 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) continue; } else if (errno != EAGAIN) { VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", - netdev_get_name(netdev_), strerror(errno)); + netdev_get_name(netdev_), ovs_strerror(errno)); } return errno; } else if (retval != size) { - VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " - "%zu) on %s", retval, size, netdev_get_name(netdev_)); + VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE"d bytes of " + "%"PRIuSIZE") on %s", retval, size, netdev_get_name(netdev_)); return EMSGSIZE; } else { return 0; @@ -581,14 +1142,9 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) * expected to do additional queuing of packets. Thus, this function is * unlikely to ever be used. It is included for completeness. */ static void -netdev_linux_send_wait(struct netdev *netdev_) +netdev_linux_send_wait(struct netdev *netdev) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->fd < 0) { - /* Nothing to do. */ - } else if (strcmp(netdev_get_type(netdev_), "tap")) { - poll_fd_wait(netdev->fd, POLLOUT); - } else { + if (is_tap_netdev(netdev)) { /* TAP device always accepts packets.*/ poll_immediate_wake(); } @@ -600,41 +1156,86 @@ static int netdev_linux_set_etheraddr(struct netdev *netdev_, const uint8_t mac[ETH_ADDR_LEN]) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + enum netdev_flags old_flags = 0; int error; - if (!(netdev_dev->cache_valid & VALID_ETHERADDR) - || !eth_addr_equals(netdev_dev->etheraddr, mac)) { - error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac); - if (!error) { - netdev_dev->cache_valid |= VALID_ETHERADDR; - memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); + ovs_mutex_lock(&netdev->mutex); + + if (netdev->cache_valid & VALID_ETHERADDR) { + error = netdev->ether_addr_error; + if (error || eth_addr_equals(netdev->etheraddr, mac)) { + goto exit; } - } else { - error = 0; + netdev->cache_valid &= ~VALID_ETHERADDR; + } + + /* Tap devices must be brought down before setting the address. */ + if (is_tap_netdev(netdev_)) { + update_flags(netdev, NETDEV_UP, 0, &old_flags); + } + error = set_etheraddr(netdev_get_name(netdev_), mac); + if (!error || error == ENODEV) { + netdev->ether_addr_error = error; + netdev->cache_valid |= VALID_ETHERADDR; + if (!error) { + memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN); + } + } + + if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { + update_flags(netdev, 0, NETDEV_UP, &old_flags); } + +exit: + ovs_mutex_unlock(&netdev->mutex); return error; } -/* Returns a pointer to 'netdev''s MAC address. The caller must not modify or - * free the returned buffer. */ +/* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ static int netdev_linux_get_etheraddr(const struct netdev *netdev_, uint8_t mac[ETH_ADDR_LEN]) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) { - int error = get_etheraddr(netdev_get_name(netdev_), - netdev_dev->etheraddr); - if (error) { - return error; - } - netdev_dev->cache_valid |= VALID_ETHERADDR; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + if (!(netdev->cache_valid & VALID_ETHERADDR)) { + netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_), + netdev->etheraddr); + netdev->cache_valid |= VALID_ETHERADDR; } - memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); - return 0; + + error = netdev->ether_addr_error; + if (!error) { + memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN); + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) +{ + int error; + + if (!(netdev->cache_valid & VALID_MTU)) { + struct ifreq ifr; + + netdev->netdev_mtu_error = af_inet_ifreq_ioctl( + netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); + netdev->mtu = ifr.ifr_mtu; + netdev->cache_valid |= VALID_MTU; + } + + error = netdev->netdev_mtu_error; + if (!error) { + *mtup = netdev->mtu; + } + + return error; } /* Returns the maximum size of transmitted (and received) packets on 'netdev', @@ -643,193 +1244,375 @@ netdev_linux_get_etheraddr(const struct netdev *netdev_, static int netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - if (!(netdev_dev->cache_valid & VALID_MTU)) { - struct ifreq ifr; - int error; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; - error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, - SIOCGIFMTU, "SIOCGIFMTU"); - if (error) { - return error; + ovs_mutex_lock(&netdev->mutex); + error = netdev_linux_get_mtu__(netdev, mtup); + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +/* Sets the maximum size of transmitted (MTU) for given device using linux + * networking ioctl interface. + */ +static int +netdev_linux_set_mtu(const struct netdev *netdev_, int mtu) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct ifreq ifr; + int error; + + ovs_mutex_lock(&netdev->mutex); + if (netdev->cache_valid & VALID_MTU) { + error = netdev->netdev_mtu_error; + if (error || netdev->mtu == mtu) { + goto exit; } - netdev_dev->mtu = ifr.ifr_mtu; - netdev_dev->cache_valid |= VALID_MTU; + netdev->cache_valid &= ~VALID_MTU; } - *mtup = netdev_dev->mtu; - return 0; + ifr.ifr_mtu = mtu; + error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, + SIOCSIFMTU, "SIOCSIFMTU"); + if (!error || error == ENODEV) { + netdev->netdev_mtu_error = error; + netdev->mtu = ifr.ifr_mtu; + netdev->cache_valid |= VALID_MTU; + } +exit: + ovs_mutex_unlock(&netdev->mutex); + return error; } /* Returns the ifindex of 'netdev', if successful, as a positive number. * On failure, returns a negative errno value. */ static int -netdev_linux_get_ifindex(const struct netdev *netdev) +netdev_linux_get_ifindex(const struct netdev *netdev_) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int ifindex, error; - error = get_ifindex(netdev, &ifindex); + ovs_mutex_lock(&netdev->mutex); + error = get_ifindex(netdev_, &ifindex); + ovs_mutex_unlock(&netdev->mutex); + return error ? -error : ifindex; } static int netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - int error = 0; - char *fn = NULL; - int fd = -1; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (!(netdev_dev->cache_valid & VALID_CARRIER)) { - char line[8]; - int retval; + ovs_mutex_lock(&netdev->mutex); + if (netdev->miimon_interval > 0) { + *carrier = netdev->miimon; + } else { + *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; + } + ovs_mutex_unlock(&netdev->mutex); - fn = xasprintf("/sys/class/net/%s/carrier", - netdev_get_name(netdev_)); - fd = open(fn, O_RDONLY); - if (fd < 0) { - error = errno; - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); - goto exit; - } + return 0; +} - retval = read(fd, line, sizeof line); - if (retval < 0) { - error = errno; - if (error == EINVAL) { - /* This is the normal return value when we try to check carrier - * if the network device is not up. */ - } else { - VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error)); - } - goto exit; - } else if (retval == 0) { - error = EPROTO; - VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); - goto exit; - } +static long long int +netdev_linux_get_carrier_resets(const struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + long long int carrier_resets; - if (line[0] != '0' && line[0] != '1') { - error = EPROTO; - VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", - fn, line[0]); - goto exit; - } - netdev_dev->carrier = line[0] != '0'; - netdev_dev->cache_valid |= VALID_CARRIER; - } - *carrier = netdev_dev->carrier; - error = 0; + ovs_mutex_lock(&netdev->mutex); + carrier_resets = netdev->carrier_resets; + ovs_mutex_unlock(&netdev->mutex); + + return carrier_resets; +} + +static int +netdev_linux_do_miimon(const char *name, 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 = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); + memcpy(data, &ifr.ifr_data, sizeof *data); -exit: - if (fd >= 0) { - close(fd); - } - free(fn); 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. */ -static bool -check_for_working_netlink_stats(void) -{ - /* Decide on the netdev_get_stats() implementation to use. Netlink is - * preferable, so if that works, we'll use it. */ - int ifindex = do_get_ifindex("lo"); - if (ifindex < 0) { - VLOG_WARN("failed to get ifindex for lo, " - "obtaining netdev stats from proc"); - return false; +static int +netdev_linux_get_miimon(const char *name, bool *miimon) +{ + struct mii_ioctl_data data; + int error; + + *miimon = false; + + memset(&data, 0, sizeof data); + error = netdev_linux_do_miimon(name, 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(name, SIOCGMIIREG, "SIOCGMIIREG", + &data); + + if (!error) { + *miimon = !!(data.val_out & BMSR_LSTATUS); + } else { + VLOG_WARN_RL(&rl, "%s: failed to query MII", name); + } } else { - struct netdev_stats stats; - int error = get_stats_via_netlink(ifindex, &stats); + struct ethtool_cmd ecmd; + + VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", + name); + + COVERAGE_INC(netdev_get_ethtool); + memset(&ecmd, 0, sizeof ecmd); + error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, + "ETHTOOL_GLINK"); if (!error) { - VLOG_DBG("obtaining netdev stats via rtnetlink"); - return true; + struct ethtool_value eval; + + memcpy(&eval, &ecmd, sizeof eval); + *miimon = !!eval.data; } else { - VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " - "via proc (you are probably running a pre-2.6.19 " - "kernel)", strerror(error)); - return false; + VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); + } + } + + return error; +} + +static int +netdev_linux_set_miimon_interval(struct netdev *netdev_, + long long int interval) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); + interval = interval > 0 ? MAX(interval, 100) : 0; + if (netdev->miimon_interval != interval) { + int junk; + + if (interval && !netdev->miimon_interval) { + atomic_add(&miimon_cnt, 1, &junk); + } else if (!interval && netdev->miimon_interval) { + atomic_sub(&miimon_cnt, 1, &junk); } + + netdev->miimon_interval = interval; + timer_set_expired(&netdev->miimon_timer); } + ovs_mutex_unlock(&netdev->mutex); + + return 0; } -/* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */ static void -netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev) -{ - 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_linux_miimon_run(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); + bool miimon; + + ovs_mutex_lock(&dev->mutex); + if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) { + netdev_linux_get_miimon(dev->up.name, &miimon); + if (miimon != dev->miimon) { + dev->miimon = miimon; + netdev_linux_changed(dev, dev->ifi_flags, 0); } + + timer_set_duration(&dev->miimon_timer, dev->miimon_interval); } + ovs_mutex_unlock(&dev->mutex); + netdev_close(netdev); + } + + shash_destroy(&device_shash); +} - netdev_dev->cache_valid |= VALID_IS_PSEUDO; +static void +netdev_linux_miimon_wait(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); + + ovs_mutex_lock(&dev->mutex); + if (dev->miimon_interval > 0) { + timer_wait(&dev->miimon_timer); + } + ovs_mutex_unlock(&dev->mutex); + netdev_close(netdev); } + shash_destroy(&device_shash); } static void swap_uint64(uint64_t *a, uint64_t *b) { - *a ^= *b; - *b ^= *a; - *a ^= *b; + uint64_t tmp = *a; + *a = *b; + *b = tmp; } -/* Retrieves current device stats for 'netdev'. +/* Copies 'src' into 'dst', performing format conversion in the process. * - * XXX All of the members of struct netdev_stats are 64 bits wide, but on - * 32-bit architectures the Linux network stats are only 32 bits. */ + * 'src' is allowed to be misaligned. */ +static void +netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, + const struct ovs_vport_stats *src) +{ + dst->rx_packets = get_unaligned_u64(&src->rx_packets); + dst->tx_packets = get_unaligned_u64(&src->tx_packets); + dst->rx_bytes = get_unaligned_u64(&src->rx_bytes); + dst->tx_bytes = get_unaligned_u64(&src->tx_bytes); + dst->rx_errors = get_unaligned_u64(&src->rx_errors); + dst->tx_errors = get_unaligned_u64(&src->tx_errors); + dst->rx_dropped = get_unaligned_u64(&src->rx_dropped); + dst->tx_dropped = get_unaligned_u64(&src->tx_dropped); + dst->multicast = 0; + dst->collisions = 0; + dst->rx_length_errors = 0; + dst->rx_over_errors = 0; + dst->rx_crc_errors = 0; + dst->rx_frame_errors = 0; + dst->rx_fifo_errors = 0; + dst->rx_missed_errors = 0; + dst->tx_aborted_errors = 0; + dst->tx_carrier_errors = 0; + dst->tx_fifo_errors = 0; + dst->tx_heartbeat_errors = 0; + dst->tx_window_errors = 0; +} + static int -netdev_linux_get_stats(const struct netdev *netdev_, - struct netdev_stats *stats) +get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - static int use_netlink_stats = -1; + struct dpif_linux_vport reply; + struct ofpbuf *buf; int error; - COVERAGE_INC(netdev_get_stats); + error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf); + if (error) { + return error; + } else if (!reply.stats) { + ofpbuf_delete(buf); + return EOPNOTSUPP; + } + + netdev_stats_from_ovs_vport_stats(stats, reply.stats); + + ofpbuf_delete(buf); + + return 0; +} + +static void +get_stats_via_vport(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (!netdev->vport_stats_error || + !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { + int error; - if (use_netlink_stats < 0) { - use_netlink_stats = check_for_working_netlink_stats(); + error = get_stats_via_vport__(netdev_, stats); + if (error && error != ENOENT) { + VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " + "(%s)", + netdev_get_name(netdev_), ovs_strerror(error)); + } + netdev->vport_stats_error = error; + netdev->cache_valid |= VALID_VPORT_STAT_ERROR; } - if (use_netlink_stats) { - int ifindex; +} - error = get_ifindex(netdev_, &ifindex); - if (!error) { - error = get_stats_via_netlink(ifindex, stats); +/* Retrieves current device stats for 'netdev-linux'. */ +static int +netdev_linux_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_stats dev_stats; + int error; + + ovs_mutex_lock(&netdev->mutex); + get_stats_via_vport(netdev_, stats); + error = get_stats_via_netlink(netdev_, &dev_stats); + if (error) { + if (!netdev->vport_stats_error) { + error = 0; } + } else if (netdev->vport_stats_error) { + /* stats not available from OVS then use ioctl stats. */ + *stats = dev_stats; } else { - error = get_stats_via_proc(netdev_get_name(netdev_), stats); + stats->rx_errors += dev_stats.rx_errors; + stats->tx_errors += dev_stats.tx_errors; + stats->rx_dropped += dev_stats.rx_dropped; + stats->tx_dropped += dev_stats.tx_dropped; + stats->multicast += dev_stats.multicast; + stats->collisions += dev_stats.collisions; + stats->rx_length_errors += dev_stats.rx_length_errors; + stats->rx_over_errors += dev_stats.rx_over_errors; + stats->rx_crc_errors += dev_stats.rx_crc_errors; + stats->rx_frame_errors += dev_stats.rx_frame_errors; + stats->rx_fifo_errors += dev_stats.rx_fifo_errors; + stats->rx_missed_errors += dev_stats.rx_missed_errors; + stats->tx_aborted_errors += dev_stats.tx_aborted_errors; + stats->tx_carrier_errors += dev_stats.tx_carrier_errors; + stats->tx_fifo_errors += dev_stats.tx_fifo_errors; + stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; + stats->tx_window_errors += dev_stats.tx_window_errors; } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +/* Retrieves current device stats for 'netdev-tap' netdev or + * netdev-internal. */ +static int +netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_stats dev_stats; + int error; - /* If this port is an internal port then the transmit and receive stats - * will appear to be swapped relative to the other ports since we are the - * one sending the data, not a remote computer. For consistency, we swap - * them back here. */ - netdev_linux_update_is_pseudo(netdev_dev); - if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) { + ovs_mutex_lock(&netdev->mutex); + get_stats_via_vport(netdev_, stats); + error = get_stats_via_netlink(netdev_, &dev_stats); + if (error) { + if (!netdev->vport_stats_error) { + error = 0; + } + } else if (netdev->vport_stats_error) { + /* Transmit and receive stats will appear to be swapped relative to the + * other ports since we are the one sending the data, not a remote + * computer. For consistency, we swap them back here. This does not + * apply if we are getting stats from the vport layer because it always + * tracks stats from the perspective of the switch. */ + + *stats = dev_stats; swap_uint64(&stats->rx_packets, &stats->tx_packets); swap_uint64(&stats->rx_bytes, &stats->tx_bytes); swap_uint64(&stats->rx_errors, &stats->tx_errors); @@ -845,434 +1628,782 @@ netdev_linux_get_stats(const struct netdev *netdev_, stats->tx_fifo_errors = 0; stats->tx_heartbeat_errors = 0; stats->tx_window_errors = 0; + } else { + stats->rx_dropped += dev_stats.tx_dropped; + stats->tx_dropped += dev_stats.rx_dropped; + + stats->rx_errors += dev_stats.tx_errors; + stats->tx_errors += dev_stats.rx_errors; + + stats->multicast += dev_stats.multicast; + stats->collisions += dev_stats.collisions; } + ovs_mutex_unlock(&netdev->mutex); return error; } static int -netdev_linux_set_stats(struct netdev *netdev, - const struct netdev_stats *stats) +netdev_internal_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - struct internal_dev_stats dp_dev_stats; - struct ifreq ifr; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; - /* We must reject this call if 'netdev' is not an Open vSwitch internal - * port, because the ioctl that we are about to execute is in the "device - * private ioctls" range, which means that executing it on a device that - * is not the type we expect could do any random thing. - * - * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE - * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h. I guess - * DaveM is a little behind on that.) */ - netdev_linux_update_is_pseudo(netdev_dev); - if (!netdev_dev->is_internal) { - return EOPNOTSUPP; - } + ovs_mutex_lock(&netdev->mutex); + get_stats_via_vport(netdev_, stats); + error = netdev->vport_stats_error; + ovs_mutex_unlock(&netdev->mutex); - /* This actually only sets the *offset* that the dp_dev applies, but in our - * usage for fake bond devices the dp_dev never has any traffic of it own - * so it has the same effect. */ - dp_dev_stats.rx_packets = stats->rx_packets; - dp_dev_stats.rx_bytes = stats->rx_bytes; - dp_dev_stats.tx_packets = stats->tx_packets; - dp_dev_stats.tx_bytes = stats->tx_bytes; - ifr.ifr_data = (void *) &dp_dev_stats; - return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, - INTERNAL_DEV_SET_STATS, - "INTERNAL_DEV_SET_STATS"); + return error; } -/* Stores the features supported by 'netdev' into each of '*current', - * '*advertised', '*supported', and '*peer' that are non-null. Each value is a - * 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, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) +netdev_internal_set_stats(struct netdev *netdev, + const struct netdev_stats *stats) +{ + struct ovs_vport_stats vport_stats; + struct dpif_linux_vport vport; + int err; + + vport_stats.rx_packets = stats->rx_packets; + vport_stats.tx_packets = stats->tx_packets; + vport_stats.rx_bytes = stats->rx_bytes; + vport_stats.tx_bytes = stats->tx_bytes; + vport_stats.rx_errors = stats->rx_errors; + vport_stats.tx_errors = stats->tx_errors; + vport_stats.rx_dropped = stats->rx_dropped; + vport_stats.tx_dropped = stats->tx_dropped; + + dpif_linux_vport_init(&vport); + vport.cmd = OVS_VPORT_CMD_SET; + vport.name = netdev_get_name(netdev); + vport.stats = &vport_stats; + + err = dpif_linux_vport_transact(&vport, NULL, NULL); + + /* If the vport layer doesn't know about the device, that doesn't mean it + * doesn't exist (after all were able to open it when netdev_open() was + * called), it just means that it isn't attached and we'll be getting + * stats a different way. */ + if (err == ENODEV) { + err = EOPNOTSUPP; + } + + return err; +} + +static void +netdev_linux_read_features(struct netdev_linux *netdev) { struct ethtool_cmd ecmd; + uint32_t speed; int error; + if (netdev->cache_valid & VALID_FEATURES) { + return; + } + + COVERAGE_INC(netdev_get_ethtool); memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, + error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { - return error; + goto out; } /* Supported features. */ - *supported = 0; + netdev->supported = 0; if (ecmd.supported & SUPPORTED_10baseT_Half) { - *supported |= OFPPF_10MB_HD; + netdev->supported |= NETDEV_F_10MB_HD; } if (ecmd.supported & SUPPORTED_10baseT_Full) { - *supported |= OFPPF_10MB_FD; + netdev->supported |= NETDEV_F_10MB_FD; } if (ecmd.supported & SUPPORTED_100baseT_Half) { - *supported |= OFPPF_100MB_HD; + netdev->supported |= NETDEV_F_100MB_HD; } if (ecmd.supported & SUPPORTED_100baseT_Full) { - *supported |= OFPPF_100MB_FD; + netdev->supported |= NETDEV_F_100MB_FD; } if (ecmd.supported & SUPPORTED_1000baseT_Half) { - *supported |= OFPPF_1GB_HD; + netdev->supported |= NETDEV_F_1GB_HD; } if (ecmd.supported & SUPPORTED_1000baseT_Full) { - *supported |= OFPPF_1GB_FD; + netdev->supported |= NETDEV_F_1GB_FD; } if (ecmd.supported & SUPPORTED_10000baseT_Full) { - *supported |= OFPPF_10GB_FD; + netdev->supported |= NETDEV_F_10GB_FD; } if (ecmd.supported & SUPPORTED_TP) { - *supported |= OFPPF_COPPER; + netdev->supported |= NETDEV_F_COPPER; } if (ecmd.supported & SUPPORTED_FIBRE) { - *supported |= OFPPF_FIBER; + netdev->supported |= NETDEV_F_FIBER; } if (ecmd.supported & SUPPORTED_Autoneg) { - *supported |= OFPPF_AUTONEG; + netdev->supported |= NETDEV_F_AUTONEG; } if (ecmd.supported & SUPPORTED_Pause) { - *supported |= OFPPF_PAUSE; + netdev->supported |= NETDEV_F_PAUSE; } if (ecmd.supported & SUPPORTED_Asym_Pause) { - *supported |= OFPPF_PAUSE_ASYM; + netdev->supported |= NETDEV_F_PAUSE_ASYM; } /* Advertised features. */ - *advertised = 0; + netdev->advertised = 0; if (ecmd.advertising & ADVERTISED_10baseT_Half) { - *advertised |= OFPPF_10MB_HD; + netdev->advertised |= NETDEV_F_10MB_HD; } if (ecmd.advertising & ADVERTISED_10baseT_Full) { - *advertised |= OFPPF_10MB_FD; + netdev->advertised |= NETDEV_F_10MB_FD; } if (ecmd.advertising & ADVERTISED_100baseT_Half) { - *advertised |= OFPPF_100MB_HD; + netdev->advertised |= NETDEV_F_100MB_HD; } if (ecmd.advertising & ADVERTISED_100baseT_Full) { - *advertised |= OFPPF_100MB_FD; + netdev->advertised |= NETDEV_F_100MB_FD; } if (ecmd.advertising & ADVERTISED_1000baseT_Half) { - *advertised |= OFPPF_1GB_HD; + netdev->advertised |= NETDEV_F_1GB_HD; } if (ecmd.advertising & ADVERTISED_1000baseT_Full) { - *advertised |= OFPPF_1GB_FD; + netdev->advertised |= NETDEV_F_1GB_FD; } if (ecmd.advertising & ADVERTISED_10000baseT_Full) { - *advertised |= OFPPF_10GB_FD; + netdev->advertised |= NETDEV_F_10GB_FD; } if (ecmd.advertising & ADVERTISED_TP) { - *advertised |= OFPPF_COPPER; + netdev->advertised |= NETDEV_F_COPPER; } if (ecmd.advertising & ADVERTISED_FIBRE) { - *advertised |= OFPPF_FIBER; + netdev->advertised |= NETDEV_F_FIBER; } if (ecmd.advertising & ADVERTISED_Autoneg) { - *advertised |= OFPPF_AUTONEG; + netdev->advertised |= NETDEV_F_AUTONEG; } if (ecmd.advertising & ADVERTISED_Pause) { - *advertised |= OFPPF_PAUSE; + netdev->advertised |= NETDEV_F_PAUSE; } if (ecmd.advertising & ADVERTISED_Asym_Pause) { - *advertised |= OFPPF_PAUSE_ASYM; + netdev->advertised |= NETDEV_F_PAUSE_ASYM; } /* Current settings. */ - if (ecmd.speed == SPEED_10) { - *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD; - } else if (ecmd.speed == SPEED_100) { - *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD; - } else if (ecmd.speed == SPEED_1000) { - *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD; - } else if (ecmd.speed == SPEED_10000) { - *current = OFPPF_10GB_FD; + speed = ecmd.speed; + if (speed == SPEED_10) { + netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; + } else if (speed == SPEED_100) { + netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; + } else if (speed == SPEED_1000) { + netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; + } else if (speed == SPEED_10000) { + netdev->current = NETDEV_F_10GB_FD; + } else if (speed == 40000) { + netdev->current = NETDEV_F_40GB_FD; + } else if (speed == 100000) { + netdev->current = NETDEV_F_100GB_FD; + } else if (speed == 1000000) { + netdev->current = NETDEV_F_1TB_FD; } else { - *current = 0; + netdev->current = 0; } if (ecmd.port == PORT_TP) { - *current |= OFPPF_COPPER; + netdev->current |= NETDEV_F_COPPER; } else if (ecmd.port == PORT_FIBRE) { - *current |= OFPPF_FIBER; + netdev->current |= NETDEV_F_FIBER; } if (ecmd.autoneg) { - *current |= OFPPF_AUTONEG; + netdev->current |= NETDEV_F_AUTONEG; } - /* Peer advertisements. */ - *peer = 0; /* XXX */ +out: + netdev->cache_valid |= VALID_FEATURES; + netdev->get_features_error = error; +} - return 0; +/* Stores the features supported by 'netdev' into of '*current', '*advertised', + * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits. + * Returns 0 if successful, otherwise a positive errno value. */ +static int +netdev_linux_get_features(const struct netdev *netdev_, + enum netdev_features *current, + enum netdev_features *advertised, + enum netdev_features *supported, + enum netdev_features *peer) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + netdev_linux_read_features(netdev); + if (!netdev->get_features_error) { + *current = netdev->current; + *advertised = netdev->advertised; + *supported = netdev->supported; + *peer = 0; /* XXX */ + } + error = netdev->get_features_error; + ovs_mutex_unlock(&netdev->mutex); + + return error; } /* Set the features advertised by 'netdev' to 'advertise'. */ static int -netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) +netdev_linux_set_advertisements(struct netdev *netdev_, + enum netdev_features advertise) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct ethtool_cmd ecmd; int error; + ovs_mutex_lock(&netdev->mutex); + + COVERAGE_INC(netdev_get_ethtool); memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, + error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { - return error; + goto exit; } ecmd.advertising = 0; - if (advertise & OFPPF_10MB_HD) { + if (advertise & NETDEV_F_10MB_HD) { ecmd.advertising |= ADVERTISED_10baseT_Half; } - if (advertise & OFPPF_10MB_FD) { + if (advertise & NETDEV_F_10MB_FD) { ecmd.advertising |= ADVERTISED_10baseT_Full; } - if (advertise & OFPPF_100MB_HD) { + if (advertise & NETDEV_F_100MB_HD) { ecmd.advertising |= ADVERTISED_100baseT_Half; } - if (advertise & OFPPF_100MB_FD) { + if (advertise & NETDEV_F_100MB_FD) { ecmd.advertising |= ADVERTISED_100baseT_Full; } - if (advertise & OFPPF_1GB_HD) { + if (advertise & NETDEV_F_1GB_HD) { ecmd.advertising |= ADVERTISED_1000baseT_Half; } - if (advertise & OFPPF_1GB_FD) { + if (advertise & NETDEV_F_1GB_FD) { ecmd.advertising |= ADVERTISED_1000baseT_Full; } - if (advertise & OFPPF_10GB_FD) { + if (advertise & NETDEV_F_10GB_FD) { ecmd.advertising |= ADVERTISED_10000baseT_Full; } - if (advertise & OFPPF_COPPER) { + if (advertise & NETDEV_F_COPPER) { ecmd.advertising |= ADVERTISED_TP; } - if (advertise & OFPPF_FIBER) { + if (advertise & NETDEV_F_FIBER) { ecmd.advertising |= ADVERTISED_FIBRE; } - if (advertise & OFPPF_AUTONEG) { + if (advertise & NETDEV_F_AUTONEG) { ecmd.advertising |= ADVERTISED_Autoneg; } - if (advertise & OFPPF_PAUSE) { + if (advertise & NETDEV_F_PAUSE) { ecmd.advertising |= ADVERTISED_Pause; } - if (advertise & OFPPF_PAUSE_ASYM) { + if (advertise & NETDEV_F_PAUSE_ASYM) { ecmd.advertising |= ADVERTISED_Asym_Pause; } - return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, - ETHTOOL_SSET, "ETHTOOL_SSET"); + COVERAGE_INC(netdev_set_ethtool); + error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, + ETHTOOL_SSET, "ETHTOOL_SSET"); + +exit: + ovs_mutex_unlock(&netdev->mutex); + return error; } -/* If 'netdev_name' is the name of a VLAN network device (e.g. one created with - * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device - * and returns 0. Otherwise returns a errno value (specifically ENOENT if - * 'netdev_name' is the name of a network device that is not a VLAN device) and - * sets '*vlan_vid' to -1. */ +/* Attempts to set input rate limiting (policing) policy. Returns 0 if + * successful, otherwise a positive errno value. */ static int -netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) +netdev_linux_set_policing(struct netdev *netdev_, + uint32_t kbits_rate, uint32_t kbits_burst) { - const char *netdev_name = netdev_get_name(netdev); - struct ds line = DS_EMPTY_INITIALIZER; - FILE *stream = NULL; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + const char *netdev_name = netdev_get_name(netdev_); int error; - char *fn; - COVERAGE_INC(netdev_get_vlan_vid); - fn = xasprintf("/proc/net/vlan/%s", netdev_name); - stream = fopen(fn, "r"); - if (!stream) { - error = errno; - goto done; - } + kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ + : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ + : kbits_burst); /* Stick with user-specified value. */ - if (ds_get_line(&line, stream)) { - if (ferror(stream)) { - error = errno; - VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno)); - } else { - error = EPROTO; - VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn); + ovs_mutex_lock(&netdev->mutex); + if (netdev->cache_valid & VALID_POLICING) { + error = netdev->netdev_policing_error; + if (error || (netdev->kbits_rate == kbits_rate && + netdev->kbits_burst == kbits_burst)) { + /* Assume that settings haven't changed since we last set them. */ + goto out; } - goto done; + netdev->cache_valid &= ~VALID_POLICING; } - if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) { - error = EPROTO; - VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"", - fn, ds_cstr(&line)); - goto done; + COVERAGE_INC(netdev_set_policing); + /* Remove any existing ingress qdisc. */ + error = tc_add_del_ingress_qdisc(netdev_, false); + if (error) { + VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", + netdev_name, ovs_strerror(error)); + goto out; } - error = 0; + if (kbits_rate) { + error = tc_add_del_ingress_qdisc(netdev_, true); + if (error) { + VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", + netdev_name, ovs_strerror(error)); + goto out; + } -done: - free(fn); - if (stream) { - fclose(stream); + error = tc_add_policer(netdev_, kbits_rate, kbits_burst); + if (error){ + VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", + netdev_name, ovs_strerror(error)); + goto out; + } } - ds_destroy(&line); - if (error) { - *vlan_vid = -1; + + netdev->kbits_rate = kbits_rate; + netdev->kbits_burst = kbits_burst; + +out: + if (!error || error == ENODEV) { + netdev->netdev_policing_error = error; + netdev->cache_valid |= VALID_POLICING; } + ovs_mutex_unlock(&netdev->mutex); return error; } -#define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress" -#define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1" - -/* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a - * positive errno value. - * - * This function is equivalent to running - * /sbin/tc qdisc del dev %s handle ffff: ingress - * but it is much, much faster. - */ static int -netdev_linux_remove_policing(struct netdev *netdev) +netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, + struct sset *types) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - const char *netdev_name = netdev_get_name(netdev); - - struct ofpbuf request; - struct ofpbuf *reply; - struct tcmsg *tcmsg; - struct nl_sock *rtnl_sock; - int ifindex; - int error; + const struct tc_ops *const *opsp; - error = get_ifindex(netdev, &ifindex); - if (error) { - return error; + for (opsp = tcs; *opsp != NULL; opsp++) { + const struct tc_ops *ops = *opsp; + if (ops->tc_install && ops->ovs_name[0] != '\0') { + sset_add(types, ops->ovs_name); + } } + return 0; +} - error = get_rtnl_sock(&rtnl_sock); - if (error) { - return error; +static const struct tc_ops * +tc_lookup_ovs_name(const char *name) +{ + const struct tc_ops *const *opsp; + + for (opsp = tcs; *opsp != NULL; opsp++) { + const struct tc_ops *ops = *opsp; + if (!strcmp(name, ops->ovs_name)) { + return ops; + } } + return NULL; +} - ofpbuf_init(&request, 0); - nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *tcmsg, - RTM_DELQDISC, NLM_F_REQUEST); - tcmsg = ofpbuf_put_zeros(&request, sizeof *tcmsg); - tcmsg->tcm_family = AF_UNSPEC; - tcmsg->tcm_ifindex = ifindex; - tcmsg->tcm_handle = 0xffff0000; - tcmsg->tcm_parent = TC_H_INGRESS; - nl_msg_put_string(&request, TCA_KIND, "ingress"); - nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); - error = nl_sock_transact(rtnl_sock, &request, &reply); - ofpbuf_uninit(&request); - ofpbuf_delete(reply); - if (error && error != ENOENT && error != EINVAL) { - VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", - netdev_name, strerror(error)); - return error; +static const struct tc_ops * +tc_lookup_linux_name(const char *name) +{ + const struct tc_ops *const *opsp; + + for (opsp = tcs; *opsp != NULL; opsp++) { + const struct tc_ops *ops = *opsp; + if (ops->linux_name && !strcmp(name, ops->linux_name)) { + return ops; + } + } + return NULL; +} + +static struct tc_queue * +tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, + size_t hash) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct tc_queue *queue; + + HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->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)); +} - netdev_dev->kbits_rate = 0; - netdev_dev->kbits_burst = 0; - netdev_dev->cache_valid |= VALID_POLICING; +static int +netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, + const char *type, + struct netdev_qos_capabilities *caps) +{ + const struct tc_ops *ops = tc_lookup_ovs_name(type); + if (!ops) { + return EOPNOTSUPP; + } + caps->n_queues = ops->n_queues; return 0; } -/* Attempts to set input rate limiting (policing) policy. */ static int -netdev_linux_set_policing(struct netdev *netdev, - uint32_t kbits_rate, uint32_t kbits_burst) +netdev_linux_get_qos(const struct netdev *netdev_, + const char **typep, struct smap *details) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - const char *netdev_name = netdev_get_name(netdev); - char command[1024]; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; - COVERAGE_INC(netdev_set_policing); + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + *typep = netdev->tc->ops->ovs_name; + error = (netdev->tc->ops->qdisc_get + ? netdev->tc->ops->qdisc_get(netdev_, details) + : 0); + } + ovs_mutex_unlock(&netdev->mutex); - kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ - : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ - : kbits_burst); /* Stick with user-specified value. */ + return error; +} - if (netdev_dev->cache_valid & VALID_POLICING - && netdev_dev->kbits_rate == kbits_rate - && netdev_dev->kbits_burst == kbits_burst) { - /* Assume that settings haven't changed since we last set them. */ - return 0; +static int +netdev_linux_set_qos(struct netdev *netdev_, + const char *type, const struct smap *details) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + const struct tc_ops *new_ops; + int error; + + new_ops = tc_lookup_ovs_name(type); + if (!new_ops || !new_ops->tc_install) { + return EOPNOTSUPP; } - netdev_linux_remove_policing(netdev); - if (kbits_rate) { - snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name); - if (system(command) != 0) { - VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name); - return -1; + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (error) { + goto exit; + } + + if (new_ops == netdev->tc->ops) { + error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; + } else { + /* Delete existing qdisc. */ + error = tc_del_qdisc(netdev_); + if (error) { + goto exit; + } + ovs_assert(netdev->tc == NULL); + + /* Install new qdisc. */ + error = new_ops->tc_install(netdev_, details); + ovs_assert((error == 0) == (netdev->tc != NULL)); + } + +exit: + ovs_mutex_unlock(&netdev->mutex); + return error; +} + +static int +netdev_linux_get_queue(const struct netdev *netdev_, + unsigned int queue_id, struct smap *details) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + error = (queue + ? netdev->tc->ops->class_get(netdev_, queue, details) + : ENOENT); + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_linux_set_queue(struct netdev *netdev_, + unsigned int queue_id, const struct smap *details) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + error = (queue_id < netdev->tc->ops->n_queues + && netdev->tc->ops->class_set + ? netdev->tc->ops->class_set(netdev_, queue_id, details) + : EINVAL); + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + if (netdev->tc->ops->class_delete) { + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + error = (queue + ? netdev->tc->ops->class_delete(netdev_, queue) + : ENOENT); + } else { + error = EINVAL; } + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_linux_get_queue_stats(const struct netdev *netdev_, + unsigned int queue_id, + struct netdev_queue_stats *stats) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + if (netdev->tc->ops->class_get_stats) { + const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + if (queue) { + stats->created = queue->created; + error = netdev->tc->ops->class_get_stats(netdev_, queue, + stats); + } else { + error = ENOENT; + } + } else { + error = EOPNOTSUPP; + } + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +struct queue_dump_state { + struct nl_dump dump; + struct ofpbuf buf; +}; + +static bool +start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state) +{ + 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(&state->dump, NETLINK_ROUTE, &request); + ofpbuf_uninit(&request); + + ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); + return true; +} - snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name, - kbits_rate, kbits_burst); - if (system(command) != 0) { - VLOG_WARN_RL(&rl, "%s: problem configuring policing", - netdev_name); - return -1; +static int +finish_queue_dump(struct queue_dump_state *state) +{ + ofpbuf_uninit(&state->buf); + return nl_dump_done(&state->dump); +} + +struct netdev_linux_queue_state { + unsigned int *queues; + size_t cur_queue; + size_t n_queues; +}; + +static int +netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) +{ + const struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + if (netdev->tc->ops->class_get) { + struct netdev_linux_queue_state *state; + struct tc_queue *queue; + size_t i; + + *statep = state = xmalloc(sizeof *state); + state->n_queues = hmap_count(&netdev->tc->queues); + state->cur_queue = 0; + state->queues = xmalloc(state->n_queues * sizeof *state->queues); + + i = 0; + HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { + state->queues[i++] = queue->queue_id; + } + } else { + error = EOPNOTSUPP; } + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} - netdev_dev->kbits_rate = kbits_rate; - netdev_dev->kbits_burst = kbits_burst; - netdev_dev->cache_valid |= VALID_POLICING; +static int +netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, + unsigned int *queue_idp, struct smap *details) +{ + const struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_linux_queue_state *state = state_; + int error = EOF; + + ovs_mutex_lock(&netdev->mutex); + while (state->cur_queue < state->n_queues) { + unsigned int queue_id = state->queues[state->cur_queue++]; + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + + if (queue) { + *queue_idp = queue_id; + error = netdev->tc->ops->class_get(netdev_, queue, details); + break; + } } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, + void *state_) +{ + struct netdev_linux_queue_state *state = state_; + free(state->queues); + free(state); return 0; } +static int +netdev_linux_dump_queue_stats(const struct netdev *netdev_, + netdev_dump_queue_stats_cb *cb, void *aux) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = tc_query_qdisc(netdev_); + if (!error) { + struct queue_dump_state state; + + if (!netdev->tc->ops->class_dump_stats) { + error = EOPNOTSUPP; + } else if (!start_queue_dump(netdev_, &state)) { + error = ENODEV; + } else { + struct ofpbuf msg; + int retval; + + while (nl_dump_next(&state.dump, &msg, &state.buf)) { + retval = netdev->tc->ops->class_dump_stats(netdev_, &msg, + cb, aux); + if (retval) { + error = retval; + } + } + + retval = finish_queue_dump(&state); + if (retval) { + error = retval; + } + } + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + static int netdev_linux_get_in4(const struct netdev *netdev_, struct in_addr *address, struct in_addr *netmask) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - - if (!(netdev_dev->cache_valid & VALID_IN4)) { - int error; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; - error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address, + ovs_mutex_lock(&netdev->mutex); + if (!(netdev->cache_valid & VALID_IN4)) { + error = netdev_linux_get_ipv4(netdev_, &netdev->address, SIOCGIFADDR, "SIOCGIFADDR"); - if (error) { - return error; + if (!error) { + error = netdev_linux_get_ipv4(netdev_, &netdev->netmask, + SIOCGIFNETMASK, "SIOCGIFNETMASK"); + if (!error) { + netdev->cache_valid |= VALID_IN4; + } } + } else { + error = 0; + } - error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask, - SIOCGIFNETMASK, "SIOCGIFNETMASK"); - if (error) { - return error; + if (!error) { + if (netdev->address.s_addr != INADDR_ANY) { + *address = netdev->address; + *netmask = netdev->netmask; + } else { + error = EADDRNOTAVAIL; } - - netdev_dev->cache_valid |= VALID_IN4; } - *address = netdev_dev->address; - *netmask = netdev_dev->netmask; - return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; + ovs_mutex_unlock(&netdev->mutex); + + return error; } static int netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, struct in_addr netmask) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); if (!error) { - netdev_dev->cache_valid |= VALID_IN4; - netdev_dev->address = address; - netdev_dev->netmask = netmask; + netdev->cache_valid |= VALID_IN4; + netdev->address = address; + netdev->netmask = netmask; if (address.s_addr != INADDR_ANY) { error = do_set_addr(netdev_, SIOCSIFNETMASK, "SIOCSIFNETMASK", netmask); } } + ovs_mutex_unlock(&netdev->mutex); + return error; } @@ -1282,14 +2413,14 @@ parse_if_inet6_line(const char *line, { uint8_t *s6 = in6->s6_addr; #define X8 "%2"SCNx8 - return sscanf(line, - " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 - "%*x %*x %*x %*x %16s\n", - &s6[0], &s6[1], &s6[2], &s6[3], - &s6[4], &s6[5], &s6[6], &s6[7], - &s6[8], &s6[9], &s6[10], &s6[11], - &s6[12], &s6[13], &s6[14], &s6[15], - ifname) == 17; + return ovs_scan(line, + " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 + "%*x %*x %*x %*x %16s\n", + &s6[0], &s6[1], &s6[2], &s6[3], + &s6[4], &s6[5], &s6[6], &s6[7], + &s6[8], &s6[9], &s6[10], &s6[11], + &s6[12], &s6[13], &s6[14], &s6[15], + ifname); } /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if @@ -1297,32 +2428,35 @@ parse_if_inet6_line(const char *line, static int netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - if (!(netdev_dev->cache_valid & VALID_IN6)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); + if (!(netdev->cache_valid & VALID_IN6)) { FILE *file; char line[128]; - netdev_dev->in6 = in6addr_any; + netdev->in6 = in6addr_any; file = fopen("/proc/net/if_inet6", "r"); 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->in6 = in6_tmp; break; } } fclose(file); } - netdev_dev->cache_valid |= VALID_IN6; + netdev->cache_valid |= VALID_IN6; } - *in6 = netdev_dev->in6; + *in6 = netdev->in6; + ovs_mutex_unlock(&netdev->mutex); + return 0; } @@ -1344,11 +2478,10 @@ do_set_addr(struct netdev *netdev, 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); - make_in4_sockaddr(&ifr.ifr_addr, addr); - return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, - ioctl_name); + make_in4_sockaddr(&ifr.ifr_addr, addr); + return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, + ioctl_name); } /* Adds 'router' as a default IP gateway. */ @@ -1364,10 +2497,9 @@ netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) 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; + error = af_inet_ioctl(SIOCADDRT, &rt); if (error) { - VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); } return error; } @@ -1384,7 +2516,7 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, *netdev_name = NULL; stream = fopen(fn, "r"); if (stream == NULL) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); return errno; } @@ -1392,17 +2524,16 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, while (fgets(line, sizeof line, stream)) { if (++ln >= 2) { char iface[17]; - uint32_t dest, gateway, mask; + ovs_be32 dest, gateway, mask; int refcnt, metric, mtu; unsigned int flags, use, window, irtt; - if (sscanf(line, - "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 - " %d %u %u\n", - iface, &dest, &gateway, &flags, &refcnt, - &use, &metric, &mask, &mtu, &window, &irtt) != 11) { - - VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", + if (!ovs_scan(line, + "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 + " %d %u %u\n", + iface, &dest, &gateway, &flags, &refcnt, + &use, &metric, &mask, &mtu, &window, &irtt)) { + VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", fn, ln, line); continue; } @@ -1412,7 +2543,7 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, } /* 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) { @@ -1433,33 +2564,74 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, return ENXIO; } +static int +netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error = 0; + + ovs_mutex_lock(&netdev->mutex); + if (!(netdev->cache_valid & VALID_DRVINFO)) { + struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; + + COVERAGE_INC(netdev_get_ethtool); + memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); + error = netdev_linux_do_ethtool(netdev->up.name, + cmd, + ETHTOOL_GDRVINFO, + "ETHTOOL_GDRVINFO"); + if (!error) { + netdev->cache_valid |= VALID_DRVINFO; + } + } + + if (!error) { + smap_add(smap, "driver_name", netdev->drvinfo.driver); + smap_add(smap, "driver_version", netdev->drvinfo.version); + smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, + struct smap *smap) +{ + smap_add(smap, "driver_name", "openvswitch"); + return 0; +} + /* 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, * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ static int netdev_linux_arp_lookup(const struct netdev *netdev, - uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) + ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) { struct arpreq r; struct sockaddr_in sin; 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; + retval = af_inet_ioctl(SIOCGARP, &r); if (!retval) { memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); } else if (retval != ENXIO) { VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", - netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval)); + netdev_get_name(netdev), IP_ARGS(ip), + ovs_strerror(retval)); } return retval; } @@ -1474,6 +2646,9 @@ nd_to_iff_flags(enum netdev_flags nd) if (nd & NETDEV_PROMISC) { iff |= IFF_PROMISC; } + if (nd & NETDEV_LOOPBACK) { + iff |= IFF_LOOPBACK; + } return iff; } @@ -1487,378 +2662,1998 @@ iff_to_nd_flags(int iff) if (iff & IFF_PROMISC) { nd |= NETDEV_PROMISC; } + if (iff & IFF_LOOPBACK) { + nd |= NETDEV_LOOPBACK; + } return nd; } static int -netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, - enum netdev_flags on, enum netdev_flags *old_flagsp) +update_flags(struct netdev_linux *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) + OVS_REQUIRES(netdev->mutex) { int old_flags, new_flags; - int error; + int error = 0; - error = get_flags(netdev, &old_flags); - if (!error) { - *old_flagsp = iff_to_nd_flags(old_flags); - new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); - if (new_flags != old_flags) { - error = set_flags(netdev, new_flags); - } + old_flags = netdev->ifi_flags; + *old_flagsp = iff_to_nd_flags(old_flags); + new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); + if (new_flags != old_flags) { + error = set_flags(netdev_get_name(&netdev->up), new_flags); + get_flags(&netdev->up, &netdev->ifi_flags); } + return error; } -static void -poll_notify(struct list *list) +static int +netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) { - struct netdev_linux_notifier *notifier; - LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) { - struct netdev_notifier *n = ¬ifier->notifier; - n->cb(n); - } + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = update_flags(netdev, off, on, old_flagsp); + ovs_mutex_unlock(&netdev->mutex); + + return error; } -static void -netdev_linux_poll_cb(const struct rtnetlink_change *change, - void *aux OVS_UNUSED) -{ - if (change) { - struct list *list = shash_find_data(&netdev_linux_notifiers, - change->ifname); - if (list) { - poll_notify(list); - } - } else { - struct shash_node *node; - SHASH_FOR_EACH (node, &netdev_linux_notifiers) { - poll_notify(node->data); - } - } +#define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, SET_STATS, \ + GET_FEATURES, GET_STATUS) \ +{ \ + NAME, \ + \ + NULL, \ + netdev_linux_run, \ + netdev_linux_wait, \ + \ + netdev_linux_alloc, \ + CONSTRUCT, \ + netdev_linux_destruct, \ + netdev_linux_dealloc, \ + NULL, /* get_config */ \ + NULL, /* set_config */ \ + NULL, /* get_tunnel_config */ \ + \ + netdev_linux_send, \ + netdev_linux_send_wait, \ + \ + netdev_linux_set_etheraddr, \ + netdev_linux_get_etheraddr, \ + netdev_linux_get_mtu, \ + netdev_linux_set_mtu, \ + netdev_linux_get_ifindex, \ + netdev_linux_get_carrier, \ + netdev_linux_get_carrier_resets, \ + netdev_linux_set_miimon_interval, \ + GET_STATS, \ + SET_STATS, \ + \ + GET_FEATURES, \ + netdev_linux_set_advertisements, \ + \ + 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_queue_dump_start, \ + netdev_linux_queue_dump_next, \ + netdev_linux_queue_dump_done, \ + 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, \ + GET_STATUS, \ + netdev_linux_arp_lookup, \ + \ + netdev_linux_update_flags, \ + \ + netdev_linux_rxq_alloc, \ + netdev_linux_rxq_construct, \ + netdev_linux_rxq_destruct, \ + netdev_linux_rxq_dealloc, \ + netdev_linux_rxq_recv, \ + netdev_linux_rxq_wait, \ + netdev_linux_rxq_drain, \ } -static int -netdev_linux_poll_add(struct netdev *netdev, - void (*cb)(struct netdev_notifier *), void *aux, - struct netdev_notifier **notifierp) -{ - const char *netdev_name = netdev_get_name(netdev); - struct netdev_linux_notifier *notifier; - struct list *list; +const struct netdev_class netdev_linux_class = + NETDEV_LINUX_CLASS( + "system", + netdev_linux_construct, + netdev_linux_get_stats, + NULL, /* set_stats */ + netdev_linux_get_features, + netdev_linux_get_status); + +const struct netdev_class netdev_tap_class = + NETDEV_LINUX_CLASS( + "tap", + netdev_linux_construct_tap, + netdev_tap_get_stats, + NULL, /* set_stats */ + netdev_linux_get_features, + netdev_linux_get_status); + +const struct netdev_class netdev_internal_class = + NETDEV_LINUX_CLASS( + "internal", + netdev_linux_construct, + netdev_internal_get_stats, + netdev_internal_set_stats, + NULL, /* get_features */ + netdev_internal_get_status); + +/* HTB traffic control class. */ - if (shash_is_empty(&netdev_linux_notifiers)) { - int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier, - netdev_linux_poll_cb, NULL); - if (error) { - return error; - } - } +#define HTB_N_QUEUES 0xf000 - list = shash_find_data(&netdev_linux_notifiers, netdev_name); - if (!list) { - list = xmalloc(sizeof *list); - list_init(list); - shash_add(&netdev_linux_notifiers, netdev_name, list); - } +struct htb { + struct tc tc; + unsigned int max_rate; /* In bytes/s. */ +}; - notifier = xmalloc(sizeof *notifier); - netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); - list_push_back(list, ¬ifier->node); - *notifierp = ¬ifier->notifier; - return 0; -} +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. */ + unsigned int priority; /* Lower values are higher priorities. */ +}; -static void -netdev_linux_poll_remove(struct netdev_notifier *notifier_) +static struct htb * +htb_get__(const struct netdev *netdev_) { - struct netdev_linux_notifier *notifier = - CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier); - struct list *list; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + return CONTAINER_OF(netdev->tc, struct htb, tc); +} + +static void +htb_install__(struct netdev *netdev_, uint64_t max_rate) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct htb *htb; + + htb = xmalloc(sizeof *htb); + tc_init(&htb->tc, &tc_ops_htb); + htb->max_rate = max_rate; + + netdev->tc = &htb->tc; +} + +/* Create an HTB qdisc. + * + * Equivalent to "tc qdisc add dev root handle 1: htb default 1". */ +static int +htb_setup_qdisc__(struct netdev *netdev) +{ + size_t opt_offset; + struct tc_htb_glob opt; + struct ofpbuf request; + struct tcmsg *tcmsg; + + 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; + + nl_msg_put_string(&request, TCA_KIND, "htb"); + + memset(&opt, 0, sizeof opt); + opt.rate2quantum = 10; + opt.version = 3; + opt.defcls = 1; + + opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); + nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); + nl_msg_end_nested(&request, opt_offset); + + return tc_transact(&request, NULL); +} + +/* Equivalent to "tc class replace classid parent htb + * rate bps ceil bps burst b prio ". */ +static int +htb_setup_class__(struct netdev *netdev, unsigned int handle, + unsigned int parent, struct htb_class *class) +{ + size_t opt_offset; + struct tc_htb_opt opt; + struct ofpbuf request; + struct tcmsg *tcmsg; + int error; + int mtu; + + error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); + if (error) { + VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", + netdev_get_name(netdev)); + return error; + } + + memset(&opt, 0, sizeof opt); + tc_fill_rate(&opt.rate, class->min_rate, mtu); + tc_fill_rate(&opt.ceil, class->max_rate, mtu); + opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); + opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); + 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; + + nl_msg_put_string(&request, TCA_KIND, "htb"); + opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); + nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); + tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); + tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); + 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=%u max_rate=%u burst=%u prio=%u (%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, + class->burst, class->priority, ovs_strerror(error)); + } + return error; +} + +/* Parses Netlink attributes in 'options' for HTB parameters and stores a + * description of them into 'details'. The description complies with the + * specification given in the vswitch database documentation for linux-htb + * queue details. */ +static int +htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) +{ + static const struct nl_policy tca_htb_policy[] = { + [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, + .min_len = sizeof(struct tc_htb_opt) }, + }; + + struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; + const struct tc_htb_opt *htb; + + if (!nl_parse_nested(nl_options, tca_htb_policy, + attrs, ARRAY_SIZE(tca_htb_policy))) { + VLOG_WARN_RL(&rl, "failed to parse HTB class options"); + return EPROTO; + } + + htb = nl_attr_get(attrs[TCA_HTB_PARMS]); + class->min_rate = htb->rate.rate; + class->max_rate = htb->ceil.rate; + class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); + class->priority = htb->prio; + return 0; +} + +static int +htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, + struct htb_class *options, + struct netdev_queue_stats *stats) +{ + struct nlattr *nl_options; + unsigned int handle; + int error; + + error = tc_parse_class(tcmsg, &handle, &nl_options, stats); + if (!error && queue_id) { + unsigned int major = tc_get_major(handle); + unsigned int minor = tc_get_minor(handle); + if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { + *queue_id = minor - 1; + } else { + error = EPROTO; + } + } + if (!error && options) { + error = htb_parse_tca_options__(nl_options, options); + } + return error; +} + +static void +htb_parse_qdisc_details__(struct netdev *netdev_, + const struct smap *details, struct htb_class *hc) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + const char *max_rate_s; + + max_rate_s = smap_get(details, "max-rate"); + hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; + if (!hc->max_rate) { + enum netdev_features current; + + netdev_linux_read_features(netdev); + current = !netdev->get_features_error ? netdev->current : 0; + hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; + } + hc->min_rate = hc->max_rate; + hc->burst = 0; + hc->priority = 0; +} + +static int +htb_parse_class_details__(struct netdev *netdev, + const struct smap *details, struct htb_class *hc) +{ + const struct htb *htb = htb_get__(netdev); + const char *min_rate_s = smap_get(details, "min-rate"); + const char *max_rate_s = smap_get(details, "max-rate"); + const char *burst_s = smap_get(details, "burst"); + const char *priority_s = smap_get(details, "priority"); + int mtu, error; + + error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); + if (error) { + VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", + netdev_get_name(netdev)); + return error; + } + + /* 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 */ + hc->max_rate = (max_rate_s + ? strtoull(max_rate_s, NULL, 10) / 8 + : htb->max_rate); + hc->max_rate = MAX(hc->max_rate, hc->min_rate); + hc->max_rate = MIN(hc->max_rate, htb->max_rate); + + /* burst + * + * According to hints in the documentation that I've read, it is important + * that 'burst' be at least as big as the largest frame that might be + * transmitted. Also, making 'burst' a bit bigger than necessary is OK, + * but having it a bit too small is a problem. Since netdev_get_mtu() + * 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. */ + hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; + hc->burst = MAX(hc->burst, mtu + 64); + + /* priority */ + hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; + + return 0; +} + +static int +htb_query_class__(const struct netdev *netdev, unsigned int handle, + unsigned int parent, struct htb_class *options, + struct netdev_queue_stats *stats) +{ + struct ofpbuf *reply; + int error; + + error = tc_query_class(netdev, handle, parent, &reply); + if (!error) { + error = htb_parse_tcmsg__(reply, NULL, options, stats); + ofpbuf_delete(reply); + } + return error; +} + +static int +htb_tc_install(struct netdev *netdev, const struct smap *details) +{ + int error; + + error = htb_setup_qdisc__(netdev); + if (!error) { + struct htb_class hc; + + htb_parse_qdisc_details__(netdev, details, &hc); + error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &hc); + if (!error) { + htb_install__(netdev, hc.max_rate); + } + } + 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; + + queue = tc_find_queue__(netdev, queue_id, hash); + if (queue) { + hcp = htb_class_cast__(queue); + } else { + hcp = xmalloc(sizeof *hcp); + queue = &hcp->tc_queue; + queue->queue_id = queue_id; + queue->created = time_msec(); + hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); + } + + 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 ofpbuf msg; + struct queue_dump_state state; + struct htb_class hc; + + /* Get qdisc options. */ + hc.max_rate = 0; + htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); + htb_install__(netdev, hc.max_rate); + + /* Get queues. */ + if (!start_queue_dump(netdev, &state)) { + return ENODEV; + } + while (nl_dump_next(&state.dump, &msg, &state.buf)) { + unsigned int queue_id; + + if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { + htb_update_queue__(netdev, queue_id, &hc); + } + } + finish_queue_dump(&state); + + return 0; +} + +static void +htb_tc_destroy(struct tc *tc) +{ + struct htb *htb = CONTAINER_OF(tc, struct htb, tc); + struct htb_class *hc, *next; + + 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); + free(htb); +} + +static int +htb_qdisc_get(const struct netdev *netdev, struct smap *details) +{ + const struct htb *htb = htb_get__(netdev); + smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); + return 0; +} + +static int +htb_qdisc_set(struct netdev *netdev, const struct smap *details) +{ + struct htb_class hc; + int error; + + htb_parse_qdisc_details__(netdev, details, &hc); + error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &hc); + if (!error) { + htb_get__(netdev)->max_rate = hc.max_rate; + } + return error; +} + +static int +htb_class_get(const struct netdev *netdev OVS_UNUSED, + const struct tc_queue *queue, struct smap *details) +{ + const struct htb_class *hc = htb_class_cast__(queue); + + smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); + if (hc->min_rate != hc->max_rate) { + smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); + } + smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); + if (hc->priority) { + smap_add_format(details, "priority", "%u", hc->priority); + } + return 0; +} + +static int +htb_class_set(struct netdev *netdev, unsigned int queue_id, + const struct smap *details) +{ + struct htb_class hc; + int error; + + error = htb_parse_class_details__(netdev, details, &hc); + if (error) { + return error; + } + + error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), + tc_make_handle(1, 0xfffe), &hc); + if (error) { + return error; + } + + htb_update_queue__(netdev, queue_id, &hc); + return 0; +} + +static int +htb_class_delete(struct netdev *netdev, struct tc_queue *queue) +{ + struct htb_class *hc = htb_class_cast__(queue); + struct htb *htb = htb_get__(netdev); + int error; + + 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); + } + return error; +} + +static int +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->queue_id + 1), + tc_make_handle(1, 0xfffe), NULL, stats); +} + +static int +htb_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 <= HTB_N_QUEUES) { + (*cb)(minor - 1, &stats, aux); + } + return 0; +} + +static const struct tc_ops tc_ops_htb = { + "htb", /* linux_name */ + "linux-htb", /* ovs_name */ + HTB_N_QUEUES, /* n_queues */ + htb_tc_install, + htb_tc_load, + htb_tc_destroy, + htb_qdisc_get, + htb_qdisc_set, + htb_class_get, + htb_class_set, + htb_class_delete, + htb_class_get_stats, + htb_class_dump_stats +}; + +/* "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_linux *netdev = netdev_linux_cast(netdev_); + return CONTAINER_OF(netdev->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_linux *netdev = netdev_linux_cast(netdev_); + struct hfsc *hfsc; + + hfsc = xmalloc(sizeof *hfsc); + tc_init(&hfsc->tc, &tc_ops_hfsc); + hfsc->max_rate = max_rate; + netdev->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; + queue->created = time_msec(); + 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 smap *details, + struct hfsc_class *class) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + uint32_t max_rate; + const char *max_rate_s; + + max_rate_s = smap_get(details, "max-rate"); + max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; + + if (!max_rate) { + enum netdev_features current; + + netdev_linux_read_features(netdev); + current = !netdev->get_features_error ? netdev->current : 0; + max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; + } + + class->min_rate = max_rate; + class->max_rate = max_rate; +} + +static int +hfsc_parse_class_details__(struct netdev *netdev, + const struct smap *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 = smap_get(details, "min-rate"); + max_rate_s = smap_get(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 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 parent classid hfsc + * sc rate ul 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, ovs_strerror(error)); + } + + return error; +} + +static int +hfsc_tc_install(struct netdev *netdev, const struct smap *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 queue_dump_state state; + 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, &state)) { + return ENODEV; + } + + while (nl_dump_next(&state.dump, &msg, &state.buf)) { + unsigned int queue_id; + + if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { + hfsc_update_queue__(netdev, queue_id, &hc); + } + } + + finish_queue_dump(&state); + 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 smap *details) +{ + const struct hfsc *hfsc; + hfsc = hfsc_get__(netdev); + smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); + return 0; +} + +static int +hfsc_qdisc_set(struct netdev *netdev, const struct smap *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 smap *details) +{ + const struct hfsc_class *hc; + + hc = hfsc_class_cast__(queue); + smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); + if (hc->min_rate != hc->max_rate) { + smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); + } + return 0; +} + +static int +hfsc_class_set(struct netdev *netdev, unsigned int queue_id, + const struct smap *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 */ +}; + +/* "linux-default" traffic control class. + * + * This class represents the default, unnamed Linux qdisc. It corresponds to + * the "" (empty string) QoS type in the OVS database. */ + +static void +default_install__(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); + + /* Nothing but a tc class implementation is allowed to write to a tc. This + * class never does that, so we can legitimately use a const tc object. */ + netdev->tc = CONST_CAST(struct tc *, &tc); +} + +static int +default_tc_install(struct netdev *netdev, + const struct smap *details OVS_UNUSED) +{ + default_install__(netdev); + return 0; +} + +static int +default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) +{ + default_install__(netdev); + return 0; +} + +static const struct tc_ops tc_ops_default = { + NULL, /* linux_name */ + "", /* ovs_name */ + 0, /* n_queues */ + default_tc_install, + default_tc_load, + NULL, /* tc_destroy */ + NULL, /* qdisc_get */ + NULL, /* qdisc_set */ + NULL, /* class_get */ + NULL, /* class_set */ + NULL, /* class_delete */ + NULL, /* class_get_stats */ + NULL /* class_dump_stats */ +}; + +/* "linux-other" traffic control class. + * + * */ + +static int +other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); + + /* Nothing but a tc class implementation is allowed to write to a tc. This + * class never does that, so we can legitimately use a const tc object. */ + netdev->tc = CONST_CAST(struct tc *, &tc); + return 0; +} + +static const struct tc_ops tc_ops_other = { + NULL, /* linux_name */ + "linux-other", /* ovs_name */ + 0, /* n_queues */ + NULL, /* tc_install */ + other_tc_load, + NULL, /* tc_destroy */ + NULL, /* qdisc_get */ + NULL, /* qdisc_set */ + NULL, /* class_get */ + NULL, /* class_set */ + NULL, /* class_delete */ + NULL, /* class_get_stats */ + NULL /* class_dump_stats */ +}; + +/* Traffic control. */ + +/* Number of kernel "tc" ticks per second. */ +static double ticks_per_s; + +/* Number of kernel "jiffies" per second. This is used for the purpose of + * computing buffer sizes. Generally kernel qdiscs need to be able to buffer + * one jiffy's worth of data. + * + * There are two possibilities here: + * + * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the + * approximate range of 100 to 1024. That means that we really need to + * make sure that the qdisc can buffer that much data. + * + * - 'buffer_hz' is an absurdly large number. That means that the kernel + * has finely granular timers and there's no need to fudge additional room + * for buffers. (There's no extra effort needed to implement that: the + * large 'buffer_hz' is used as a divisor, so practically any number will + * come out as 0 in the division. Small integer results in the case of + * really high dividends won't have any real effect anyhow.) + */ +static unsigned int buffer_hz; + +/* Returns tc handle 'major':'minor'. */ +static unsigned int +tc_make_handle(unsigned int major, unsigned int minor) +{ + return TC_H_MAKE(major << 16, minor); +} + +/* Returns the major number from 'handle'. */ +static unsigned int +tc_get_major(unsigned int handle) +{ + return TC_H_MAJ(handle) >> 16; +} + +/* Returns the minor number from 'handle'. */ +static unsigned int +tc_get_minor(unsigned int handle) +{ + return TC_H_MIN(handle); +} + +static struct tcmsg * +tc_make_request(const struct netdev *netdev, int type, unsigned int flags, + struct ofpbuf *request) +{ + struct tcmsg *tcmsg; + int ifindex; + int error; + + error = get_ifindex(netdev, &ifindex); + if (error) { + return NULL; + } + + ofpbuf_init(request, 512); + nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags); + tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg); + tcmsg->tcm_family = AF_UNSPEC; + tcmsg->tcm_ifindex = ifindex; + /* Caller should fill in tcmsg->tcm_handle. */ + /* Caller should fill in tcmsg->tcm_parent. */ + + return tcmsg; +} + +static int +tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) +{ + int error = nl_transact(NETLINK_ROUTE, request, replyp); + ofpbuf_uninit(request); + return error; +} + +/* Adds or deletes a root ingress qdisc on 'netdev'. We use this for + * policing configuration. + * + * This function is equivalent to running the following when 'add' is true: + * /sbin/tc qdisc add dev handle ffff: ingress + * + * This function is equivalent to running the following when 'add' is false: + * /sbin/tc qdisc del dev handle ffff: ingress + * + * The configuration and stats may be seen with the following command: + * /sbin/tc -s qdisc show dev + * + * Returns 0 if successful, otherwise a positive errno value. + */ +static int +tc_add_del_ingress_qdisc(struct netdev *netdev, bool add) +{ + struct ofpbuf request; + struct tcmsg *tcmsg; + int error; + int type = add ? RTM_NEWQDISC : RTM_DELQDISC; + int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0; + + tcmsg = tc_make_request(netdev, type, flags, &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"); + nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); + + error = tc_transact(&request, NULL); + if (error) { + /* If we're deleting the qdisc, don't worry about some of the + * error conditions. */ + if (!add && (error == ENOENT || error == EINVAL)) { + return 0; + } + return error; + } + + return 0; +} + +/* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size + * of 'kbits_burst'. + * + * This function is equivalent to running: + * /sbin/tc filter add dev parent ffff: protocol all prio 49 + * basic police rate kbit burst k + * mtu 65535 drop + * + * The configuration and stats may be seen with the following command: + * /sbin/tc -s filter show eth0 parent ffff: + * + * Returns 0 if successful, otherwise a positive errno value. + */ +static int +tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst) +{ + struct tc_police tc_police; + struct ofpbuf request; + struct tcmsg *tcmsg; + size_t basic_offset; + size_t police_offset; + int error; + int mtu = 65535; + + memset(&tc_police, 0, sizeof tc_police); + tc_police.action = TC_POLICE_SHOT; + tc_police.mtu = mtu; + tc_fill_rate(&tc_police.rate, (kbits_rate * 1000)/8, mtu); + tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate, + kbits_burst * 1024); + + tcmsg = tc_make_request(netdev, RTM_NEWTFILTER, + NLM_F_EXCL | NLM_F_CREATE, &request); + if (!tcmsg) { + return ENODEV; + } + tcmsg->tcm_parent = tc_make_handle(0xffff, 0); + tcmsg->tcm_info = tc_make_handle(49, + (OVS_FORCE uint16_t) htons(ETH_P_ALL)); + + nl_msg_put_string(&request, TCA_KIND, "basic"); + basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); + police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); + nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); + tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); + nl_msg_end_nested(&request, police_offset); + nl_msg_end_nested(&request, basic_offset); + + error = tc_transact(&request, NULL); + if (error) { + return error; + } + + return 0; +} + +static void +read_psched(void) +{ + /* The values in psched are not individually very meaningful, but they are + * important. The tables below show some values seen in the wild. + * + * Some notes: + * + * - "c" has always been a constant 1000000 since at least Linux 2.4.14. + * (Before that, there are hints that it was 1000000000.) + * + * - "d" can be unrealistically large, see the comment on 'buffer_hz' + * above. + * + * /proc/net/psched + * ----------------------------------- + * [1] 000c8000 000f4240 000f4240 00000064 + * [2] 000003e8 00000400 000f4240 3b9aca00 + * [3] 000003e8 00000400 000f4240 3b9aca00 + * [4] 000003e8 00000400 000f4240 00000064 + * [5] 000003e8 00000040 000f4240 3b9aca00 + * [6] 000003e8 00000040 000f4240 000000f9 + * + * a b c d ticks_per_s buffer_hz + * ------- --------- ---------- ------------- ----------- ------------- + * [1] 819,200 1,000,000 1,000,000 100 819,200 100 + * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 + * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 + * [4] 1,000 1,024 1,000,000 100 976,562 100 + * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 + * [6] 1,000 64 1,000,000 249 15,625,000 249 + * + * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p + * [2] 2.6.26-1-686-bigmem from Debian lenny + * [3] 2.6.26-2-sparc64 from Debian lenny + * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p + * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion + * [6] 2.6.34 from kernel.org on KVM + */ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static const char fn[] = "/proc/net/psched"; + unsigned int a, b, c, d; + FILE *stream; + + if (!ovsthread_once_start(&once)) { + return; + } + + ticks_per_s = 1.0; + buffer_hz = 100; + + stream = fopen(fn, "r"); + if (!stream) { + VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); + goto exit; + } + + if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { + VLOG_WARN("%s: read failed", fn); + fclose(stream); + goto exit; + } + VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); + fclose(stream); - /* Remove 'notifier' from its list. */ - list = list_remove(¬ifier->node); - if (list_is_empty(list)) { - /* The list is now empty. Remove it from the hash and free it. */ - const char *netdev_name = netdev_get_name(notifier->notifier.netdev); - shash_delete(&netdev_linux_notifiers, - shash_find(&netdev_linux_notifiers, netdev_name)); - free(list); + if (!a || !c) { + VLOG_WARN("%s: invalid scheduler parameters", fn); + goto exit; } - free(notifier); - /* If that was the last notifier, unregister. */ - if (shash_is_empty(&netdev_linux_notifiers)) { - rtnetlink_notifier_unregister(&netdev_linux_poll_notifier); + ticks_per_s = (double) a * c / b; + if (c == 1000000) { + buffer_hz = d; + } else { + VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", + fn, a, b, c, d); } + VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); + +exit: + ovsthread_once_done(&once); +} + +/* Returns the number of bytes that can be transmitted in 'ticks' ticks at a + * rate of 'rate' bytes per second. */ +static unsigned int +tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) +{ + read_psched(); + return (rate * ticks) / ticks_per_s; +} + +/* Returns the number of ticks that it would take to transmit 'size' bytes at a + * rate of 'rate' bytes per second. */ +static unsigned int +tc_bytes_to_ticks(unsigned int rate, unsigned int size) +{ + read_psched(); + return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; } -const struct netdev_class netdev_linux_class = { - "system", +/* Returns the number of bytes that need to be reserved for qdisc buffering at + * a transmission rate of 'rate' bytes per second. */ +static unsigned int +tc_buffer_per_jiffy(unsigned int rate) +{ + read_psched(); + return rate / buffer_hz; +} - netdev_linux_init, - netdev_linux_run, - netdev_linux_wait, +/* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, + * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, + * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or + * stores NULL into it if it is absent. + * + * '*kind' and '*options' point into 'msg', so they are owned by whoever owns + * 'msg'. + * + * Returns 0 if successful, otherwise a positive errno value. */ +static int +tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, + struct nlattr **options) +{ + static const struct nl_policy tca_policy[] = { + [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, + [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, + }; + struct nlattr *ta[ARRAY_SIZE(tca_policy)]; - netdev_linux_create_system, - netdev_linux_destroy, - NULL, /* reconfigure */ + if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), + tca_policy, ta, ARRAY_SIZE(ta))) { + VLOG_WARN_RL(&rl, "failed to parse qdisc message"); + goto error; + } - netdev_linux_open, - netdev_linux_close, + if (kind) { + *kind = nl_attr_get_string(ta[TCA_KIND]); + } - netdev_linux_enumerate, + if (options) { + *options = ta[TCA_OPTIONS]; + } - netdev_linux_recv, - netdev_linux_recv_wait, - netdev_linux_drain, + return 0; - netdev_linux_send, - netdev_linux_send_wait, +error: + if (kind) { + *kind = NULL; + } + if (options) { + *options = NULL; + } + return EPROTO; +} - 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_linux_set_stats, +/* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the + * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute + * into '*options', and its queue statistics into '*stats'. Any of the output + * arguments may be null. + * + * Returns 0 if successful, otherwise a positive errno value. */ +static int +tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, + struct nlattr **options, struct netdev_queue_stats *stats) +{ + static const struct nl_policy tca_policy[] = { + [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, + [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, + }; + struct nlattr *ta[ARRAY_SIZE(tca_policy)]; - netdev_linux_get_features, - netdev_linux_set_advertisements, - netdev_linux_get_vlan_vid, - netdev_linux_set_policing, + if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), + tca_policy, ta, ARRAY_SIZE(ta))) { + VLOG_WARN_RL(&rl, "failed to parse class message"); + goto error; + } - 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, + if (handlep) { + struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); + *handlep = tc->tcm_handle; + } - netdev_linux_update_flags, + if (options) { + *options = ta[TCA_OPTIONS]; + } - netdev_linux_poll_add, - netdev_linux_poll_remove, -}; + if (stats) { + const struct gnet_stats_queue *gsq; + struct gnet_stats_basic gsb; + + static const struct nl_policy stats_policy[] = { + [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, + .min_len = sizeof gsb }, + [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, + .min_len = sizeof *gsq }, + }; + struct nlattr *sa[ARRAY_SIZE(stats_policy)]; + + if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, + sa, ARRAY_SIZE(sa))) { + VLOG_WARN_RL(&rl, "failed to parse class stats"); + goto error; + } -const struct netdev_class netdev_tap_class = { - "tap", + /* Alignment issues screw up the length of struct gnet_stats_basic on + * some arch/bitsize combinations. Newer versions of Linux have a + * struct gnet_stats_basic_packed, but we can't depend on that. The + * easiest thing to do is just to make a copy. */ + memset(&gsb, 0, sizeof gsb); + memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), + MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); + stats->tx_bytes = gsb.bytes; + stats->tx_packets = gsb.packets; + + gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); + stats->tx_errors = gsq->drops; + } - netdev_linux_init, - netdev_linux_run, - netdev_linux_wait, + return 0; - netdev_linux_create_tap, - netdev_linux_destroy, - NULL, /* reconfigure */ +error: + if (options) { + *options = NULL; + } + if (stats) { + memset(stats, 0, sizeof *stats); + } + return EPROTO; +} - netdev_linux_open, - netdev_linux_close, +/* Queries the kernel for class with identifier 'handle' and parent 'parent' + * on 'netdev'. */ +static int +tc_query_class(const struct netdev *netdev, + unsigned int handle, unsigned int parent, + struct ofpbuf **replyp) +{ + struct ofpbuf request; + struct tcmsg *tcmsg; + int error; - NULL, /* enumerate */ + tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); + if (!tcmsg) { + return ENODEV; + } + tcmsg->tcm_handle = handle; + tcmsg->tcm_parent = parent; - netdev_linux_recv, - netdev_linux_recv_wait, - netdev_linux_drain, + error = tc_transact(&request, replyp); + if (error) { + VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", + netdev_get_name(netdev), + tc_get_major(handle), tc_get_minor(handle), + tc_get_major(parent), tc_get_minor(parent), + ovs_strerror(error)); + } + return error; +} - netdev_linux_send, - netdev_linux_send_wait, +/* Equivalent to "tc class del dev handle ". */ +static int +tc_delete_class(const struct netdev *netdev, unsigned int handle) +{ + struct ofpbuf request; + struct tcmsg *tcmsg; + int error; - 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 */ + tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); + if (!tcmsg) { + return ENODEV; + } + tcmsg->tcm_handle = handle; + tcmsg->tcm_parent = 0; - netdev_linux_get_features, - netdev_linux_set_advertisements, - netdev_linux_get_vlan_vid, - netdev_linux_set_policing, + error = tc_transact(&request, NULL); + if (error) { + VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", + netdev_get_name(netdev), + tc_get_major(handle), tc_get_minor(handle), + ovs_strerror(error)); + } + return error; +} - 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, +/* Equivalent to "tc qdisc del dev root". */ +static int +tc_del_qdisc(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct ofpbuf request; + struct tcmsg *tcmsg; + int error; - netdev_linux_update_flags, + 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; - netdev_linux_poll_add, - netdev_linux_poll_remove, -}; + error = tc_transact(&request, NULL); + if (error == EINVAL) { + /* EINVAL probably means that the default qdisc was in use, in which + * case we've accomplished our purpose. */ + error = 0; + } + if (!error && netdev->tc) { + if (netdev->tc->ops->tc_destroy) { + netdev->tc->ops->tc_destroy(netdev->tc); + } + netdev->tc = NULL; + } + return error; +} - +/* If 'netdev''s qdisc type and parameters are not yet known, queries the + * kernel to determine what they are. Returns 0 if successful, otherwise a + * positive errno value. */ static int -get_stats_via_netlink(int ifindex, struct netdev_stats *stats) +tc_query_qdisc(const struct netdev *netdev_) { - /* Policy for RTNLGRP_LINK messages. + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct ofpbuf request, *qdisc; + const struct tc_ops *ops; + struct tcmsg *tcmsg; + int load_error; + int error; + + if (netdev->tc) { + return 0; + } + + /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have + * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before + * 2.6.35 without that fix backported to it. * - * There are *many* more fields in these messages, but currently we only - * care about these fields. */ - static const struct nl_policy rtnlgrp_link_policy[] = { - [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false }, - [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, - .min_len = sizeof(struct rtnl_link_stats) }, - }; + * To avoid the OOPS, we must not make a request that would attempt to dump + * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a + * few others. There are a few ways that I can see to do this, but most of + * them seem to be racy (and if you lose the race the kernel OOPSes). The + * technique chosen here is to assume that any non-default qdisc that we + * create will have a class with handle 1:0. The built-in qdiscs only have + * a class with handle 0:0. + * + * 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; - struct nl_sock *rtnl_sock; - struct ofpbuf request; - struct ofpbuf *reply; - struct ifinfomsg *ifi; - const struct rtnl_link_stats *rtnl_stats; - struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; + /* Figure out what tc class to instantiate. */ + error = tc_transact(&request, &qdisc); + if (!error) { + const char *kind; + + error = tc_parse_qdisc(qdisc, &kind, NULL); + if (error) { + ops = &tc_ops_other; + } else { + ops = tc_lookup_linux_name(kind); + if (!ops) { + static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); + VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind); + + ops = &tc_ops_other; + } + } + } else if (error == ENOENT) { + /* Either it's a built-in qdisc, or it's a qdisc set up by some + * other entity that doesn't have a handle 1:0. We will assume + * that it's the system default qdisc. */ + ops = &tc_ops_default; + error = 0; + } else { + /* Who knows? Maybe the device got deleted. */ + VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); + ops = &tc_ops_other; + } + + /* Instantiate it. */ + load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); + ovs_assert((load_error == 0) == (netdev->tc != NULL)); + ofpbuf_delete(qdisc); + + return error ? error : load_error; +} + +/* Linux traffic control uses tables with 256 entries ("rtab" tables) to + approximate the time to transmit packets of various lengths. For an MTU of + 256 or less, each entry is exact; for an MTU of 257 through 512, each entry + represents two possible packet lengths; for a MTU of 513 through 1024, four + possible lengths; and so on. + + Returns, for the specified 'mtu', the number of bits that packet lengths + need to be shifted right to fit within such a 256-entry table. */ +static int +tc_calc_cell_log(unsigned int mtu) +{ + int cell_log; + + if (!mtu) { + mtu = ETH_PAYLOAD_MAX; + } + mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; + + for (cell_log = 0; mtu >= 256; cell_log++) { + mtu >>= 1; + } + + return cell_log; +} + +/* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU + * of 'mtu'. */ +static void +tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) +{ + memset(rate, 0, sizeof *rate); + rate->cell_log = tc_calc_cell_log(mtu); + /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ + /* rate->cell_align = 0; */ /* distro headers. */ + rate->mpu = ETH_TOTAL_MIN; + rate->rate = Bps; +} + +/* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink + * attribute of the specified "type". + * + * See tc_calc_cell_log() above for a description of "rtab"s. */ +static void +tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) +{ + uint32_t *rtab; + unsigned int i; + + rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); + for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { + unsigned packet_size = (i + 1) << rate->cell_log; + if (packet_size < rate->mpu) { + packet_size = rate->mpu; + } + rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); + } +} + +/* 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.) */ +static int +tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) +{ + unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; + return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); +} + +/* Linux-only functions declared in netdev-linux.h */ + +/* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If + * 'enable' is true, the bit is set. Otherwise, it is cleared. */ +int +netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, + const char *flag_name, bool enable) +{ + const char *netdev_name = netdev_get_name(netdev); + struct ethtool_value evalue; + uint32_t new_flags; int error; - error = get_rtnl_sock(&rtnl_sock); + COVERAGE_INC(netdev_get_ethtool); + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); if (error) { return error; } - ofpbuf_init(&request, 0); - nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi, - RTM_GETLINK, NLM_F_REQUEST); - ifi = ofpbuf_put_zeros(&request, sizeof *ifi); - ifi->ifi_family = PF_UNSPEC; - ifi->ifi_index = ifindex; - error = nl_sock_transact(rtnl_sock, &request, &reply); - ofpbuf_uninit(&request); + COVERAGE_INC(netdev_set_ethtool); + evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); if (error) { return error; } - if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), - rtnlgrp_link_policy, - attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { - ofpbuf_delete(reply); - return EPROTO; + COVERAGE_INC(netdev_get_ethtool); + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); + if (error) { + return error; } - if (!attrs[IFLA_STATS]) { - VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); - ofpbuf_delete(reply); - return EPROTO; + if (new_flags != evalue.data) { + VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " + "device %s failed", enable ? "enable" : "disable", + flag_name, netdev_name); + return EOPNOTSUPP; } - rtnl_stats = nl_attr_get(attrs[IFLA_STATS]); - stats->rx_packets = rtnl_stats->rx_packets; - stats->tx_packets = rtnl_stats->tx_packets; - stats->rx_bytes = rtnl_stats->rx_bytes; - stats->tx_bytes = rtnl_stats->tx_bytes; - stats->rx_errors = rtnl_stats->rx_errors; - stats->tx_errors = rtnl_stats->tx_errors; - stats->rx_dropped = rtnl_stats->rx_dropped; - stats->tx_dropped = rtnl_stats->tx_dropped; - stats->multicast = rtnl_stats->multicast; - stats->collisions = rtnl_stats->collisions; - stats->rx_length_errors = rtnl_stats->rx_length_errors; - stats->rx_over_errors = rtnl_stats->rx_over_errors; - stats->rx_crc_errors = rtnl_stats->rx_crc_errors; - stats->rx_frame_errors = rtnl_stats->rx_frame_errors; - stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors; - stats->rx_missed_errors = rtnl_stats->rx_missed_errors; - stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors; - stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors; - stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors; - stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors; - stats->tx_window_errors = rtnl_stats->tx_window_errors; - - ofpbuf_delete(reply); - return 0; } + +/* Utility functions. */ + +/* Copies 'src' into 'dst', performing format conversion in the process. */ +static void +netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, + const struct rtnl_link_stats *src) +{ + dst->rx_packets = src->rx_packets; + dst->tx_packets = src->tx_packets; + dst->rx_bytes = src->rx_bytes; + dst->tx_bytes = src->tx_bytes; + dst->rx_errors = src->rx_errors; + dst->tx_errors = src->tx_errors; + dst->rx_dropped = src->rx_dropped; + dst->tx_dropped = src->tx_dropped; + dst->multicast = src->multicast; + dst->collisions = src->collisions; + dst->rx_length_errors = src->rx_length_errors; + dst->rx_over_errors = src->rx_over_errors; + dst->rx_crc_errors = src->rx_crc_errors; + dst->rx_frame_errors = src->rx_frame_errors; + dst->rx_fifo_errors = src->rx_fifo_errors; + dst->rx_missed_errors = src->rx_missed_errors; + dst->tx_aborted_errors = src->tx_aborted_errors; + dst->tx_carrier_errors = src->tx_carrier_errors; + dst->tx_fifo_errors = src->tx_fifo_errors; + dst->tx_heartbeat_errors = src->tx_heartbeat_errors; + dst->tx_window_errors = src->tx_window_errors; +} static int -get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) +get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats) { - static const char fn[] = "/proc/net/dev"; - char line[1024]; - FILE *stream; - int ln; + struct ofpbuf request; + struct ofpbuf *reply; + int error; - stream = fopen(fn, "r"); - if (!stream) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); - return errno; + ofpbuf_init(&request, 0); + nl_msg_put_nlmsghdr(&request, + sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), + RTM_GETLINK, NLM_F_REQUEST); + ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); + nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_)); + error = nl_transact(NETLINK_ROUTE, &request, &reply); + ofpbuf_uninit(&request); + if (error) { + return error; } - ln = 0; - while (fgets(line, sizeof line, stream)) { - if (++ln >= 3) { - char devname[16]; -#define X64 "%"SCNu64 - if (sscanf(line, - " %15[^:]:" - X64 X64 X64 X64 X64 X64 X64 "%*u" - X64 X64 X64 X64 X64 X64 X64 "%*u", - devname, - &stats->rx_bytes, - &stats->rx_packets, - &stats->rx_errors, - &stats->rx_dropped, - &stats->rx_fifo_errors, - &stats->rx_frame_errors, - &stats->multicast, - &stats->tx_bytes, - &stats->tx_packets, - &stats->tx_errors, - &stats->tx_dropped, - &stats->tx_fifo_errors, - &stats->collisions, - &stats->tx_carrier_errors) != 15) { - VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); - } else if (!strcmp(devname, netdev_name)) { - stats->rx_length_errors = UINT64_MAX; - stats->rx_over_errors = UINT64_MAX; - stats->rx_crc_errors = UINT64_MAX; - stats->rx_missed_errors = UINT64_MAX; - stats->tx_aborted_errors = UINT64_MAX; - stats->tx_heartbeat_errors = UINT64_MAX; - stats->tx_window_errors = UINT64_MAX; - fclose(stream); - return 0; - } + if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) { + const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS); + if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) { + netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a)); + error = 0; + } else { + VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); + error = EPROTO; } + } else { + VLOG_WARN_RL(&rl, "short RTM_GETLINK reply"); + error = EPROTO; } - VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); - fclose(stream); - return ENODEV; + + + ofpbuf_delete(reply); + return error; } - + static int -get_flags(const struct netdev *netdev, int *flags) +get_flags(const struct netdev *dev, unsigned int *flags) { struct ifreq ifr; int error; - error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS, - "SIOCGIFFLAGS"); - *flags = ifr.ifr_flags; + *flags = 0; + error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); + if (!error) { + *flags = ifr.ifr_flags; + } return error; } static int -set_flags(struct netdev *netdev, int flags) +set_flags(const char *name, unsigned int flags) { struct ifreq ifr; ifr.ifr_flags = flags; - return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS, - "SIOCSIFFLAGS"); + return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); } static int do_get_ifindex(const char *netdev_name) { struct ifreq ifr; + int error; - 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) { + + error = af_inet_ioctl(SIOCGIFINDEX, &ifr); + if (error) { VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", - netdev_name, strerror(errno)); - return -errno; + netdev_name, ovs_strerror(error)); + return -error; } return ifr.ifr_ifindex; } @@ -1866,19 +4661,23 @@ do_get_ifindex(const char *netdev_name) static int get_ifindex(const struct netdev *netdev_, int *ifindexp) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - *ifindexp = 0; - if (!(netdev_dev->cache_valid & VALID_IFINDEX)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (!(netdev->cache_valid & VALID_IFINDEX)) { int ifindex = do_get_ifindex(netdev_get_name(netdev_)); + if (ifindex < 0) { - return -ifindex; + netdev->get_ifindex_error = -ifindex; + netdev->ifindex = 0; + } else { + netdev->get_ifindex_error = 0; + netdev->ifindex = ifindex; } - netdev_dev->cache_valid |= VALID_IFINDEX; - netdev_dev->ifindex = ifindex; + netdev->cache_valid |= VALID_IFINDEX; } - *ifindexp = netdev_dev->ifindex; - return 0; + + *ifindexp = netdev->ifindex; + return netdev->get_ifindex_error; } static int @@ -1886,14 +4685,20 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) { struct ifreq ifr; int hwaddr_family; + int error; 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", - netdev_name, strerror(errno)); - return errno; + error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); + if (error) { + /* ENODEV probably means that a vif disappeared asynchronously and + * hasn't been removed from the database yet, so reduce the log level + * to INFO for that case. */ + VLOG(error == ENODEV ? VLL_INFO : VLL_ERR, + "ioctl(SIOCGIFHWADDR) on %s device failed: %s", + netdev_name, ovs_strerror(error)); + return error; } hwaddr_family = ifr.ifr_hwaddr.sa_family; if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { @@ -1905,22 +4710,23 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) } static int -set_etheraddr(const char *netdev_name, int hwaddr_family, +set_etheraddr(const char *netdev_name, const uint8_t mac[ETH_ADDR_LEN]) { struct ifreq ifr; + int error; memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - ifr.ifr_hwaddr.sa_family = hwaddr_family; + ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); COVERAGE_INC(netdev_set_hwaddr); - if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { + error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); + if (error) { VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; + netdev_name, ovs_strerror(error)); } - return 0; + return error; } static int @@ -1928,38 +4734,24 @@ netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, int cmd, const char *cmd_name) { struct ifreq ifr; + int error; 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; - COVERAGE_INC(netdev_ethtool); - if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { - return 0; - } else { - if (errno != EOPNOTSUPP) { + error = af_inet_ioctl(SIOCETHTOOL, &ifr); + if (error) { + if (error != EOPNOTSUPP) { VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " - "failed: %s", cmd_name, name, strerror(errno)); + "failed: %s", cmd_name, name, ovs_strerror(error)); } else { /* The device doesn't support this operation. That's pretty * common, so there's no point in logging anything. */ } - return errno; - } -} - -static int -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); - if (ioctl(af_inet_sock, cmd, ifr) == -1) { - VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, - strerror(errno)); - return errno; } - return 0; + return error; } static int @@ -1970,33 +4762,37 @@ netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, int error; ifr.ifr_addr.sa_family = AF_INET; - error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); + error = af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); if (!error) { - const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; + const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, + &ifr.ifr_addr); *ip = sin->sin_addr; } return error; } -/* Obtains a Netlink routing socket that is not subscribed to any multicast - * groups. Returns 0 if successful, otherwise a positive errno value. Stores - * the socket in '*rtnl_sockp' if successful, otherwise a null pointer. */ +/* Returns an AF_PACKET raw socket or a negative errno value. */ static int -get_rtnl_sock(struct nl_sock **rtnl_sockp) +af_packet_sock(void) { - static struct nl_sock *sock; - int error; - - if (!sock) { - error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &sock); - if (error) { - VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", - strerror(error)); + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static int sock; + + if (ovsthread_once_start(&once)) { + sock = socket(AF_PACKET, SOCK_RAW, 0); + if (sock >= 0) { + int error = set_nonblocking(sock); + if (error) { + close(sock); + sock = -error; + } + } else { + sock = -errno; + VLOG_ERR("failed to create packet socket: %s", + ovs_strerror(errno)); } - } else { - error = 0; + ovsthread_once_done(&once); } - *rtnl_sockp = sock; - return error; + return sock; }