X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev.c;h=c21afef829ca22d1bca8e475813df654dc5f9955;hb=a4fdb0f3bdbfff1924afefd19239260ed429c193;hp=bed480f71159e3f734952b37fe7aec5da8d1e9ae;hpb=0b9eaeb9112743044d9bb564bea9299f00269291;p=sliver-openvswitch.git diff --git a/lib/netdev.c b/lib/netdev.c index bed480f71..45a416520 100644 --- a/lib/netdev.c +++ b/lib/netdev.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -17,596 +17,615 @@ #include #include "netdev.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 "coverage.h" +#include "dpif.h" #include "dynamic-string.h" #include "fatal-signal.h" +#include "hash.h" #include "list.h" -#include "netlink.h" +#include "netdev-dpdk.h" +#include "netdev-provider.h" +#include "netdev-vport.h" #include "ofpbuf.h" #include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" -#include "socket-util.h" +#include "shash.h" +#include "smap.h" +#include "sset.h" #include "svec.h" +#include "vlog.h" -/* linux/if.h defines IFF_LOWER_UP, net/if.h doesn't. - * net/if.h defines if_nameindex(), linux/if.h doesn't. - * We can't include both headers, so define IFF_LOWER_UP ourselves. */ -#ifndef IFF_LOWER_UP -#define IFF_LOWER_UP 0x10000 -#endif - -/* These were introduced in Linux 2.6.14, so they might be missing if we have - * old headers. */ -#ifndef ADVERTISED_Pause -#define ADVERTISED_Pause (1 << 13) -#endif -#ifndef ADVERTISED_Asym_Pause -#define ADVERTISED_Asym_Pause (1 << 14) -#endif +VLOG_DEFINE_THIS_MODULE(netdev); -#define THIS_MODULE VLM_netdev -#include "vlog.h" +COVERAGE_DEFINE(netdev_received); +COVERAGE_DEFINE(netdev_sent); +COVERAGE_DEFINE(netdev_add_router); +COVERAGE_DEFINE(netdev_get_stats); -struct netdev { - struct list node; - char *name; - - /* File descriptors. For ordinary network devices, the two fds below are - * the same; for tap devices, they differ. */ - int netdev_fd; /* Network device. */ - int tap_fd; /* TAP character device, if any, otherwise the - * network device. */ - - /* Cached network device information. */ - int ifindex; /* -1 if not known. */ - uint8_t etheraddr[ETH_ADDR_LEN]; - struct in6_addr in6; - int speed; - int mtu; - int txqlen; - int hwaddr_family; - - int save_flags; /* Initial device flags. */ - int changed_flags; /* Flags that we changed. */ +struct netdev_saved_flags { + struct netdev *netdev; + struct list node; /* In struct netdev's saved_flags_list. */ + enum netdev_flags saved_flags; + enum netdev_flags saved_values; }; -/* Policy for RTNLGRP_LINK messages. +/* Protects 'netdev_shash' and the mutable members of struct netdev. */ +static struct ovs_mutex netdev_mutex = OVS_MUTEX_INITIALIZER; + +/* All created network devices. */ +static struct shash netdev_shash OVS_GUARDED_BY(netdev_mutex) + = SHASH_INITIALIZER(&netdev_shash); + +/* Protects 'netdev_classes' against insertions or deletions. * - * There are *many* more fields in these messages, but currently we only care - * about interface names. */ -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) }, + * This is a recursive mutex to allow recursive acquisition when calling into + * providers. For example, netdev_run() calls into provider 'run' functions, + * which might reasonably want to call one of the netdev functions that takes + * netdev_class_mutex. */ +static struct ovs_mutex netdev_class_mutex OVS_ACQ_BEFORE(netdev_mutex); + +/* Contains 'struct netdev_registered_class'es. */ +static struct hmap netdev_classes OVS_GUARDED_BY(netdev_class_mutex) + = HMAP_INITIALIZER(&netdev_classes); + +struct netdev_registered_class { + struct hmap_node hmap_node; /* In 'netdev_classes', by class->type. */ + const struct netdev_class *class; + atomic_int ref_cnt; /* Number of 'struct netdev's of this class. */ }; -/* All open network devices. */ -static struct list netdev_list = LIST_INITIALIZER(&netdev_list); +/* 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); -/* An AF_INET socket (used for ioctl operations). */ -static int af_inet_sock = -1; +static void restore_all_flags(void *aux OVS_UNUSED); +void update_device_args(struct netdev *, const struct shash *args); -/* NETLINK_ROUTE socket. */ -static struct nl_sock *rtnl_sock; +int +netdev_n_rxq(const struct netdev *netdev) +{ + return netdev->n_rxq; +} -/* 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 use_netlink_stats; +bool +netdev_is_pmd(const struct netdev *netdev) +{ + return !strcmp(netdev->netdev_class->type, "dpdk"); +} -/* 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 void +netdev_class_mutex_initialize(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + + if (ovsthread_once_start(&once)) { + ovs_mutex_init_recursive(&netdev_class_mutex); + ovsthread_once_done(&once); + } +} -static void init_netdev(void); -static int do_open_netdev(const char *name, int ethertype, int tap_fd, - struct netdev **netdev_); -static int restore_flags(struct netdev *netdev); -static int get_flags(const char *netdev_name, int *flagsp); -static int set_flags(const char *netdev_name, int flags); -static int do_get_ifindex(const char *netdev_name); -static int get_ifindex(const struct netdev *, int *ifindexp); -static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN], - int *hwaddr_familyp); -static int set_etheraddr(const char *netdev_name, int hwaddr_family, - const uint8_t[ETH_ADDR_LEN]); - -/* Obtains the IPv6 address for 'name' into 'in6'. */ static void -get_ipv6_address(const char *name, struct in6_addr *in6) +netdev_initialize(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - FILE *file; - char line[128]; + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + + if (ovsthread_once_start(&once)) { + netdev_class_mutex_initialize(); + + fatal_signal_add_hook(restore_all_flags, NULL, NULL, true); + netdev_vport_patch_register(); + +#ifdef __linux__ + netdev_register_provider(&netdev_linux_class); + netdev_register_provider(&netdev_internal_class); + netdev_register_provider(&netdev_tap_class); + netdev_vport_tunnel_register(); +#endif +#if defined(__FreeBSD__) || defined(__NetBSD__) + netdev_register_provider(&netdev_tap_class); + netdev_register_provider(&netdev_bsd_class); +#endif + netdev_dpdk_register(); - file = fopen("/proc/net/if_inet6", "r"); - if (file == NULL) { - /* This most likely indicates that the host doesn't have IPv6 support, - * so it's not really a failure condition.*/ - *in6 = in6addr_any; - return; + ovsthread_once_done(&once); } +} + +/* Performs periodic work needed by all the various kinds of netdevs. + * + * If your program opens any netdevs, it must call this function within its + * main poll loop. */ +void +netdev_run(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) +{ + struct netdev_registered_class *rc; - while (fgets(line, sizeof line, file)) { - uint8_t *s6 = in6->s6_addr; - char ifname[16 + 1]; - -#define X8 "%2"SCNx8 - if (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 - && !strcmp(name, ifname)) - { - fclose(file); - return; + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + if (rc->class->run) { + rc->class->run(); } } - *in6 = in6addr_any; + ovs_mutex_unlock(&netdev_class_mutex); +} - fclose(file); +/* Arranges for poll_block() to wake up when netdev_run() needs to be called. + * + * If your program opens any netdevs, it must call this function within its + * main poll loop. */ +void +netdev_wait(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) +{ + struct netdev_registered_class *rc; + + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + if (rc->class->wait) { + rc->class->wait(); + } + } + ovs_mutex_unlock(&netdev_class_mutex); } -static int -do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd, - int cmd, const char *cmd_name) +static struct netdev_registered_class * +netdev_lookup_class(const char *type) + OVS_REQ_RDLOCK(netdev_class_mutex) { - struct ifreq ifr; + struct netdev_registered_class *rc; - memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); - ifr.ifr_data = (caddr_t) ecmd; + HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0), + &netdev_classes) { + if (!strcmp(type, rc->class->type)) { + return rc; + } + } + return NULL; +} - ecmd->cmd = cmd; - COVERAGE_INC(netdev_ethtool); - if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) { - return 0; +/* Initializes and registers a new netdev provider. After successful + * registration, new netdevs of that type can be opened using netdev_open(). */ +int +netdev_register_provider(const struct netdev_class *new_class) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) +{ + int error; + + netdev_class_mutex_initialize(); + ovs_mutex_lock(&netdev_class_mutex); + if (netdev_lookup_class(new_class->type)) { + VLOG_WARN("attempted to register duplicate netdev provider: %s", + new_class->type); + error = EEXIST; } else { - if (errno != EOPNOTSUPP) { - VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " - "failed: %s", cmd_name, netdev->name, - strerror(errno)); + error = new_class->init ? new_class->init() : 0; + if (!error) { + struct netdev_registered_class *rc; + + rc = xmalloc(sizeof *rc); + hmap_insert(&netdev_classes, &rc->hmap_node, + hash_string(new_class->type, 0)); + rc->class = new_class; + atomic_init(&rc->ref_cnt, 0); } else { - /* The device doesn't support this operation. That's pretty - * common, so there's no point in logging anything. */ + VLOG_ERR("failed to initialize %s network device class: %s", + new_class->type, ovs_strerror(error)); } - return errno; } + ovs_mutex_unlock(&netdev_class_mutex); + + return error; } -static int -do_get_features(struct netdev *netdev, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) +/* Unregisters a netdev provider. 'type' must have been previously + * registered and not currently be in use by any netdevs. After unregistration + * new netdevs of that type cannot be opened using netdev_open(). */ +int +netdev_unregister_provider(const char *type) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - struct ethtool_cmd ecmd; + struct netdev_registered_class *rc; int error; - *current = 0; - *supported = 0; - *advertised = 0; - *peer = 0; - - memset(&ecmd, 0, sizeof ecmd); - error = do_ethtool(netdev, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); - if (error) { - return error; - } + ovs_mutex_lock(&netdev_class_mutex); + rc = netdev_lookup_class(type); + if (!rc) { + VLOG_WARN("attempted to unregister a netdev provider that is not " + "registered: %s", type); + error = EAFNOSUPPORT; + } else { + int ref_cnt; - if (ecmd.supported & SUPPORTED_10baseT_Half) { - *supported |= OFPPF_10MB_HD; - } - if (ecmd.supported & SUPPORTED_10baseT_Full) { - *supported |= OFPPF_10MB_FD; - } - if (ecmd.supported & SUPPORTED_100baseT_Half) { - *supported |= OFPPF_100MB_HD; - } - if (ecmd.supported & SUPPORTED_100baseT_Full) { - *supported |= OFPPF_100MB_FD; - } - if (ecmd.supported & SUPPORTED_1000baseT_Half) { - *supported |= OFPPF_1GB_HD; - } - if (ecmd.supported & SUPPORTED_1000baseT_Full) { - *supported |= OFPPF_1GB_FD; - } - if (ecmd.supported & SUPPORTED_10000baseT_Full) { - *supported |= OFPPF_10GB_FD; - } - if (ecmd.supported & SUPPORTED_TP) { - *supported |= OFPPF_COPPER; - } - if (ecmd.supported & SUPPORTED_FIBRE) { - *supported |= OFPPF_FIBER; - } - if (ecmd.supported & SUPPORTED_Autoneg) { - *supported |= OFPPF_AUTONEG; - } - if (ecmd.supported & SUPPORTED_Pause) { - *supported |= OFPPF_PAUSE; - } - if (ecmd.supported & SUPPORTED_Asym_Pause) { - *supported |= OFPPF_PAUSE_ASYM; + atomic_read(&rc->ref_cnt, &ref_cnt); + if (!ref_cnt) { + hmap_remove(&netdev_classes, &rc->hmap_node); + free(rc); + error = 0; + } else { + VLOG_WARN("attempted to unregister in use netdev provider: %s", + type); + error = EBUSY; + } } + ovs_mutex_unlock(&netdev_class_mutex); - /* Set the advertised features */ - if (ecmd.advertising & ADVERTISED_10baseT_Half) { - *advertised |= OFPPF_10MB_HD; - } - if (ecmd.advertising & ADVERTISED_10baseT_Full) { - *advertised |= OFPPF_10MB_FD; - } - if (ecmd.advertising & ADVERTISED_100baseT_Half) { - *advertised |= OFPPF_100MB_HD; - } - if (ecmd.advertising & ADVERTISED_100baseT_Full) { - *advertised |= OFPPF_100MB_FD; - } - if (ecmd.advertising & ADVERTISED_1000baseT_Half) { - *advertised |= OFPPF_1GB_HD; - } - if (ecmd.advertising & ADVERTISED_1000baseT_Full) { - *advertised |= OFPPF_1GB_FD; - } - if (ecmd.advertising & ADVERTISED_10000baseT_Full) { - *advertised |= OFPPF_10GB_FD; - } - if (ecmd.advertising & ADVERTISED_TP) { - *advertised |= OFPPF_COPPER; - } - if (ecmd.advertising & ADVERTISED_FIBRE) { - *advertised |= OFPPF_FIBER; - } - if (ecmd.advertising & ADVERTISED_Autoneg) { - *advertised |= OFPPF_AUTONEG; - } - if (ecmd.advertising & ADVERTISED_Pause) { - *advertised |= OFPPF_PAUSE; - } - if (ecmd.advertising & ADVERTISED_Asym_Pause) { - *advertised |= OFPPF_PAUSE_ASYM; - } + return error; +} - /* Set the current features */ - 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; - } +/* Clears 'types' and enumerates the types of all currently registered netdev + * providers into it. The caller must first initialize the sset. */ +void +netdev_enumerate_types(struct sset *types) + OVS_EXCLUDED(netdev_mutex) +{ + struct netdev_registered_class *rc; - if (ecmd.port == PORT_TP) { - *current |= OFPPF_COPPER; - } - else if (ecmd.port == PORT_FIBRE) { - *current |= OFPPF_FIBER; - } + netdev_initialize(); + sset_clear(types); - if (ecmd.autoneg) { - *current |= OFPPF_AUTONEG; + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + sset_add(types, rc->class->type); } - return 0; + ovs_mutex_unlock(&netdev_class_mutex); } -/* Opens the network device named 'name' (e.g. "eth0") and returns zero if - * successful, otherwise a positive errno value. On success, sets '*netdevp' - * to the new network device, otherwise to null. +/* Check that the network device name is not the same as any of the registered + * vport providers' dpif_port name (dpif_port is NULL if the vport provider + * does not define it) or the datapath internal port name (e.g. ovs-system). * - * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to - * capture frames of that type received on the device. It may also be one of - * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those - * categories. */ -int -netdev_open(const char *name, int ethertype, struct netdev **netdevp) + * Returns true if there is a name conflict, false otherwise. */ +bool +netdev_is_reserved_name(const char *name) + OVS_EXCLUDED(netdev_mutex) { - if (!strncmp(name, "tap:", 4)) { - return netdev_open_tap(name + 4, netdevp); - } else { - return do_open_netdev(name, ethertype, -1, netdevp); + struct netdev_registered_class *rc; + + netdev_initialize(); + + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class); + if (dpif_port && !strcmp(dpif_port, name)) { + ovs_mutex_unlock(&netdev_class_mutex); + return true; + } + } + ovs_mutex_unlock(&netdev_class_mutex); + + if (!strncmp(name, "ovs-", 4)) { + struct sset types; + const char *type; + + sset_init(&types); + dp_enumerate_types(&types); + SSET_FOR_EACH (type, &types) { + if (!strcmp(name+4, type)) { + sset_destroy(&types); + return true; + } + } + sset_destroy(&types); } + + return false; } -/* Opens a TAP virtual network device. If 'name' is a nonnull, non-empty - * string, attempts to assign that name to the TAP device (failing if the name - * is already in use); otherwise, a name is automatically assigned. Returns - * zero if successful, otherwise a positive errno value. On success, sets - * '*netdevp' to the new network device, otherwise to null. */ +/* Opens the network device named 'name' (e.g. "eth0") of the specified 'type' + * (e.g. "system") and returns zero if successful, otherwise a positive errno + * value. On success, sets '*netdevp' to the new network device, otherwise to + * null. + * + * Some network devices may need to be configured (with netdev_set_config()) + * before they can be used. */ int -netdev_open_tap(const char *name, struct netdev **netdevp) +netdev_open(const char *name, const char *type, struct netdev **netdevp) + OVS_EXCLUDED(netdev_mutex) { - static const char tap_dev[] = "/dev/net/tun"; - struct ifreq ifr; + struct netdev *netdev; int error; - int tap_fd; - tap_fd = open(tap_dev, O_RDWR); - if (tap_fd < 0) { - ovs_error(errno, "opening \"%s\" failed", tap_dev); - return errno; - } + netdev_initialize(); + + ovs_mutex_lock(&netdev_class_mutex); + ovs_mutex_lock(&netdev_mutex); + netdev = shash_find_data(&netdev_shash, name); + if (!netdev) { + struct netdev_registered_class *rc; + + rc = netdev_lookup_class(type && type[0] ? type : "system"); + if (rc) { + netdev = rc->class->alloc(); + if (netdev) { + memset(netdev, 0, sizeof *netdev); + netdev->netdev_class = rc->class; + netdev->name = xstrdup(name); + netdev->change_seq = 1; + netdev->node = shash_add(&netdev_shash, name, netdev); + + /* By default enable one rx queue per netdev. */ + if (netdev->netdev_class->rxq_alloc) { + netdev->n_rxq = 1; + } else { + netdev->n_rxq = 0; + } + list_init(&netdev->saved_flags_list); - memset(&ifr, 0, sizeof ifr); - ifr.ifr_flags = IFF_TAP | IFF_NO_PI; - if (name) { - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - } - if (ioctl(tap_fd, TUNSETIFF, &ifr) < 0) { - int error = errno; - ovs_error(error, "ioctl(TUNSETIFF) on \"%s\" failed", tap_dev); - close(tap_fd); - return error; - } + error = rc->class->construct(netdev); + if (!error) { + int old_ref_cnt; - error = set_nonblocking(tap_fd); - if (error) { - ovs_error(error, "set_nonblocking on \"%s\" failed", tap_dev); - close(tap_fd); - return error; + atomic_add(&rc->ref_cnt, 1, &old_ref_cnt); + netdev_change_seq_changed(netdev); + } else { + free(netdev->name); + ovs_assert(list_is_empty(&netdev->saved_flags_list)); + shash_delete(&netdev_shash, netdev->node); + rc->class->dealloc(netdev); + } + } else { + error = ENOMEM; + } + } else { + VLOG_WARN("could not create netdev %s of unknown type %s", + name, type); + error = EAFNOSUPPORT; + } + } else { + error = 0; } - error = do_open_netdev(ifr.ifr_name, NETDEV_ETH_TYPE_NONE, tap_fd, - netdevp); - if (error) { - close(tap_fd); + ovs_mutex_unlock(&netdev_mutex); + ovs_mutex_unlock(&netdev_class_mutex); + + if (!error) { + netdev->ref_cnt++; + *netdevp = netdev; + } else { + *netdevp = NULL; } return error; } -static int -do_open_netdev(const char *name, int ethertype, int tap_fd, - struct netdev **netdev_) -{ - int netdev_fd; - struct sockaddr_ll sll; - struct ifreq ifr; - int ifindex = -1; - uint8_t etheraddr[ETH_ADDR_LEN]; - struct in6_addr in6; - int mtu; - int txqlen; - int hwaddr_family; - int error; - struct netdev *netdev; +/* Returns a reference to 'netdev_' for the caller to own. Returns null if + * 'netdev_' is null. */ +struct netdev * +netdev_ref(const struct netdev *netdev_) + OVS_EXCLUDED(netdev_mutex) +{ + struct netdev *netdev = CONST_CAST(struct netdev *, netdev_); - init_netdev(); - *netdev_ = NULL; - COVERAGE_INC(netdev_open); - - /* Create raw socket. */ - netdev_fd = socket(PF_PACKET, SOCK_RAW, - htons(ethertype == NETDEV_ETH_TYPE_NONE ? 0 - : ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL - : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 - : ethertype)); - if (netdev_fd < 0) { - return errno; + if (netdev) { + ovs_mutex_lock(&netdev_mutex); + ovs_assert(netdev->ref_cnt > 0); + netdev->ref_cnt++; + ovs_mutex_unlock(&netdev_mutex); } + return netdev; +} - if (ethertype != NETDEV_ETH_TYPE_NONE) { - /* Set non-blocking mode. */ - error = set_nonblocking(netdev_fd); - if (error) { - goto error_already_set; - } +/* Reconfigures the device 'netdev' with 'args'. 'args' may be empty + * or NULL if none are needed. */ +int +netdev_set_config(struct netdev *netdev, const struct smap *args) + OVS_EXCLUDED(netdev_mutex) +{ + if (netdev->netdev_class->set_config) { + const struct smap no_args = SMAP_INITIALIZER(&no_args); + int error; - /* Get ethernet device index. */ - ifindex = do_get_ifindex(name); - if (ifindex < 0) { - return -ifindex; + error = netdev->netdev_class->set_config(netdev, + args ? args : &no_args); + if (error) { + VLOG_WARN("%s: could not set configuration (%s)", + netdev_get_name(netdev), ovs_strerror(error)); } + return error; + } else if (args && !smap_is_empty(args)) { + VLOG_WARN("%s: arguments provided to device that is not configurable", + netdev_get_name(netdev)); + } + return 0; +} - /* Bind to specific ethernet device. */ - memset(&sll, 0, sizeof sll); - sll.sll_family = AF_PACKET; - sll.sll_ifindex = ifindex; - if (bind(netdev_fd, (struct sockaddr *) &sll, sizeof sll) < 0) { - VLOG_ERR("bind to %s failed: %s", name, strerror(errno)); - goto error; - } +/* Returns the current configuration for 'netdev' in 'args'. The caller must + * have already initialized 'args' with smap_init(). Returns 0 on success, in + * which case 'args' will be filled with 'netdev''s configuration. On failure + * returns a positive errno value, in which case 'args' will be empty. + * + * The caller owns 'args' and its contents and must eventually free them with + * smap_destroy(). */ +int +netdev_get_config(const struct netdev *netdev, struct smap *args) + OVS_EXCLUDED(netdev_mutex) +{ + int 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); + smap_clear(args); + if (netdev->netdev_class->get_config) { + error = netdev->netdev_class->get_config(netdev, args); if (error) { - goto error_already_set; + smap_clear(args); } + } else { + error = 0; } - /* Get MAC address. */ - error = get_etheraddr(name, etheraddr, &hwaddr_family); - if (error) { - goto error_already_set; - } - - /* Get MTU. */ - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - if (ioctl(netdev_fd, SIOCGIFMTU, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCGIFMTU) on %s device failed: %s", - name, strerror(errno)); - goto error; - } - mtu = ifr.ifr_mtu; + return error; +} - /* Get TX queue length. */ - if (ioctl(netdev_fd, SIOCGIFTXQLEN, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCGIFTXQLEN) on %s device failed: %s", - name, strerror(errno)); - goto error; - } - txqlen = ifr.ifr_qlen; - - get_ipv6_address(name, &in6); - - /* Allocate network device. */ - netdev = xmalloc(sizeof *netdev); - netdev->name = xstrdup(name); - netdev->ifindex = ifindex; - netdev->txqlen = txqlen; - netdev->hwaddr_family = hwaddr_family; - netdev->netdev_fd = netdev_fd; - netdev->tap_fd = tap_fd < 0 ? netdev_fd : tap_fd; - memcpy(netdev->etheraddr, etheraddr, sizeof etheraddr); - netdev->mtu = mtu; - netdev->in6 = in6; - - /* Save flags to restore at close or exit. */ - error = get_flags(netdev->name, &netdev->save_flags); - if (error) { - goto error_already_set; +const struct netdev_tunnel_config * +netdev_get_tunnel_config(const struct netdev *netdev) + OVS_EXCLUDED(netdev_mutex) +{ + if (netdev->netdev_class->get_tunnel_config) { + return netdev->netdev_class->get_tunnel_config(netdev); + } else { + return NULL; } - netdev->changed_flags = 0; - fatal_signal_block(); - list_push_back(&netdev_list, &netdev->node); - fatal_signal_unblock(); - - /* Success! */ - *netdev_ = netdev; - return 0; +} -error: - error = errno; -error_already_set: - close(netdev_fd); - if (tap_fd >= 0) { - close(tap_fd); +static void +netdev_unref(struct netdev *dev) + OVS_RELEASES(netdev_mutex) +{ + ovs_assert(dev->ref_cnt); + if (!--dev->ref_cnt) { + const struct netdev_class *class = dev->netdev_class; + struct netdev_registered_class *rc; + int old_ref_cnt; + + dev->netdev_class->destruct(dev); + + shash_delete(&netdev_shash, dev->node); + free(dev->name); + dev->netdev_class->dealloc(dev); + ovs_mutex_unlock(&netdev_mutex); + + ovs_mutex_lock(&netdev_class_mutex); + rc = netdev_lookup_class(class->type); + atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt); + ovs_assert(old_ref_cnt > 0); + ovs_mutex_unlock(&netdev_class_mutex); + } else { + ovs_mutex_unlock(&netdev_mutex); } - return error; } /* Closes and destroys 'netdev'. */ void netdev_close(struct netdev *netdev) + OVS_EXCLUDED(netdev_mutex) { if (netdev) { - /* Bring down interface and drop promiscuous mode, if we brought up - * the interface or enabled promiscuous mode. */ - int error; - fatal_signal_block(); - error = restore_flags(netdev); - list_remove(&netdev->node); - fatal_signal_unblock(); - if (error) { - VLOG_WARN("failed to restore network device flags on %s: %s", - netdev->name, strerror(error)); - } - - /* Free. */ - free(netdev->name); - close(netdev->netdev_fd); - if (netdev->netdev_fd != netdev->tap_fd) { - close(netdev->tap_fd); - } - free(netdev); + ovs_mutex_lock(&netdev_mutex); + netdev_unref(netdev); } } -/* Pads 'buffer' out with zero-bytes to the minimum valid length of an - * Ethernet packet, if necessary. */ -static void -pad_to_minimum_length(struct ofpbuf *buffer) +/* Parses 'netdev_name_', which is of the form [type@]name into its component + * pieces. 'name' and 'type' must be freed by the caller. */ +void +netdev_parse_name(const char *netdev_name_, char **name, char **type) { - if (buffer->size < ETH_TOTAL_MIN) { - ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size); + char *netdev_name = xstrdup(netdev_name_); + char *separator; + + separator = strchr(netdev_name, '@'); + if (separator) { + *separator = '\0'; + *type = netdev_name; + *name = xstrdup(separator + 1); + } else { + *name = netdev_name; + *type = xstrdup("system"); } } -/* Attempts to receive a packet from 'netdev' into 'buffer', which the caller - * must have initialized with sufficient room for the packet. The space - * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN - * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()). - * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN - * need not be included.) +/* Attempts to open a netdev_rxq handle for obtaining packets received on + * 'netdev'. On success, returns 0 and stores a nonnull 'netdev_rxq *' into + * '*rxp'. On failure, returns a positive errno value and stores NULL into + * '*rxp'. * - * If a packet is successfully retrieved, returns 0. In this case 'buffer' is - * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a - * positive errno value. Returns EAGAIN immediately if no packet is ready to - * be returned. - */ + * Some kinds of network devices might not support receiving packets. This + * function returns EOPNOTSUPP in that case.*/ int -netdev_recv(struct netdev *netdev, struct ofpbuf *buffer) -{ - ssize_t n_bytes; - - assert(buffer->size == 0); - assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN); - do { - n_bytes = read(netdev->tap_fd, - ofpbuf_tail(buffer), ofpbuf_tailroom(buffer)); - } while (n_bytes < 0 && errno == EINTR); - if (n_bytes < 0) { - if (errno != EAGAIN) { - VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", - netdev->name, strerror(errno)); +netdev_rxq_open(struct netdev *netdev, struct netdev_rxq **rxp, int id) + OVS_EXCLUDED(netdev_mutex) +{ + int error; + + if (netdev->netdev_class->rxq_alloc && id < netdev->n_rxq) { + struct netdev_rxq *rx = netdev->netdev_class->rxq_alloc(); + if (rx) { + rx->netdev = netdev; + rx->queue_id = id; + error = netdev->netdev_class->rxq_construct(rx); + if (!error) { + ovs_mutex_lock(&netdev_mutex); + netdev->ref_cnt++; + ovs_mutex_unlock(&netdev_mutex); + + *rxp = rx; + return 0; + } + netdev->netdev_class->rxq_dealloc(rx); + } else { + error = ENOMEM; } - return errno; } else { - COVERAGE_INC(netdev_received); - buffer->size += n_bytes; - - /* When the kernel internally sends out an Ethernet frame on an - * interface, it gives us a copy *before* padding the frame to the - * minimum length. Thus, when it sends out something like an ARP - * request, we see a too-short frame. So pad it out to the minimum - * length. */ - pad_to_minimum_length(buffer); - return 0; + error = EOPNOTSUPP; } + + *rxp = NULL; + return error; } -/* 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'. */ +/* Closes 'rx'. */ void -netdev_recv_wait(struct netdev *netdev) +netdev_rxq_close(struct netdev_rxq *rx) + OVS_EXCLUDED(netdev_mutex) { - poll_fd_wait(netdev->tap_fd, POLLIN); + if (rx) { + struct netdev *netdev = rx->netdev; + netdev->netdev_class->rxq_destruct(rx); + netdev->netdev_class->rxq_dealloc(rx); + netdev_close(netdev); + } } -/* Discards all packets waiting to be received from 'netdev'. */ +/* Attempts to receive batch of packets from 'rx'. + * + * Returns EAGAIN immediately if no packet is ready to be received. + * + * Returns EMSGSIZE, and discards the packet, if the received packet is longer + * than 'ofpbuf_tailroom(buffer)'. + * + * It is advised that the tailroom of 'buffer' should be + * VLAN_HEADER_LEN bytes longer than the MTU to allow space for an + * out-of-band VLAN header to be added to the packet. At the very least, + * 'buffer' must have at least ETH_TOTAL_MIN bytes of tailroom. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ int -netdev_drain(struct netdev *netdev) +netdev_rxq_recv(struct netdev_rxq *rx, struct ofpbuf **buffers, int *cnt) { - if (netdev->tap_fd != netdev->netdev_fd) { - drain_fd(netdev->tap_fd, netdev->txqlen); - return 0; - } else { - return drain_rcvbuf(netdev->netdev_fd); + int retval; + + retval = rx->netdev->netdev_class->rxq_recv(rx, buffers, cnt); + if (!retval) { + COVERAGE_INC(netdev_received); } + return retval; +} + +/* Arranges for poll_block() to wake up when a packet is ready to be received + * on 'rx'. */ +void +netdev_rxq_wait(struct netdev_rxq *rx) +{ + rx->netdev->netdev_class->rxq_wait(rx); +} + +/* Discards any packets ready to be received on 'rx'. */ +int +netdev_rxq_drain(struct netdev_rxq *rx) +{ + return (rx->netdev->netdev_class->rxq_drain + ? rx->netdev->netdev_class->rxq_drain(rx) + : 0); } /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive @@ -614,39 +633,25 @@ netdev_drain(struct netdev *netdev) * immediately. Returns EMSGSIZE if a partial packet was transmitted or if * the packet is too big or too small to transmit on the device. * - * The caller retains ownership of 'buffer' in all cases. + * To retain ownership of 'buffer' caller can set may_steal to false. * * The kernel maintains a packet transmission queue, so the caller is not - * expected to do additional queuing of packets. */ + * expected to do additional queuing of packets. + * + * Some network devices may not implement support for this function. In such + * cases this function will always return EOPNOTSUPP. */ int -netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) -{ - ssize_t n_bytes; - - do { - n_bytes = write(netdev->tap_fd, buffer->data, buffer->size); - } while (n_bytes < 0 && errno == EINTR); - - if (n_bytes < 0) { - /* The Linux AF_PACKET implementation never blocks waiting for room - * for packets, instead returning ENOBUFS. Translate this into EAGAIN - * for the caller. */ - if (errno == ENOBUFS) { - return EAGAIN; - } else if (errno != EAGAIN) { - VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", - netdev->name, strerror(errno)); - } - return errno; - } else if (n_bytes != buffer->size) { - VLOG_WARN_RL(&rl, - "send partial Ethernet packet (%d bytes of %zu) on %s", - (int) n_bytes, buffer->size, netdev->name); - return EMSGSIZE; - } else { +netdev_send(struct netdev *netdev, struct ofpbuf *buffer, bool may_steal) +{ + int error; + + error = (netdev->netdev_class->send + ? netdev->netdev_class->send(netdev, buffer, may_steal) + : EOPNOTSUPP); + if (!error) { COVERAGE_INC(netdev_sent); - return 0; } + return error; } /* Registers with the poll loop to wake up from the next call to poll_block() @@ -659,11 +664,8 @@ netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) void netdev_send_wait(struct netdev *netdev) { - if (netdev->tap_fd == netdev->netdev_fd) { - poll_fd_wait(netdev->tap_fd, POLLOUT); - } else { - /* TAP device always accepts packets.*/ - poll_immediate_wake(); + if (netdev->netdev_class->send_wait) { + netdev->netdev_class->send_wait(netdev); } } @@ -672,26 +674,16 @@ netdev_send_wait(struct netdev *netdev) int netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN]) { - int error = set_etheraddr(netdev->name, netdev->hwaddr_family, mac); - if (!error) { - memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN); - } - return error; + return netdev->netdev_class->set_etheraddr(netdev, mac); } +/* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the + * the MAC address into 'mac'. On failure, returns a positive errno value and + * clears 'mac' to all-zeros. */ int -netdev_nodev_set_etheraddr(const char *name, const uint8_t mac[ETH_ADDR_LEN]) +netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN]) { - init_netdev(); - return set_etheraddr(name, ARPHRD_ETHER, mac); -} - -/* Returns a pointer to 'netdev''s MAC address. The caller must not modify or - * free the returned buffer. */ -const uint8_t * -netdev_get_etheraddr(const struct netdev *netdev) -{ - return netdev->etheraddr; + return netdev->netdev_class->get_etheraddr(netdev, mac); } /* Returns the name of the network device that 'netdev' represents, @@ -702,157 +694,199 @@ netdev_get_name(const struct netdev *netdev) return netdev->name; } -/* Returns the maximum size of transmitted (and received) packets on 'netdev', - * in bytes, not including the hardware header; thus, this is typically 1500 - * bytes for Ethernet devices. */ +/* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted + * (and received) packets, in bytes, not including the hardware header; thus, + * this is typically 1500 bytes for Ethernet devices. + * + * If successful, returns 0 and stores the MTU size in '*mtup'. Returns + * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not). + * On other failure, returns a positive errno value. On failure, sets '*mtup' + * to 0. */ int -netdev_get_mtu(const struct netdev *netdev) +netdev_get_mtu(const struct netdev *netdev, int *mtup) { - return netdev->mtu; -} + const struct netdev_class *class = netdev->netdev_class; + int 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. On failure, all of the - * passed-in values are set to 0. */ -int -netdev_get_features(struct netdev *netdev, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) -{ - uint32_t dummy[4]; - return do_get_features(netdev, - current ? current : &dummy[0], - advertised ? advertised : &dummy[1], - supported ? supported : &dummy[2], - peer ? peer : &dummy[3]); + error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP; + if (error) { + *mtup = 0; + if (error != EOPNOTSUPP) { + VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: " + "%s", netdev_get_name(netdev), ovs_strerror(error)); + } + } + return error; } +/* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted + * (and received) packets, in bytes. + * + * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an + * MTU (as e.g. some tunnels do not). On other failure, returns a positive + * errno value. */ int -netdev_set_advertisements(struct netdev *netdev, uint32_t advertise) +netdev_set_mtu(const struct netdev *netdev, int mtu) { - struct ethtool_cmd ecmd; + const struct netdev_class *class = netdev->netdev_class; int error; - memset(&ecmd, 0, sizeof ecmd); - error = do_ethtool(netdev, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); - if (error) { - return error; + error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP; + if (error && error != EOPNOTSUPP) { + VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s", + netdev_get_name(netdev), ovs_strerror(error)); } - ecmd.advertising = 0; - if (advertise & OFPPF_10MB_HD) { - ecmd.advertising |= ADVERTISED_10baseT_Half; - } - if (advertise & OFPPF_10MB_FD) { - ecmd.advertising |= ADVERTISED_10baseT_Full; - } - if (advertise & OFPPF_100MB_HD) { - ecmd.advertising |= ADVERTISED_100baseT_Half; - } - if (advertise & OFPPF_100MB_FD) { - ecmd.advertising |= ADVERTISED_100baseT_Full; - } - if (advertise & OFPPF_1GB_HD) { - ecmd.advertising |= ADVERTISED_1000baseT_Half; - } - if (advertise & OFPPF_1GB_FD) { - ecmd.advertising |= ADVERTISED_1000baseT_Full; - } - if (advertise & OFPPF_10GB_FD) { - ecmd.advertising |= ADVERTISED_10000baseT_Full; - } - if (advertise & OFPPF_COPPER) { - ecmd.advertising |= ADVERTISED_TP; - } - if (advertise & OFPPF_FIBER) { - ecmd.advertising |= ADVERTISED_FIBRE; - } - if (advertise & OFPPF_AUTONEG) { - ecmd.advertising |= ADVERTISED_Autoneg; - } - if (advertise & OFPPF_PAUSE) { - ecmd.advertising |= ADVERTISED_Pause; - } - if (advertise & OFPPF_PAUSE_ASYM) { - ecmd.advertising |= ADVERTISED_Asym_Pause; - } - return do_ethtool(netdev, &ecmd, ETHTOOL_SSET, "ETHTOOL_SSET"); + return error; } -/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address - * and '*mask' to the netmask (if they are non-null) and returns true. - * Otherwise, returns false. */ -bool -netdev_nodev_get_in4(const char *netdev_name, struct in_addr *in4, - struct in_addr *mask) +/* Returns the ifindex of 'netdev', if successful, as a positive number. On + * failure, returns a negative errno value. + * + * The desired semantics of the ifindex value are a combination of those + * specified by POSIX for if_nametoindex() and by SNMP for ifIndex. An ifindex + * value should be unique within a host and remain stable at least until + * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber" + * but many systems do not follow this rule anyhow. + * + * Some network devices may not implement support for this function. In such + * cases this function will always return -EOPNOTSUPP. + */ +int +netdev_get_ifindex(const struct netdev *netdev) { - struct ifreq ifr; - struct in_addr ip = { INADDR_ANY }; + int (*get_ifindex)(const struct netdev *); - init_netdev(); + get_ifindex = netdev->netdev_class->get_ifindex; - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - ifr.ifr_addr.sa_family = AF_INET; - COVERAGE_INC(netdev_get_in4); - if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) { - struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; - ip = sin->sin_addr; - } else { - VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFADDR) failed: %s", - netdev_name, strerror(errno)); + return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP; +} + +/* 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. On failure, all of the + * passed-in values are set to 0. + * + * Some network devices may not implement support for this function. In such + * cases this function will always return EOPNOTSUPP. */ +int +netdev_get_features(const struct netdev *netdev, + enum netdev_features *current, + enum netdev_features *advertised, + enum netdev_features *supported, + enum netdev_features *peer) +{ + int (*get_features)(const struct netdev *netdev, + enum netdev_features *current, + enum netdev_features *advertised, + enum netdev_features *supported, + enum netdev_features *peer); + enum netdev_features dummy[4]; + int error; + + if (!current) { + current = &dummy[0]; } - if (in4) { - *in4 = ip; + if (!advertised) { + advertised = &dummy[1]; + } + if (!supported) { + supported = &dummy[2]; + } + if (!peer) { + peer = &dummy[3]; } - if (mask) { - if (ioctl(af_inet_sock, SIOCGIFNETMASK, &ifr) == 0) { - struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; - *mask = sin->sin_addr; - } else { - VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFNETMASK) failed: %s", - netdev_name, strerror(errno)); - } + get_features = netdev->netdev_class->get_features; + error = get_features + ? get_features(netdev, current, advertised, supported, + peer) + : EOPNOTSUPP; + if (error) { + *current = *advertised = *supported = *peer = 0; } + return error; +} - return ip.s_addr != INADDR_ANY; +/* Returns the maximum speed of a network connection that has the NETDEV_F_* + * bits in 'features', in bits per second. If no bits that indicate a speed + * are set in 'features', returns 'default_bps'. */ +uint64_t +netdev_features_to_bps(enum netdev_features features, + uint64_t default_bps) +{ + enum { + F_1000000MB = NETDEV_F_1TB_FD, + F_100000MB = NETDEV_F_100GB_FD, + F_40000MB = NETDEV_F_40GB_FD, + F_10000MB = NETDEV_F_10GB_FD, + F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD, + F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD, + F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD + }; + + return ( features & F_1000000MB ? UINT64_C(1000000000000) + : features & F_100000MB ? UINT64_C(100000000000) + : features & F_40000MB ? UINT64_C(40000000000) + : features & F_10000MB ? UINT64_C(10000000000) + : features & F_1000MB ? UINT64_C(1000000000) + : features & F_100MB ? UINT64_C(100000000) + : features & F_10MB ? UINT64_C(10000000) + : default_bps); } +/* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link + * are set in 'features', otherwise false. */ bool -netdev_get_in4(const struct netdev *netdev, struct in_addr *in4, struct - in_addr *mask) +netdev_features_is_full_duplex(enum netdev_features features) { - return netdev_nodev_get_in4(netdev->name, in4, mask); + return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD + | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD + | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0; } -static void -make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) +/* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if + * successful, otherwise a positive errno value. */ +int +netdev_set_advertisements(struct netdev *netdev, + enum netdev_features advertise) { - struct sockaddr_in sin; - memset(&sin, 0, sizeof sin); - sin.sin_family = AF_INET; - sin.sin_addr = addr; - sin.sin_port = 0; - - memset(sa, 0, sizeof *sa); - memcpy(sa, &sin, sizeof sin); + return (netdev->netdev_class->set_advertisements + ? netdev->netdev_class->set_advertisements( + netdev, advertise) + : EOPNOTSUPP); } -static int -do_set_addr(struct netdev *netdev, int sock, - int ioctl_nr, const char *ioctl_name, struct in_addr addr) +/* If 'netdev' has an assigned IPv4 address, sets '*address' to that address + * and '*netmask' to its netmask and returns 0. Otherwise, returns a positive + * errno value and sets '*address' to 0 (INADDR_ANY). + * + * The following error values have well-defined meanings: + * + * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address. + * + * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'. + * + * 'address' or 'netmask' or both may be null, in which case the address or + * netmask is not reported. */ +int +netdev_get_in4(const struct netdev *netdev, + struct in_addr *address_, struct in_addr *netmask_) { - struct ifreq ifr; + struct in_addr address; + struct in_addr netmask; int error; - strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); - make_in4_sockaddr(&ifr.ifr_addr, addr); - COVERAGE_INC(netdev_set_in4); - error = ioctl(sock, ioctl_nr, &ifr) < 0 ? errno : 0; - if (error) { - VLOG_WARN("ioctl(%s): %s", ioctl_name, strerror(error)); + error = (netdev->netdev_class->get_in4 + ? netdev->netdev_class->get_in4(netdev, + &address, &netmask) + : EOPNOTSUPP); + if (address_) { + address_->s_addr = error ? 0 : address.s_addr; + } + if (netmask_) { + netmask_->s_addr = error ? 0 : netmask.s_addr; } return error; } @@ -863,840 +897,773 @@ do_set_addr(struct netdev *netdev, int sock, int netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask) { - int error; - - error = do_set_addr(netdev, af_inet_sock, - SIOCSIFADDR, "SIOCSIFADDR", addr); - if (!error && addr.s_addr != INADDR_ANY) { - error = do_set_addr(netdev, af_inet_sock, - SIOCSIFNETMASK, "SIOCSIFNETMASK", mask); - } - return error; + return (netdev->netdev_class->set_in4 + ? netdev->netdev_class->set_in4(netdev, addr, mask) + : EOPNOTSUPP); } -/* Adds 'router' as a default IP gateway. */ +/* Obtains ad IPv4 address from device name and save the address in + * in4. Returns 0 if successful, otherwise a positive errno value. + */ int -netdev_add_router(struct in_addr router) +netdev_get_in4_by_name(const char *device_name, struct in_addr *in4) { - struct in_addr any = { INADDR_ANY }; - struct rtentry rt; + struct netdev *netdev; int error; - memset(&rt, 0, sizeof rt); - make_in4_sockaddr(&rt.rt_dst, any); - 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 = netdev_open(device_name, "system", &netdev); if (error) { - VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + in4->s_addr = htonl(0); + return error; } + + error = netdev_get_in4(netdev, in4, NULL); + netdev_close(netdev); return error; } -/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if - * 'in6' is non-null) and returns true. Otherwise, returns false. */ -bool -netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6) +/* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds + * to 'netdev'. */ +int +netdev_add_router(struct netdev *netdev, struct in_addr router) { - if (in6) { - *in6 = netdev->in6; + COVERAGE_INC(netdev_add_router); + return (netdev->netdev_class->add_router + ? netdev->netdev_class->add_router(netdev, router) + : EOPNOTSUPP); +} + +/* Looks up the next hop for 'host' for the TCP/IP stack that corresponds to + * 'netdev'. If a route cannot not be determined, sets '*next_hop' to 0, + * '*netdev_name' to null, and returns a positive errno value. Otherwise, if a + * next hop is found, stores the next hop gateway's address (0 if 'host' is on + * a directly connected network) in '*next_hop' and a copy of the name of the + * device to reach 'host' in '*netdev_name', and returns 0. The caller is + * responsible for freeing '*netdev_name' (by calling free()). */ +int +netdev_get_next_hop(const struct netdev *netdev, + const struct in_addr *host, struct in_addr *next_hop, + char **netdev_name) +{ + int error = (netdev->netdev_class->get_next_hop + ? netdev->netdev_class->get_next_hop( + host, next_hop, netdev_name) + : EOPNOTSUPP); + if (error) { + next_hop->s_addr = 0; + *netdev_name = NULL; } - return memcmp(&netdev->in6, &in6addr_any, sizeof netdev->in6) != 0; + return error; } -/* Obtains the current flags for 'netdev' and stores them into '*flagsp'. - * Returns 0 if successful, otherwise a positive errno value. On failure, - * stores 0 into '*flagsp'. */ +/* Populates 'smap' with status information. + * + * Populates 'smap' with 'netdev' specific status information. This + * information may be used to populate the status column of the Interface table + * as defined in ovs-vswitchd.conf.db(5). */ int -netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp) +netdev_get_status(const struct netdev *netdev, struct smap *smap) { - return netdev_nodev_get_flags(netdev->name, flagsp); + return (netdev->netdev_class->get_status + ? netdev->netdev_class->get_status(netdev, smap) + : EOPNOTSUPP); } -static int -nd_to_iff_flags(enum netdev_flags nd) +/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and + * returns 0. Otherwise, returns a positive errno value and sets '*in6' to + * all-zero-bits (in6addr_any). + * + * The following error values have well-defined meanings: + * + * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address. + * + * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'. + * + * 'in6' may be null, in which case the address itself is not reported. */ +int +netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6) { - int iff = 0; - if (nd & NETDEV_UP) { - iff |= IFF_UP; - } - if (nd & NETDEV_PROMISC) { - iff |= IFF_PROMISC; + struct in6_addr dummy; + int error; + + error = (netdev->netdev_class->get_in6 + ? netdev->netdev_class->get_in6(netdev, + in6 ? in6 : &dummy) + : EOPNOTSUPP); + if (error && in6) { + memset(in6, 0, sizeof *in6); } - return iff; + return error; } /* On 'netdev', turns off the flags in 'off' and then turns on the flags in - * 'on'. If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. Returns 0 if - * successful, otherwise a positive errno value. */ + * 'on'. Returns 0 if successful, otherwise a positive errno value. */ static int do_update_flags(struct netdev *netdev, enum netdev_flags off, - enum netdev_flags on, bool permanent) + enum netdev_flags on, enum netdev_flags *old_flagsp, + struct netdev_saved_flags **sfp) + OVS_EXCLUDED(netdev_mutex) { - int old_flags, new_flags; + struct netdev_saved_flags *sf = NULL; + enum netdev_flags old_flags; int error; - error = get_flags(netdev->name, &old_flags); + error = netdev->netdev_class->update_flags(netdev, off & ~on, on, + &old_flags); if (error) { - return error; + VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s", + off || on ? "set" : "get", netdev_get_name(netdev), + ovs_strerror(error)); + old_flags = 0; + } else if ((off || on) && sfp) { + enum netdev_flags new_flags = (old_flags & ~off) | on; + enum netdev_flags changed_flags = old_flags ^ new_flags; + if (changed_flags) { + ovs_mutex_lock(&netdev_mutex); + *sfp = sf = xmalloc(sizeof *sf); + sf->netdev = netdev; + list_push_front(&netdev->saved_flags_list, &sf->node); + sf->saved_flags = changed_flags; + sf->saved_values = changed_flags & new_flags; + + netdev->ref_cnt++; + ovs_mutex_unlock(&netdev_mutex); + } } - new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); - if (!permanent) { - netdev->changed_flags |= new_flags ^ old_flags; + if (old_flagsp) { + *old_flagsp = old_flags; } - if (new_flags != old_flags) { - error = set_flags(netdev->name, new_flags); + if (sfp) { + *sfp = sf; } + return error; } +/* Obtains the current flags for 'netdev' and stores them into '*flagsp'. + * Returns 0 if successful, otherwise a positive errno value. On failure, + * stores 0 into '*flagsp'. */ +int +netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp) +{ + struct netdev *netdev = CONST_CAST(struct netdev *, netdev_); + return do_update_flags(netdev, 0, 0, flagsp, NULL); +} + /* Sets the flags for 'netdev' to 'flags'. - * If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. * Returns 0 if successful, otherwise a positive errno value. */ int netdev_set_flags(struct netdev *netdev, enum netdev_flags flags, - bool permanent) + struct netdev_saved_flags **sfp) { - return do_update_flags(netdev, -1, flags, permanent); + return do_update_flags(netdev, -1, flags, NULL, sfp); } -/* Turns on the specified 'flags' on 'netdev'. - * If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. - * Returns 0 if successful, otherwise a positive errno value. */ +/* Turns on the specified 'flags' on 'netdev': + * + * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly + * allocated 'struct netdev_saved_flags *' that may be passed to + * netdev_restore_flags() to restore the original values of 'flags' on + * 'netdev' (this will happen automatically at program termination if + * netdev_restore_flags() is never called) , or to NULL if no flags were + * actually changed. + * + * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets + * '*sfp' to NULL. */ int netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags, - bool permanent) + struct netdev_saved_flags **sfp) { - return do_update_flags(netdev, 0, flags, permanent); + return do_update_flags(netdev, 0, flags, NULL, sfp); } -/* Turns off the specified 'flags' on 'netdev'. - * If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. - * Returns 0 if successful, otherwise a positive errno value. */ +/* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for + * details of the interface. */ int netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags, - bool permanent) + struct netdev_saved_flags **sfp) { - return do_update_flags(netdev, flags, 0, permanent); + return do_update_flags(netdev, flags, 0, NULL, sfp); } -/* 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'. */ -int -netdev_nodev_arp_lookup(const char *netdev_name, uint32_t ip, - uint8_t mac[ETH_ADDR_LEN]) +/* Restores the flags that were saved in 'sf', and destroys 'sf'. + * Does nothing if 'sf' is NULL. */ +void +netdev_restore_flags(struct netdev_saved_flags *sf) + OVS_EXCLUDED(netdev_mutex) { - struct arpreq r; - struct sockaddr_in *pa; - int retval; + if (sf) { + struct netdev *netdev = sf->netdev; + enum netdev_flags old_flags; - init_netdev(); - - memset(&r, 0, sizeof r); - pa = (struct sockaddr_in *) &r.arp_pa; - pa->sin_family = AF_INET; - pa->sin_addr.s_addr = ip; - pa->sin_port = 0; - r.arp_ha.sa_family = ARPHRD_ETHER; - r.arp_flags = 0; - strncpy(r.arp_dev, netdev_name, sizeof r.arp_dev); - COVERAGE_INC(netdev_arp_lookup); - retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; - 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_name, IP_ARGS(&ip), strerror(retval)); + netdev->netdev_class->update_flags(netdev, + sf->saved_flags & sf->saved_values, + sf->saved_flags & ~sf->saved_values, + &old_flags); + + ovs_mutex_lock(&netdev_mutex); + list_remove(&sf->node); + free(sf); + netdev_unref(netdev); } - return retval; } +/* 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 no ARP table entry for 'ip' on 'netdev'. */ int -netdev_arp_lookup(const struct netdev *netdev, uint32_t ip, - uint8_t mac[ETH_ADDR_LEN]) +netdev_arp_lookup(const struct netdev *netdev, + ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) { - return netdev_nodev_arp_lookup(netdev->name, ip, mac); + int error = (netdev->netdev_class->arp_lookup + ? netdev->netdev_class->arp_lookup(netdev, ip, mac) + : EOPNOTSUPP); + if (error) { + memset(mac, 0, ETH_ADDR_LEN); + } + return error; } -static int -get_stats_via_netlink(int ifindex, struct netdev_stats *stats) +/* Returns true if carrier is active (link light is on) on 'netdev'. */ +bool +netdev_get_carrier(const struct netdev *netdev) { - struct ofpbuf request; - struct ofpbuf *reply; - struct ifinfomsg *ifi; - const struct rtnl_link_stats *rtnl_stats; - struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; int error; + enum netdev_flags flags; + bool carrier; - 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); - 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; + netdev_get_flags(netdev, &flags); + if (!(flags & NETDEV_UP)) { + return false; } - if (!attrs[IFLA_STATS]) { - VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); - ofpbuf_delete(reply); - return EPROTO; + if (!netdev->netdev_class->get_carrier) { + return true; } - 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; -} - -static int -get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) -{ - static const char fn[] = "/proc/net/dev"; - char line[1024]; - FILE *stream; - int ln; - - stream = fopen(fn, "r"); - if (!stream) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); - return errno; + error = netdev->netdev_class->get_carrier(netdev, &carrier); + if (error) { + VLOG_DBG("%s: failed to get network device carrier status, assuming " + "down: %s", netdev_get_name(netdev), ovs_strerror(error)); + carrier = false; } - 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; - } - } - } - VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); - fclose(stream); - return ENODEV; + return carrier; } -int -netdev_get_carrier(const struct netdev *netdev, bool *carrier) +/* Returns the number of times 'netdev''s carrier has changed. */ +long long int +netdev_get_carrier_resets(const struct netdev *netdev) { - return netdev_nodev_get_carrier(netdev->name, carrier); + return (netdev->netdev_class->get_carrier_resets + ? netdev->netdev_class->get_carrier_resets(netdev) + : 0); } +/* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for + * link status instead of checking 'netdev''s carrier. 'netdev''s MII + * registers will be polled once ever 'interval' milliseconds. If 'netdev' + * does not support MII, another method may be used as a fallback. If + * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to + * its normal behavior. + * + * Returns 0 if successful, otherwise a positive errno value. */ int -netdev_nodev_get_carrier(const char *netdev_name, bool *carrier) +netdev_set_miimon_interval(struct netdev *netdev, long long int interval) { - char line[8]; - int retval; - int error; - char *fn; - int fd; - - *carrier = false; - - fn = xasprintf("/sys/class/net/%s/carrier", netdev_name); - fd = open(fn, O_RDONLY); - if (fd < 0) { - error = errno; - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); - goto exit; - } - - 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_close; - } else if (retval == 0) { - error = EPROTO; - VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); - goto exit_close; - } - - 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_close; - } - *carrier = line[0] != '0'; - error = 0; - -exit_close: - close(fd); -exit: - free(fn); - return error; + return (netdev->netdev_class->set_miimon_interval + ? netdev->netdev_class->set_miimon_interval(netdev, interval) + : EOPNOTSUPP); } +/* Retrieves current device stats for 'netdev'. */ int netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats) { int error; COVERAGE_INC(netdev_get_stats); - if (use_netlink_stats) { - int ifindex; - - error = get_ifindex(netdev, &ifindex); - if (!error) { - error = get_stats_via_netlink(ifindex, stats); - } - } else { - error = get_stats_via_proc(netdev->name, stats); - } - + error = (netdev->netdev_class->get_stats + ? netdev->netdev_class->get_stats(netdev, stats) + : EOPNOTSUPP); if (error) { memset(stats, 0xff, sizeof *stats); } 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" -/* We redirect stderr to /dev/null because we often want to remove all - * traffic control configuration on a port so its in a known state. If - * this done when there is no such configuration, tc complains, so we just - * always ignore it. - */ -#define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null" - -/* Attempts to set input rate limiting (policing) policy. */ +/* Attempts to change the stats for 'netdev' to those provided in 'stats'. + * Returns 0 if successful, otherwise a positive errno value. + * + * This will probably fail for most network devices. Some devices might only + * allow setting their stats to 0. */ int -netdev_nodev_set_policing(const char *netdev_name, uint32_t kbits_rate, - uint32_t kbits_burst) +netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats) { - char command[1024]; - - init_netdev(); - - COVERAGE_INC(netdev_set_policing); - if (kbits_rate) { - if (!kbits_burst) { - /* Default to 10 kilobits if not specified. */ - kbits_burst = 10; - } - - /* xxx This should be more careful about only adding if it - * xxx actually exists, as opposed to always deleting it. */ - snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name); - if (system(command) == -1) { - VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name); - } - - 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; - } - - 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; - } - } else { - snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name); - if (system(command) == -1) { - VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name); - } - } - - return 0; + return (netdev->netdev_class->set_stats + ? netdev->netdev_class->set_stats(netdev, stats) + : EOPNOTSUPP); } +/* Attempts to set input rate limiting (policing) policy, such that up to + * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst + * size of 'kbits' kb. */ int netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate, uint32_t kbits_burst) { - return netdev_nodev_set_policing(netdev->name, kbits_rate, kbits_burst); + return (netdev->netdev_class->set_policing + ? netdev->netdev_class->set_policing(netdev, + kbits_rate, kbits_burst) + : EOPNOTSUPP); } -/* Initializes 'svec' with a list of the names of all known network devices. */ -void -netdev_enumerate(struct svec *svec) +/* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it + * empty if 'netdev' does not support QoS. Any names added to 'types' should + * be documented as valid for the "type" column in the "QoS" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * Every network device supports disabling QoS with a type of "", but this type + * will not be added to 'types'. + * + * The caller must initialize 'types' (e.g. with sset_init()) before calling + * this function. The caller is responsible for destroying 'types' (e.g. with + * sset_destroy()) when it is no longer needed. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +netdev_get_qos_types(const struct netdev *netdev, struct sset *types) { - struct if_nameindex *names; - - svec_init(svec); - names = if_nameindex(); - if (names) { - size_t i; + const struct netdev_class *class = netdev->netdev_class; + return (class->get_qos_types + ? class->get_qos_types(netdev, types) + : 0); +} - for (i = 0; names[i].if_name != NULL; i++) { - svec_add(svec, names[i].if_name); +/* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS, + * which should be "" or one of the types returned by netdev_get_qos_types() + * for 'netdev'. Returns 0 if successful, otherwise a positive errno value. + * On success, initializes 'caps' with the QoS capabilities; on failure, clears + * 'caps' to all zeros. */ +int +netdev_get_qos_capabilities(const struct netdev *netdev, const char *type, + struct netdev_qos_capabilities *caps) +{ + const struct netdev_class *class = netdev->netdev_class; + + if (*type) { + int retval = (class->get_qos_capabilities + ? class->get_qos_capabilities(netdev, type, caps) + : EOPNOTSUPP); + if (retval) { + memset(caps, 0, sizeof *caps); } - if_freenameindex(names); + return retval; } else { - VLOG_WARN("could not obtain list of network device names: %s", - strerror(errno)); + /* Every netdev supports turning off QoS. */ + memset(caps, 0, sizeof *caps); + return 0; } } -/* Attempts to locate a device based on its IPv4 address. The caller - * may provide a hint as to the device by setting 'netdev_name' to a - * likely device name. This string must be malloc'd, since if it is - * not correct then it will be freed. If there is no hint, then - * 'netdev_name' must be the NULL pointer. +/* Obtains the number of queues supported by 'netdev' for the specified 'type' + * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores + * the number of queues (zero on failure) in '*n_queuesp'. * - * If the device is found, the return value will be true and 'netdev_name' - * contains the device's name as a string, which the caller is responsible - * for freeing. If the device is not found, the return value is false. */ -bool -netdev_find_dev_by_in4(const struct in_addr *in4, char **netdev_name) + * This is just a simple wrapper around netdev_get_qos_capabilities(). */ +int +netdev_get_n_queues(const struct netdev *netdev, + const char *type, unsigned int *n_queuesp) { - int i; - struct in_addr dev_in4; - struct svec dev_list; + struct netdev_qos_capabilities caps; + int retval; - /* Check the hint first. */ - if (*netdev_name && (netdev_nodev_get_in4(*netdev_name, &dev_in4, NULL)) - && (dev_in4.s_addr == in4->s_addr)) { - return true; - } + retval = netdev_get_qos_capabilities(netdev, type, &caps); + *n_queuesp = caps.n_queues; + return retval; +} - free(*netdev_name); - *netdev_name = NULL; - netdev_enumerate(&dev_list); +/* Queries 'netdev' about its currently configured form of QoS. If successful, + * stores the name of the current form of QoS into '*typep', stores any details + * of configuration as string key-value pairs in 'details', and returns 0. On + * failure, sets '*typep' to NULL and returns a positive errno value. + * + * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'. + * + * The caller must initialize 'details' as an empty smap (e.g. with + * smap_init()) before calling this function. The caller must free 'details' + * when it is no longer needed (e.g. with smap_destroy()). + * + * The caller must not modify or free '*typep'. + * + * '*typep' will be one of the types returned by netdev_get_qos_types() for + * 'netdev'. The contents of 'details' should be documented as valid for + * '*typep' in the "other_config" column in the "QoS" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */ +int +netdev_get_qos(const struct netdev *netdev, + const char **typep, struct smap *details) +{ + const struct netdev_class *class = netdev->netdev_class; + int retval; - for (i=0; is_addr)) { - *netdev_name = xstrdup(dev_list.names[i]); - svec_destroy(&dev_list); - return true; + if (class->get_qos) { + retval = class->get_qos(netdev, typep, details); + if (retval) { + *typep = NULL; + smap_clear(details); } + return retval; + } else { + /* 'netdev' doesn't support QoS, so report that QoS is disabled. */ + *typep = ""; + return 0; } - - svec_destroy(&dev_list); - return false; } -/* Looks up the next hop for 'ip'. If the next hop can be found, the - * address is stored in 'next_hop'. If a gateway is not required to - * reach 'ip', zero is stored in 'next_hop'. In either case, zero is - * returned and a copy of the name of the device to reach 'ip' is stored - * in 'netdev_name', which the caller is responsible for freeing. If a - * route could not be determined, a positive errno is returned. */ +/* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type' + * with details of configuration from 'details'. Returns 0 if successful, + * otherwise a positive errno value. On error, the previous QoS configuration + * is retained. + * + * When this function changes the type of QoS (not just 'details'), this also + * resets all queue configuration for 'netdev' to their defaults (which depend + * on the specific type of QoS). Otherwise, the queue configuration for + * 'netdev' is unchanged. + * + * 'type' should be "" (to disable QoS) or one of the types returned by + * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be + * documented as valid for the given 'type' in the "other_config" column in the + * "QoS" table in vswitchd/vswitch.xml (which is built as + * ovs-vswitchd.conf.db(8)). + * + * NULL may be specified for 'details' if there are no configuration + * details. */ int -netdev_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, - char **netdev_name) -{ - static const char fn[] = "/proc/net/route"; - FILE *stream; - char line[256]; - int ln; - - *netdev_name = NULL; - stream = fopen(fn, "r"); - if (stream == NULL) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); - return errno; - } +netdev_set_qos(struct netdev *netdev, + const char *type, const struct smap *details) +{ + const struct netdev_class *class = netdev->netdev_class; - ln = 0; - while (fgets(line, sizeof line, stream)) { - if (++ln >= 2) { - char iface[17]; - uint32_t 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", - fn, ln, line); - continue; - } - if (!(flags & RTF_UP)) { - /* Skip routes that aren't up. */ - continue; - } + if (!type) { + type = ""; + } - /* The output of 'dest', 'mask', and 'gateway' were given in - * network byte order, so we don't need need any endian - * conversions here. */ - if ((dest & mask) == (host->s_addr & mask)) { - if (!gateway) { - /* The host is directly reachable. */ - next_hop->s_addr = 0; - } else { - /* To reach the host, we must go through a gateway. */ - next_hop->s_addr = gateway; - } - *netdev_name = xstrdup(iface); - fclose(stream); - return 0; - } + if (class->set_qos) { + if (!details) { + static const struct smap empty = SMAP_INITIALIZER(&empty); + details = ∅ } + return class->set_qos(netdev, type, details); + } else { + return *type ? EOPNOTSUPP : 0; } - - fclose(stream); - return ENXIO; } -/* Obtains the current flags for the network device named 'netdev_name' and - * stores them into '*flagsp'. Returns 0 if successful, otherwise a positive - * errno value. On error, stores 0 into '*flagsp'. +/* Queries 'netdev' for information about the queue numbered 'queue_id'. If + * successful, adds that information as string key-value pairs to 'details'. + * Returns 0 if successful, otherwise a positive errno value. + * + * 'queue_id' must be less than the number of queues supported by 'netdev' for + * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)). * - * If only device flags are needed, this is more efficient than calling - * netdev_open(), netdev_get_flags(), netdev_close(). */ + * The returned contents of 'details' should be documented as valid for the + * given 'type' in the "other_config" column in the "Queue" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * The caller must initialize 'details' (e.g. with smap_init()) before calling + * this function. The caller must free 'details' when it is no longer needed + * (e.g. with smap_destroy()). */ int -netdev_nodev_get_flags(const char *netdev_name, enum netdev_flags *flagsp) +netdev_get_queue(const struct netdev *netdev, + unsigned int queue_id, struct smap *details) { - int error, flags; - - init_netdev(); - - *flagsp = 0; - error = get_flags(netdev_name, &flags); - if (error) { - return error; - } + const struct netdev_class *class = netdev->netdev_class; + int retval; - if (flags & IFF_UP) { - *flagsp |= NETDEV_UP; - } - if (flags & IFF_PROMISC) { - *flagsp |= NETDEV_PROMISC; + retval = (class->get_queue + ? class->get_queue(netdev, queue_id, details) + : EOPNOTSUPP); + if (retval) { + smap_clear(details); } - return 0; + return retval; } +/* Configures the queue numbered 'queue_id' on 'netdev' with the key-value + * string pairs in 'details'. The contents of 'details' should be documented + * as valid for the given 'type' in the "other_config" column in the "Queue" + * table in vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * Returns 0 if successful, otherwise a positive errno value. On failure, the + * given queue's configuration should be unmodified. + * + * 'queue_id' must be less than the number of queues supported by 'netdev' for + * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)). + * + * This function does not modify 'details', and the caller retains ownership of + * it. */ int -netdev_nodev_get_etheraddr(const char *netdev_name, uint8_t mac[6]) +netdev_set_queue(struct netdev *netdev, + unsigned int queue_id, const struct smap *details) { - init_netdev(); + const struct netdev_class *class = netdev->netdev_class; + return (class->set_queue + ? class->set_queue(netdev, queue_id, details) + : EOPNOTSUPP); +} - return get_etheraddr(netdev_name, mac, NULL); +/* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds + * of QoS may have a fixed set of queues, in which case attempts to delete them + * will fail with EOPNOTSUPP. + * + * Returns 0 if successful, otherwise a positive errno value. On failure, the + * given queue will be unmodified. + * + * 'queue_id' must be less than the number of queues supported by 'netdev' for + * the current form of QoS (e.g. as returned by + * netdev_get_n_queues(netdev)). */ +int +netdev_delete_queue(struct netdev *netdev, unsigned int queue_id) +{ + const struct netdev_class *class = netdev->netdev_class; + return (class->delete_queue + ? class->delete_queue(netdev, queue_id) + : EOPNOTSUPP); } -/* 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. */ +/* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and + * fills 'stats' with the queue's statistics; individual members of 'stats' may + * be set to all-1-bits if the statistic is unavailable. On failure, returns a + * positive errno value and fills 'stats' with values indicating unsupported + * statistics. */ int -netdev_get_vlan_vid(const char *netdev_name, int *vlan_vid) +netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id, + struct netdev_queue_stats *stats) { - struct ds line = DS_EMPTY_INITIALIZER; - FILE *stream = NULL; - 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; - } + const struct netdev_class *class = netdev->netdev_class; + int retval; - 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); - } - goto done; + retval = (class->get_queue_stats + ? class->get_queue_stats(netdev, queue_id, stats) + : EOPNOTSUPP); + if (retval) { + stats->tx_bytes = UINT64_MAX; + stats->tx_packets = UINT64_MAX; + stats->tx_errors = UINT64_MAX; + stats->created = LLONG_MIN; } + return retval; +} - 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; +/* Initializes 'dump' to begin dumping the queues in a netdev. + * + * This function provides no status indication. An error status for the entire + * dump operation is provided when it is completed by calling + * netdev_queue_dump_done(). + */ +void +netdev_queue_dump_start(struct netdev_queue_dump *dump, + const struct netdev *netdev) +{ + dump->netdev = netdev_ref(netdev); + if (netdev->netdev_class->queue_dump_start) { + dump->error = netdev->netdev_class->queue_dump_start(netdev, + &dump->state); + } else { + dump->error = EOPNOTSUPP; } +} - error = 0; +/* Attempts to retrieve another queue from 'dump', which must have been + * initialized with netdev_queue_dump_start(). On success, stores a new queue + * ID into '*queue_id', fills 'details' with configuration details for the + * queue, and returns true. On failure, returns false. + * + * Queues are not necessarily dumped in increasing order of queue ID (or any + * other predictable order). + * + * Failure might indicate an actual error or merely that the last queue has + * been dumped. An error status for the entire dump operation is provided when + * it is completed by calling netdev_queue_dump_done(). + * + * The returned contents of 'details' should be documented as valid for the + * given 'type' in the "other_config" column in the "Queue" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * The caller must initialize 'details' (e.g. with smap_init()) before calling + * this function. This function will clear and replace its contents. The + * caller must free 'details' when it is no longer needed (e.g. with + * smap_destroy()). */ +bool +netdev_queue_dump_next(struct netdev_queue_dump *dump, + unsigned int *queue_id, struct smap *details) +{ + const struct netdev *netdev = dump->netdev; -done: - free(fn); - if (stream) { - fclose(stream); + if (dump->error) { + return false; } - ds_destroy(&line); - if (error) { - *vlan_vid = -1; + + dump->error = netdev->netdev_class->queue_dump_next(netdev, dump->state, + queue_id, details); + + if (dump->error) { + netdev->netdev_class->queue_dump_done(netdev, dump->state); + return false; } - return error; + return true; } - -static void restore_all_flags(void *aux); -/* Set up a signal hook to restore network device flags on program - * termination. */ -static void -init_netdev(void) +/* Completes queue table dump operation 'dump', which must have been + * initialized with netdev_queue_dump_start(). Returns 0 if the dump operation + * was error-free, otherwise a positive errno value describing the problem. */ +int +netdev_queue_dump_done(struct netdev_queue_dump *dump) { - static bool inited; - if (!inited) { - int ifindex; - int error; + const struct netdev *netdev = dump->netdev; + if (!dump->error && netdev->netdev_class->queue_dump_done) { + dump->error = netdev->netdev_class->queue_dump_done(netdev, + dump->state); + } + netdev_close(dump->netdev); + return dump->error == EOF ? 0 : dump->error; +} - inited = true; +/* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID, + * its statistics, and the 'aux' specified by the caller. The order of + * iteration is unspecified, but (when successful) each queue is visited + * exactly once. + * + * Calling this function may be more efficient than calling + * netdev_get_queue_stats() for every queue. + * + * 'cb' must not modify or free the statistics passed in. + * + * Returns 0 if successful, otherwise a positive errno value. On error, some + * configured queues may not have been included in the iteration. */ +int +netdev_dump_queue_stats(const struct netdev *netdev, + netdev_dump_queue_stats_cb *cb, void *aux) +{ + const struct netdev_class *class = netdev->netdev_class; + return (class->dump_queue_stats + ? class->dump_queue_stats(netdev, cb, aux) + : EOPNOTSUPP); +} - fatal_signal_add_hook(restore_all_flags, NULL, true); + +/* Returns the class type of 'netdev'. + * + * The caller must not free the returned value. */ +const char * +netdev_get_type(const struct netdev *netdev) +{ + return netdev->netdev_class->type; +} - af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); - if (af_inet_sock < 0) { - ovs_fatal(errno, "socket(AF_INET)"); - } +/* Returns the class associated with 'netdev'. */ +const struct netdev_class * +netdev_get_class(const struct netdev *netdev) +{ + return netdev->netdev_class; +} - error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); - if (error) { - ovs_fatal(error, "socket(AF_NETLINK, NETLINK_ROUTE)"); - } +/* Returns the netdev with 'name' or NULL if there is none. + * + * The caller must free the returned netdev with netdev_close(). */ +struct netdev * +netdev_from_name(const char *name) + OVS_EXCLUDED(netdev_mutex) +{ + struct netdev *netdev; - /* Decide on the netdev_get_stats() implementation to use. Netlink is - * preferable, so if that works, we'll use it. */ - ifindex = do_get_ifindex("lo"); - if (ifindex < 0) { - VLOG_WARN("failed to get ifindex for lo, " - "obtaining netdev stats from proc"); - use_netlink_stats = false; - } else { - struct netdev_stats stats; - error = get_stats_via_netlink(ifindex, &stats); - if (!error) { - VLOG_DBG("obtaining netdev stats via rtnetlink"); - use_netlink_stats = true; - } else { - VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " - "via proc (you are probably running a pre-2.6.19 " - "kernel)", strerror(error)); - use_netlink_stats = false; - } - } + ovs_mutex_lock(&netdev_mutex); + netdev = shash_find_data(&netdev_shash, name); + if (netdev) { + netdev->ref_cnt++; } + ovs_mutex_unlock(&netdev_mutex); + + return netdev; } -/* Restore the network device flags on 'netdev' to those that were active - * before we changed them. Returns 0 if successful, otherwise a positive - * errno value. +/* Fills 'device_list' with devices that match 'netdev_class'. * - * To avoid reentry, the caller must ensure that fatal signals are blocked. */ -static int -restore_flags(struct netdev *netdev) + * The caller is responsible for initializing and destroying 'device_list' and + * must close each device on the list. */ +void +netdev_get_devices(const struct netdev_class *netdev_class, + struct shash *device_list) + OVS_EXCLUDED(netdev_mutex) { - struct ifreq ifr; - int restore_flags; + struct shash_node *node; - /* Get current flags. */ - strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); - COVERAGE_INC(netdev_get_flags); - if (ioctl(netdev->netdev_fd, SIOCGIFFLAGS, &ifr) < 0) { - return errno; - } + ovs_mutex_lock(&netdev_mutex); + SHASH_FOR_EACH (node, &netdev_shash) { + struct netdev *dev = node->data; - /* Restore flags that we might have changed, if necessary. */ - restore_flags = netdev->changed_flags & (IFF_PROMISC | IFF_UP); - if ((ifr.ifr_flags ^ netdev->save_flags) & restore_flags) { - ifr.ifr_flags &= ~restore_flags; - ifr.ifr_flags |= netdev->save_flags & restore_flags; - COVERAGE_INC(netdev_set_flags); - if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) { - return errno; + if (dev->netdev_class == netdev_class) { + dev->ref_cnt++; + shash_add(device_list, node->name, node->data); } } - - return 0; + ovs_mutex_unlock(&netdev_mutex); } -/* Retores all the flags on all network devices that we modified. Called from - * a signal handler, so it does not attempt to report error conditions. */ -static void -restore_all_flags(void *aux UNUSED) +const char * +netdev_get_type_from_name(const char *name) { - struct netdev *netdev; - LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) { - restore_flags(netdev); - } -} - -static int -get_flags(const char *netdev_name, int *flags) -{ - struct ifreq ifr; - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - COVERAGE_INC(netdev_get_flags); - if (ioctl(af_inet_sock, SIOCGIFFLAGS, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; - } - *flags = ifr.ifr_flags; - return 0; + struct netdev *dev = netdev_from_name(name); + const char *type = dev ? netdev_get_type(dev) : NULL; + netdev_close(dev); + return type; } - -static int -set_flags(const char *netdev_name, int flags) -{ - struct ifreq ifr; - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - ifr.ifr_flags = flags; - COVERAGE_INC(netdev_set_flags); - if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; - } - return 0; + +struct netdev * +netdev_rxq_get_netdev(const struct netdev_rxq *rx) +{ + ovs_assert(rx->netdev->ref_cnt > 0); + return rx->netdev; } -static int -do_get_ifindex(const char *netdev_name) +const char * +netdev_rxq_get_name(const struct netdev_rxq *rx) { - struct ifreq ifr; - - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - COVERAGE_INC(netdev_get_ifindex); - if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { - VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", - netdev_name, strerror(errno)); - return -errno; - } - return ifr.ifr_ifindex; + return netdev_get_name(netdev_rxq_get_netdev(rx)); } -static int -get_ifindex(const struct netdev *netdev, int *ifindexp) +static void +restore_all_flags(void *aux OVS_UNUSED) { - *ifindexp = 0; - if (netdev->ifindex < 0) { - int ifindex = do_get_ifindex(netdev->name); - if (ifindex < 0) { - return -ifindex; + struct shash_node *node; + + SHASH_FOR_EACH (node, &netdev_shash) { + struct netdev *netdev = node->data; + const struct netdev_saved_flags *sf; + enum netdev_flags saved_values; + enum netdev_flags saved_flags; + + saved_values = saved_flags = 0; + LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) { + saved_flags |= sf->saved_flags; + saved_values &= ~sf->saved_flags; + saved_values |= sf->saved_flags & sf->saved_values; } - ((struct netdev *) netdev)->ifindex = ifindex; - } - *ifindexp = netdev->ifindex; - return 0; -} + if (saved_flags) { + enum netdev_flags old_flags; -static int -get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN], - int *hwaddr_familyp) -{ - struct ifreq ifr; - - memset(&ifr, 0, sizeof ifr); - strncpy(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; - } - if (hwaddr_familyp) { - int hwaddr_family = ifr.ifr_hwaddr.sa_family; - *hwaddr_familyp = hwaddr_family; - if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { - VLOG_WARN("%s device has unknown hardware address family %d", - netdev_name, hwaddr_family); + netdev->netdev_class->update_flags(netdev, + saved_flags & saved_values, + saved_flags & ~saved_values, + &old_flags); } } - memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); - return 0; } -static int -set_etheraddr(const char *netdev_name, int hwaddr_family, - const uint8_t mac[ETH_ADDR_LEN]) -{ - struct ifreq ifr; - - memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - ifr.ifr_hwaddr.sa_family = hwaddr_family; - memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); - COVERAGE_INC(netdev_set_hwaddr); - if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; - } - return 0; +uint64_t +netdev_get_change_seq(const struct netdev *netdev) +{ + return netdev->change_seq; }