X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=e86a160c8e17da10d506ab277950e75e0faf1448;hb=79c9f2ee7883b52860c76c3730725f5731402874;hp=0dff5082bfe9a5e8e1d1afcf4a09f9944e4934a3;hpb=a740f0de5bec1e62dfa64173e6925e7872c66560;p=openvswitch diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 0dff5082..e86a160c 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009 Nicira Networks. + * Copyright (c) 2009, 2010 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -50,6 +50,7 @@ #include "netlink.h" #include "ofpbuf.h" #include "openflow/openflow.h" +#include "openvswitch/internal_dev.h" #include "openvswitch/gre.h" #include "packets.h" #include "poll-loop.h" @@ -58,10 +59,6 @@ #include "shash.h" #include "svec.h" -#ifndef GRE_IOCTL_ONLY -#include -#endif - #define THIS_MODULE VLM_netdev_linux #include "vlog.h" @@ -74,30 +71,8 @@ #define ADVERTISED_Asym_Pause (1 << 14) #endif -struct tap_state { - int tap_fd; /* File descriptor for TAP device. */ -}; - -/* Provider-specific netdev object. Netdev objects are devices that are - * created by the netdev library through a netdev_create() call. */ -struct netdev_obj_linux { - struct netdev_obj netdev_obj; - union { - struct tap_state tap; - } state; -}; - -struct netdev_linux { - struct netdev netdev; - - /* 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. */ - - struct netdev_linux_cache *cache; -}; +static struct rtnetlink_notifier netdev_linux_cache_notifier; +static int cache_notifier_refcount; enum { VALID_IFINDEX = 1 << 0, @@ -106,49 +81,48 @@ enum { VALID_IN6 = 1 << 3, VALID_MTU = 1 << 4, VALID_CARRIER = 1 << 5, - VALID_IS_INTERNAL = 1 << 6 + VALID_IS_PSEUDO = 1 << 6 /* Represents is_internal and is_tap. */ }; -/* Cached network device information. */ -struct netdev_linux_cache { +struct tap_state { + int fd; +}; + +struct patch_state { + char *peer; +}; + +struct netdev_dev_linux { + struct netdev_dev netdev_dev; + struct shash_node *shash_node; - unsigned int valid; - int ref_cnt; + unsigned int cache_valid; + /* The following are figured out "on demand" only. They are only valid + * when the corresponding VALID_* bit in 'cache_valid' is set. */ int ifindex; uint8_t etheraddr[ETH_ADDR_LEN]; struct in_addr address, netmask; struct in6_addr in6; int mtu; int carrier; - bool is_internal; + bool is_internal; /* Is this an openvswitch internal device? */ + bool is_tap; /* Is this a tuntap device? */ + + union { + struct tap_state tap; + struct patch_state patch; + } state; }; -static struct shash cache_map = SHASH_INITIALIZER(&cache_map); -static struct rtnetlink_notifier netdev_linux_cache_notifier; +struct netdev_linux { + struct netdev netdev; + int fd; +}; /* An AF_INET socket (used for ioctl operations). */ static int af_inet_sock = -1; -struct gre_config { - uint32_t local_ip; - uint32_t remote_ip; - uint32_t in_key; - uint32_t out_key; - bool have_in_key; - bool have_out_key; - bool in_csum; - bool out_csum; -}; - -static struct { - union { - struct nl_sock *nl_sock; - int ioctl_fd; - }; - bool use_ioctl; -} gre_descriptors; - struct netdev_linux_notifier { struct netdev_notifier notifier; struct list node; @@ -162,10 +136,12 @@ static struct rtnetlink_notifier netdev_linux_poll_notifier; * additional log messages. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); -static int netdev_linux_do_ethtool(struct netdev *, struct ethtool_cmd *, +static int netdev_linux_init(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 struct netdev *, struct ifreq *, - 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); @@ -181,20 +157,28 @@ static int set_etheraddr(const char *netdev_name, int hwaddr_family, 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 struct netdev_obj_linux * -netdev_obj_linux_cast(const struct netdev_obj *netdev_obj) +static bool +is_netdev_linux_class(const struct netdev_class *netdev_class) { - /* xxx Gross hack! Until we get the class 'type' issue resolved. */ -#if 0 - netdev_obj_assert_class(netdev_obj, &netdev_linux_class); -#endif - return CONTAINER_OF(netdev_obj, struct netdev_obj_linux, netdev_obj); + return netdev_class->init == netdev_linux_init; +} + +static struct netdev_dev_linux * +netdev_dev_linux_cast(const struct netdev_dev *netdev_dev) +{ + 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); } static struct netdev_linux * netdev_linux_cast(const struct netdev *netdev) { - netdev_assert_class(netdev, &netdev_linux_class); + 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)); + return CONTAINER_OF(netdev, struct netdev_linux, netdev); } @@ -226,169 +210,77 @@ netdev_linux_wait(void) static void netdev_linux_cache_cb(const struct rtnetlink_change *change, - void *aux UNUSED) + void *aux OVS_UNUSED) { - struct netdev_linux_cache *cache; + struct netdev_dev_linux *dev; if (change) { - cache = shash_find_data(&cache_map, change->ifname); - if (cache) { - cache->valid = 0; + 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; + } } } else { + struct shash device_shash; struct shash_node *node; - SHASH_FOR_EACH (node, &cache_map) { - cache = node->data; - cache->valid = 0; + + 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; } + shash_destroy(&device_shash); } } -/* The arguments are marked as unused to prevent warnings on platforms where - * the Netlink interface isn't supported. */ static int -setup_gre_netlink(const char *name UNUSED, struct gre_config *config UNUSED, - bool create UNUSED) +if_up(const char *name) { -#ifdef GRE_IOCTL_ONLY - return EOPNOTSUPP; -#else - int error; - struct ofpbuf request, *reply; - unsigned int nl_flags; - struct ifinfomsg ifinfomsg; - struct nlattr *linkinfo_hdr; - struct nlattr *info_data_hdr; - uint16_t iflags = 0; - uint16_t oflags = 0; - uint8_t pmtudisc = 0; - - if (!gre_descriptors.nl_sock) { - error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, - &gre_descriptors.nl_sock); - if (error) { - VLOG_WARN("couldn't create netlink socket: %s\n", strerror(error)); - gre_descriptors.nl_sock = NULL; - goto error; - } - } - - ofpbuf_init(&request, 0); - - nl_flags = NLM_F_REQUEST; - if (create) { - nl_flags |= NLM_F_CREATE|NLM_F_EXCL; - } - - /* We over-reserve space, because we do some pointer arithmetic - * and don't want the buffer address shifting under us. */ - nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 2048, RTM_NEWLINK, - nl_flags); - - memset(&ifinfomsg, 0, sizeof ifinfomsg); - ifinfomsg.ifi_family = AF_UNSPEC; - nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg); - - linkinfo_hdr = ofpbuf_tail(&request); - nl_msg_put_unspec(&request, IFLA_LINKINFO, NULL, 0); - - nl_msg_put_unspec(&request, IFLA_INFO_KIND, "gretap", 6); - - info_data_hdr = ofpbuf_tail(&request); - nl_msg_put_unspec(&request, IFLA_INFO_DATA, NULL, 0); - - /* Set flags */ - if (config->have_in_key) { - iflags |= GRE_KEY; - } - if (config->have_out_key) { - oflags |= GRE_KEY; - } - - if (config->in_csum) { - iflags |= GRE_CSUM; - } - if (config->out_csum) { - oflags |= GRE_CSUM; - } - - /* Add options */ - nl_msg_put_u32(&request, IFLA_GRE_IKEY, config->in_key); - nl_msg_put_u32(&request, IFLA_GRE_OKEY, config->out_key); - nl_msg_put_u16(&request, IFLA_GRE_IFLAGS, iflags); - nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags); - nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip); - nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip); - nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, pmtudisc); - nl_msg_put_u8(&request, IFLA_GRE_TTL, 0); - nl_msg_put_u8(&request, IFLA_GRE_TOS, 0); - - info_data_hdr->nla_len = (char *)ofpbuf_tail(&request) - - (char *)info_data_hdr; - linkinfo_hdr->nla_len = (char *)ofpbuf_tail(&request) - - (char *)linkinfo_hdr; + struct ifreq ifr; - nl_msg_put_string(&request, IFLA_IFNAME, name); + strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); + ifr.ifr_flags = IFF_UP; - error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply); - ofpbuf_uninit(&request); - if (error) { - VLOG_WARN("couldn't transact netlink socket: %s\n", strerror(error)); - goto error; + if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) { + VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s", + name, strerror(errno)); + return errno; } - ofpbuf_delete(reply); -error: - return error; -#endif + return 0; } +/* A veth may be created using the 'command' "+,". A veth may + * be destroyed by using the 'command' "-", where can be + * either side of the device. + */ static int -setup_gre_ioctl(const char *name, struct gre_config *config, bool create) +modify_veth(const char *format, ...) { - struct ip_tunnel_parm p; - struct ifreq ifr; - - memset(&p, 0, sizeof p); - - strncpy(p.name, name, IFNAMSIZ); - - p.iph.version = 4; - p.iph.ihl = 5; - p.iph.protocol = IPPROTO_GRE; - p.iph.saddr = config->local_ip; - p.iph.daddr = config->remote_ip; - - if (config->have_in_key) { - p.i_flags |= GRE_KEY; - p.i_key = config->in_key; - } - if (config->have_out_key) { - p.o_flags |= GRE_KEY; - p.o_key = config->out_key; - } + FILE *veth_file; + va_list args; + int retval; - if (config->in_csum) { - p.i_flags |= GRE_CSUM; + veth_file = fopen("/sys/class/net/veth_pairs", "w"); + if (!veth_file) { + VLOG_WARN_RL(&rl, "could not open veth device. Are you running a " + "supported XenServer with the kernel module loaded?"); + return ENODEV; } - if (config->out_csum) { - p.o_flags |= GRE_CSUM; - } - - strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ); - ifr.ifr_ifru.ifru_data = (void *)&p; + setvbuf(veth_file, NULL, _IONBF, 0); - if (!gre_descriptors.ioctl_fd) { - gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0); - if (gre_descriptors.ioctl_fd < 0) { - VLOG_WARN("couldn't create gre ioctl socket: %s\n", strerror(errno)); - gre_descriptors.ioctl_fd = 0; - return errno; - } - } + va_start(args, format); + retval = vfprintf(veth_file, format, args); + va_end(args); - if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP, - &ifr) < 0) { - VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno)); + fclose(veth_file); + if (retval < 0) { + VLOG_WARN_RL(&rl, "could not destroy patch: %s", strerror(errno)); return errno; } @@ -396,128 +288,127 @@ setup_gre_ioctl(const char *name, struct gre_config *config, bool create) } static int -setup_gre(const char *name, const struct shash *args, bool create) +create_patch(const char *name, const char *peer) { - int error; - struct in_addr in_addr; - struct shash_node *node; - struct gre_config config; - - memset(&config, 0, sizeof config); - config.in_csum = true; - config.out_csum = true; - - SHASH_FOR_EACH (node, args) { - if (!strcmp(node->name, "remote_ip")) { - if (lookup_ip(node->data, &in_addr)) { - VLOG_WARN("bad 'remote_ip' for gre device %s ", name); - } else { - config.remote_ip = in_addr.s_addr; - } - } else if (!strcmp(node->name, "local_ip")) { - if (lookup_ip(node->data, &in_addr)) { - VLOG_WARN("bad 'local_ip' for gre device %s ", name); + int retval; + struct netdev_dev *peer_nd; + + + /* Only create the veth if the peer didn't already do it. */ + peer_nd = netdev_dev_from_name(peer); + if (peer_nd) { + if (!strcmp("patch", netdev_dev_get_type(peer_nd))) { + struct netdev_dev_linux *ndl = netdev_dev_linux_cast(peer_nd); + if (!strcmp(name, ndl->state.patch.peer)) { + return 0; } else { - config.local_ip = in_addr.s_addr; - } - } else if (!strcmp(node->name, "key")) { - config.have_in_key = true; - config.have_out_key = true; - config.in_key = htonl(atoi(node->data)); - config.out_key = htonl(atoi(node->data)); - } else if (!strcmp(node->name, "in_key")) { - config.have_in_key = true; - config.in_key = htonl(atoi(node->data)); - } else if (!strcmp(node->name, "out_key")) { - config.have_out_key = true; - config.out_key = htonl(atoi(node->data)); - } else if (!strcmp(node->name, "csum")) { - if (!strcmp(node->data, "false")) { - config.in_csum = false; - config.out_csum = false; + VLOG_WARN_RL(&rl, "peer '%s' already paired with '%s'", + peer, ndl->state.patch.peer); + return EINVAL; } } else { - VLOG_WARN("unknown gre argument '%s'", node->name); + VLOG_WARN_RL(&rl, "peer '%s' exists and is not a patch", peer); + return EINVAL; } } - if (!config.remote_ip) { - VLOG_WARN("gre type requires valid 'remote_ip' argument"); - error = EINVAL; - goto error; + retval = modify_veth("+%s,%s", name, peer); + if (retval) { + return retval; } - if (!gre_descriptors.use_ioctl) { - error = setup_gre_netlink(name, &config, create); - if (error == EOPNOTSUPP) { - gre_descriptors.use_ioctl = true; - } + retval = if_up(name); + if (retval) { + return retval; } - if (gre_descriptors.use_ioctl) { - error = setup_gre_ioctl(name, &config, create); + + retval = if_up(peer); + if (retval) { + return retval; } -error: - return error; + return 0; } static int -if_up(const char *name) +setup_patch(const char *name, const struct shash *args, char **peer_) { - struct ifreq ifr; + const char *peer; - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - ifr.ifr_flags = IFF_UP; + peer = shash_find_data(args, "peer"); + if (!peer) { + VLOG_WARN("patch type requires valid 'peer' argument"); + return EINVAL; + } - if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) { - VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s", - name, strerror(errno)); - return errno; + if (shash_count(args) > 1) { + VLOG_WARN("patch type takes only a 'peer' argument"); + return EINVAL; } - return 0; + if (strlen(peer) >= IFNAMSIZ) { + VLOG_WARN_RL(&rl, "patch 'peer' arg too long"); + return EINVAL; + } + + *peer_ = xstrdup(peer); + return create_patch(name, peer); } -/* Creates the netdev object of 'type' with 'name'. */ +/* Creates the netdev device of 'type' with 'name'. */ static int -netdev_linux_create_system(const char *name, const char *type UNUSED, - const struct shash *args, bool created) +netdev_linux_create_system(const char *name, const char *type OVS_UNUSED, + const struct shash *args, struct netdev_dev **netdev_devp) { - struct netdev_obj_linux *netdev_obj; + struct netdev_dev_linux *netdev_dev; + int error; if (!shash_is_empty(args)) { - VLOG_WARN("arguments for system devices should be empty"); + VLOG_WARN("%s: arguments for system devices should be empty", name); + } + + if (!cache_notifier_refcount) { + error = rtnetlink_notifier_register(&netdev_linux_cache_notifier, + netdev_linux_cache_cb, NULL); + if (error) { + return error; + } } + cache_notifier_refcount++; - netdev_obj = xcalloc(1, sizeof *netdev_obj); - netdev_obj_init(&netdev_obj->netdev_obj, name, &netdev_linux_class, - created); + 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; } +/* For most types of netdevs we open the device for each call of + * netdev_open(). However, this is not the case with tap devices, + * since it is only possible to open the device once. In this + * situation we share a single file descriptor, and consequently + * 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 UNUSED, - const struct shash *args, bool created) +netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED, + const struct shash *args, struct netdev_dev **netdev_devp) { - struct netdev_obj_linux *netdev_obj; + struct netdev_dev_linux *netdev_dev; struct tap_state *state; static const char tap_dev[] = "/dev/net/tun"; struct ifreq ifr; int error; if (!shash_is_empty(args)) { - VLOG_WARN("arguments for TAP devices should be empty"); + VLOG_WARN("%s: arguments for TAP devices should be empty", name); } - netdev_obj = xcalloc(1, sizeof *netdev_obj); - netdev_obj_init(&netdev_obj->netdev_obj, name, &netdev_tap_class, created); - - state = &netdev_obj->state.tap; + netdev_dev = xzalloc(sizeof *netdev_dev); + state = &netdev_dev->state.tap; /* Open tap device. */ - state->tap_fd = open(tap_dev, O_RDWR); - if (state->tap_fd < 0) { + state->fd = open(tap_dev, O_RDWR); + if (state->fd < 0) { error = errno; VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); goto error; @@ -526,7 +417,7 @@ netdev_linux_create_tap(const char *name, const char *type UNUSED, /* Create tap device. */ ifr.ifr_flags = IFF_TAP | IFF_NO_PI; strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - if (ioctl(state->tap_fd, TUNSETIFF, &ifr) == -1) { + if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { VLOG_WARN("%s: creating tap device failed: %s", name, strerror(errno)); error = errno; @@ -534,215 +425,109 @@ netdev_linux_create_tap(const char *name, const char *type UNUSED, } /* Make non-blocking. */ - error = set_nonblocking(state->tap_fd); - if (error) { - goto error; - } - - error = if_up(name); + error = set_nonblocking(state->fd); if (error) { goto error; } + netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class); + *netdev_devp = &netdev_dev->netdev_dev; return 0; error: - netdev_destroy(name); + free(netdev_dev); return error; } static int -netdev_linux_create_gre(const char *name, const char *type UNUSED, - const struct shash *args, bool created) +netdev_linux_create_patch(const char *name, const char *type OVS_UNUSED, + const struct shash *args, struct netdev_dev **netdev_devp) { - struct netdev_obj_linux *netdev_obj; + struct netdev_dev_linux *netdev_dev; + char *peer = NULL; int error; - netdev_obj = xcalloc(1, sizeof *netdev_obj); - netdev_obj_init(&netdev_obj->netdev_obj, name, &netdev_gre_class, created); - - error = setup_gre(name, args, true); + error = setup_patch(name, args, &peer); if (error) { - goto error; + free(peer); + return error; } - error = if_up(name); - if (error) { - goto error; - } + netdev_dev = xzalloc(sizeof *netdev_dev); + netdev_dev->state.patch.peer = peer; + netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_patch_class); + *netdev_devp = &netdev_dev->netdev_dev; return 0; - -error: - netdev_destroy(name); - return error; -} - -static int -netdev_linux_reconfigure_gre(struct netdev_obj *netdev_obj_, - const struct shash *args) -{ - struct netdev_obj_linux *netdev_obj = netdev_obj_linux_cast(netdev_obj_); - const char *name = netdev_obj_get_name(&netdev_obj->netdev_obj); - - return setup_gre(name, args, false); } -/* The arguments are marked as unused to prevent warnings on platforms where - * the Netlink interface isn't supported. */ -static int -destroy_gre_netlink(struct netdev_obj_linux *netdev_obj UNUSED) +static void +destroy_tap(struct netdev_dev_linux *netdev_dev) { -#ifdef GRE_IOCTL_ONLY - return EOPNOTSUPP; -#else - const char *name = netdev_obj_get_name(&netdev_obj->netdev_obj); - int error; - struct ofpbuf request, *reply; - struct ifinfomsg ifinfomsg; - int ifindex; - - ofpbuf_init(&request, 0); - - nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK, - NLM_F_REQUEST); - - memset(&ifinfomsg, 0, sizeof ifinfomsg); - ifinfomsg.ifi_family = AF_UNSPEC; - nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg); - - ifindex = do_get_ifindex(name); - nl_msg_put_u32(&request, IFLA_LINK, ifindex); + struct tap_state *state = &netdev_dev->state.tap; - nl_msg_put_string(&request, IFLA_IFNAME, name); - - error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply); - ofpbuf_uninit(&request); - if (error) { - VLOG_WARN("couldn't transact netlink socket: %s\n", strerror(error)); - goto error; + if (state->fd >= 0) { + close(state->fd); } - ofpbuf_delete(reply); - -error: - return 0; -#endif } -static int -destroy_gre_ioctl(struct netdev_obj_linux *netdev_obj) +static void +destroy_patch(struct netdev_dev_linux *netdev_dev) { - const char *name = netdev_obj_get_name(&netdev_obj->netdev_obj); - struct ip_tunnel_parm p; - struct ifreq ifr; - - memset(&p, 0, sizeof p); - strncpy(p.name, name, IFNAMSIZ); + const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev); + struct patch_state *state = &netdev_dev->state.patch; - strncpy(ifr.ifr_name, name, IFNAMSIZ); - ifr.ifr_ifru.ifru_data = (void *)&p; - - if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) { - VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno)); - return errno; + /* Only destroy veth if 'peer' doesn't exist as an existing netdev. */ + if (!netdev_dev_from_name(state->peer)) { + modify_veth("-%s", name); } - - return 0; + free(state->peer); } +/* Destroys the netdev device 'netdev_dev_'. */ static void -destroy_tap(struct netdev_obj_linux *netdev_obj) +netdev_linux_destroy(struct netdev_dev *netdev_dev_) { - struct tap_state *state = &netdev_obj->state.tap; - if (state->tap_fd >= 0) { - close(state->tap_fd); - } -} + struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); + const char *type = netdev_dev_get_type(netdev_dev_); -/* Destroys the netdev object 'netdev_obj_'. */ -static void -netdev_linux_destroy(struct netdev_obj *netdev_obj_) -{ - struct netdev_obj_linux *netdev_obj = netdev_obj_linux_cast(netdev_obj_); - const char *type = netdev_obj_get_type(netdev_obj_); + if (!strcmp(type, "system")) { + cache_notifier_refcount--; - if (!strcmp(type, "tap")) { - destroy_tap(netdev_obj); - } else if (!strcmp(type, "gre")) { - if (gre_descriptors.use_ioctl) { - destroy_gre_ioctl(netdev_obj); - } else { - destroy_gre_netlink(netdev_obj); + if (!cache_notifier_refcount) { + rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); } + } else if (!strcmp(type, "tap")) { + destroy_tap(netdev_dev); + } else if (!strcmp(type, "patch")) { + destroy_patch(netdev_dev); } - free(netdev_obj); + + free(netdev_dev); } static int -netdev_linux_open(const char *name, int ethertype, struct netdev **netdevp) +netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, + struct netdev **netdevp) { + struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); struct netdev_linux *netdev; enum netdev_flags flags; int error; /* Allocate network device. */ netdev = xzalloc(sizeof *netdev); - netdev_init(&netdev->netdev, name, &netdev_linux_class); - netdev->netdev_fd = -1; - netdev->tap_fd = -1; - netdev->cache = shash_find_data(&cache_map, name); - if (!netdev->cache) { - if (shash_is_empty(&cache_map)) { - int error = rtnetlink_notifier_register( - &netdev_linux_cache_notifier, netdev_linux_cache_cb, NULL); - if (error) { - netdev_close(&netdev->netdev); - return error; - } - } - netdev->cache = xmalloc(sizeof *netdev->cache); - netdev->cache->shash_node = shash_add(&cache_map, name, - netdev->cache); - netdev->cache->valid = 0; - netdev->cache->ref_cnt = 0; - } - netdev->cache->ref_cnt++; - - if (!strcmp(netdev_get_type(&netdev->netdev), "tap")) { - static const char tap_dev[] = "/dev/net/tun"; - struct ifreq ifr; - - /* Open tap device. */ - 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; - } - - /* Create tap device. */ - ifr.ifr_flags = IFF_TAP | IFF_NO_PI; - strncpy(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)); - error = errno; - goto error; - } - - /* Make non-blocking. */ - error = set_nonblocking(netdev->tap_fd); - if (error) { - goto error; - } - } + netdev->fd = -1; + netdev_init(&netdev->netdev, netdev_dev_); error = netdev_get_flags(&netdev->netdev, &flags); if (error == ENODEV) { goto error; } - if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) { + if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) { + netdev->fd = netdev_dev->state.tap.fd; + } else if (ethertype != NETDEV_ETH_TYPE_NONE) { struct sockaddr_ll sll; int protocol; int ifindex; @@ -751,17 +536,14 @@ netdev_linux_open(const char *name, int ethertype, struct netdev **netdevp) protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 : ethertype); - netdev->netdev_fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); - if (netdev->netdev_fd < 0) { + netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); + if (netdev->fd < 0) { error = errno; goto error; } - if (netdev->tap_fd < 0) { - netdev->tap_fd = netdev->netdev_fd; - } /* Set non-blocking mode. */ - error = set_nonblocking(netdev->netdev_fd); + error = set_nonblocking(netdev->fd); if (error) { goto error; } @@ -776,10 +558,11 @@ netdev_linux_open(const char *name, int ethertype, struct netdev **netdevp) memset(&sll, 0, sizeof sll); sll.sll_family = AF_PACKET; sll.sll_ifindex = ifindex; - if (bind(netdev->netdev_fd, + if (bind(netdev->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { error = errno; - VLOG_ERR("bind to %s failed: %s", name, strerror(error)); + VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), + strerror(error)); goto error; } @@ -787,7 +570,7 @@ netdev_linux_open(const char *name, int ethertype, struct netdev **netdevp) * 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->netdev_fd); + error = drain_rcvbuf(netdev->fd); if (error) { goto error; } @@ -797,7 +580,7 @@ netdev_linux_open(const char *name, int ethertype, struct netdev **netdevp) return 0; error: - netdev_close(&netdev->netdev); + netdev_uninit(&netdev->netdev, true); return error; } @@ -807,19 +590,8 @@ netdev_linux_close(struct netdev *netdev_) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->cache && !--netdev->cache->ref_cnt) { - shash_delete(&cache_map, netdev->cache->shash_node); - free(netdev->cache); - - if (shash_is_empty(&cache_map)) { - rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); - } - } - if (netdev->netdev_fd >= 0) { - close(netdev->netdev_fd); - } - if (netdev->tap_fd >= 0 && netdev->netdev_fd != netdev->tap_fd) { - close(netdev->tap_fd); + if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) { + close(netdev->fd); } free(netdev); } @@ -851,13 +623,13 @@ netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->tap_fd < 0) { + if (netdev->fd < 0) { /* Device was opened with NETDEV_ETH_TYPE_NONE. */ return -EAGAIN; } for (;;) { - ssize_t retval = read(netdev->tap_fd, data, size); + ssize_t retval = read(netdev->fd, data, size); if (retval >= 0) { return retval; } else if (errno != EINTR) { @@ -876,8 +648,8 @@ static void netdev_linux_recv_wait(struct netdev *netdev_) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->tap_fd >= 0) { - poll_fd_wait(netdev->tap_fd, POLLIN); + if (netdev->fd >= 0) { + poll_fd_wait(netdev->fd, POLLIN); } } @@ -886,19 +658,19 @@ static int netdev_linux_drain(struct netdev *netdev_) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) { + if (netdev->fd < 0) { return 0; - } else if (netdev->tap_fd != netdev->netdev_fd) { + } else if (!strcmp(netdev_get_type(netdev_), "tap")) { struct ifreq ifr; - int error = netdev_linux_do_ioctl(netdev_, &ifr, + int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); if (error) { return error; } - drain_fd(netdev->tap_fd, ifr.ifr_qlen); + drain_fd(netdev->fd, ifr.ifr_qlen); return 0; } else { - return drain_rcvbuf(netdev->netdev_fd); + return drain_rcvbuf(netdev->fd); } } @@ -918,12 +690,12 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. */ - if (netdev->tap_fd < 0) { + if (netdev->fd < 0) { return EPIPE; } for (;;) { - ssize_t retval = write(netdev->tap_fd, data, size); + ssize_t retval = write(netdev->fd, data, size); if (retval < 0) { /* The Linux AF_PACKET implementation never blocks waiting for room * for packets, instead returning ENOBUFS. Translate this into @@ -958,10 +730,10 @@ static void netdev_linux_send_wait(struct netdev *netdev_) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) { + if (netdev->fd < 0) { /* Nothing to do. */ - } else if (netdev->tap_fd == netdev->netdev_fd) { - poll_fd_wait(netdev->tap_fd, POLLOUT); + } else if (strcmp(netdev_get_type(netdev_), "tap")) { + poll_fd_wait(netdev->fd, POLLOUT); } else { /* TAP device always accepts packets.*/ poll_immediate_wake(); @@ -974,15 +746,16 @@ static int netdev_linux_set_etheraddr(struct netdev *netdev_, const uint8_t mac[ETH_ADDR_LEN]) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); int error; - if (!(netdev->cache->valid & VALID_ETHERADDR) - || !eth_addr_equals(netdev->cache->etheraddr, mac)) { + 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->cache->valid |= VALID_ETHERADDR; - memcpy(netdev->cache->etheraddr, mac, ETH_ADDR_LEN); + netdev_dev->cache_valid |= VALID_ETHERADDR; + memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); } } else { error = 0; @@ -996,16 +769,17 @@ static int netdev_linux_get_etheraddr(const struct netdev *netdev_, uint8_t mac[ETH_ADDR_LEN]) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (!(netdev->cache->valid & VALID_ETHERADDR)) { + 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->cache->etheraddr); + netdev_dev->etheraddr); if (error) { return error; } - netdev->cache->valid |= VALID_ETHERADDR; + netdev_dev->cache_valid |= VALID_ETHERADDR; } - memcpy(mac, netdev->cache->etheraddr, ETH_ADDR_LEN); + memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); return 0; } @@ -1015,19 +789,21 @@ netdev_linux_get_etheraddr(const struct netdev *netdev_, static int netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (!(netdev->cache->valid & VALID_MTU)) { + 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; - error = netdev_linux_do_ioctl(netdev_, &ifr, SIOCGIFMTU, "SIOCGIFMTU"); + error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, + SIOCGIFMTU, "SIOCGIFMTU"); if (error) { return error; } - netdev->cache->mtu = ifr.ifr_mtu; - netdev->cache->valid |= VALID_MTU; + netdev_dev->mtu = ifr.ifr_mtu; + netdev_dev->cache_valid |= VALID_MTU; } - *mtup = netdev->cache->mtu; + *mtup = netdev_dev->mtu; return 0; } @@ -1045,16 +821,18 @@ netdev_linux_get_ifindex(const struct netdev *netdev) static int netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); int error = 0; char *fn = NULL; int fd = -1; - if (!(netdev->cache->valid & VALID_CARRIER)) { + if (!(netdev_dev->cache_valid & VALID_CARRIER)) { char line[8]; int retval; - fn = xasprintf("/sys/class/net/%s/carrier", netdev_get_name(netdev_)); + fn = xasprintf("/sys/class/net/%s/carrier", + netdev_get_name(netdev_)); fd = open(fn, O_RDONLY); if (fd < 0) { error = errno; @@ -1084,10 +862,10 @@ netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) fn, line[0]); goto exit; } - netdev->cache->carrier = line[0] != '0'; - netdev->cache->valid |= VALID_CARRIER; + netdev_dev->carrier = line[0] != '0'; + netdev_dev->cache_valid |= VALID_CARRIER; } - *carrier = netdev->cache->carrier; + *carrier = netdev_dev->carrier; error = 0; exit: @@ -1126,14 +904,45 @@ check_for_working_netlink_stats(void) } } +/* 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_dev->cache_valid |= VALID_IS_PSEUDO; + } +} + /* Retrieves current device stats for 'netdev'. * * 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. */ static int -netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) +netdev_linux_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); static int use_netlink_stats = -1; int error; struct netdev_stats raw_stats; @@ -1141,28 +950,8 @@ netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) COVERAGE_INC(netdev_get_stats); - if (!(netdev->cache->valid & VALID_IS_INTERNAL)) { - netdev->cache->is_internal = (netdev->tap_fd != -1); - - if (!netdev->cache->is_internal) { - struct ethtool_drvinfo drvinfo; - - memset(&drvinfo, 0, sizeof drvinfo); - error = netdev_linux_do_ethtool(&netdev->netdev, - (struct ethtool_cmd *)&drvinfo, - ETHTOOL_GDRVINFO, - "ETHTOOL_GDRVINFO"); - - if (!error) { - netdev->cache->is_internal = !strcmp(drvinfo.driver, - "openvswitch"); - } - } - - netdev->cache->valid |= VALID_IS_INTERNAL; - } - - if (netdev->cache->is_internal) { + netdev_linux_update_is_pseudo(netdev_dev); + if (netdev_dev->is_internal) { collect_stats = &raw_stats; } @@ -1172,19 +961,19 @@ netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) if (use_netlink_stats) { int ifindex; - error = get_ifindex(&netdev->netdev, &ifindex); + error = get_ifindex(netdev_, &ifindex); if (!error) { error = get_stats_via_netlink(ifindex, collect_stats); } } else { - error = get_stats_via_proc(netdev->netdev.name, collect_stats); + error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats); } /* 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. */ - if (netdev->cache->is_internal) { + if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) { stats->rx_packets = raw_stats.tx_packets; stats->tx_packets = raw_stats.rx_packets; stats->rx_bytes = raw_stats.tx_bytes; @@ -1211,6 +1000,41 @@ netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) return error; } +static int +netdev_linux_set_stats(struct netdev *netdev, + const 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; + + /* 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; + } + + /* 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"); +} + /* 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 @@ -1224,7 +1048,7 @@ netdev_linux_get_features(struct netdev *netdev, int error; memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev, &ecmd, + error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { return error; @@ -1345,7 +1169,7 @@ netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) int error; memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev, &ecmd, + error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { return error; @@ -1388,7 +1212,7 @@ netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) if (advertise & OFPPF_PAUSE_ASYM) { ecmd.advertising |= ADVERTISED_Asym_Pause; } - return netdev_linux_do_ethtool(netdev, &ecmd, + return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, ETHTOOL_SSET, "ETHTOOL_SSET"); } @@ -1466,8 +1290,8 @@ netdev_linux_set_policing(struct netdev *netdev, COVERAGE_INC(netdev_set_policing); if (kbits_rate) { if (!kbits_burst) { - /* Default to 10 kilobits if not specified. */ - kbits_burst = 10; + /* Default to 1000 kilobits if not specified. */ + kbits_burst = 1000; } /* xxx This should be more careful about only adding if it @@ -1504,26 +1328,28 @@ static int netdev_linux_get_in4(const struct netdev *netdev_, struct in_addr *address, struct in_addr *netmask) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (!(netdev->cache->valid & VALID_IN4)) { + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); + + if (!(netdev_dev->cache_valid & VALID_IN4)) { int error; - error = netdev_linux_get_ipv4(netdev_, &netdev->cache->address, + error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address, SIOCGIFADDR, "SIOCGIFADDR"); if (error) { return error; } - error = netdev_linux_get_ipv4(netdev_, &netdev->cache->netmask, + error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask, SIOCGIFNETMASK, "SIOCGIFNETMASK"); if (error) { return error; } - netdev->cache->valid |= VALID_IN4; + netdev_dev->cache_valid |= VALID_IN4; } - *address = netdev->cache->address; - *netmask = netdev->cache->netmask; + *address = netdev_dev->address; + *netmask = netdev_dev->netmask; return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; } @@ -1531,14 +1357,15 @@ static int netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, struct in_addr netmask) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); int error; error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); if (!error) { - netdev->cache->valid |= VALID_IN4; - netdev->cache->address = address; - netdev->cache->netmask = netmask; + netdev_dev->cache_valid |= VALID_IN4; + netdev_dev->address = address; + netdev_dev->netmask = netmask; if (address.s_addr != INADDR_ANY) { error = do_set_addr(netdev_, SIOCSIFNETMASK, "SIOCSIFNETMASK", netmask); @@ -1568,12 +1395,13 @@ parse_if_inet6_line(const char *line, static int netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (!(netdev->cache->valid & VALID_IN6)) { + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); + if (!(netdev_dev->cache_valid & VALID_IN6)) { FILE *file; char line[128]; - netdev->cache->in6 = in6addr_any; + netdev_dev->in6 = in6addr_any; file = fopen("/proc/net/if_inet6", "r"); if (file != NULL) { @@ -1584,15 +1412,15 @@ netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) if (parse_if_inet6_line(line, &in6, ifname) && !strcmp(name, ifname)) { - netdev->cache->in6 = in6; + netdev_dev->in6 = in6; break; } } fclose(file); } - netdev->cache->valid |= VALID_IN6; + netdev_dev->cache_valid |= VALID_IN6; } - *in6 = netdev->cache->in6; + *in6 = netdev_dev->in6; return 0; } @@ -1614,14 +1442,16 @@ 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->name, sizeof ifr.ifr_name); + 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, &ifr, ioctl_nr, ioctl_name); + + return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, + ioctl_name); } /* Adds 'router' as a default IP gateway. */ static int -netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router) +netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) { struct in_addr any = { INADDR_ANY }; struct rtentry rt; @@ -1710,24 +1540,24 @@ netdev_linux_arp_lookup(const struct netdev *netdev, uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) { struct arpreq r; - struct sockaddr_in *pa; + struct sockaddr_in sin; int retval; 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; + 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->name, sizeof r.arp_dev); + strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); COVERAGE_INC(netdev_arp_lookup); retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; if (!retval) { 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_get_name(netdev), IP_ARGS(&ip), strerror(retval)); } return retval; } @@ -1788,7 +1618,7 @@ poll_notify(struct list *list) static void netdev_linux_poll_cb(const struct rtnetlink_change *change, - void *aux UNUSED) + void *aux OVS_UNUSED) { if (change) { struct list *list = shash_find_data(&netdev_linux_notifiers, @@ -1860,7 +1690,7 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) } const struct netdev_class netdev_linux_class = { - "system", /* type */ + "system", netdev_linux_init, netdev_linux_run, @@ -1888,6 +1718,7 @@ const struct netdev_class netdev_linux_class = { netdev_linux_get_ifindex, netdev_linux_get_carrier, netdev_linux_get_stats, + netdev_linux_set_stats, netdev_linux_get_features, netdev_linux_set_advertisements, @@ -1908,11 +1739,11 @@ const struct netdev_class netdev_linux_class = { }; const struct netdev_class netdev_tap_class = { - "tap", /* type */ + "tap", netdev_linux_init, - NULL, /* run */ - NULL, /* wait */ + netdev_linux_run, + netdev_linux_wait, netdev_linux_create_tap, netdev_linux_destroy, @@ -1921,7 +1752,7 @@ const struct netdev_class netdev_tap_class = { netdev_linux_open, netdev_linux_close, - netdev_linux_enumerate, + NULL, /* enumerate */ netdev_linux_recv, netdev_linux_recv_wait, @@ -1936,6 +1767,7 @@ const struct netdev_class netdev_tap_class = { netdev_linux_get_ifindex, netdev_linux_get_carrier, netdev_linux_get_stats, + NULL, /* set_stats */ netdev_linux_get_features, netdev_linux_set_advertisements, @@ -1955,21 +1787,21 @@ const struct netdev_class netdev_tap_class = { netdev_linux_poll_remove, }; -const struct netdev_class netdev_gre_class = { - "gre", /* type */ +const struct netdev_class netdev_patch_class = { + "patch", netdev_linux_init, - NULL, /* run */ - NULL, /* wait */ + netdev_linux_run, + netdev_linux_wait, - netdev_linux_create_gre, + netdev_linux_create_patch, netdev_linux_destroy, - netdev_linux_reconfigure_gre, + NULL, /* reconfigure */ netdev_linux_open, netdev_linux_close, - netdev_linux_enumerate, + NULL, /* enumerate */ netdev_linux_recv, netdev_linux_recv_wait, @@ -1984,6 +1816,7 @@ const struct netdev_class netdev_gre_class = { netdev_linux_get_ifindex, netdev_linux_get_carrier, netdev_linux_get_stats, + NULL, /* set_stats */ netdev_linux_get_features, netdev_linux_set_advertisements, @@ -2002,6 +1835,7 @@ const struct netdev_class netdev_gre_class = { netdev_linux_poll_add, netdev_linux_poll_remove, }; + static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats) @@ -2150,7 +1984,8 @@ get_flags(const struct netdev *netdev, int *flags) struct ifreq ifr; int error; - error = netdev_linux_do_ioctl(netdev, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); + error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS, + "SIOCGIFFLAGS"); *flags = ifr.ifr_flags; return error; } @@ -2161,7 +1996,8 @@ set_flags(struct netdev *netdev, int flags) struct ifreq ifr; ifr.ifr_flags = flags; - return netdev_linux_do_ioctl(netdev, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); + return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS, + "SIOCSIFFLAGS"); } static int @@ -2182,17 +2018,18 @@ do_get_ifindex(const char *netdev_name) static int get_ifindex(const struct netdev *netdev_, int *ifindexp) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); *ifindexp = 0; - if (!(netdev->cache->valid & VALID_IFINDEX)) { + if (!(netdev_dev->cache_valid & VALID_IFINDEX)) { int ifindex = do_get_ifindex(netdev_get_name(netdev_)); if (ifindex < 0) { return -ifindex; } - netdev->cache->valid |= VALID_IFINDEX; - netdev->cache->ifindex = ifindex; + netdev_dev->cache_valid |= VALID_IFINDEX; + netdev_dev->ifindex = ifindex; } - *ifindexp = netdev->cache->ifindex; + *ifindexp = netdev_dev->ifindex; return 0; } @@ -2239,13 +2076,13 @@ set_etheraddr(const char *netdev_name, int hwaddr_family, } static int -netdev_linux_do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd, +netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, int cmd, const char *cmd_name) { struct ifreq ifr; memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); + strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); ifr.ifr_data = (caddr_t) ecmd; ecmd->cmd = cmd; @@ -2255,8 +2092,7 @@ netdev_linux_do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd, } else { if (errno != EOPNOTSUPP) { VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " - "failed: %s", cmd_name, netdev->name, - strerror(errno)); + "failed: %s", cmd_name, name, strerror(errno)); } else { /* The device doesn't support this operation. That's pretty * common, so there's no point in logging anything. */ @@ -2266,13 +2102,13 @@ netdev_linux_do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd, } static int -netdev_linux_do_ioctl(const struct netdev *netdev, struct ifreq *ifr, - int cmd, const char *cmd_name) +netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, + const char *cmd_name) { - strncpy(ifr->ifr_name, netdev_get_name(netdev), sizeof ifr->ifr_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", - netdev_get_name(netdev), cmd_name, strerror(errno)); + VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, + strerror(errno)); return errno; } return 0; @@ -2286,7 +2122,7 @@ 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, &ifr, cmd, cmd_name); + error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); if (!error) { const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; *ip = sin->sin_addr;