diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 0973b5d95..e075c8b02 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -53,6 +53,7 @@ struct dpif_linux { int local_ifindex; /* Ifindex of local port. */ struct svec changed_ports; /* Ports that have changed. */ struct linux_netdev_notifier port_notifier; + bool change_error; }; static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); @@ -73,18 +74,6 @@ dpif_linux_cast(const struct dpif *dpif) return CONTAINER_OF(dpif, struct dpif_linux, dpif); } -static void -dpif_linux_run(void) -{ - linux_netdev_notifier_run(); -} - -static void -dpif_linux_wait(void) -{ - linux_netdev_notifier_wait(); -} - static int dpif_linux_enumerate(struct svec *all_dps) { @@ -285,26 +274,24 @@ static int dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - int error; - error = linux_netdev_notifier_get_error(&dpif->port_notifier); - if (!error) { - if (!dpif->changed_ports.n) { - return EAGAIN; - } - *devnamep = dpif->changed_ports.names[--dpif->changed_ports.n]; - } else { + if (dpif->change_error) { + dpif->change_error = false; svec_clear(&dpif->changed_ports); + return ENOBUFS; + } else if (dpif->changed_ports.n) { + *devnamep = dpif->changed_ports.names[--dpif->changed_ports.n]; + return 0; + } else { + return EAGAIN; } - return error; } static void dpif_linux_port_poll_wait(const struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (dpif->changed_ports.n - || linux_netdev_notifier_peek_error(&dpif->port_notifier)) { + if (dpif->changed_ports.n || dpif->change_error) { poll_immediate_wake(); } else { linux_netdev_notifier_wait(); @@ -452,8 +439,8 @@ dpif_linux_recv_wait(struct dpif *dpif_) const struct dpif_class dpif_linux_class = { "", /* This is the default class. */ "linux", - dpif_linux_run, - dpif_linux_wait, + NULL, + NULL, dpif_linux_enumerate, dpif_linux_open, dpif_linux_close, @@ -725,6 +712,7 @@ open_minor(int minor, struct dpif **dpifp) dpif->minor = minor; dpif->local_ifindex = 0; svec_init(&dpif->changed_ports); + dpif->change_error = false; *dpifp = &dpif->dpif; } else { free(dpif); @@ -743,15 +731,19 @@ dpif_linux_port_changed(const struct linux_netdev_change *change, void *dpif_) { struct dpif_linux *dpif = dpif_; - if (change->master_ifindex == dpif->local_ifindex - && (change->nlmsg_type == RTM_NEWLINK - || change->nlmsg_type == RTM_DELLINK)) - { - /* Our datapath changed, either adding a new port or deleting an - * existing one. */ - if (!svec_contains(&dpif->changed_ports, change->ifname)) { - svec_add(&dpif->changed_ports, change->ifname); - svec_sort(&dpif->changed_ports); + if (change) { + if (change->master_ifindex == dpif->local_ifindex + && (change->nlmsg_type == RTM_NEWLINK + || change->nlmsg_type == RTM_DELLINK)) + { + /* Our datapath changed, either adding a new port or deleting an + * existing one. */ + if (!svec_contains(&dpif->changed_ports, change->ifname)) { + svec_add(&dpif->changed_ports, change->ifname); + svec_sort(&dpif->changed_ports); + } } + } else { + dpif->change_error = true; } } diff --git a/lib/dpif.c b/lib/dpif.c index 73fa4b983..14b424e65 100644 --- a/lib/dpif.c +++ b/lib/dpif.c @@ -65,8 +65,8 @@ static void check_rw_odp_flow(struct odp_flow *); /* Performs periodic work needed by all the various kinds of dpifs. * - * If your program opens any dpifs, it must call this function within its main - * poll loop. */ + * If your program opens any dpifs, it must call both this function and + * netdev_run() within its main poll loop. */ void dp_run(void) { @@ -81,8 +81,8 @@ dp_run(void) /* Arranges for poll_block() to wake up when dp_run() needs to be called. * - * If your program opens any dpifs, it must call this function within its main - * poll loop. */ + * If your program opens any dpifs, it must call both this function and + * netdev_wait() within its main poll loop. */ void dp_wait(void) { diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index c753e28a2..6189bf790 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -18,20 +18,1584 @@ #include "netdev-linux.h" +#include #include -#include +#include +#include +#include +#include +#include +#include #include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include #include +#include +#include +#include #include "coverage.h" +#include "dynamic-string.h" +#include "fatal-signal.h" +#include "netdev-provider.h" #include "netlink.h" #include "ofpbuf.h" +#include "openflow/openflow.h" +#include "packets.h" +#include "poll-loop.h" +#include "socket-util.h" +#include "shash.h" +#include "svec.h" #define THIS_MODULE VLM_netdev_linux #include "vlog.h" + +/* 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 +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; +}; + +enum { + VALID_IFINDEX = 1 << 0, + VALID_ETHERADDR = 1 << 1, + VALID_IN4 = 1 << 2, + VALID_IN6 = 1 << 3, + VALID_MTU = 1 << 4, + VALID_CARRIER = 1 << 5 +}; + +/* Cached network device information. */ +struct netdev_linux_cache { + struct shash_node *shash_node; + unsigned int valid; + int ref_cnt; + + int ifindex; + uint8_t etheraddr[ETH_ADDR_LEN]; + struct in_addr in4; + struct in6_addr in6; + int mtu; + int carrier; +}; + +static struct shash cache_map = SHASH_INITIALIZER(&cache_map); +static struct linux_netdev_notifier netdev_linux_cache_notifier; + +/* Policy for RTNLGRP_LINK messages. + * + * 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_MASTER] = { .type = NL_A_U32, .optional = true }, + [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, + .min_len = sizeof(struct rtnl_link_stats) }, +}; + +/* An AF_INET socket (used for ioctl operations). */ +static int af_inet_sock = -1; + +struct netdev_linux_notifier { + struct netdev_notifier notifier; + struct list node; +}; + +static struct shash netdev_linux_notifiers = + SHASH_INITIALIZER(&netdev_linux_notifiers); +static struct linux_netdev_notifier netdev_linux_poll_notifier; + +/* This is set pretty low because we probably won't learn anything from the + * additional log messages. */ +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + +static int netdev_linux_do_ethtool(struct netdev *, 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 get_flags(const struct netdev *, int *flagsp); +static int set_flags(struct netdev *, int flags); +static int do_get_ifindex(const char *netdev_name); +static int get_ifindex(const struct netdev *, int *ifindexp); +static int do_set_addr(struct netdev *netdev, + int ioctl_nr, const char *ioctl_name, + struct in_addr addr); +static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); +static int set_etheraddr(const char *netdev_name, int hwaddr_family, + const uint8_t[ETH_ADDR_LEN]); +static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); +static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); + +static struct netdev_linux * +netdev_linux_cast(const struct netdev *netdev) +{ + netdev_assert_class(netdev, &netdev_linux_class); + return CONTAINER_OF(netdev, struct netdev_linux, netdev); +} + +static int +netdev_linux_init(void) +{ + static int status = -1; + if (status < 0) { + af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); + status = af_inet_sock >= 0 ? 0 : errno; + if (status) { + VLOG_ERR("failed to create inet socket: %s", strerror(status)); + } + } + return status; +} + +static void +netdev_linux_run(void) +{ + linux_netdev_notifier_run(); +} + +static void +netdev_linux_wait(void) +{ + linux_netdev_notifier_wait(); +} + +static void +netdev_linux_cache_cb(const struct linux_netdev_change *change, + void *aux UNUSED) +{ + struct netdev_linux_cache *cache; + if (change) { + cache = shash_find_data(&cache_map, change->ifname); + if (cache) { + cache->valid = 0; + } + } else { + struct shash_node *node; + SHASH_FOR_EACH (node, &cache_map) { + cache = node->data; + cache->valid = 0; + } + } +} + +static int +netdev_linux_open(const char *name, char *suffix, int ethertype, + struct netdev **netdevp) +{ + struct netdev_linux *netdev; + enum netdev_flags flags; + int error; + + /* Allocate network device. */ + netdev = xcalloc(1, sizeof *netdev); + netdev_init(&netdev->netdev, suffix, &netdev_linux_class); + netdev->netdev_fd = -1; + netdev->tap_fd = -1; + netdev->cache = shash_find_data(&cache_map, suffix); + if (!netdev->cache) { + if (shash_is_empty(&cache_map)) { + int error = linux_netdev_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, suffix, + netdev->cache); + netdev->cache->valid = 0; + netdev->cache->ref_cnt = 0; + } + netdev->cache->ref_cnt++; + + if (!strncmp(name, "tap:", 4)) { + 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; + error = netdev_linux_do_ioctl(&netdev->netdev, &ifr, + TUNSETIFF, "TUNSETIFF"); + if (error) { + goto error; + } + + /* Make non-blocking. */ + error = set_nonblocking(netdev->tap_fd); + if (error) { + goto error; + } + } + + error = netdev_get_flags(&netdev->netdev, &flags); + if (error == ENODEV) { + goto error; + } + + if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) { + struct sockaddr_ll sll; + int protocol; + int ifindex; + + /* Create file descriptor. */ + 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) { + 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); + if (error) { + goto error; + } + + /* Get ethernet device index. */ + error = get_ifindex(&netdev->netdev, &ifindex); + if (error) { + goto error; + } + + /* Bind to specific ethernet device. */ + memset(&sll, 0, sizeof sll); + sll.sll_family = AF_PACKET; + sll.sll_ifindex = ifindex; + if (bind(netdev->netdev_fd, + (struct sockaddr *) &sll, sizeof sll) < 0) { + error = errno; + VLOG_ERR("bind to %s failed: %s", suffix, strerror(error)); + goto error; + } + + /* Between the socket() and bind() calls above, the socket receives all + * packets of the requested type on all system interfaces. We do not + * want to receive that data, but there is no way to avoid it. So we + * must now drain out the receive queue. */ + error = drain_rcvbuf(netdev->netdev_fd); + if (error) { + goto error; + } + } + + *netdevp = &netdev->netdev; + return 0; + +error: + netdev_close(&netdev->netdev); + return error; +} + +/* Closes and destroys 'netdev'. */ +static void +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)) { + linux_netdev_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); + } + free(netdev); +} + +/* Initializes 'svec' with a list of the names of all known network devices. */ +static int +netdev_linux_enumerate(struct svec *svec) +{ + struct if_nameindex *names; + + names = if_nameindex(); + if (names) { + size_t i; + + for (i = 0; names[i].if_name != NULL; i++) { + svec_add(svec, names[i].if_name); + } + if_freenameindex(names); + return 0; + } else { + VLOG_WARN("could not obtain list of network device names: %s", + strerror(errno)); + return errno; + } +} + +static int +netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (netdev->tap_fd < 0) { + /* Device was opened with NETDEV_ETH_TYPE_NONE. */ + return EAGAIN; + } + + for (;;) { + ssize_t retval = read(netdev->tap_fd, data, size); + if (retval >= 0) { + return retval; + } else if (errno != EINTR) { + if (errno != EAGAIN) { + VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", + strerror(errno), netdev_get_name(netdev_)); + } + return errno; + } + } +} + +/* Registers with the poll loop to wake up from the next call to poll_block() + * when a packet is ready to be received with netdev_recv() on 'netdev'. */ +static void +netdev_linux_recv_wait(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (netdev->tap_fd >= 0) { + poll_fd_wait(netdev->tap_fd, POLLIN); + } +} + +/* Discards all packets waiting to be received from 'netdev'. */ +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) { + return 0; + } else if (netdev->tap_fd != netdev->netdev_fd) { + struct ifreq ifr; + int error = netdev_linux_do_ioctl(netdev_, &ifr, + SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); + if (error) { + return error; + } + drain_fd(netdev->tap_fd, ifr.ifr_qlen); + return 0; + } else { + return drain_rcvbuf(netdev->netdev_fd); + } +} + +/* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive + * errno value. Returns EAGAIN without blocking if the packet cannot be queued + * 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. + * + * The kernel maintains a packet transmission queue, so the caller is not + * expected to do additional queuing of packets. */ +static int +netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. + */ + if (netdev->tap_fd < 0) { + return EPIPE; + } + + for (;;) { + ssize_t retval = write(netdev->tap_fd, data, size); + if (retval < 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 == EINTR) { + continue; + } else if (errno != EAGAIN) { + VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", + netdev_get_name(netdev_), strerror(errno)); + } + return errno; + } else if (retval != size) { + VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " + "%zu) on %s", retval, size, netdev_get_name(netdev_)); + return EMSGSIZE; + } else { + return 0; + } + } +} + +/* Registers with the poll loop to wake up from the next call to poll_block() + * when the packet transmission queue has sufficient room to transmit a packet + * with netdev_send(). + * + * The kernel maintains a packet transmission queue, so the client is not + * expected to do additional queuing of packets. Thus, this function is + * unlikely to ever be used. It is included for completeness. */ +static void +netdev_linux_send_wait(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) { + /* Nothing to do. */ + } else if (netdev->tap_fd == netdev->netdev_fd) { + poll_fd_wait(netdev->tap_fd, POLLOUT); + } else { + /* TAP device always accepts packets.*/ + poll_immediate_wake(); + } +} + +/* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, + * otherwise a positive errno value. */ +static int +netdev_linux_set_etheraddr(struct netdev *netdev_, + const uint8_t mac[ETH_ADDR_LEN]) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac); + if (!error) { + memcpy(netdev->cache->etheraddr, mac, ETH_ADDR_LEN); + } + return error; +} + +/* Returns a pointer to 'netdev''s MAC address. The caller must not modify or + * free the returned buffer. */ +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)) { + int error = get_etheraddr(netdev_get_name(netdev_), + netdev->cache->etheraddr); + if (error) { + return error; + } + netdev->cache->valid |= VALID_ETHERADDR; + } + memcpy(mac, netdev->cache->etheraddr, ETH_ADDR_LEN); + return 0; +} + +/* 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. */ +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 ifreq ifr; + int error; + + error = netdev_linux_do_ioctl(netdev_, &ifr, SIOCGIFMTU, "SIOCGIFMTU"); + if (error) { + return error; + } + netdev->cache->mtu = ifr.ifr_mtu; + netdev->cache->valid |= VALID_MTU; + } + *mtup = netdev->cache->mtu; + return 0; +} + +static int +netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error = 0; + char *fn = NULL; + int fd = -1; + + if (!(netdev->cache->valid & VALID_CARRIER)) { + char line[8]; + int retval; + + fn = xasprintf("/sys/class/net/%s/carrier", netdev_get_name(netdev_)); + fd = open(fn, O_RDONLY); + if (fd < 0) { + error = errno; + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); + goto exit; + } + + retval = read(fd, line, sizeof line); + if (retval < 0) { + error = errno; + if (error == EINVAL) { + /* This is the normal return value when we try to check carrier + * if the network device is not up. */ + } else { + VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error)); + } + goto exit; + } else if (retval == 0) { + error = EPROTO; + VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); + goto exit; + } + + if (line[0] != '0' && line[0] != '1') { + error = EPROTO; + VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", + fn, line[0]); + goto exit; + } + netdev->cache->carrier = line[0] != '0'; + netdev->cache->valid |= VALID_CARRIER; + } + *carrier = netdev->cache->carrier; + error = 0; + +exit: + if (fd >= 0) { + close(fd); + } + free(fn); + return error; +} + +/* Check whether we can we use RTM_GETLINK to get network device statistics. + * In pre-2.6.19 kernels, this was only available if wireless extensions were + * enabled. */ +static bool +check_for_working_netlink_stats(void) +{ + /* Decide on the netdev_get_stats() implementation to use. Netlink is + * preferable, so if that works, we'll use it. */ + int ifindex = do_get_ifindex("lo"); + if (ifindex < 0) { + VLOG_WARN("failed to get ifindex for lo, " + "obtaining netdev stats from proc"); + return false; + } else { + struct netdev_stats stats; + int error = get_stats_via_netlink(ifindex, &stats); + if (!error) { + VLOG_DBG("obtaining netdev stats via rtnetlink"); + return 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)); + return false; + } + } +} + +/* 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) +{ + static int use_netlink_stats = -1; + int error; + + COVERAGE_INC(netdev_get_stats); + if (use_netlink_stats < 0) { + use_netlink_stats = check_for_working_netlink_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); + } + return error; +} + +/* Stores the features supported by 'netdev' into each of '*current', + * '*advertised', '*supported', and '*peer' that are non-null. Each value is a + * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if + * successful, otherwise a positive errno value. On failure, all of the + * passed-in values are set to 0. */ +static int +netdev_linux_get_features(struct netdev *netdev, + uint32_t *current, uint32_t *advertised, + uint32_t *supported, uint32_t *peer) +{ + struct ethtool_cmd ecmd; + int error; + + memset(&ecmd, 0, sizeof ecmd); + error = netdev_linux_do_ethtool(netdev, &ecmd, + ETHTOOL_GSET, "ETHTOOL_GSET"); + if (error) { + return error; + } + + /* Supported features. */ + *supported = 0; + 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; + } + + /* Advertised features. */ + *advertised = 0; + 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; + } + + /* Current settings. */ + if (ecmd.speed == SPEED_10) { + *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD; + } else if (ecmd.speed == SPEED_100) { + *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD; + } else if (ecmd.speed == SPEED_1000) { + *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD; + } else if (ecmd.speed == SPEED_10000) { + *current = OFPPF_10GB_FD; + } else { + *current = 0; + } + + if (ecmd.port == PORT_TP) { + *current |= OFPPF_COPPER; + } else if (ecmd.port == PORT_FIBRE) { + *current |= OFPPF_FIBER; + } + + if (ecmd.autoneg) { + *current |= OFPPF_AUTONEG; + } + + /* Peer advertisements. */ + *peer = 0; /* XXX */ + + return 0; +} + +/* Set the features advertised by 'netdev' to 'advertise'. */ +static int +netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) +{ + struct ethtool_cmd ecmd; + int error; + + memset(&ecmd, 0, sizeof ecmd); + error = netdev_linux_do_ethtool(netdev, &ecmd, + ETHTOOL_GSET, "ETHTOOL_GSET"); + if (error) { + return 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 netdev_linux_do_ethtool(netdev, &ecmd, + ETHTOOL_SSET, "ETHTOOL_SSET"); +} + +/* 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. */ +static int +netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) +{ + const char *netdev_name = netdev_get_name(netdev); + 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; + } + + 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; + } + + 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; + } + + error = 0; + +done: + free(fn); + if (stream) { + fclose(stream); + } + ds_destroy(&line); + if (error) { + *vlan_vid = -1; + } + 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. */ +static int +netdev_linux_set_policing(struct netdev *netdev, + uint32_t kbits_rate, uint32_t kbits_burst) +{ + const char *netdev_name = netdev_get_name(netdev); + char command[1024]; + + 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; +} + +/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if + * 'in4' is non-null) and returns true. Otherwise, returns false. */ +static int +netdev_linux_get_in4(const struct netdev *netdev_, struct in_addr *in4) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache->valid & VALID_IN4)) { + const struct sockaddr_in *sin; + struct ifreq ifr; + int error; + + ifr.ifr_addr.sa_family = AF_INET; + error = netdev_linux_do_ioctl(netdev_, &ifr, + SIOCGIFADDR, "SIOCGIFADDR"); + if (error) { + return error; + } + + sin = (struct sockaddr_in *) &ifr.ifr_addr; + netdev->cache->in4 = sin->sin_addr; + netdev->cache->valid |= VALID_IN4; + } + *in4 = netdev->cache->in4; + return in4->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; +} + +/* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If + * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. Returns a + * positive errno value. */ +static int +netdev_linux_set_in4(struct netdev *netdev_, struct in_addr addr, + struct in_addr mask) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", addr); + if (!error) { + netdev->cache->valid |= VALID_IN4; + netdev->cache->in4 = addr; + if (addr.s_addr != INADDR_ANY) { + error = do_set_addr(netdev_, SIOCSIFNETMASK, + "SIOCSIFNETMASK", mask); + } + } + return error; +} + +static bool +parse_if_inet6_line(const char *line, + struct in6_addr *in6, char ifname[16 + 1]) +{ + uint8_t *s6 = in6->s6_addr; +#define X8 "%2"SCNx8 + return sscanf(line, + " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 + "%*x %*x %*x %*x %16s\n", + &s6[0], &s6[1], &s6[2], &s6[3], + &s6[4], &s6[5], &s6[6], &s6[7], + &s6[8], &s6[9], &s6[10], &s6[11], + &s6[12], &s6[13], &s6[14], &s6[15], + ifname) == 17; +} + +/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if + * 'in6' is non-null) and returns true. Otherwise, returns false. */ +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)) { + FILE *file; + char line[128]; + + netdev->cache->in6 = in6addr_any; + + file = fopen("/proc/net/if_inet6", "r"); + if (file != NULL) { + const char *name = netdev_get_name(netdev_); + while (fgets(line, sizeof line, file)) { + struct in6_addr in6; + char ifname[16 + 1]; + if (parse_if_inet6_line(line, &in6, ifname) + && !strcmp(name, ifname)) + { + netdev->cache->in6 = in6; + break; + } + } + fclose(file); + } + netdev->cache->valid |= VALID_IN6; + } + *in6 = netdev->cache->in6; + return 0; +} + +static void +make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) +{ + 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); +} + +static int +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); + make_in4_sockaddr(&ifr.ifr_addr, addr); + return netdev_linux_do_ioctl(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) +{ + struct in_addr any = { INADDR_ANY }; + struct rtentry rt; + 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; + if (error) { + VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + } + return error; +} + +/* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be + * successfully retrieved, it stores the corresponding MAC address in 'mac' and + * returns 0. Otherwise, it returns a positive errno value; in particular, + * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ +static int +netdev_linux_arp_lookup(const struct netdev *netdev, + uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) +{ + struct arpreq r; + struct sockaddr_in *pa; + 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; + 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)); + } + return retval; +} + +static int +nd_to_iff_flags(enum netdev_flags nd) +{ + int iff = 0; + if (nd & NETDEV_UP) { + iff |= IFF_UP; + } + if (nd & NETDEV_PROMISC) { + iff |= IFF_PROMISC; + } + return iff; +} + +static int +iff_to_nd_flags(int iff) +{ + enum netdev_flags nd = 0; + if (iff & IFF_UP) { + nd |= NETDEV_UP; + } + if (iff & IFF_PROMISC) { + nd |= NETDEV_PROMISC; + } + return nd; +} + +static int +netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) +{ + int old_flags, new_flags; + int error; + + error = get_flags(netdev, &old_flags); + if (!error) { + *old_flagsp = iff_to_nd_flags(old_flags); + new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); + if (new_flags != old_flags) { + error = set_flags(netdev, new_flags); + } + } + return error; +} + +static void +poll_notify(struct list *list) +{ + struct netdev_linux_notifier *notifier; + LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) { + struct netdev_notifier *n = ¬ifier->notifier; + n->cb(n); + } +} + +static void +netdev_linux_poll_cb(const struct linux_netdev_change *change, + void *aux UNUSED) +{ + if (change) { + struct list *list = shash_find_data(&netdev_linux_notifiers, + change->ifname); + if (list) { + poll_notify(list); + } + } else { + struct shash_node *node; + SHASH_FOR_EACH (node, &netdev_linux_notifiers) { + poll_notify(node->data); + } + } +} + +static int +netdev_linux_poll_add(struct netdev *netdev, + void (*cb)(struct netdev_notifier *), void *aux, + struct netdev_notifier **notifierp) +{ + const char *netdev_name = netdev_get_name(netdev); + struct netdev_linux_notifier *notifier; + struct list *list; + + if (shash_is_empty(&netdev_linux_notifiers)) { + int error = linux_netdev_notifier_register(&netdev_linux_poll_notifier, + netdev_linux_poll_cb, NULL); + if (error) { + return error; + } + } + + list = shash_find_data(&netdev_linux_notifiers, netdev_name); + if (!list) { + list = xmalloc(sizeof *list); + list_init(list); + shash_add(&netdev_linux_notifiers, netdev_name, list); + } + + notifier = xmalloc(sizeof *notifier); + netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); + list_push_back(list, ¬ifier->node); + *notifierp = ¬ifier->notifier; + return 0; +} + +static void +netdev_linux_poll_remove(struct netdev_notifier *notifier_) +{ + struct netdev_linux_notifier *notifier = + CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier); + struct list *list; + + /* Remove 'notifier' from its list. */ + list = list_remove(¬ifier->node); + if (list_is_empty(list)) { + /* The list is now empty. Remove it from the hash and free it. */ + const char *netdev_name = netdev_get_name(notifier->notifier.netdev); + shash_delete(&netdev_linux_notifiers, + shash_find(&netdev_linux_notifiers, netdev_name)); + free(list); + } + free(notifier); + + /* If that was the last notifier, unregister. */ + if (shash_is_empty(&netdev_linux_notifiers)) { + linux_netdev_notifier_unregister(&netdev_linux_poll_notifier); + } +} + +const struct netdev_class netdev_linux_class = { + "", /* prefix */ + "linux", /* name */ + + netdev_linux_init, + netdev_linux_run, + netdev_linux_wait, + + netdev_linux_open, + netdev_linux_close, + + netdev_linux_enumerate, + + netdev_linux_recv, + netdev_linux_recv_wait, + netdev_linux_drain, + + netdev_linux_send, + netdev_linux_send_wait, + + netdev_linux_set_etheraddr, + netdev_linux_get_etheraddr, + netdev_linux_get_mtu, + netdev_linux_get_carrier, + netdev_linux_get_stats, + + netdev_linux_get_features, + netdev_linux_set_advertisements, + netdev_linux_get_vlan_vid, + netdev_linux_set_policing, + + netdev_linux_get_in4, + netdev_linux_set_in4, + netdev_linux_get_in6, + netdev_linux_add_router, + netdev_linux_arp_lookup, + + netdev_linux_update_flags, + + netdev_linux_poll_add, + netdev_linux_poll_remove, +}; + +const struct netdev_class netdev_tap_class = { + "tap", /* prefix */ + "tap", /* name */ + + netdev_linux_init, + NULL, /* run */ + NULL, /* wait */ + + netdev_linux_open, + netdev_linux_close, + + netdev_linux_enumerate, + + netdev_linux_recv, + netdev_linux_recv_wait, + netdev_linux_drain, + + netdev_linux_send, + netdev_linux_send_wait, + + netdev_linux_set_etheraddr, + netdev_linux_get_etheraddr, + netdev_linux_get_mtu, + netdev_linux_get_carrier, + netdev_linux_get_stats, + + netdev_linux_get_features, + netdev_linux_set_advertisements, + netdev_linux_get_vlan_vid, + netdev_linux_set_policing, + + netdev_linux_get_in4, + netdev_linux_set_in4, + netdev_linux_get_in6, + netdev_linux_add_router, + netdev_linux_arp_lookup, + + netdev_linux_update_flags, + + netdev_linux_poll_add, + netdev_linux_poll_remove, +}; + +static int +get_stats_via_netlink(int ifindex, struct netdev_stats *stats) +{ + static struct nl_sock *rtnl_sock; + struct ofpbuf request; + struct ofpbuf *reply; + struct ifinfomsg *ifi; + const struct rtnl_link_stats *rtnl_stats; + struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; + int error; + + if (!rtnl_sock) { + error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); + if (error) { + VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", + strerror(error)); + return error; + } + } + + ofpbuf_init(&request, 0); + nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi, + RTM_GETLINK, NLM_F_REQUEST); + ifi = ofpbuf_put_zeros(&request, sizeof *ifi); + ifi->ifi_family = PF_UNSPEC; + ifi->ifi_index = ifindex; + error = nl_sock_transact(rtnl_sock, &request, &reply); + ofpbuf_uninit(&request); + 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; + } + + if (!attrs[IFLA_STATS]) { + VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); + return EPROTO; + } + + 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; + + 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; + } + + 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; +} + +static int +get_flags(const struct netdev *netdev, int *flags) +{ + struct ifreq ifr; + int error; + + error = netdev_linux_do_ioctl(netdev, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); + *flags = ifr.ifr_flags; + return error; +} + +static int +set_flags(struct netdev *netdev, int flags) +{ + struct ifreq ifr; + + ifr.ifr_flags = flags; + return netdev_linux_do_ioctl(netdev, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); +} + +static int +do_get_ifindex(const char *netdev_name) +{ + 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; +} + +static int +get_ifindex(const struct netdev *netdev_, int *ifindexp) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + *ifindexp = 0; + if (!(netdev->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; + } + *ifindexp = netdev->cache->ifindex; + return 0; +} + +static int +get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) +{ + struct ifreq ifr; + int hwaddr_family; + + 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; + } + hwaddr_family = ifr.ifr_hwaddr.sa_family; + if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { + VLOG_WARN("%s device has unknown hardware address family %d", + netdev_name, hwaddr_family); + } + 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; +} + +static int +netdev_linux_do_ethtool(struct netdev *netdev, 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); + ifr.ifr_data = (caddr_t) ecmd; + + ecmd->cmd = cmd; + COVERAGE_INC(netdev_ethtool); + if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { + return 0; + } else { + if (errno != EOPNOTSUPP) { + VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " + "failed: %s", cmd_name, netdev->name, + strerror(errno)); + } else { + /* The device doesn't support this operation. That's pretty + * common, so there's no point in logging anything. */ + } + return errno; + } +} + +static int +netdev_linux_do_ioctl(const struct netdev *netdev, struct ifreq *ifr, + int cmd, const char *cmd_name) +{ + strncpy(ifr->ifr_name, netdev_get_name(netdev), 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)); + return errno; + } + return 0; +} + /* rtnetlink socket. */ -static struct nl_sock *rtnl_sock; +static struct nl_sock *notify_sock; /* All registered notifiers. */ static struct list all_notifiers = LIST_INITIALIZER(&all_notifiers); @@ -39,15 +1603,15 @@ static struct list all_notifiers = LIST_INITIALIZER(&all_notifiers); static void linux_netdev_report_change(const struct nlmsghdr *, const struct ifinfomsg *, struct nlattr *attrs[]); -static void linux_netdev_report_notify_error(int error); +static void linux_netdev_report_notify_error(void); int linux_netdev_notifier_register(struct linux_netdev_notifier *notifier, linux_netdev_notify_func *cb, void *aux) { - if (!rtnl_sock) { + if (!notify_sock) { int error = nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, - &rtnl_sock); + ¬ify_sock); if (error) { VLOG_WARN("could not create rtnetlink socket: %s", strerror(error)); @@ -60,7 +1624,6 @@ linux_netdev_notifier_register(struct linux_netdev_notifier *notifier, } list_push_back(&all_notifiers, ¬ifier->node); - notifier->error = 0; notifier->cb = cb; notifier->aux = aux; return 0; @@ -71,36 +1634,17 @@ linux_netdev_notifier_unregister(struct linux_netdev_notifier *notifier) { list_remove(¬ifier->node); if (list_is_empty(&all_notifiers)) { - nl_sock_destroy(rtnl_sock); - rtnl_sock = NULL; + nl_sock_destroy(notify_sock); + notify_sock = NULL; } } -int -linux_netdev_notifier_get_error(struct linux_netdev_notifier *notifier) -{ - int error = notifier->error; - notifier->error = 0; - return error; -} - -int -linux_netdev_notifier_peek_error(const struct linux_netdev_notifier *notifier) -{ - return notifier->error; -} - -static const struct nl_policy rtnlgrp_link_policy[] = { - [IFLA_IFNAME] = { .type = NL_A_STRING }, - [IFLA_MASTER] = { .type = NL_A_U32, .optional = true }, -}; - void linux_netdev_notifier_run(void) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - if (!rtnl_sock) { + if (!notify_sock) { return; } @@ -109,7 +1653,7 @@ linux_netdev_notifier_run(void) struct ofpbuf *buf; int error; - error = nl_sock_recv(rtnl_sock, &buf, false); + error = nl_sock_recv(notify_sock, &buf, false); if (!error) { if (nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg), rtnlgrp_link_policy, @@ -120,7 +1664,7 @@ linux_netdev_notifier_run(void) linux_netdev_report_change(buf->data, ifinfo, attrs); } else { VLOG_WARN_RL(&rl, "received bad rtnl message"); - linux_netdev_report_notify_error(ENOBUFS); + linux_netdev_report_notify_error(); } ofpbuf_delete(buf); } else if (error == EAGAIN) { @@ -132,7 +1676,7 @@ linux_netdev_notifier_run(void) VLOG_WARN_RL(&rl, "error reading rtnetlink socket: %s", strerror(error)); } - linux_netdev_report_notify_error(error); + linux_netdev_report_notify_error(); } } } @@ -140,8 +1684,8 @@ linux_netdev_notifier_run(void) void linux_netdev_notifier_wait(void) { - if (rtnl_sock) { - nl_sock_wait(rtnl_sock, POLLIN); + if (notify_sock) { + nl_sock_wait(notify_sock, POLLIN); } } @@ -163,21 +1707,17 @@ linux_netdev_report_change(const struct nlmsghdr *nlmsg, LIST_FOR_EACH (notifier, struct linux_netdev_notifier, node, &all_notifiers) { - if (!notifier->error) { - notifier->cb(&change, notifier->aux); - } + notifier->cb(&change, notifier->aux); } } static void -linux_netdev_report_notify_error(int error) +linux_netdev_report_notify_error(void) { struct linux_netdev_notifier *notifier; LIST_FOR_EACH (notifier, struct linux_netdev_notifier, node, &all_notifiers) { - if (error != ENOBUFS || !notifier->error) { - notifier->error = error; - } + notifier->cb(NULL, notifier->aux); } } diff --git a/lib/netdev-linux.h b/lib/netdev-linux.h index 93ddfcb63..e8e204b64 100644 --- a/lib/netdev-linux.h +++ b/lib/netdev-linux.h @@ -39,7 +39,6 @@ typedef void linux_netdev_notify_func(const struct linux_netdev_change *, struct linux_netdev_notifier { struct list node; - int error; linux_netdev_notify_func *cb; void *aux; }; @@ -47,8 +46,6 @@ struct linux_netdev_notifier { int linux_netdev_notifier_register(struct linux_netdev_notifier *, linux_netdev_notify_func *, void *aux); void linux_netdev_notifier_unregister(struct linux_netdev_notifier *); -int linux_netdev_notifier_get_error(struct linux_netdev_notifier *); -int linux_netdev_notifier_peek_error(const struct linux_netdev_notifier *); void linux_netdev_notifier_run(void); void linux_netdev_notifier_wait(void); diff --git a/lib/netdev-provider.h b/lib/netdev-provider.h new file mode 100644 index 000000000..752fd82a2 --- /dev/null +++ b/lib/netdev-provider.h @@ -0,0 +1,284 @@ +/* + * Copyright (c) 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef NETDEV_PROVIDER_H +#define NETDEV_PROVIDER_H 1 + +/* Generic interface to network devices. */ + +#include +#include "netdev.h" +#include "list.h" + +/* A network device (e.g. an Ethernet device). + * + * This structure should be treated as opaque by network device + * implementations. */ +struct netdev { + const struct netdev_class *class; + char *name; /* e.g. "eth0" */ + enum netdev_flags save_flags; /* Initial device flags. */ + enum netdev_flags changed_flags; /* Flags that we changed. */ + struct list node; /* Element in global list. */ +}; + +void netdev_init(struct netdev *, const char *name, + const struct netdev_class *); +static inline void netdev_assert_class(const struct netdev *netdev, + const struct netdev_class *class) +{ + assert(netdev->class == class); +} + +/* A network device notifier. + * + * Network device implementations should use netdev_notifier_init() to + * initialize this structure, but they may freely read its members after + * initialization. */ +struct netdev_notifier { + struct netdev *netdev; + void (*cb)(struct netdev_notifier *); + void *aux; +}; +void netdev_notifier_init(struct netdev_notifier *, struct netdev *, + void (*cb)(struct netdev_notifier *), void *aux); + +/* Network device class structure, to be defined by each implementation of a + * network device. + * + * These functions return 0 if successful or a positive errno value on failure, + * except where otherwise noted. */ +struct netdev_class { + /* Prefix for names of netdevs in this class, e.g. "ndunix:". + * + * One netdev class may have the empty string "" as its prefix, in which + * case that netdev class is associated with netdev names that do not + * contain a colon. */ + const char *prefix; + + /* Class name, for use in error messages. */ + const char *name; + + /* Called only once, at program startup. Returning an error from this + * function will prevent any network device, of any class, from being + * opened. + * + * This function may be set to null if a network device class needs no + * initialization at program startup. */ + int (*init)(void); + + /* Performs periodic work needed by netdevs of this class. May be null if + * no periodic work is necessary. */ + void (*run)(void); + + /* Arranges for poll_block() to wake up if the "run" member function needs + * to be called. May be null if nothing is needed here. */ + void (*wait)(void); + + /* Attempts to open a network device. On success, sets '*netdevp' to the + * new network device. 'name' is the full network device name provided by + * the user. This name is useful for error messages but must not be + * modified. + * + * 'suffix' is a copy of 'name' following the netdev's 'prefix'. + * + * '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 (*open)(const char *name, char *suffix, int ethertype, + struct netdev **netdevp); + + /* Closes 'netdev'. */ + void (*close)(struct netdev *netdev); + + /* Enumerates the names of all network devices of this class. + * + * The caller has already initialized 'all_names' and might already have + * added some names to it. This function should not disturb any existing + * names in 'all_names'. + * + * If this netdev class does not support enumeration, this may be a null + * pointer. */ + int (*enumerate)(struct svec *all_anmes); + + /* Attempts to receive a packet from 'netdev' into the 'size' bytes in + * 'buffer'. If successful, returns the number of bytes in the received + * packet, otherwise a negative errno value. Returns -EAGAIN immediately + * if no packet is ready to be received. */ + int (*recv)(struct netdev *netdev, void *buffer, size_t size); + + /* 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'. */ + void (*recv_wait)(struct netdev *netdev); + + /* Discards all packets waiting to be received from 'netdev'. */ + int (*drain)(struct netdev *netdev); + + /* Sends the 'size'-byte packet in 'buffer' on 'netdev'. Returns 0 if + * successful, otherwise a positive errno value. Returns EAGAIN without + * blocking if the packet cannot be queued 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. + * + * The network device is expected to maintain a packet transmission queue, + * so that the caller does not ordinarily have to do additional queuing of + * packets. */ + int (*send)(struct netdev *netdev, const void *buffer, size_t size); + + /* Registers with the poll loop to wake up from the next call to + * poll_block() when the packet transmission queue for 'netdev' has + * sufficient room to transmit a packet with netdev_send(). + * + * The network device is expected to maintain a packet transmission queue, + * so that the caller does not ordinarily have to do additional queuing of + * packets. Thus, this function is unlikely to ever be useful. */ + void (*send_wait)(struct netdev *netdev); + + /* Sets 'netdev''s Ethernet address to 'mac' */ + int (*set_etheraddr)(struct netdev *netdev, const uint8_t mac[6]); + + /* Retrieves 'netdev''s Ethernet address into 'mac'. */ + int (*get_etheraddr)(const struct netdev *netdev, uint8_t mac[6]); + + /* Retrieves 'netdev''s MTU into '*mtup'. + * + * 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.*/ + int (*get_mtu)(const struct netdev *, int *mtup); + + /* Sets 'carrier' to true if carrier is active (link light is on) on + * 'netdev'. */ + int (*get_carrier)(const struct netdev *netdev, bool *carrier); + + /* Retrieves current device stats for 'netdev' into 'stats'. + * + * A network device that supports some statistics but not others, it should + * set the values of the unsupported statistics to all-1-bits + * (UINT64_MAX). */ + int (*get_stats)(const struct netdev *netdev, struct netdev_stats *stats); + + /* Stores the features supported by 'netdev' into each of '*current', + * '*advertised', '*supported', and '*peer'. Each value is a bitmap of + * "enum ofp_port_features" bits, in host byte order. */ + int (*get_features)(struct netdev *netdev, + uint32_t *current, uint32_t *advertised, + uint32_t *supported, uint32_t *peer); + + /* Set the features advertised by 'netdev' to 'advertise', which is a + * bitmap of "enum ofp_port_features" bits, in host byte order. + * + * This function may be set to null for a network device that does not + * support configuring advertisements. */ + int (*set_advertisements)(struct netdev *, uint32_t advertise); + + /* If 'netdev' is 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. + * + * Returns ENOENT if 'netdev_name' is the name of a network device that is + * not a VLAN device. + * + * This function should be set to null if it doesn't make any sense for + * your network device (it probably doesn't). */ + int (*get_vlan_vid)(const struct netdev *netdev, int *vlan_vid); + + /* 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. + * + * This function may be set to null if policing is not supported. */ + int (*set_policing)(struct netdev *netdev, unsigned int kbits_rate, + unsigned int kbits_burst); + + /* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address. + * + * The following error values have well-defined meanings: + * + * - EADDRNOTAVAIL: 'netdev' has no assigned IPv4 address. + * + * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ + int (*get_in4)(const struct netdev *netdev, struct in_addr *in4); + + /* Assigns 'addr' as 'netdev''s IPv4 address and 'mask' as its netmask. If + * 'addr' is INADDR_ANY, 'netdev''s IPv4 address is cleared. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ + int (*set_in4)(struct netdev *, struct in_addr addr, struct in_addr mask); + + /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address. + * + * The following error values have well-defined meanings: + * + * - EADDRNOTAVAIL: 'netdev' has no assigned IPv6 address. + * + * - EOPNOTSUPP: No IPv6 network stack attached to 'netdev'. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ + int (*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'. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ + int (*add_router)(struct netdev *netdev, struct in_addr router); + + /* Looks up the ARP table entry for 'ip' on 'netdev' and stores the + * corresponding MAC address in 'mac'. A return value of ENXIO, in + * particular, indicates that there is no ARP table entry for 'ip' on + * 'netdev'. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ + int (*arp_lookup)(const struct netdev *, uint32_t ip, uint8_t mac[6]); + + /* Retrieves the current set of flags on 'netdev' into '*old_flags'. Then, + * turns off the flags that are set to 1 in 'off' and turns on the flags + * that are set to 1 in 'on'. (No bit will be set to 1 in both 'off' and + * 'on'; that is, off & on == 0.) + * + * This function may be invoked from a signal handler. Therefore, it + * should not do anything that is not signal-safe (such as logging). */ + int (*update_flags)(struct netdev *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flags); + + /* Arranges for 'cb' to be called whenever one of the attributes of + * 'netdev' changes and sets '*notifierp' to a newly created + * netdev_notifier that represents this arrangement. The created notifier + * will have its 'netdev', 'cb', and 'aux' members set to the values of the + * corresponding parameters. */ + int (*poll_add)(struct netdev *netdev, + void (*cb)(struct netdev_notifier *), void *aux, + struct netdev_notifier **notifierp); + + /* Cancels poll notification for 'notifier'. */ + void (*poll_remove)(struct netdev_notifier *notifier); +}; + +extern const struct netdev_class netdev_linux_class; +extern const struct netdev_class netdev_tap_class; + +#endif /* netdev.h */ diff --git a/lib/netdev.c b/lib/netdev.c index 5df43c4ea..575291478 100644 --- a/lib/netdev.c +++ b/lib/netdev.c @@ -19,24 +19,7 @@ #include #include -#include -#include #include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include #include #include #include @@ -46,282 +29,94 @@ #include "dynamic-string.h" #include "fatal-signal.h" #include "list.h" -#include "netdev-linux.h" -#include "netlink.h" +#include "netdev-provider.h" #include "ofpbuf.h" -#include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" #include "shash.h" -#include "socket-util.h" #include "svec.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 - #define THIS_MODULE VLM_netdev #include "vlog.h" -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. */ -}; - -/* Policy for RTNLGRP_LINK messages. - * - * 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) }, +static const struct netdev_class *netdev_classes[] = { + &netdev_linux_class, + &netdev_tap_class, }; +enum { N_NETDEV_CLASSES = ARRAY_SIZE(netdev_classes) }; /* All open network devices. */ static struct list netdev_list = LIST_INITIALIZER(&netdev_list); -/* An AF_INET socket (used for ioctl operations). */ -static int af_inet_sock = -1; - -/* NETLINK_ROUTE socket. */ -static struct nl_sock *rtnl_sock; - -/* 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; - /* 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 init_netdev(void); -static int do_open_netdev(const char *name, int ethertype, int tap_fd, - struct netdev **netdev_); +static void restore_all_flags(void *aux); 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) +/* Attempts to initialize the netdev module. Returns 0 if successful, + * otherwise a positive errno value. + * + * Calling this function is optional. If not called explicitly, it will + * automatically be called upon the first attempt to open a network device. */ +int +netdev_initialize(void) { - FILE *file; - char line[128]; + static int status = -1; + if (status < 0) { + int i; - 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; - } + fatal_signal_add_hook(restore_all_flags, NULL, true); - 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; + status = 0; + for (i = 0; i < N_NETDEV_CLASSES; i++) { + const struct netdev_class *class = netdev_classes[i]; + if (class->init) { + int retval = class->init(); + if (retval) { + VLOG_ERR("failed to initialize %s network device " + "class: %s", class->name, strerror(retval)); + if (!status) { + status = retval; + } + } + } } } - *in6 = in6addr_any; - - fclose(file); + return status; } -static int -do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd, - int cmd, const char *cmd_name) +/* 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) { - struct ifreq ifr; - - memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); - ifr.ifr_data = (caddr_t) ecmd; - - ecmd->cmd = cmd; - COVERAGE_INC(netdev_ethtool); - if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) { - return 0; - } else { - if (errno != EOPNOTSUPP) { - VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " - "failed: %s", cmd_name, netdev->name, - strerror(errno)); - } else { - /* The device doesn't support this operation. That's pretty - * common, so there's no point in logging anything. */ + int i; + for (i = 0; i < N_NETDEV_CLASSES; i++) { + const struct netdev_class *class = netdev_classes[i]; + if (class->run) { + class->run(); } - return errno; } } -static int -do_get_features(struct netdev *netdev, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) +/* 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) { - struct ethtool_cmd ecmd; - 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; + int i; + for (i = 0; i < N_NETDEV_CLASSES; i++) { + const struct netdev_class *class = netdev_classes[i]; + if (class->wait) { + class->wait(); + } } - - 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; - } - - /* 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; - } - - /* 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; - } - - if (ecmd.port == PORT_TP) { - *current |= OFPPF_COPPER; - } - else if (ecmd.port == PORT_FIBRE) { - *current |= OFPPF_FIBER; - } - - if (ecmd.autoneg) { - *current |= OFPPF_AUTONEG; - } - return 0; } /* Opens the network device named 'name' (e.g. "eth0") and returns zero if @@ -333,171 +128,40 @@ do_get_features(struct netdev *netdev, * 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) +netdev_open(const char *name_, int ethertype, struct netdev **netdevp) { - if (strncmp(name, "tap:", 4)) { - return do_open_netdev(name, ethertype, -1, netdevp); - } else { - static const char tap_dev[] = "/dev/net/tun"; - struct ifreq ifr; - int error; - int tap_fd; + char *name = xstrdup(name_); + char *prefix, *suffix, *colon; + struct netdev *netdev = NULL; + int error; + int i; - tap_fd = open(tap_dev, O_RDWR); - if (tap_fd < 0) { - ovs_error(errno, "opening \"%s\" failed", tap_dev); - return errno; - } - - 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 = set_nonblocking(tap_fd); - if (error) { - ovs_error(error, "set_nonblocking on \"%s\" failed", tap_dev); - close(tap_fd); - return error; - } - - error = do_open_netdev(ifr.ifr_name, NETDEV_ETH_TYPE_NONE, tap_fd, - netdevp); - if (error) { - close(tap_fd); - } + error = netdev_initialize(); + if (error) { 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; - - 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; + colon = strchr(name, ':'); + if (colon) { + *colon = '\0'; + prefix = name; + suffix = colon + 1; + } else { + prefix = ""; + suffix = name; } - if (ethertype != NETDEV_ETH_TYPE_NONE) { - /* Set non-blocking mode. */ - error = set_nonblocking(netdev_fd); - if (error) { - goto error_already_set; - } - - /* Get ethernet device index. */ - ifindex = do_get_ifindex(name); - if (ifindex < 0) { - return -ifindex; - } - - /* 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; - } - - /* 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); - if (error) { - goto error_already_set; + for (i = 0; i < N_NETDEV_CLASSES; i++) { + const struct netdev_class *class = netdev_classes[i]; + if (!strcmp(prefix, class->prefix)) { + error = class->open(name_, suffix, ethertype, &netdev); + goto exit; } } + error = EAFNOSUPPORT; - /* 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; - - /* 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; - } - 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); - } +exit: + *netdevp = error ? NULL : netdev; return error; } @@ -506,9 +170,10 @@ void netdev_close(struct netdev *netdev) { if (netdev) { - /* Bring down interface and drop promiscuous mode, if we brought up - * the interface or enabled promiscuous mode. */ + char *name; int error; + + /* Restore flags that we changed, if any. */ fatal_signal_block(); error = restore_flags(netdev); list_remove(&netdev->node); @@ -519,38 +184,62 @@ netdev_close(struct netdev *netdev) } /* Free. */ - free(netdev->name); - close(netdev->netdev_fd); - if (netdev->netdev_fd != netdev->tap_fd) { - close(netdev->tap_fd); - } - free(netdev); + name = netdev->name; + netdev->class->close(netdev); + free(name); } } -/* Checks whether a network device named 'name' exists and returns true if so, - * false otherwise. */ +/* Returns true if a network device named 'name' exists and may be opened, + * otherwise false. */ bool netdev_exists(const char *name) { - struct stat s; - char *filename; + struct netdev *netdev; int error; - filename = xasprintf("/sys/class/net/%s", name); - error = stat(filename, &s); - free(filename); - return !error; + error = netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev); + if (!error) { + netdev_close(netdev); + return true; + } else { + if (error != ENODEV) { + VLOG_WARN("failed to open network device %s: %s", + name, strerror(error)); + } + return false; + } } -/* 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) +/* Initializes 'svec' with a list of the names of all known network devices. */ +int +netdev_enumerate(struct svec *svec) { - if (buffer->size < ETH_TOTAL_MIN) { - ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size); + int error; + int i; + + svec_init(svec); + + error = netdev_initialize(); + if (error) { + return error; } + + error = 0; + for (i = 0; i < N_NETDEV_CLASSES; i++) { + const struct netdev_class *class = netdev_classes[i]; + if (class->enumerate) { + int retval = class->enumerate(svec); + if (retval) { + VLOG_WARN("failed to enumerate %s network devices: %s", + class->name, strerror(retval)); + if (!error) { + error = retval; + } + } + } + } + return error; } /* Attempts to receive a packet from 'netdev' into 'buffer', which the caller @@ -568,31 +257,22 @@ pad_to_minimum_length(struct ofpbuf *buffer) int netdev_recv(struct netdev *netdev, struct ofpbuf *buffer) { - ssize_t n_bytes; + int retval; 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", - strerror(errno), netdev->name); - } - 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); + retval = netdev->class->recv(netdev, + buffer->data, ofpbuf_tailroom(buffer)); + if (retval >= 0) { + COVERAGE_INC(netdev_received); + buffer->size += retval; + if (buffer->size < ETH_TOTAL_MIN) { + ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size); + } return 0; + } else { + return -retval; } } @@ -601,19 +281,14 @@ netdev_recv(struct netdev *netdev, struct ofpbuf *buffer) void netdev_recv_wait(struct netdev *netdev) { - poll_fd_wait(netdev->tap_fd, POLLIN); + netdev->class->recv_wait(netdev); } /* Discards all packets waiting to be received from 'netdev'. */ int netdev_drain(struct netdev *netdev) { - if (netdev->tap_fd != netdev->netdev_fd) { - drain_fd(netdev->tap_fd, netdev->txqlen); - return 0; - } else { - return drain_rcvbuf(netdev->netdev_fd); - } + return netdev->class->drain(netdev); } /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive @@ -628,32 +303,11 @@ netdev_drain(struct netdev *netdev) 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 { + int error = netdev->class->send(netdev, buffer->data, buffer->size); + 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() @@ -666,12 +320,7 @@ 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(); - } + return netdev->class->send_wait(netdev); } /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, @@ -679,11 +328,7 @@ 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->class->set_etheraddr(netdev, mac); } /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the @@ -692,8 +337,7 @@ netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN]) int netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN]) { - memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN); - return 0; + return netdev->class->get_etheraddr(netdev, mac); } /* Returns the name of the network device that 'netdev' represents, @@ -714,8 +358,13 @@ netdev_get_name(const struct netdev *netdev) int netdev_get_mtu(const struct netdev *netdev, int *mtup) { - *mtup = netdev->mtu; - return 0; + int error = netdev->class->get_mtu(netdev, mtup); + if (error) { + VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s", + netdev_get_name(netdev), strerror(error)); + *mtup = ETH_PAYLOAD_MAX; + } + return error; } /* Stores the features supported by 'netdev' into each of '*current', @@ -729,123 +378,45 @@ netdev_get_features(struct netdev *netdev, 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]); + return netdev->class->get_features(netdev, + current ? current : &dummy[0], + advertised ? advertised : &dummy[1], + supported ? supported : &dummy[2], + peer ? peer : &dummy[3]); } -/* Set the features advertised by 'netdev' to 'advertise'. */ +/* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if + * successful, otherwise a positive errno value. */ int netdev_set_advertisements(struct netdev *netdev, uint32_t advertise) { - struct ethtool_cmd ecmd; - int error; - - memset(&ecmd, 0, sizeof ecmd); - error = do_ethtool(netdev, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); - if (error) { - return 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 (netdev->class->set_advertisements + ? netdev->class->set_advertisements(netdev, advertise) + : EOPNOTSUPP); } -/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if - * 'in4' is non-null) and returns 0. Otherwise, returns a positive errno value - * and sets '*in4' to INADDR_ANY (0). */ +/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address and + * returns 0. Otherwise, returns a positive errno value and sets '*in4' 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'. + * + * 'in4' may be null, in which case the address itself is not reported. */ int netdev_get_in4(const struct netdev *netdev, struct in_addr *in4) { - const char *netdev_name = netdev_get_name(netdev); - struct ifreq ifr; - struct in_addr ip = { INADDR_ANY }; + struct in_addr dummy; int error; - init_netdev(); - - 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; - error = ip.s_addr != INADDR_ANY ? 0 : EADDRNOTAVAIL; - } else { - VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFADDR) failed: %s", - netdev_name, strerror(errno)); - error = errno; - } - if (in4) { - *in4 = ip; - } - return error; -} - -static void -make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) -{ - 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); -} - -static int -do_set_addr(struct netdev *netdev, int sock, - int ioctl_nr, const char *ioctl_name, struct in_addr addr) -{ - struct ifreq ifr; - 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->class->get_in4 + ? netdev->class->get_in4(netdev, in4 ? in4 : &dummy) + : EOPNOTSUPP); + if (error && in4) { + in4->s_addr = 0; } return error; } @@ -856,86 +427,46 @@ 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->class->set_in4 + ? netdev->class->set_in4(netdev, addr, mask) + : EOPNOTSUPP); } /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds * to 'netdev'. */ int -netdev_add_router(struct netdev *netdev UNUSED, struct in_addr router) +netdev_add_router(struct netdev *netdev, struct in_addr router) { - struct in_addr any = { INADDR_ANY }; - struct rtentry rt; - 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; - if (error) { - VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); - } - return error; + return (netdev->class->add_router + ? netdev->class->add_router(netdev, router) + : EOPNOTSUPP); } -/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if - * 'in6' is non-null) and returns true. Otherwise, returns false. */ -bool +/* 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) { - if (in6) { - *in6 = netdev->in6; - } - return memcmp(&netdev->in6, &in6addr_any, sizeof netdev->in6) != 0; -} + struct in6_addr dummy; + int 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) -{ - int error, flags; - - init_netdev(); - - *flagsp = 0; - error = get_flags(netdev_get_name(netdev), &flags); - if (error) { - return error; + error = (netdev->class->get_in6 + ? netdev->class->get_in6(netdev, in6 ? in6 : &dummy) + : EOPNOTSUPP); + if (error && in6) { + memset(in6, 0, sizeof *in6); } - - if (flags & IFF_UP) { - *flagsp |= NETDEV_UP; - } - if (flags & IFF_PROMISC) { - *flagsp |= NETDEV_PROMISC; - } - return 0; -} - -static int -nd_to_iff_flags(enum netdev_flags nd) -{ - int iff = 0; - if (nd & NETDEV_UP) { - iff |= IFF_UP; - } - if (nd & NETDEV_PROMISC) { - iff |= IFF_PROMISC; - } - return iff; + return error; } /* On 'netdev', turns off the flags in 'off' and then turns on the flags in @@ -944,26 +475,44 @@ nd_to_iff_flags(enum netdev_flags nd) * 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, + bool permanent) { - int old_flags, new_flags; + enum netdev_flags old_flags; int error; - error = get_flags(netdev->name, &old_flags); + error = 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), + strerror(error)); + old_flags = 0; + } else if ((off || on) && !permanent) { + enum netdev_flags new_flags = (old_flags & ~off) | on; + enum netdev_flags changed_flags = old_flags ^ new_flags; + if (changed_flags) { + if (!netdev->changed_flags) { + netdev->save_flags = old_flags; + } + netdev->changed_flags |= changed_flags; + } } - - new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); - if (!permanent) { - netdev->changed_flags |= new_flags ^ old_flags; - } - if (new_flags != old_flags) { - error = set_flags(netdev->name, new_flags); + if (old_flagsp) { + *old_flagsp = old_flags; } 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 = (struct netdev *) netdev_; + return do_update_flags(netdev, 0, 0, flagsp, false); +} + /* 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. @@ -972,7 +521,7 @@ int netdev_set_flags(struct netdev *netdev, enum netdev_flags flags, bool permanent) { - return do_update_flags(netdev, -1, flags, permanent); + return do_update_flags(netdev, -1, flags, NULL, permanent); } /* Turns on the specified 'flags' on 'netdev'. @@ -983,7 +532,7 @@ int netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags, bool permanent) { - return do_update_flags(netdev, 0, flags, permanent); + return do_update_flags(netdev, 0, flags, NULL, permanent); } /* Turns off the specified 'flags' on 'netdev'. @@ -994,209 +543,37 @@ int netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags, bool permanent) { - return do_update_flags(netdev, flags, 0, permanent); + return do_update_flags(netdev, flags, 0, NULL, permanent); } /* 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'. */ + * 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, + uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) { - const char *netdev_name = netdev_get_name(netdev); - struct arpreq r; - struct sockaddr_in *pa; - int retval; - - 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)); - } - return retval; -} - -static int -get_stats_via_netlink(int ifindex, struct netdev_stats *stats) -{ - 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; - - 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); + int error = (netdev->class->arp_lookup + ? netdev->class->arp_lookup(netdev, ip, mac) + : EOPNOTSUPP); if (error) { - return error; + memset(mac, 0, ETH_ADDR_LEN); } - - if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), - rtnlgrp_link_policy, - attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { - ofpbuf_delete(reply); - return EPROTO; - } - - if (!attrs[IFLA_STATS]) { - VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); - return EPROTO; - } - - 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; - - return 0; + return error; } -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; - } - - 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; -} - -/* Sets 'carrier' to true if carrier is active (link light is on) on +/* Sets 'carrier' to true if carrier is active (link light is on) on * 'netdev'. */ int netdev_get_carrier(const struct netdev *netdev, bool *carrier) { - char line[8]; - int retval; - int error; - char *fn; - int fd; - - *carrier = false; - - fn = xasprintf("/sys/class/net/%s/carrier", netdev_get_name(netdev)); - fd = open(fn, O_RDONLY); - if (fd < 0) { - error = errno; - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); - goto exit; + int error = (netdev->class->get_carrier + ? netdev->class->get_carrier(netdev, carrier) + : EOPNOTSUPP); + if (error) { + *carrier = false; } - - 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; } @@ -1207,98 +584,42 @@ 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->class->get_stats + ? 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 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) { - const char *netdev_name = netdev_get_name(netdev); - 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->class->set_policing + ? 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) +/* If 'netdev' is 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. */ +int +netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) { - struct if_nameindex *names; - - svec_init(svec); - names = if_nameindex(); - if (names) { - size_t i; - - for (i = 0; names[i].if_name != NULL; i++) { - svec_add(svec, names[i].if_name); - } - if_freenameindex(names); - } else { - VLOG_WARN("could not obtain list of network device names: %s", - strerror(errno)); + int error = (netdev->class->get_vlan_vid + ? netdev->class->get_vlan_vid(netdev, vlan_vid) + : ENOENT); + if (error) { + *vlan_vid = 0; } + return error; } /* Returns a network device that has 'in4' as its IP address, if one exists, @@ -1328,209 +649,162 @@ exit: svec_destroy(&dev_list); return netdev; } - -/* If 'netdev' is 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. */ -int -netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) + +/* Initializes 'netdev' as a netdev named 'name' of the specified 'class'. + * + * This function adds 'netdev' to a netdev-owned linked list, so it is very + * important that 'netdev' only be freed after calling netdev_close(). */ +void +netdev_init(struct netdev *netdev, const char *name, + const struct netdev_class *class) { - struct ds line = DS_EMPTY_INITIALIZER; - FILE *stream = NULL; - int error; - char *fn; + netdev->class = class; + netdev->name = xstrdup(name); + netdev->save_flags = 0; + netdev->changed_flags = 0; + list_push_back(&netdev_list, &netdev->node); +} - COVERAGE_INC(netdev_get_vlan_vid); - fn = xasprintf("/proc/net/vlan/%s", netdev_get_name(netdev)); - stream = fopen(fn, "r"); - if (!stream) { - error = errno; - goto done; - } - - 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; - } - - 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; - } - - error = 0; - -done: - free(fn); - if (stream) { - fclose(stream); - } - ds_destroy(&line); - if (error) { - *vlan_vid = -1; - } - return error; +/* Initializes 'notifier' as a netdev notifier for 'netdev', for which + * notification will consist of calling 'cb', with auxiliary data 'aux'. */ +void +netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev, + void (*cb)(struct netdev_notifier *), void *aux) +{ + notifier->netdev = netdev; + notifier->cb = cb; + notifier->aux = aux; } +/* Tracks changes in the status of a set of network devices. */ struct netdev_monitor { - struct linux_netdev_notifier notifier; struct shash polled_netdevs; struct shash changed_netdevs; }; -static void netdev_monitor_change(const struct linux_netdev_change *change, - void *monitor); - -int -netdev_monitor_create(struct netdev_monitor **monitorp) +/* Creates and returns a new structure for monitor changes in the status of + * network devices. */ +struct netdev_monitor * +netdev_monitor_create(void) { - struct netdev_monitor *monitor; - int error; - - monitor = xmalloc(sizeof *monitor); - error = linux_netdev_notifier_register(&monitor->notifier, - netdev_monitor_change, monitor); - if (error) { - free(monitor); - return error; - } + struct netdev_monitor *monitor = xmalloc(sizeof *monitor); shash_init(&monitor->polled_netdevs); shash_init(&monitor->changed_netdevs); - *monitorp = monitor; - return 0; + return monitor; } +/* Destroys 'monitor'. */ void netdev_monitor_destroy(struct netdev_monitor *monitor) { if (monitor) { - linux_netdev_notifier_unregister(&monitor->notifier); + struct shash_node *node; + + SHASH_FOR_EACH (node, &monitor->polled_netdevs) { + struct netdev_notifier *notifier = node->data; + notifier->netdev->class->poll_remove(notifier); + } + shash_destroy(&monitor->polled_netdevs); + shash_destroy(&monitor->changed_netdevs); free(monitor); } } -void -netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev) +static void +netdev_monitor_cb(struct netdev_notifier *notifier) { - if (!shash_find(&monitor->polled_netdevs, netdev_get_name(netdev))) { - shash_add(&monitor->polled_netdevs, netdev_get_name(netdev), NULL); + struct netdev_monitor *monitor = notifier->aux; + const char *name = netdev_get_name(notifier->netdev); + if (!shash_find(&monitor->changed_netdevs, name)) { + shash_add(&monitor->changed_netdevs, name, NULL); } } +/* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if + * successful, otherwise a positive errno value. + * + * Adding a given 'netdev' to a monitor multiple times is equivalent to adding + * it once. */ +int +netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev) +{ + const char *netdev_name = netdev_get_name(netdev); + int error = 0; + if (!shash_find(&monitor->polled_netdevs, netdev_name) + && netdev->class->poll_add) + { + struct netdev_notifier *notifier; + error = netdev->class->poll_add(netdev, netdev_monitor_cb, monitor, + ¬ifier); + if (!error) { + assert(notifier->netdev == netdev); + shash_add(&monitor->polled_netdevs, netdev_name, notifier); + } + } + return error; +} + +/* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has + * no effect if 'netdev' is not in the set of devices monitored by + * 'monitor'.) */ void netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev) { + const char *netdev_name = netdev_get_name(netdev); struct shash_node *node; - node = shash_find(&monitor->polled_netdevs, netdev_get_name(netdev)); + node = shash_find(&monitor->polled_netdevs, netdev_name); if (node) { + /* Cancel future notifications. */ + struct netdev_notifier *notifier = node->data; + netdev->class->poll_remove(notifier); shash_delete(&monitor->polled_netdevs, node); - node = shash_find(&monitor->changed_netdevs, netdev_get_name(netdev)); + + /* Drop any pending notification. */ + node = shash_find(&monitor->changed_netdevs, netdev_name); if (node) { shash_delete(&monitor->changed_netdevs, node); } } } +/* Checks for changes to netdevs in the set monitored by 'monitor'. If any of + * the attributes (Ethernet address, carrier status, speed or peer-advertised + * speed, flags, etc.) of a network device monitored by 'monitor' has changed, + * sets '*devnamep' to the name of a device that has changed and returns 0. + * The caller is responsible for freeing '*devnamep' (with free()). + * + * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. + */ int netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep) { - int error = linux_netdev_notifier_get_error(&monitor->notifier); - *devnamep = NULL; - if (!error) { - struct shash_node *node = shash_first(&monitor->changed_netdevs); - if (!node) { - return EAGAIN; - } + struct shash_node *node = shash_first(&monitor->changed_netdevs); + if (!node) { + *devnamep = NULL; + return EAGAIN; + } else { *devnamep = xstrdup(node->name); shash_delete(&monitor->changed_netdevs, node); - } else { - shash_clear(&monitor->changed_netdevs); + return 0; } - return error; } +/* Registers with the poll loop to wake up from the next call to poll_block() + * when netdev_monitor_poll(monitor) would indicate that a device has + * changed. */ void netdev_monitor_poll_wait(const struct netdev_monitor *monitor) { - if (!shash_is_empty(&monitor->changed_netdevs) - || linux_netdev_notifier_peek_error(&monitor->notifier)) { + if (!shash_is_empty(&monitor->changed_netdevs)) { poll_immediate_wake(); } else { - linux_netdev_notifier_wait(); - } -} - -static void -netdev_monitor_change(const struct linux_netdev_change *change, void *monitor_) -{ - struct netdev_monitor *monitor = monitor_; - if (shash_find(&monitor->polled_netdevs, change->ifname) - && !shash_find(&monitor->changed_netdevs, change->ifname)) { - shash_add(&monitor->changed_netdevs, change->ifname, NULL); + /* XXX Nothing needed here for netdev_linux, but maybe other netdev + * classes need help. */ } } -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) -{ - static bool inited; - if (!inited) { - int ifindex; - int error; - - inited = true; - - fatal_signal_add_hook(restore_all_flags, NULL, true); - - af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); - if (af_inet_sock < 0) { - ovs_fatal(errno, "socket(AF_INET)"); - } - - error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); - if (error) { - ovs_fatal(error, "socket(AF_NETLINK, NETLINK_ROUTE)"); - } - - /* 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; - } - } - } -} - /* 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. @@ -1539,27 +813,13 @@ init_netdev(void) static int restore_flags(struct netdev *netdev) { - struct ifreq ifr; - int restore_flags; - - /* 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; + if (netdev->changed_flags) { + enum netdev_flags restore = netdev->save_flags & netdev->changed_flags; + enum netdev_flags old_flags; + return netdev->class->update_flags(netdev, + netdev->changed_flags & ~restore, + restore, &old_flags); } - - /* 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; - } - } - return 0; } @@ -1573,109 +833,3 @@ restore_all_flags(void *aux UNUSED) 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; -} - -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; -} - -static int -do_get_ifindex(const char *netdev_name) -{ - 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; -} - -static int -get_ifindex(const struct netdev *netdev, int *ifindexp) -{ - *ifindexp = 0; - if (netdev->ifindex < 0) { - int ifindex = do_get_ifindex(netdev->name); - if (ifindex < 0) { - return -ifindex; - } - ((struct netdev *) netdev)->ifindex = ifindex; - } - *ifindexp = netdev->ifindex; - return 0; -} - -static int -get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN], - int *hwaddr_familyp) -{ - struct ifreq ifr; - - *hwaddr_familyp = 0; - 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); - } - } - 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; -} diff --git a/lib/netdev.h b/lib/netdev.h index 5c98b1569..b66d7bc0b 100644 --- a/lib/netdev.h +++ b/lib/netdev.h @@ -44,6 +44,9 @@ enum netdev_pseudo_ethertype { NETDEV_ETH_TYPE_802_2 /* Receive all IEEE 802.2 frames. */ }; +/* Network device statistics. + * + * Values of unsupported statistics are set to all-1-bits (UINT64_MAX). */ struct netdev_stats { uint64_t rx_packets; /* Total packets received. */ uint64_t tx_packets; /* Total packets transmitted. */ @@ -74,47 +77,57 @@ struct netdev_stats { struct netdev; +int netdev_initialize(void); +void netdev_run(void); +void netdev_wait(void); + int netdev_open(const char *name, int ethertype, struct netdev **); void netdev_close(struct netdev *); -bool netdev_exists(const char *netdev_name); +bool netdev_exists(const char *name); + +int netdev_enumerate(struct svec *); + +const char *netdev_get_name(const struct netdev *); +int netdev_get_mtu(const struct netdev *, int *mtup); int netdev_recv(struct netdev *, struct ofpbuf *); void netdev_recv_wait(struct netdev *); int netdev_drain(struct netdev *); + int netdev_send(struct netdev *, const struct ofpbuf *); void netdev_send_wait(struct netdev *); + int netdev_set_etheraddr(struct netdev *, const uint8_t mac[6]); int netdev_get_etheraddr(const struct netdev *, uint8_t mac[6]); -const char *netdev_get_name(const struct netdev *); -int netdev_get_mtu(const struct netdev *, int *mtup); + +int netdev_get_carrier(const struct netdev *, bool *carrier); int netdev_get_features(struct netdev *, uint32_t *current, uint32_t *advertised, uint32_t *supported, uint32_t *peer); int netdev_set_advertisements(struct netdev *, uint32_t advertise); + int netdev_get_in4(const struct netdev *, struct in_addr *); int netdev_set_in4(struct netdev *, struct in_addr addr, struct in_addr mask); +int netdev_get_in6(const struct netdev *, struct in6_addr *); int netdev_add_router(struct netdev *, struct in_addr router); -bool netdev_get_in6(const struct netdev *, struct in6_addr *); +int netdev_arp_lookup(const struct netdev *, uint32_t ip, uint8_t mac[6]); + int netdev_get_flags(const struct netdev *, enum netdev_flags *); int netdev_set_flags(struct netdev *, enum netdev_flags, bool permanent); int netdev_turn_flags_on(struct netdev *, enum netdev_flags, bool permanent); int netdev_turn_flags_off(struct netdev *, enum netdev_flags, bool permanent); -int netdev_arp_lookup(const struct netdev *, uint32_t ip, uint8_t mac[6]); -int netdev_get_carrier(const struct netdev *, bool *carrier); + int netdev_get_stats(const struct netdev *, struct netdev_stats *); int netdev_set_policing(struct netdev *, uint32_t kbits_rate, uint32_t kbits_burst); -void netdev_enumerate(struct svec *); +int netdev_get_vlan_vid(const struct netdev *, int *vlan_vid); struct netdev *netdev_find_dev_by_in4(const struct in_addr *); -int netdev_get_vlan_vid(const struct netdev *, int *vlan_vid); - -struct netdev_monitor; -int netdev_monitor_create(struct netdev_monitor **); +struct netdev_monitor *netdev_monitor_create(void); void netdev_monitor_destroy(struct netdev_monitor *); -void netdev_monitor_add(struct netdev_monitor *, struct netdev *); +int netdev_monitor_add(struct netdev_monitor *, struct netdev *); void netdev_monitor_remove(struct netdev_monitor *, struct netdev *); int netdev_monitor_poll(struct netdev_monitor *, char **devnamep); void netdev_monitor_poll_wait(const struct netdev_monitor *); diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 56432194c..55eb2c28a 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -259,7 +259,6 @@ int ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, struct ofproto **ofprotop) { - struct netdev_monitor *netdev_monitor; struct odp_stats stats; struct ofproto *p; struct dpif *dpif; @@ -290,15 +289,6 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, dpif_flow_flush(dpif); dpif_recv_purge(dpif); - /* Arrange to monitor datapath ports for status changes. */ - error = netdev_monitor_create(&netdev_monitor); - if (error) { - VLOG_ERR("failed to starting monitoring datapath %s: %s", - datapath, strerror(error)); - dpif_close(dpif); - return error; - } - /* Initialize settings. */ p = xcalloc(1, sizeof *p); p->fallback_dpid = pick_fallback_dpid(); @@ -310,7 +300,7 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, /* Initialize datapath. */ p->dpif = dpif; - p->netdev_monitor = netdev_monitor; + p->netdev_monitor = netdev_monitor_create(); port_array_init(&p->ports); shash_init(&p->port_by_name); p->max_ports = stats.max_ports; diff --git a/utilities/ovs-openflowd.c b/utilities/ovs-openflowd.c index f60dea5a4..5dd77c056 100644 --- a/utilities/ovs-openflowd.c +++ b/utilities/ovs-openflowd.c @@ -195,10 +195,12 @@ main(int argc, char *argv[]) } unixctl_server_run(unixctl); dp_run(); + netdev_run(); ofproto_wait(ofproto); unixctl_server_wait(unixctl); dp_wait(); + netdev_wait(); poll_block(); } diff --git a/vswitchd/ovs-brcompatd.c b/vswitchd/ovs-brcompatd.c index 306de136f..70570e925 100644 --- a/vswitchd/ovs-brcompatd.c +++ b/vswitchd/ovs-brcompatd.c @@ -38,7 +38,6 @@ #include "coverage.h" #include "daemon.h" #include "dirs.h" -#include "dpif.h" #include "dynamic-string.h" #include "fatal-signal.h" #include "fault.h" @@ -911,6 +910,7 @@ main(int argc, char *argv[]) for (;;) { unixctl_server_run(unixctl); brc_recv_update(); + netdev_run(); /* If 'prune_timeout' is non-zero, we actively prune from the * config file any 'bridge..port' entries that are no @@ -932,6 +932,7 @@ main(int argc, char *argv[]) nl_sock_wait(brc_sock, POLLIN); unixctl_server_wait(unixctl); + netdev_wait(); poll_block(); } diff --git a/vswitchd/ovs-vswitchd.c b/vswitchd/ovs-vswitchd.c index 3ee29c771..a7b43f7b4 100644 --- a/vswitchd/ovs-vswitchd.c +++ b/vswitchd/ovs-vswitchd.c @@ -32,6 +32,7 @@ #include "fault.h" #include "leak-checker.h" #include "mgmt.h" +#include "netdev.h" #include "ovs-vswitchd.h" #include "poll-loop.h" #include "proc-net-compat.h" @@ -100,6 +101,7 @@ main(int argc, char *argv[]) } unixctl_server_run(unixctl); dp_run(); + netdev_run(); if (need_reconfigure) { poll_immediate_wake(); @@ -109,6 +111,7 @@ main(int argc, char *argv[]) bridge_wait(); unixctl_server_wait(unixctl); dp_wait(); + netdev_wait(); poll_block(); }