X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev.c;h=9a91e2f628706a15abfd6884dfbb2c4f71ef8658;hb=refs%2Fheads%2Ffor-nox%2F0.4;hp=82325616a546a0aa4ac25a01497fcc460eaa07c5;hpb=9dc091a35fe3420eff2942014555a6d561d3378d;p=sliver-openvswitch.git diff --git a/lib/netdev.c b/lib/netdev.c index 82325616a..9a91e2f62 100644 --- a/lib/netdev.c +++ b/lib/netdev.c @@ -17,7 +17,7 @@ * The above copyright notice and this permission notice shall be * included in all copies or substantial portions of the Software. * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * THE SOFTWRE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS @@ -31,15 +31,19 @@ * derivatives without specific, written prior permission. */ +#include #include "netdev.h" #include #include +#include #include #include +#include #include #include #include +#include #include #include #include @@ -48,18 +52,20 @@ #include #include #include +#include #include #include #include #include -#include "list.h" #include "fatal-signal.h" -#include "buffer.h" -#include "openflow.h" +#include "list.h" +#include "ofpbuf.h" +#include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" #include "socket-util.h" +#include "svec.h" #define THIS_MODULE VLM_netdev #include "vlog.h" @@ -67,44 +73,49 @@ struct netdev { struct list node; char *name; - int fd; + + /* 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; uint8_t etheraddr[ETH_ADDR_LEN]; + struct in6_addr in6; int speed; int mtu; - uint32_t features; - struct in_addr in4; - struct in6_addr in6; - int save_flags; + int txqlen; + int hwaddr_family; + + /* Bitmaps of OFPPF_* that describe features. All bits disabled if + * unsupported or unavailable. */ + uint32_t curr; /* Current features. */ + uint32_t advertised; /* Features being advertised by the port. */ + uint32_t supported; /* Features supported by the port. */ + uint32_t peer; /* Features advertised by the peer. */ + + int save_flags; /* Initial device flags. */ + int changed_flags; /* Flags that we changed. */ }; +/* 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; +/* 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 int restore_flags(struct netdev *netdev); -static int get_flags(const struct netdev *, int *flagsp); -static int set_flags(struct netdev *, int flags); - -/* Obtains the IPv4 address for 'name' into 'in4'. Returns true if - * successful. */ -static bool -get_ipv4_address(const char *name, struct in_addr *in4) -{ - struct ifreq ifr; - - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - ifr.ifr_addr.sa_family = AF_INET; - if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) { - struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; - *in4 = sin->sin_addr; - } else { - in4->s_addr = INADDR_ANY; - } - - return true; -} +static int get_flags(const struct netdev *, int fd, int *flagsp); +static int set_flags(struct netdev *, int fd, int flags); /* Obtains the IPv6 address for 'name' into 'in6'. */ static void @@ -135,6 +146,7 @@ get_ipv6_address(const char *name, struct in6_addr *in6) ifname) == 17 && !strcmp(name, ifname)) { + fclose(file); return; } } @@ -149,8 +161,10 @@ do_ethtool(struct netdev *netdev) struct ifreq ifr; struct ethtool_cmd ecmd; - netdev->speed = 0; - netdev->features = 0; + netdev->curr = 0; + netdev->supported = 0; + netdev->advertised = 0; + netdev->peer = 0; memset(&ifr, 0, sizeof ifr); strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); @@ -158,50 +172,109 @@ do_ethtool(struct netdev *netdev) memset(&ecmd, 0, sizeof ecmd); ecmd.cmd = ETHTOOL_GSET; - if (ioctl(netdev->fd, SIOCETHTOOL, &ifr) == 0) { + if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) { if (ecmd.supported & SUPPORTED_10baseT_Half) { - netdev->features |= OFPPF_10MB_HD; + netdev->supported |= OFPPF_10MB_HD; } if (ecmd.supported & SUPPORTED_10baseT_Full) { - netdev->features |= OFPPF_10MB_FD; + netdev->supported |= OFPPF_10MB_FD; } if (ecmd.supported & SUPPORTED_100baseT_Half) { - netdev->features |= OFPPF_100MB_HD; + netdev->supported |= OFPPF_100MB_HD; } if (ecmd.supported & SUPPORTED_100baseT_Full) { - netdev->features |= OFPPF_100MB_FD; + netdev->supported |= OFPPF_100MB_FD; } if (ecmd.supported & SUPPORTED_1000baseT_Half) { - netdev->features |= OFPPF_1GB_HD; + netdev->supported |= OFPPF_1GB_HD; } if (ecmd.supported & SUPPORTED_1000baseT_Full) { - netdev->features |= OFPPF_1GB_FD; + netdev->supported |= OFPPF_1GB_FD; } - /* 10Gbps half-duplex doesn't exist... */ if (ecmd.supported & SUPPORTED_10000baseT_Full) { - netdev->features |= OFPPF_10GB_FD; + netdev->supported |= OFPPF_10GB_FD; } + if (ecmd.supported & SUPPORTED_TP) { + netdev->supported |= OFPPF_COPPER; + } + if (ecmd.supported & SUPPORTED_FIBRE) { + netdev->supported |= OFPPF_FIBER; + } + if (ecmd.supported & SUPPORTED_Autoneg) { + netdev->supported |= OFPPF_AUTONEG; + } +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14) + if (ecmd.supported & SUPPORTED_Pause) { + netdev->supported |= OFPPF_PAUSE; + } + if (ecmd.supported & SUPPORTED_Asym_Pause) { + netdev->supported |= OFPPF_PAUSE_ASYM; + } +#endif /* kernel >= 2.6.14 */ - switch (ecmd.speed) { - case SPEED_10: - netdev->speed = 10; - break; - - case SPEED_100: - netdev->speed = 100; - break; + /* Set the advertised features */ + if (ecmd.advertising & ADVERTISED_10baseT_Half) { + netdev->advertised |= OFPPF_10MB_HD; + } + if (ecmd.advertising & ADVERTISED_10baseT_Full) { + netdev->advertised |= OFPPF_10MB_FD; + } + if (ecmd.advertising & ADVERTISED_100baseT_Half) { + netdev->advertised |= OFPPF_100MB_HD; + } + if (ecmd.advertising & ADVERTISED_100baseT_Full) { + netdev->advertised |= OFPPF_100MB_FD; + } + if (ecmd.advertising & ADVERTISED_1000baseT_Half) { + netdev->advertised |= OFPPF_1GB_HD; + } + if (ecmd.advertising & ADVERTISED_1000baseT_Full) { + netdev->advertised |= OFPPF_1GB_FD; + } + if (ecmd.advertising & ADVERTISED_10000baseT_Full) { + netdev->advertised |= OFPPF_10GB_FD; + } + if (ecmd.advertising & ADVERTISED_TP) { + netdev->advertised |= OFPPF_COPPER; + } + if (ecmd.advertising & ADVERTISED_FIBRE) { + netdev->advertised |= OFPPF_FIBER; + } + if (ecmd.advertising & ADVERTISED_Autoneg) { + netdev->advertised |= OFPPF_AUTONEG; + } +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14) + if (ecmd.advertising & ADVERTISED_Pause) { + netdev->advertised |= OFPPF_PAUSE; + } + if (ecmd.advertising & ADVERTISED_Asym_Pause) { + netdev->advertised |= OFPPF_PAUSE_ASYM; + } +#endif /* kernel >= 2.6.14 */ - case SPEED_1000: - netdev->speed = 1000; - break; + /* Set the current features */ + if (ecmd.speed == SPEED_10) { + netdev->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD; + } + else if (ecmd.speed == SPEED_100) { + netdev->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD; + } + else if (ecmd.speed == SPEED_1000) { + netdev->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD; + } + else if (ecmd.speed == SPEED_10000) { + netdev->curr = OFPPF_10GB_FD; + } - case SPEED_2500: - netdev->speed = 2500; - break; + if (ecmd.port == PORT_TP) { + netdev->curr |= OFPPF_COPPER; + } + else if (ecmd.port == PORT_FIBRE) { + netdev->curr |= OFPPF_FIBER; + } - case SPEED_10000: - netdev->speed = 10000; - break; + if (ecmd.autoneg) { + netdev->curr |= OFPPF_AUTONEG; } } else { VLOG_DBG("ioctl(SIOCETHTOOL) failed: %s", strerror(errno)); @@ -209,7 +282,7 @@ do_ethtool(struct netdev *netdev) } /* Opens the network device named 'name' (e.g. "eth0") and returns zero if - * successful, otherwise a positive errno value. On success, sets '*netdev' + * successful, otherwise a positive errno value. On success, sets '*netdevp' * to the new network device, otherwise to null. * * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to @@ -217,41 +290,110 @@ do_ethtool(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 **netdev_) +netdev_open(const char *name, int ethertype, struct netdev **netdevp) { - int fd; - struct sockaddr sa; + if (!strncmp(name, "tap:", 4)) { + return netdev_open_tap(name + 4, netdevp); + } else { + return do_open_netdev(name, ethertype, -1, netdevp); + } +} + +/* Opens a TAP virtual network device. If 'name' is a nonnull, non-empty + * string, attempts to assign that name to the TAP device (failing if the name + * is already in use); otherwise, a name is automatically assigned. Returns + * zero if successful, otherwise a positive errno value. On success, sets + * '*netdevp' to the new network device, otherwise to null. */ +int +netdev_open_tap(const char *name, struct netdev **netdevp) +{ + static const char tap_dev[] = "/dev/net/tun"; + struct ifreq ifr; + int error; + int tap_fd; + + tap_fd = open(tap_dev, O_RDWR); + if (tap_fd < 0) { + ofp_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; + ofp_error(error, "ioctl(TUNSETIFF) on \"%s\" failed", tap_dev); + close(tap_fd); + return error; + } + + error = set_nonblocking(tap_fd); + if (error) { + ofp_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); + } + 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; unsigned int ifindex; uint8_t etheraddr[ETH_ADDR_LEN]; - struct in_addr in4; struct in6_addr in6; int mtu; + int txqlen; + int hwaddr_family; int error; struct netdev *netdev; - *netdev_ = NULL; init_netdev(); + *netdev_ = NULL; - /* Create raw socket. - * - * We have to use SOCK_PACKET, despite its deprecation, because only - * SOCK_PACKET lets us set the hardware source address of outgoing - * packets. */ - fd = socket(PF_PACKET, SOCK_PACKET, - 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 (fd < 0) { + /* 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; } + /* Set non-blocking mode. */ + error = set_nonblocking(netdev_fd); + if (error) { + goto error_already_set; + } + + /* Get ethernet device index. */ + strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); + if (ioctl(netdev_fd, SIOCGIFINDEX, &ifr) < 0) { + VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s", + name, strerror(errno)); + goto error; + } + ifindex = ifr.ifr_ifindex; + /* Bind to specific ethernet device. */ - memset(&sa, 0, sizeof sa); - sa.sa_family = AF_UNSPEC; - strncpy((char *) sa.sa_data, name, sizeof sa.sa_data); - if (bind(fd, &sa, sizeof sa) < 0) { + 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; } @@ -261,64 +403,64 @@ netdev_open(const char *name, int ethertype, struct netdev **netdev_) * 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(fd); + error = drain_rcvbuf(netdev_fd); if (error) { goto error; } } - /* Get ethernet device index. */ - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - if (ioctl(fd, SIOCGIFINDEX, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCGIFINDEX) on %s device failed: %s", - name, strerror(errno)); - goto error; - } - ifindex = ifr.ifr_ifindex; - /* Get MAC address. */ - if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) { + if (ioctl(netdev_fd, SIOCGIFHWADDR, &ifr) < 0) { VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s", name, strerror(errno)); goto error; } - if (ifr.ifr_hwaddr.sa_family != AF_UNSPEC - && ifr.ifr_hwaddr.sa_family != ARPHRD_ETHER) { + 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", - name, (int) ifr.ifr_hwaddr.sa_family); + name, hwaddr_family); } memcpy(etheraddr, ifr.ifr_hwaddr.sa_data, sizeof etheraddr); /* Get MTU. */ - if (ioctl(fd, SIOCGIFMTU, &ifr) < 0) { + 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; - if (!get_ipv4_address(name, &in4)) { + /* 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->fd = fd; + 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->in4 = in4; netdev->in6 = in6; /* Get speed, features. */ do_ethtool(netdev); /* Save flags to restore at close or exit. */ - error = get_flags(netdev, &netdev->save_flags); + error = get_flags(netdev, netdev_fd, &netdev->save_flags); if (error) { - goto preset_error; + goto error_already_set; } + netdev->changed_flags = 0; fatal_signal_block(); list_push_back(&netdev_list, &netdev->node); fatal_signal_unblock(); @@ -329,8 +471,11 @@ netdev_open(const char *name, int ethertype, struct netdev **netdev_) error: error = errno; -preset_error: - close(fd); +error_already_set: + close(netdev_fd); + if (tap_fd >= 0) { + close(tap_fd); + } return error; } @@ -353,7 +498,10 @@ netdev_close(struct netdev *netdev) /* Free. */ free(netdev->name); - close(netdev->fd); + close(netdev->netdev_fd); + if (netdev->netdev_fd != netdev->tap_fd) { + close(netdev->tap_fd); + } free(netdev); } } @@ -361,11 +509,10 @@ netdev_close(struct netdev *netdev) /* Pads 'buffer' out with zero-bytes to the minimum valid length of an * Ethernet packet, if necessary. */ static void -pad_to_minimum_length(struct buffer *buffer) +pad_to_minimum_length(struct ofpbuf *buffer) { if (buffer->size < ETH_TOTAL_MIN) { - size_t shortage = ETH_TOTAL_MIN - buffer->size; - memset(buffer_put_uninit(buffer, shortage), 0, shortage); + ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size); } } @@ -382,21 +529,20 @@ pad_to_minimum_length(struct buffer *buffer) * be returned. */ int -netdev_recv(struct netdev *netdev, struct buffer *buffer) +netdev_recv(struct netdev *netdev, struct ofpbuf *buffer) { ssize_t n_bytes; assert(buffer->size == 0); - assert(buffer_tailroom(buffer) >= ETH_TOTAL_MIN); + assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN); do { - n_bytes = recv(netdev->fd, - buffer_tail(buffer), buffer_tailroom(buffer), - MSG_DONTWAIT); + 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("error receiving Ethernet packet on %s: %s", - strerror(errno), netdev->name); + VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", + strerror(errno), netdev->name); } return errno; } else { @@ -417,14 +563,19 @@ netdev_recv(struct netdev *netdev, struct buffer *buffer) void netdev_recv_wait(struct netdev *netdev) { - poll_fd_wait(netdev->fd, POLLIN); + poll_fd_wait(netdev->tap_fd, POLLIN); } /* Discards all packets waiting to be received from 'netdev'. */ -void +int netdev_drain(struct netdev *netdev) { - drain_rcvbuf(netdev->fd); + if (netdev->tap_fd != netdev->netdev_fd) { + drain_fd(netdev->tap_fd, netdev->txqlen); + return 0; + } else { + return drain_rcvbuf(netdev->netdev_fd); + } } /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive @@ -437,28 +588,12 @@ netdev_drain(struct netdev *netdev) * The kernel maintains a packet transmission queue, so the caller is not * expected to do additional queuing of packets. */ int -netdev_send(struct netdev *netdev, const struct buffer *buffer) +netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) { ssize_t n_bytes; - const struct eth_header *eh; - struct sockaddr_pkt spkt; - - /* Pull out the Ethernet header. */ - if (buffer->size < ETH_HEADER_LEN) { - VLOG_WARN("cannot send %zu-byte frame on %s", - buffer->size, netdev->name); - return EMSGSIZE; - } - eh = buffer_at_assert(buffer, 0, sizeof *eh); - - /* Construct packet sockaddr, which SOCK_PACKET requires. */ - spkt.spkt_family = AF_PACKET; - strncpy((char *) spkt.spkt_device, netdev->name, sizeof spkt.spkt_device); - spkt.spkt_protocol = eh->eth_type; do { - n_bytes = sendto(netdev->fd, buffer->data, buffer->size, 0, - (const struct sockaddr *) &spkt, sizeof spkt); + n_bytes = write(netdev->tap_fd, buffer->data, buffer->size); } while (n_bytes < 0 && errno == EINTR); if (n_bytes < 0) { @@ -468,13 +603,14 @@ netdev_send(struct netdev *netdev, const struct buffer *buffer) if (errno == ENOBUFS) { return EAGAIN; } else if (errno != EAGAIN) { - VLOG_WARN("error sending Ethernet packet on %s: %s", - netdev->name, strerror(errno)); + 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("send partial Ethernet packet (%d bytes of %zu) on %s", - (int) n_bytes, buffer->size, netdev->name); + VLOG_WARN_RL(&rl, + "send partial Ethernet packet (%d bytes of %zu) on %s", + (int) n_bytes, buffer->size, netdev->name); return EMSGSIZE; } else { return 0; @@ -491,7 +627,32 @@ netdev_send(struct netdev *netdev, const struct buffer *buffer) void netdev_send_wait(struct netdev *netdev) { - poll_fd_wait(netdev->fd, POLLOUT); + 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. */ +int +netdev_set_etheraddr(struct netdev *netdev, 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 = netdev->hwaddr_family; + memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); + if (ioctl(netdev->netdev_fd, SIOCSIFHWADDR, &ifr) < 0) { + VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", + netdev->name, strerror(errno)); + return errno; + } + memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN); + return 0; } /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or @@ -519,38 +680,150 @@ netdev_get_mtu(const struct netdev *netdev) return netdev->mtu; } -/* Returns the current speed of the network device that 'netdev' represents, in - * megabits per second, or 0 if the speed is unknown. */ +/* Checks the link status. Returns 1 or 0 to indicate the link is active + * or not, respectively. Any other return value indicates an error. */ int -netdev_get_speed(const struct netdev *netdev) +netdev_get_link_status(const struct netdev *netdev) { - return netdev->speed; + struct ifreq ifr; + struct ethtool_value edata; + + memset(&ifr, 0, sizeof ifr); + strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); + ifr.ifr_data = (caddr_t) &edata; + + memset(&edata, 0, sizeof edata); + edata.cmd = ETHTOOL_GLINK; + if (ioctl(netdev->netdev_fd, SIOCETHTOOL, &ifr) == 0) { + if (edata.data) { + return 1; + } else { + return 0; + } + } + + return -1; } -/* Returns the features supported by 'netdev', as a bitmap of bits from enum - * ofp_phy_port, in host byte order. */ +/* Returns the features supported by 'netdev' of type 'type', as a bitmap + * of bits from enum ofp_phy_features, in host byte order. */ uint32_t -netdev_get_features(const struct netdev *netdev) +netdev_get_features(struct netdev *netdev, int type) { - return netdev->features; + do_ethtool(netdev); + switch (type) { + case NETDEV_FEAT_CURRENT: + return netdev->curr; + case NETDEV_FEAT_ADVERTISED: + return netdev->advertised; + case NETDEV_FEAT_SUPPORTED: + return netdev->supported; + case NETDEV_FEAT_PEER: + return netdev->peer; + default: + VLOG_WARN("Unknown feature type: %d\n", type); + return 0; + } } -/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address and - * returns true. Otherwise, returns false. */ +/* If 'netdev' has an assigned IPv4 address, sets '*in4' to that address (if + * 'in4' is non-null) and returns true. Otherwise, returns false. */ bool netdev_get_in4(const struct netdev *netdev, struct in_addr *in4) { - *in4 = netdev->in4; - return in4->s_addr != INADDR_ANY; + struct ifreq ifr; + struct in_addr ip = { INADDR_ANY }; + + strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); + ifr.ifr_addr.sa_family = AF_INET; + if (ioctl(af_inet_sock, SIOCGIFADDR, &ifr) == 0) { + struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; + ip = sin->sin_addr; + } else { + VLOG_DBG_RL(&rl, "%s: ioctl(SIOCGIFADDR) failed: %s", + netdev->name, strerror(errno)); + } + if (in4) { + *in4 = ip; + } + return ip.s_addr != INADDR_ANY; +} + +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); + error = ioctl(sock, ioctl_nr, &ifr) < 0 ? errno : 0; + if (error) { + VLOG_WARN("ioctl(%s): %s", ioctl_name, strerror(error)); + } + return error; +} + +/* 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. */ +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; +} + +/* Adds 'router' as a default gateway for 'netdev''s IP address. */ +int +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; + error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; + if (error) { + VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + } + return error; } -/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and - * returns true. Otherwise, returns false. */ +/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if + * 'in6' is non-null) and returns true. Otherwise, returns false. */ bool netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6) { - *in6 = netdev->in6; - return memcmp(in6, &in6addr_any, sizeof *in6) != 0; + if (in6) { + *in6 = netdev->in6; + } + return memcmp(&netdev->in6, &in6addr_any, sizeof netdev->in6) != 0; } /* Obtains the current flags for 'netdev' and stores them into '*flagsp'. @@ -560,7 +833,7 @@ netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp) { int error, flags; - error = get_flags(netdev, &flags); + error = get_flags(netdev, netdev->netdev_fd, &flags); if (error) { return error; } @@ -575,32 +848,78 @@ netdev_get_flags(const struct netdev *netdev, enum netdev_flags *flagsp) return 0; } -/* Sets the flags for 'netdev' to 'nd_flags'. - * Returns 0 if successful, otherwise a positive errno value. */ -int -netdev_set_flags(struct netdev *netdev, enum netdev_flags nd_flags) +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; +} + +/* On 'netdev', turns off the flags in 'off' and then turns on the flags in + * 'on'. If 'permanent' is true, the changes will persist; otherwise, they + * will be reverted when 'netdev' is closed or the program exits. Returns 0 if + * successful, otherwise a positive errno value. */ +static int +do_update_flags(struct netdev *netdev, int fd, enum netdev_flags off, + enum netdev_flags on, bool permanent) { int old_flags, new_flags; int error; - error = get_flags(netdev, &old_flags); + error = get_flags(netdev, fd, &old_flags); if (error) { return error; } - new_flags = old_flags & ~(IFF_UP | IFF_PROMISC); - if (nd_flags & NETDEV_UP) { - new_flags |= IFF_UP; - } - if (nd_flags & NETDEV_PROMISC) { - new_flags |= IFF_PROMISC; + 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, new_flags); + error = set_flags(netdev, fd, new_flags); } return error; } +/* Sets the flags for 'netdev' to 'flags'. + * If 'permanent' is true, the changes will persist; otherwise, they + * will be reverted when 'netdev' is closed or the program exits. + * Returns 0 if successful, otherwise a positive errno value. */ +int +netdev_set_flags(struct netdev *netdev, enum netdev_flags flags, + bool permanent) +{ + return do_update_flags(netdev, netdev->netdev_fd, -1, flags, permanent); +} + +/* Turns on the specified 'flags' on 'netdev'. + * If 'permanent' is true, the changes will persist; otherwise, they + * will be reverted when 'netdev' is closed or the program exits. + * Returns 0 if successful, otherwise a positive errno value. */ +int +netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags, + bool permanent) +{ + return do_update_flags(netdev, netdev->netdev_fd, 0, flags, permanent); +} + +/* Turns off the specified 'flags' on 'netdev'. + * If 'permanent' is true, the changes will persist; otherwise, they + * will be reverted when 'netdev' is closed or the program exits. + * Returns 0 if successful, otherwise a positive errno value. */ +int +netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags, + bool permanent) +{ + return do_update_flags(netdev, netdev->netdev_fd, flags, 0, 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, @@ -625,11 +944,32 @@ netdev_arp_lookup(const struct netdev *netdev, if (!retval) { memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); } else if (retval != ENXIO) { - VLOG_WARN("%s: could not look up ARP entry for "IP_FMT": %s", - netdev->name, IP_ARGS(&ip), strerror(retval)); + VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", + netdev->name, IP_ARGS(&ip), strerror(retval)); } return retval; } + +/* Initializes 'svec' with a list of the names of all known network devices. */ +void +netdev_enumerate(struct svec *svec) +{ + 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)); + } +} static void restore_all_flags(void *aux); @@ -641,10 +981,10 @@ init_netdev(void) static bool inited; if (!inited) { inited = true; - fatal_signal_add_hook(restore_all_flags, NULL); + fatal_signal_add_hook(restore_all_flags, NULL, true); af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); if (af_inet_sock < 0) { - fatal(errno, "socket(AF_INET)"); + ofp_fatal(errno, "socket(AF_INET)"); } } } @@ -658,18 +998,20 @@ 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); - if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) { + if (ioctl(netdev->netdev_fd, SIOCGIFFLAGS, &ifr) < 0) { return errno; } /* Restore flags that we might have changed, if necessary. */ - if ((ifr.ifr_flags ^ netdev->save_flags) & (IFF_PROMISC | IFF_UP)) { - ifr.ifr_flags &= ~(IFF_PROMISC | IFF_UP); - ifr.ifr_flags |= netdev->save_flags & (IFF_PROMISC | IFF_UP); - if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) { + 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; + if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) { return errno; } } @@ -689,11 +1031,11 @@ restore_all_flags(void *aux UNUSED) } static int -get_flags(const struct netdev *netdev, int *flags) +get_flags(const struct netdev *netdev, int fd, int *flags) { struct ifreq ifr; strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); - if (ioctl(netdev->fd, SIOCGIFFLAGS, &ifr) < 0) { + if (ioctl(fd, SIOCGIFFLAGS, &ifr) < 0) { VLOG_ERR("ioctl(SIOCGIFFLAGS) on %s device failed: %s", netdev->name, strerror(errno)); return errno; @@ -703,12 +1045,12 @@ get_flags(const struct netdev *netdev, int *flags) } static int -set_flags(struct netdev *netdev, int flags) +set_flags(struct netdev *netdev, int fd, int flags) { struct ifreq ifr; strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); ifr.ifr_flags = flags; - if (ioctl(netdev->fd, SIOCSIFFLAGS, &ifr) < 0) { + if (ioctl(netdev->netdev_fd, SIOCSIFFLAGS, &ifr) < 0) { VLOG_ERR("ioctl(SIOCSIFFLAGS) on %s device failed: %s", netdev->name, strerror(errno)); return errno;