X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=5c7065fe1ab31309be55567ef9d10c5372b683ec;hb=3e912ffcbb;hp=b548996b9fbc487b7a101ca1ea05a68ba1f37f9b;hpb=b5d57fc87925cb3c029de19d0a94de5ca07ae28e;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index b548996b9..5c7065fe1 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. + * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -22,6 +22,7 @@ #include #include #include +#include #include #include #include @@ -32,7 +33,6 @@ #include #include #include -#include #include #include #include @@ -49,6 +49,7 @@ #include "coverage.h" #include "dpif-linux.h" +#include "dpif-netdev.h" #include "dynamic-string.h" #include "fatal-signal.h" #include "hash.h" @@ -60,6 +61,7 @@ #include "netlink.h" #include "ofpbuf.h" #include "openflow/openflow.h" +#include "ovs-atomic.h" #include "packets.h" #include "poll-loop.h" #include "rtnetlink-link.h" @@ -105,8 +107,35 @@ COVERAGE_DEFINE(netdev_set_ethtool); #define TC_RTAB_SIZE 1024 #endif -static struct nln_notifier *netdev_linux_cache_notifier = NULL; -static int cache_notifier_refcount; +/* Linux 2.6.21 introduced struct tpacket_auxdata. + * Linux 2.6.27 added the tp_vlan_tci member. + * Linux 3.0 defined TP_STATUS_VLAN_VALID. + * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined + * TP_STATUS_VLAN_TPID_VALID. + * + * With all this churn it's easiest to unconditionally define a replacement + * structure that has everything we want. + */ +#ifndef PACKET_AUXDATA +#define PACKET_AUXDATA 8 +#endif +#ifndef TP_STATUS_VLAN_VALID +#define TP_STATUS_VLAN_VALID (1 << 4) +#endif +#ifndef TP_STATUS_VLAN_TPID_VALID +#define TP_STATUS_VLAN_TPID_VALID (1 << 6) +#endif +#undef tpacket_auxdata +#define tpacket_auxdata rpl_tpacket_auxdata +struct tpacket_auxdata { + uint32_t tp_status; + uint32_t tp_len; + uint32_t tp_snaplen; + uint16_t tp_mac; + uint16_t tp_net; + uint16_t tp_vlan_tci; + uint16_t tp_vlan_tpid; +}; enum { VALID_IFINDEX = 1 << 0, @@ -119,10 +148,6 @@ enum { VALID_DRVINFO = 1 << 7, VALID_FEATURES = 1 << 8, }; - -struct tap_state { - int fd; -}; /* Traffic control. */ @@ -147,6 +172,7 @@ struct tc { struct tc_queue { struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ unsigned int queue_id; /* OpenFlow queue ID. */ + long long int created; /* Time queue was created, in msecs. */ }; /* A particular kind of traffic control. Each implementation generally maps to @@ -357,9 +383,10 @@ static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); struct netdev_linux { struct netdev up; - struct shash_node *shash_node; + /* Protects all members below. */ + struct ovs_mutex mutex; + unsigned int cache_valid; - unsigned int change_seq; bool miimon; /* Link status of last poll. */ long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ @@ -387,44 +414,40 @@ struct netdev_linux { enum netdev_features current; /* Cached from ETHTOOL_GSET. */ enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ - enum netdev_features peer; /* Cached from ETHTOOL_GSET. */ struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ struct tc *tc; - union { - struct tap_state tap; - } state; + /* For devices of class netdev_tap_class only. */ + int tap_fd; }; -struct netdev_rx_linux { - struct netdev_rx up; +struct netdev_rxq_linux { + struct netdev_rxq up; bool is_tap; int fd; }; -static const struct netdev_rx_class netdev_rx_linux_class; - -/* Sockets used for ioctl operations. */ -static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */ - -/* A Netlink routing socket that is not subscribed to any multicast groups. */ -static struct nl_sock *rtnl_sock; - /* This is set pretty low because we probably won't learn anything from the * additional log messages. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); -static int netdev_linux_init(void); +/* Polling miimon status for all ports causes performance degradation when + * handling a large number of ports. If there are no devices using miimon, then + * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). */ +static atomic_int miimon_cnt = ATOMIC_VAR_INIT(0); + +static void netdev_linux_run(void); static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, int cmd, const char *cmd_name); -static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, - const char *cmd_name); static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, int cmd, const char *cmd_name); static int get_flags(const struct netdev *, unsigned int *flags); static int set_flags(const char *, unsigned int flags); +static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) + OVS_REQUIRES(netdev->mutex); static int do_get_ifindex(const char *netdev_name); static int get_ifindex(const struct netdev *, int *ifindexp); static int do_set_addr(struct netdev *netdev, @@ -432,16 +455,17 @@ static int do_set_addr(struct netdev *netdev, 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, const uint8_t[ETH_ADDR_LEN]); -static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); -static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); +static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *); static int af_packet_sock(void); +static bool netdev_linux_miimon_enabled(void); static void netdev_linux_miimon_run(void); static void netdev_linux_miimon_wait(void); +static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup); static bool is_netdev_linux_class(const struct netdev_class *netdev_class) { - return netdev_class->init == netdev_linux_init; + return netdev_class->run == netdev_linux_run; } static bool @@ -458,59 +482,139 @@ netdev_linux_cast(const struct netdev *netdev) return CONTAINER_OF(netdev, struct netdev_linux, up); } -static struct netdev_rx_linux * -netdev_rx_linux_cast(const struct netdev_rx *rx) +static struct netdev_rxq_linux * +netdev_rxq_linux_cast(const struct netdev_rxq *rx) { - netdev_rx_assert_class(rx, &netdev_rx_linux_class); - return CONTAINER_OF(rx, struct netdev_rx_linux, up); + ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev))); + return CONTAINER_OF(rx, struct netdev_rxq_linux, up); } -static int -netdev_linux_init(void) -{ - static int status = -1; - if (status < 0) { - /* Create AF_INET socket. */ - 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)); - } +static void netdev_linux_update(struct netdev_linux *netdev, + const struct rtnetlink_link_change *) + OVS_REQUIRES(netdev->mutex); +static void netdev_linux_changed(struct netdev_linux *netdev, + unsigned int ifi_flags, unsigned int mask) + OVS_REQUIRES(netdev->mutex); + +/* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK changes, or NULL + * if no such socket could be created. */ +static struct nl_sock * +netdev_linux_notify_sock(void) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static struct nl_sock *sock; + + if (ovsthread_once_start(&once)) { + int error; - /* Create rtnetlink socket. */ - if (!status) { - status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); - if (status) { - VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", - strerror(status)); + error = nl_sock_create(NETLINK_ROUTE, &sock); + if (!error) { + error = nl_sock_join_mcgroup(sock, RTNLGRP_LINK); + if (error) { + nl_sock_destroy(sock); + sock = NULL; } } + ovsthread_once_done(&once); } - return status; + + return sock; +} + +static bool +netdev_linux_miimon_enabled(void) +{ + int miimon; + + atomic_read(&miimon_cnt, &miimon); + return miimon > 0; } static void netdev_linux_run(void) { - rtnetlink_link_run(); - netdev_linux_miimon_run(); + struct nl_sock *sock; + int error; + + if (netdev_linux_miimon_enabled()) { + netdev_linux_miimon_run(); + } + + sock = netdev_linux_notify_sock(); + if (!sock) { + return; + } + + do { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + uint64_t buf_stub[4096 / 8]; + struct ofpbuf buf; + + ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); + error = nl_sock_recv(sock, &buf, false); + if (!error) { + struct rtnetlink_link_change change; + + if (rtnetlink_link_parse(&buf, &change)) { + struct netdev *netdev_ = netdev_from_name(change.ifname); + if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); + netdev_linux_update(netdev, &change); + ovs_mutex_unlock(&netdev->mutex); + } + netdev_close(netdev_); + } + } else if (error == ENOBUFS) { + struct shash device_shash; + struct shash_node *node; + + nl_sock_drain(sock); + + shash_init(&device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev *netdev_ = node->data; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + unsigned int flags; + + ovs_mutex_lock(&netdev->mutex); + get_flags(netdev_, &flags); + netdev_linux_changed(netdev, flags, 0); + ovs_mutex_unlock(&netdev->mutex); + + netdev_close(netdev_); + } + shash_destroy(&device_shash); + } else if (error != EAGAIN) { + VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)", + ovs_strerror(error)); + } + ofpbuf_uninit(&buf); + } while (!error); } static void netdev_linux_wait(void) { - rtnetlink_link_wait(); - netdev_linux_miimon_wait(); + struct nl_sock *sock; + + if (netdev_linux_miimon_enabled()) { + netdev_linux_miimon_wait(); + } + sock = netdev_linux_notify_sock(); + if (sock) { + nl_sock_wait(sock, POLLIN); + } } static void netdev_linux_changed(struct netdev_linux *dev, unsigned int ifi_flags, unsigned int mask) + OVS_REQUIRES(dev->mutex) { - dev->change_seq++; - if (!dev->change_seq) { - dev->change_seq++; - } + netdev_change_seq_changed(&dev->up); if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { dev->carrier_resets++; @@ -523,6 +627,7 @@ netdev_linux_changed(struct netdev_linux *dev, static void netdev_linux_update(struct netdev_linux *dev, const struct rtnetlink_link_change *change) + OVS_REQUIRES(dev->mutex) { if (change->nlmsg_type == RTM_NEWLINK) { /* Keep drv-info */ @@ -550,86 +655,32 @@ netdev_linux_update(struct netdev_linux *dev, } } -static void -netdev_linux_cache_cb(const struct rtnetlink_link_change *change, - void *aux OVS_UNUSED) -{ - struct netdev_linux *dev; - if (change) { - struct netdev *base_dev = netdev_from_name(change->ifname); - if (base_dev && is_netdev_linux_class(netdev_get_class(base_dev))) { - netdev_linux_update(netdev_linux_cast(base_dev), change); - } - } else { - struct shash device_shash; - struct shash_node *node; - - shash_init(&device_shash); - netdev_get_devices(&netdev_linux_class, &device_shash); - SHASH_FOR_EACH (node, &device_shash) { - unsigned int flags; - - dev = node->data; - - get_flags(&dev->up, &flags); - netdev_linux_changed(dev, flags, 0); - } - shash_destroy(&device_shash); - } -} - -static int -cache_notifier_ref(void) +static struct netdev * +netdev_linux_alloc(void) { - if (!cache_notifier_refcount) { - ovs_assert(!netdev_linux_cache_notifier); - - netdev_linux_cache_notifier = - rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL); - - if (!netdev_linux_cache_notifier) { - return EINVAL; - } - } - cache_notifier_refcount++; - - return 0; + struct netdev_linux *netdev = xzalloc(sizeof *netdev); + return &netdev->up; } static void -cache_notifier_unref(void) +netdev_linux_common_construct(struct netdev_linux *netdev) { - ovs_assert(cache_notifier_refcount > 0); - if (!--cache_notifier_refcount) { - ovs_assert(netdev_linux_cache_notifier); - rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier); - netdev_linux_cache_notifier = NULL; - } + ovs_mutex_init(&netdev->mutex); } /* Creates system and internal devices. */ static int -netdev_linux_create(const struct netdev_class *class, const char *name, - struct netdev **netdevp) +netdev_linux_construct(struct netdev *netdev_) { - struct netdev_linux *netdev; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - error = cache_notifier_ref(); - if (error) { - return error; - } + netdev_linux_common_construct(netdev); - netdev = xzalloc(sizeof *netdev); - netdev->change_seq = 1; - netdev_init(&netdev->up, name, class); error = get_flags(&netdev->up, &netdev->ifi_flags); if (error == ENODEV) { - if (class != &netdev_internal_class) { + if (netdev->up.netdev_class != &netdev_internal_class) { /* The device does not exist, so don't allow it to be opened. */ - netdev_uninit(&netdev->up, false); - cache_notifier_unref(); - free(netdev); return ENODEV; } else { /* "Internal" netdevs have to be created as netdev objects before @@ -639,7 +690,6 @@ netdev_linux_create(const struct netdev_class *class, const char *name, } } - *netdevp = &netdev->up; return 0; } @@ -650,71 +700,49 @@ netdev_linux_create(const struct netdev_class *class, const char *name, * buffers, across all readers. Therefore once data is read it will * be unavailable to other reads for tap devices. */ static int -netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, - const char *name, struct netdev **netdevp) +netdev_linux_construct_tap(struct netdev *netdev_) { - struct netdev_linux *netdev; - struct tap_state *state; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); static const char tap_dev[] = "/dev/net/tun"; + const char *name = netdev_->name; struct ifreq ifr; int error; - netdev = xzalloc(sizeof *netdev); - state = &netdev->state.tap; - - error = cache_notifier_ref(); - if (error) { - goto error; - } + netdev_linux_common_construct(netdev); /* Open tap device. */ - state->fd = open(tap_dev, O_RDWR); - if (state->fd < 0) { + netdev->tap_fd = open(tap_dev, O_RDWR); + if (netdev->tap_fd < 0) { error = errno; - VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); - goto error_unref_notifier; + VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); + return error; } /* Create tap device. */ ifr.ifr_flags = IFF_TAP | IFF_NO_PI; ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { + if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { VLOG_WARN("%s: creating tap device failed: %s", name, - strerror(errno)); + ovs_strerror(errno)); error = errno; - goto error_unref_notifier; + goto error_close; } /* Make non-blocking. */ - error = set_nonblocking(state->fd); + error = set_nonblocking(netdev->tap_fd); if (error) { - goto error_unref_notifier; + goto error_close; } - netdev_init(&netdev->up, name, &netdev_tap_class); - *netdevp = &netdev->up; return 0; -error_unref_notifier: - cache_notifier_unref(); -error: - free(netdev); +error_close: + close(netdev->tap_fd); return error; } static void -destroy_tap(struct netdev_linux *netdev) -{ - struct tap_state *state = &netdev->state.tap; - - if (state->fd >= 0) { - close(state->fd); - } -} - -/* Destroys the netdev device 'netdev_'. */ -static void -netdev_linux_destroy(struct netdev *netdev_) +netdev_linux_destruct(struct netdev *netdev_) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); @@ -722,39 +750,78 @@ netdev_linux_destroy(struct netdev *netdev_) netdev->tc->ops->tc_destroy(netdev->tc); } - if (netdev_get_class(netdev_) == &netdev_tap_class) { - destroy_tap(netdev); + if (netdev_get_class(netdev_) == &netdev_tap_class + && netdev->tap_fd >= 0) + { + close(netdev->tap_fd); + } + + if (netdev->miimon_interval > 0) { + int junk; + atomic_sub(&miimon_cnt, 1, &junk); } + + ovs_mutex_destroy(&netdev->mutex); +} + +static void +netdev_linux_dealloc(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); free(netdev); +} - cache_notifier_unref(); +static struct netdev_rxq * +netdev_linux_rxq_alloc(void) +{ + struct netdev_rxq_linux *rx = xzalloc(sizeof *rx); + return &rx->up; } static int -netdev_linux_rx_open(struct netdev *netdev_, struct netdev_rx **rxp) +netdev_linux_rxq_construct(struct netdev_rxq *rxq_) { + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + struct netdev *netdev_ = rx->up.netdev; struct netdev_linux *netdev = netdev_linux_cast(netdev_); - bool is_tap = is_tap_netdev(netdev_); - struct netdev_rx_linux *rx; int error; - int fd; - if (is_tap) { - fd = netdev->state.tap.fd; + ovs_mutex_lock(&netdev->mutex); + rx->is_tap = is_tap_netdev(netdev_); + if (rx->is_tap) { + rx->fd = netdev->tap_fd; } else { struct sockaddr_ll sll; - int ifindex; + int ifindex, val; + /* Result of tcpdump -dd inbound */ + static const struct sock_filter filt[] = { + { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ + { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ + { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ + { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ + }; + static const struct sock_fprog fprog = { + ARRAY_SIZE(filt), (struct sock_filter *) filt + }; /* Create file descriptor. */ - fd = socket(PF_PACKET, SOCK_RAW, 0); - if (fd < 0) { + rx->fd = socket(PF_PACKET, SOCK_RAW, 0); + if (rx->fd < 0) { + error = errno; + VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); + goto error; + } + + val = 1; + if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) { error = errno; - VLOG_ERR("failed to create raw socket (%s)", strerror(error)); + VLOG_ERR("%s: failed to mark socket for auxdata (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); goto error; } /* Set non-blocking mode. */ - error = set_nonblocking(fd); + error = set_nonblocking(rx->fd); if (error) { goto error; } @@ -769,81 +836,202 @@ netdev_linux_rx_open(struct netdev *netdev_, struct netdev_rx **rxp) memset(&sll, 0, sizeof sll); sll.sll_family = AF_PACKET; sll.sll_ifindex = ifindex; - sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL); - if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) { + sll.sll_protocol = htons(ETH_P_ALL); + if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { error = errno; VLOG_ERR("%s: failed to bind raw socket (%s)", - netdev_get_name(netdev_), strerror(error)); + netdev_get_name(netdev_), ovs_strerror(error)); goto error; } - } - rx = xmalloc(sizeof *rx); - netdev_rx_init(&rx->up, netdev_, &netdev_rx_linux_class); - rx->is_tap = is_tap; - rx->fd = fd; + /* Filter for only inbound packets. */ + error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, + sizeof fprog); + if (error) { + error = errno; + VLOG_ERR("%s: failed to attach filter (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); + goto error; + } + } + ovs_mutex_unlock(&netdev->mutex); - *rxp = &rx->up; return 0; error: - if (fd >= 0) { - close(fd); + if (rx->fd >= 0) { + close(rx->fd); } + ovs_mutex_unlock(&netdev->mutex); return error; } static void -netdev_rx_linux_destroy(struct netdev_rx *rx_) +netdev_linux_rxq_destruct(struct netdev_rxq *rxq_) { - struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); if (!rx->is_tap) { close(rx->fd); } +} + +static void +netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_) +{ + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + free(rx); } +static ovs_be16 +auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux) +{ + if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) { + return htons(aux->tp_vlan_tpid); + } else { + return htons(ETH_TYPE_VLAN); + } +} + +static bool +auxdata_has_vlan_tci(const struct tpacket_auxdata *aux) +{ + return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID; +} + static int -netdev_rx_linux_recv(struct netdev_rx *rx_, void *data, size_t size) +netdev_linux_rxq_recv_sock(int fd, struct ofpbuf *buffer) { - struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + size_t size; ssize_t retval; + struct iovec iov; + struct cmsghdr *cmsg; + union { + struct cmsghdr cmsg; + char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; + } cmsg_buffer; + struct msghdr msgh; + + /* Reserve headroom for a single VLAN tag */ + ofpbuf_reserve(buffer, VLAN_HEADER_LEN); + size = ofpbuf_tailroom(buffer); + + iov.iov_base = ofpbuf_data(buffer); + iov.iov_len = size; + msgh.msg_name = NULL; + msgh.msg_namelen = 0; + msgh.msg_iov = &iov; + msgh.msg_iovlen = 1; + msgh.msg_control = &cmsg_buffer; + msgh.msg_controllen = sizeof cmsg_buffer; + msgh.msg_flags = 0; do { - retval = (rx->is_tap - ? read(rx->fd, data, size) - : recv(rx->fd, data, size, MSG_TRUNC)); + retval = recvmsg(fd, &msgh, MSG_TRUNC); } while (retval < 0 && errno == EINTR); - if (retval > size) { - return -EMSGSIZE; - } else if (retval >= 0) { - return retval; - } else { - if (errno != EAGAIN) { + if (retval < 0) { + return errno; + } else if (retval > size) { + return EMSGSIZE; + } + + ofpbuf_set_size(buffer, ofpbuf_size(buffer) + retval); + + for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) { + const struct tpacket_auxdata *aux; + + if (cmsg->cmsg_level != SOL_PACKET + || cmsg->cmsg_type != PACKET_AUXDATA + || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) { + continue; + } + + aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg)); + if (auxdata_has_vlan_tci(aux)) { + if (retval < ETH_HEADER_LEN) { + return EINVAL; + } + + eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux), + htons(aux->tp_vlan_tci)); + break; + } + } + + return 0; +} + +static int +netdev_linux_rxq_recv_tap(int fd, struct ofpbuf *buffer) +{ + ssize_t retval; + size_t size = ofpbuf_tailroom(buffer); + + do { + retval = read(fd, ofpbuf_data(buffer), size); + } while (retval < 0 && errno == EINTR); + + if (retval < 0) { + return errno; + } else if (retval > size) { + return EMSGSIZE; + } + + ofpbuf_set_size(buffer, ofpbuf_size(buffer) + retval); + return 0; +} + +static int +netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct ofpbuf **packet, int *c) +{ + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); + struct netdev *netdev = rx->up.netdev; + struct ofpbuf *buffer; + ssize_t retval; + int mtu; + + if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { + mtu = ETH_PAYLOAD_MAX; + } + + buffer = ofpbuf_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu, DP_NETDEV_HEADROOM); + + retval = (rx->is_tap + ? netdev_linux_rxq_recv_tap(rx->fd, buffer) + : netdev_linux_rxq_recv_sock(rx->fd, buffer)); + + if (retval) { + if (retval != EAGAIN && retval != EMSGSIZE) { VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", - strerror(errno), netdev_rx_get_name(rx_)); + ovs_strerror(errno), netdev_rxq_get_name(rxq_)); } - return -errno; + ofpbuf_delete(buffer); + } else { + dp_packet_pad(buffer); + packet[0] = buffer; + *c = 1; } + + return retval; } static void -netdev_rx_linux_wait(struct netdev_rx *rx_) +netdev_linux_rxq_wait(struct netdev_rxq *rxq_) { - struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); poll_fd_wait(rx->fd, POLLIN); } static int -netdev_rx_linux_drain(struct netdev_rx *rx_) +netdev_linux_rxq_drain(struct netdev_rxq *rxq_) { - struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); if (rx->is_tap) { struct ifreq ifr; - int error = netdev_linux_do_ioctl(netdev_rx_get_name(rx_), &ifr, - SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); + int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr, + SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); if (error) { return error; } @@ -864,8 +1052,11 @@ netdev_rx_linux_drain(struct netdev_rx *rx_) * The kernel maintains a packet transmission queue, so the caller is not * expected to do additional queuing of packets. */ static int -netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) +netdev_linux_send(struct netdev *netdev_, struct ofpbuf *pkt, bool may_steal) { + const void *data = ofpbuf_data(pkt); + size_t size = ofpbuf_size(pkt); + for (;;) { ssize_t retval; @@ -875,7 +1066,6 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) struct msghdr msg; struct iovec iov; int ifindex; - int error; int sock; sock = af_packet_sock(); @@ -883,9 +1073,9 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) return -sock; } - error = get_ifindex(netdev_, &ifindex); - if (error) { - return error; + ifindex = netdev_get_ifindex(netdev_); + if (ifindex < 0) { + return -ifindex; } /* We don't bother setting most fields in sockaddr_ll because the @@ -910,10 +1100,15 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) /* Use the tap fd to send to this device. This is essential for * tap devices, because packets sent to a tap device with an * AF_PACKET socket will loop back to be *received* again on the - * tap device. */ + * tap device. This doesn't occur on other interface types + * because we attach a socket filter to the rx socket. */ struct netdev_linux *netdev = netdev_linux_cast(netdev_); - retval = write(netdev->state.tap.fd, data, size); + retval = write(netdev->tap_fd, data, size); + } + + if (may_steal) { + ofpbuf_delete(pkt); } if (retval < 0) { @@ -926,12 +1121,12 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) continue; } else if (errno != EAGAIN) { VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", - netdev_get_name(netdev_), strerror(errno)); + netdev_get_name(netdev_), ovs_strerror(errno)); } return errno; } else if (retval != size) { - VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " - "%zu) on %s", retval, size, netdev_get_name(netdev_)); + VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE"d bytes of " + "%"PRIuSIZE") on %s", retval, size, netdev_get_name(netdev_)); return EMSGSIZE; } else { return 0; @@ -962,26 +1157,22 @@ netdev_linux_set_etheraddr(struct netdev *netdev_, const uint8_t mac[ETH_ADDR_LEN]) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - struct netdev_saved_flags *sf = NULL; + enum netdev_flags old_flags = 0; int error; + ovs_mutex_lock(&netdev->mutex); + if (netdev->cache_valid & VALID_ETHERADDR) { - if (netdev->ether_addr_error) { - return netdev->ether_addr_error; - } - if (eth_addr_equals(netdev->etheraddr, mac)) { - return 0; + error = netdev->ether_addr_error; + if (error || eth_addr_equals(netdev->etheraddr, mac)) { + goto exit; } netdev->cache_valid &= ~VALID_ETHERADDR; } /* Tap devices must be brought down before setting the address. */ if (is_tap_netdev(netdev_)) { - enum netdev_flags flags; - - if (!netdev_get_flags(netdev_, &flags) && (flags & NETDEV_UP)) { - netdev_turn_flags_off(netdev_, NETDEV_UP, &sf); - } + update_flags(netdev, NETDEV_UP, 0, &old_flags); } error = set_etheraddr(netdev_get_name(netdev_), mac); if (!error || error == ENODEV) { @@ -992,8 +1183,12 @@ netdev_linux_set_etheraddr(struct netdev *netdev_, } } - netdev_restore_flags(sf); + if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { + update_flags(netdev, 0, NETDEV_UP, &old_flags); + } +exit: + ovs_mutex_unlock(&netdev->mutex); return error; } @@ -1003,45 +1198,60 @@ netdev_linux_get_etheraddr(const struct netdev *netdev_, uint8_t mac[ETH_ADDR_LEN]) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + ovs_mutex_lock(&netdev->mutex); if (!(netdev->cache_valid & VALID_ETHERADDR)) { - int error = get_etheraddr(netdev_get_name(netdev_), - netdev->etheraddr); - - netdev->ether_addr_error = error; + netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_), + netdev->etheraddr); netdev->cache_valid |= VALID_ETHERADDR; } - if (!netdev->ether_addr_error) { + error = netdev->ether_addr_error; + if (!error) { memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN); } + ovs_mutex_unlock(&netdev->mutex); - return netdev->ether_addr_error; + return error; } -/* 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) +netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + if (!(netdev->cache_valid & VALID_MTU)) { struct ifreq ifr; - int error; - error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, - SIOCGIFMTU, "SIOCGIFMTU"); - - netdev->netdev_mtu_error = error; + netdev->netdev_mtu_error = af_inet_ifreq_ioctl( + netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); netdev->mtu = ifr.ifr_mtu; netdev->cache_valid |= VALID_MTU; } - if (!netdev->netdev_mtu_error) { + error = netdev->netdev_mtu_error; + if (!error) { *mtup = netdev->mtu; } - return netdev->netdev_mtu_error; + + return error; +} + +/* 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_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = netdev_linux_get_mtu__(netdev, mtup); + ovs_mutex_unlock(&netdev->mutex); + + return error; } /* Sets the maximum size of transmitted (MTU) for given device using linux @@ -1054,34 +1264,39 @@ netdev_linux_set_mtu(const struct netdev *netdev_, int mtu) struct ifreq ifr; int error; + ovs_mutex_lock(&netdev->mutex); if (netdev->cache_valid & VALID_MTU) { - if (netdev->netdev_mtu_error) { - return netdev->netdev_mtu_error; - } - if (netdev->mtu == mtu) { - return 0; + error = netdev->netdev_mtu_error; + if (error || netdev->mtu == mtu) { + goto exit; } netdev->cache_valid &= ~VALID_MTU; } ifr.ifr_mtu = mtu; - error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, - SIOCSIFMTU, "SIOCSIFMTU"); + error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, + SIOCSIFMTU, "SIOCSIFMTU"); if (!error || error == ENODEV) { netdev->netdev_mtu_error = error; netdev->mtu = ifr.ifr_mtu; netdev->cache_valid |= VALID_MTU; } +exit: + ovs_mutex_unlock(&netdev->mutex); return error; } /* Returns the ifindex of 'netdev', if successful, as a positive number. * On failure, returns a negative errno value. */ static int -netdev_linux_get_ifindex(const struct netdev *netdev) +netdev_linux_get_ifindex(const struct netdev *netdev_) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int ifindex, error; - error = get_ifindex(netdev, &ifindex); + ovs_mutex_lock(&netdev->mutex); + error = get_ifindex(netdev_, &ifindex); + ovs_mutex_unlock(&netdev->mutex); + return error ? -error : ifindex; } @@ -1090,19 +1305,28 @@ netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + ovs_mutex_lock(&netdev->mutex); if (netdev->miimon_interval > 0) { *carrier = netdev->miimon; } else { *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; } + ovs_mutex_unlock(&netdev->mutex); return 0; } static long long int -netdev_linux_get_carrier_resets(const struct netdev *netdev) +netdev_linux_get_carrier_resets(const struct netdev *netdev_) { - return netdev_linux_cast(netdev)->carrier_resets; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + long long int carrier_resets; + + ovs_mutex_lock(&netdev->mutex); + carrier_resets = netdev->carrier_resets; + ovs_mutex_unlock(&netdev->mutex); + + return carrier_resets; } static int @@ -1114,7 +1338,7 @@ netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, memset(&ifr, 0, sizeof ifr); memcpy(&ifr.ifr_data, data, sizeof *data); - error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name); + error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); memcpy(data, &ifr.ifr_data, sizeof *data); return error; @@ -1170,11 +1394,21 @@ netdev_linux_set_miimon_interval(struct netdev *netdev_, { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + ovs_mutex_lock(&netdev->mutex); interval = interval > 0 ? MAX(interval, 100) : 0; if (netdev->miimon_interval != interval) { + int junk; + + if (interval && !netdev->miimon_interval) { + atomic_add(&miimon_cnt, 1, &junk); + } else if (!interval && netdev->miimon_interval) { + atomic_sub(&miimon_cnt, 1, &junk); + } + netdev->miimon_interval = interval; timer_set_expired(&netdev->miimon_timer); } + ovs_mutex_unlock(&netdev->mutex); return 0; } @@ -1188,20 +1422,22 @@ netdev_linux_miimon_run(void) shash_init(&device_shash); netdev_get_devices(&netdev_linux_class, &device_shash); SHASH_FOR_EACH (node, &device_shash) { - struct netdev_linux *dev = node->data; + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); bool miimon; - if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) { - continue; - } + ovs_mutex_lock(&dev->mutex); + if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) { + netdev_linux_get_miimon(dev->up.name, &miimon); + if (miimon != dev->miimon) { + dev->miimon = miimon; + netdev_linux_changed(dev, dev->ifi_flags, 0); + } - netdev_linux_get_miimon(dev->up.name, &miimon); - if (miimon != dev->miimon) { - dev->miimon = miimon; - netdev_linux_changed(dev, dev->ifi_flags, 0); + timer_set_duration(&dev->miimon_timer, dev->miimon_interval); } - - timer_set_duration(&dev->miimon_timer, dev->miimon_interval); + ovs_mutex_unlock(&dev->mutex); + netdev_close(netdev); } shash_destroy(&device_shash); @@ -1216,43 +1452,19 @@ netdev_linux_miimon_wait(void) shash_init(&device_shash); netdev_get_devices(&netdev_linux_class, &device_shash); SHASH_FOR_EACH (node, &device_shash) { - struct netdev_linux *dev = node->data; + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); + ovs_mutex_lock(&dev->mutex); if (dev->miimon_interval > 0) { timer_wait(&dev->miimon_timer); } + ovs_mutex_unlock(&dev->mutex); + netdev_close(netdev); } shash_destroy(&device_shash); } -/* 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; - } - } -} - static void swap_uint64(uint64_t *a, uint64_t *b) { @@ -1326,43 +1538,14 @@ get_stats_via_vport(const struct netdev *netdev_, error = get_stats_via_vport__(netdev_, stats); if (error && error != ENOENT) { VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " - "(%s)", netdev_get_name(netdev_), strerror(error)); + "(%s)", + netdev_get_name(netdev_), ovs_strerror(error)); } netdev->vport_stats_error = error; netdev->cache_valid |= VALID_VPORT_STAT_ERROR; } } -static int -netdev_linux_sys_get_stats(const struct netdev *netdev_, - struct netdev_stats *stats) -{ - static int use_netlink_stats = -1; - int error; - - 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_get_name(netdev_), stats); - } - - if (error) { - VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d", - netdev_get_name(netdev_), error); - } - return error; - -} - /* Retrieves current device stats for 'netdev-linux'. */ static int netdev_linux_get_stats(const struct netdev *netdev_, @@ -1372,19 +1555,14 @@ netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats dev_stats; int error; + ovs_mutex_lock(&netdev->mutex); get_stats_via_vport(netdev_, stats); - - error = netdev_linux_sys_get_stats(netdev_, &dev_stats); - + error = get_stats_via_netlink(netdev_, &dev_stats); if (error) { - if (netdev->vport_stats_error) { - return error; - } else { - return 0; + if (!netdev->vport_stats_error) { + error = 0; } - } - - if (netdev->vport_stats_error) { + } else if (netdev->vport_stats_error) { /* stats not available from OVS then use ioctl stats. */ *stats = dev_stats; } else { @@ -1406,37 +1584,34 @@ netdev_linux_get_stats(const struct netdev *netdev_, stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; stats->tx_window_errors += dev_stats.tx_window_errors; } - return 0; + ovs_mutex_unlock(&netdev->mutex); + + return error; } /* Retrieves current device stats for 'netdev-tap' netdev or * netdev-internal. */ static int -netdev_tap_get_stats(const struct netdev *netdev_, - struct netdev_stats *stats) +netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct netdev_stats dev_stats; int error; + ovs_mutex_lock(&netdev->mutex); get_stats_via_vport(netdev_, stats); - - error = netdev_linux_sys_get_stats(netdev_, &dev_stats); + error = get_stats_via_netlink(netdev_, &dev_stats); if (error) { - if (netdev->vport_stats_error) { - return error; - } else { - return 0; + if (!netdev->vport_stats_error) { + error = 0; } - } + } else if (netdev->vport_stats_error) { + /* Transmit and receive stats will appear to be swapped relative to the + * other ports since we are the one sending the data, not a remote + * computer. For consistency, we swap them back here. This does not + * apply if we are getting stats from the vport layer because it always + * tracks stats from the perspective of the switch. */ - /* If this port is an internal port then the transmit and receive stats - * will appear to be swapped relative to the other ports since we are the - * one sending the data, not a remote computer. For consistency, we swap - * them back here. This does not apply if we are getting stats from the - * vport layer because it always tracks stats from the perspective of the - * switch. */ - if (netdev->vport_stats_error) { *stats = dev_stats; swap_uint64(&stats->rx_packets, &stats->tx_packets); swap_uint64(&stats->rx_bytes, &stats->tx_bytes); @@ -1463,7 +1638,9 @@ netdev_tap_get_stats(const struct netdev *netdev_, stats->multicast += dev_stats.multicast; stats->collisions += dev_stats.collisions; } - return 0; + ovs_mutex_unlock(&netdev->mutex); + + return error; } static int @@ -1471,9 +1648,14 @@ netdev_internal_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + ovs_mutex_lock(&netdev->mutex); get_stats_via_vport(netdev_, stats); - return netdev->vport_stats_error; + error = netdev->vport_stats_error; + ovs_mutex_unlock(&netdev->mutex); + + return error; } static int @@ -1638,18 +1820,14 @@ netdev_linux_read_features(struct netdev_linux *netdev) netdev->current |= NETDEV_F_AUTONEG; } - /* Peer advertisements. */ - netdev->peer = 0; /* XXX */ - out: netdev->cache_valid |= VALID_FEATURES; netdev->get_features_error = 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 NETDEV_* bits. Returns 0 if successful, otherwise a positive - * errno value. */ +/* Stores the features supported by 'netdev' into of '*current', '*advertised', + * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits. + * Returns 0 if successful, otherwise a positive errno value. */ static int netdev_linux_get_features(const struct netdev *netdev_, enum netdev_features *current, @@ -1658,32 +1836,39 @@ netdev_linux_get_features(const struct netdev *netdev_, enum netdev_features *peer) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + ovs_mutex_lock(&netdev->mutex); netdev_linux_read_features(netdev); - if (!netdev->get_features_error) { *current = netdev->current; *advertised = netdev->advertised; *supported = netdev->supported; - *peer = netdev->peer; + *peer = 0; /* XXX */ } - return netdev->get_features_error; + error = netdev->get_features_error; + ovs_mutex_unlock(&netdev->mutex); + + return error; } /* Set the features advertised by 'netdev' to 'advertise'. */ static int -netdev_linux_set_advertisements(struct netdev *netdev, +netdev_linux_set_advertisements(struct netdev *netdev_, enum netdev_features advertise) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct ethtool_cmd ecmd; int error; + ovs_mutex_lock(&netdev->mutex); + COVERAGE_INC(netdev_get_ethtool); memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, + error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { - return error; + goto exit; } ecmd.advertising = 0; @@ -1724,8 +1909,12 @@ netdev_linux_set_advertisements(struct netdev *netdev, ecmd.advertising |= ADVERTISED_Asym_Pause; } COVERAGE_INC(netdev_set_ethtool); - return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, - ETHTOOL_SSET, "ETHTOOL_SSET"); + error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, + ETHTOOL_SSET, "ETHTOOL_SSET"); + +exit: + ovs_mutex_unlock(&netdev->mutex); + return error; } /* Attempts to set input rate limiting (policing) policy. Returns 0 if @@ -1738,20 +1927,17 @@ netdev_linux_set_policing(struct netdev *netdev_, const char *netdev_name = netdev_get_name(netdev_); int error; - kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ : kbits_burst); /* Stick with user-specified value. */ + ovs_mutex_lock(&netdev->mutex); if (netdev->cache_valid & VALID_POLICING) { - if (netdev->netdev_policing_error) { - return netdev->netdev_policing_error; - } - - if (netdev->kbits_rate == kbits_rate && - netdev->kbits_burst == kbits_burst) { + error = netdev->netdev_policing_error; + if (error || (netdev->kbits_rate == kbits_rate && + netdev->kbits_burst == kbits_burst)) { /* Assume that settings haven't changed since we last set them. */ - return 0; + goto out; } netdev->cache_valid &= ~VALID_POLICING; } @@ -1761,7 +1947,7 @@ netdev_linux_set_policing(struct netdev *netdev_, error = tc_add_del_ingress_qdisc(netdev_, false); if (error) { VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", - netdev_name, strerror(error)); + netdev_name, ovs_strerror(error)); goto out; } @@ -1769,14 +1955,14 @@ netdev_linux_set_policing(struct netdev *netdev_, error = tc_add_del_ingress_qdisc(netdev_, true); if (error) { VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", - netdev_name, strerror(error)); + netdev_name, ovs_strerror(error)); goto out; } error = tc_add_policer(netdev_, kbits_rate, kbits_burst); if (error){ VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", - netdev_name, strerror(error)); + netdev_name, ovs_strerror(error)); goto out; } } @@ -1789,6 +1975,7 @@ out: netdev->netdev_policing_error = error; netdev->cache_valid |= VALID_POLICING; } + ovs_mutex_unlock(&netdev->mutex); return error; } @@ -1876,15 +2063,17 @@ netdev_linux_get_qos(const struct netdev *netdev_, struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; + if (!error) { + *typep = netdev->tc->ops->ovs_name; + error = (netdev->tc->ops->qdisc_get + ? netdev->tc->ops->qdisc_get(netdev_, details) + : 0); } + ovs_mutex_unlock(&netdev->mutex); - *typep = netdev->tc->ops->ovs_name; - return (netdev->tc->ops->qdisc_get - ? netdev->tc->ops->qdisc_get(netdev_, details) - : 0); + return error; } static int @@ -1900,27 +2089,30 @@ netdev_linux_set_qos(struct netdev *netdev_, return EOPNOTSUPP; } + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); if (error) { - return error; + goto exit; } if (new_ops == netdev->tc->ops) { - return new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; + error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; } else { /* Delete existing qdisc. */ error = tc_del_qdisc(netdev_); if (error) { - return error; + goto exit; } ovs_assert(netdev->tc == NULL); /* Install new qdisc. */ error = new_ops->tc_install(netdev_, details); ovs_assert((error == 0) == (netdev->tc != NULL)); - - return error; } + +exit: + ovs_mutex_unlock(&netdev->mutex); + return error; } static int @@ -1930,15 +2122,17 @@ netdev_linux_get_queue(const struct netdev *netdev_, struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; - } else { + if (!error) { struct tc_queue *queue = tc_find_queue(netdev_, queue_id); - return (queue + error = (queue ? netdev->tc->ops->class_get(netdev_, queue, details) : ENOENT); } + ovs_mutex_unlock(&netdev->mutex); + + return error; } static int @@ -1948,15 +2142,17 @@ netdev_linux_set_queue(struct netdev *netdev_, struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; - } else if (queue_id >= netdev->tc->ops->n_queues - || !netdev->tc->ops->class_set) { - return EINVAL; + if (!error) { + error = (queue_id < netdev->tc->ops->n_queues + && netdev->tc->ops->class_set + ? netdev->tc->ops->class_set(netdev_, queue_id, details) + : EINVAL); } + ovs_mutex_unlock(&netdev->mutex); - return netdev->tc->ops->class_set(netdev_, queue_id, details); + return error; } static int @@ -1965,17 +2161,21 @@ netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; - } else if (!netdev->tc->ops->class_delete) { - return EINVAL; - } else { - struct tc_queue *queue = tc_find_queue(netdev_, queue_id); - return (queue - ? netdev->tc->ops->class_delete(netdev_, queue) - : ENOENT); + if (!error) { + if (netdev->tc->ops->class_delete) { + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + error = (queue + ? netdev->tc->ops->class_delete(netdev_, queue) + : ENOENT); + } else { + error = EINVAL; + } } + ovs_mutex_unlock(&netdev->mutex); + + return error; } static int @@ -1986,21 +2186,34 @@ netdev_linux_get_queue_stats(const struct netdev *netdev_, struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; - } else if (!netdev->tc->ops->class_get_stats) { - return EOPNOTSUPP; - } else { - const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); - return (queue - ? netdev->tc->ops->class_get_stats(netdev_, queue, stats) - : ENOENT); + if (!error) { + if (netdev->tc->ops->class_get_stats) { + const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + if (queue) { + stats->created = queue->created; + error = netdev->tc->ops->class_get_stats(netdev_, queue, + stats); + } else { + error = ENOENT; + } + } else { + error = EOPNOTSUPP; + } } + ovs_mutex_unlock(&netdev->mutex); + + return error; } +struct queue_dump_state { + struct nl_dump dump; + struct ofpbuf buf; +}; + static bool -start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) +start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state) { struct ofpbuf request; struct tcmsg *tcmsg; @@ -2010,44 +2223,91 @@ start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) return false; } tcmsg->tcm_parent = 0; - nl_dump_start(dump, rtnl_sock, &request); + nl_dump_start(&state->dump, NETLINK_ROUTE, &request); ofpbuf_uninit(&request); + + ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); return true; } static int -netdev_linux_dump_queues(const struct netdev *netdev_, - netdev_dump_queues_cb *cb, void *aux) +finish_queue_dump(struct queue_dump_state *state) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - struct tc_queue *queue, *next_queue; - struct smap details; - int last_error; + ofpbuf_uninit(&state->buf); + return nl_dump_done(&state->dump); +} + +struct netdev_linux_queue_state { + unsigned int *queues; + size_t cur_queue; + size_t n_queues; +}; + +static int +netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) +{ + const struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; - } else if (!netdev->tc->ops->class_get) { - return EOPNOTSUPP; + if (!error) { + if (netdev->tc->ops->class_get) { + struct netdev_linux_queue_state *state; + struct tc_queue *queue; + size_t i; + + *statep = state = xmalloc(sizeof *state); + state->n_queues = hmap_count(&netdev->tc->queues); + state->cur_queue = 0; + state->queues = xmalloc(state->n_queues * sizeof *state->queues); + + i = 0; + HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { + state->queues[i++] = queue->queue_id; + } + } else { + error = EOPNOTSUPP; + } } + ovs_mutex_unlock(&netdev->mutex); - last_error = 0; - smap_init(&details); - HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node, - &netdev->tc->queues) { - smap_clear(&details); + return error; +} - error = netdev->tc->ops->class_get(netdev_, queue, &details); - if (!error) { - (*cb)(queue->queue_id, &details, aux); - } else { - last_error = error; +static int +netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, + unsigned int *queue_idp, struct smap *details) +{ + const struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_linux_queue_state *state = state_; + int error = EOF; + + ovs_mutex_lock(&netdev->mutex); + while (state->cur_queue < state->n_queues) { + unsigned int queue_id = state->queues[state->cur_queue++]; + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + + if (queue) { + *queue_idp = queue_id; + error = netdev->tc->ops->class_get(netdev_, queue, details); + break; } } - smap_destroy(&details); + ovs_mutex_unlock(&netdev->mutex); - return last_error; + return error; +} + +static int +netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, + void *state_) +{ + struct netdev_linux_queue_state *state = state_; + + free(state->queues); + free(state); + return 0; } static int @@ -2055,31 +2315,38 @@ netdev_linux_dump_queue_stats(const struct netdev *netdev_, netdev_dump_queue_stats_cb *cb, void *aux) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - struct nl_dump dump; - struct ofpbuf msg; - int last_error; int error; + ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); - if (error) { - return error; - } else if (!netdev->tc->ops->class_dump_stats) { - return EOPNOTSUPP; - } + if (!error) { + struct queue_dump_state state; - last_error = 0; - if (!start_queue_dump(netdev_, &dump)) { - return ENODEV; - } - while (nl_dump_next(&dump, &msg)) { - error = netdev->tc->ops->class_dump_stats(netdev_, &msg, cb, aux); - if (error) { - last_error = error; + if (!netdev->tc->ops->class_dump_stats) { + error = EOPNOTSUPP; + } else if (!start_queue_dump(netdev_, &state)) { + error = ENODEV; + } else { + struct ofpbuf msg; + int retval; + + while (nl_dump_next(&state.dump, &msg, &state.buf)) { + retval = netdev->tc->ops->class_dump_stats(netdev_, &msg, + cb, aux); + if (retval) { + error = retval; + } + } + + retval = finish_queue_dump(&state); + if (retval) { + error = retval; + } } } + ovs_mutex_unlock(&netdev->mutex); - error = nl_dump_done(&dump); - return error ? error : last_error; + return error; } static int @@ -2087,27 +2354,34 @@ netdev_linux_get_in4(const struct netdev *netdev_, struct in_addr *address, struct in_addr *netmask) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + ovs_mutex_lock(&netdev->mutex); if (!(netdev->cache_valid & VALID_IN4)) { - int error; - error = netdev_linux_get_ipv4(netdev_, &netdev->address, SIOCGIFADDR, "SIOCGIFADDR"); - if (error) { - return error; + if (!error) { + error = netdev_linux_get_ipv4(netdev_, &netdev->netmask, + SIOCGIFNETMASK, "SIOCGIFNETMASK"); + if (!error) { + netdev->cache_valid |= VALID_IN4; + } } + } else { + error = 0; + } - error = netdev_linux_get_ipv4(netdev_, &netdev->netmask, - SIOCGIFNETMASK, "SIOCGIFNETMASK"); - if (error) { - return error; + if (!error) { + if (netdev->address.s_addr != INADDR_ANY) { + *address = netdev->address; + *netmask = netdev->netmask; + } else { + error = EADDRNOTAVAIL; } - - netdev->cache_valid |= VALID_IN4; } - *address = netdev->address; - *netmask = netdev->netmask; - return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; + ovs_mutex_unlock(&netdev->mutex); + + return error; } static int @@ -2117,6 +2391,7 @@ netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; + ovs_mutex_lock(&netdev->mutex); error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); if (!error) { netdev->cache_valid |= VALID_IN4; @@ -2127,6 +2402,8 @@ netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, "SIOCSIFNETMASK", netmask); } } + ovs_mutex_unlock(&netdev->mutex); + return error; } @@ -2136,14 +2413,14 @@ parse_if_inet6_line(const char *line, { uint8_t *s6 = in6->s6_addr; #define X8 "%2"SCNx8 - return sscanf(line, - " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 - "%*x %*x %*x %*x %16s\n", - &s6[0], &s6[1], &s6[2], &s6[3], - &s6[4], &s6[5], &s6[6], &s6[7], - &s6[8], &s6[9], &s6[10], &s6[11], - &s6[12], &s6[13], &s6[14], &s6[15], - ifname) == 17; + return ovs_scan(line, + " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 + "%*x %*x %*x %*x %16s\n", + &s6[0], &s6[1], &s6[2], &s6[3], + &s6[4], &s6[5], &s6[6], &s6[7], + &s6[8], &s6[9], &s6[10], &s6[11], + &s6[12], &s6[13], &s6[14], &s6[15], + ifname); } /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if @@ -2152,6 +2429,8 @@ static int netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + ovs_mutex_lock(&netdev->mutex); if (!(netdev->cache_valid & VALID_IN6)) { FILE *file; char line[128]; @@ -2176,6 +2455,8 @@ netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) netdev->cache_valid |= VALID_IN6; } *in6 = netdev->in6; + ovs_mutex_unlock(&netdev->mutex); + return 0; } @@ -2197,11 +2478,10 @@ do_set_addr(struct netdev *netdev, int ioctl_nr, const char *ioctl_name, struct in_addr addr) { struct ifreq ifr; - ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); - make_in4_sockaddr(&ifr.ifr_addr, addr); - return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, - ioctl_name); + make_in4_sockaddr(&ifr.ifr_addr, addr); + return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, + ioctl_name); } /* Adds 'router' as a default IP gateway. */ @@ -2217,9 +2497,9 @@ netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) make_in4_sockaddr(&rt.rt_gateway, router); make_in4_sockaddr(&rt.rt_genmask, any); rt.rt_flags = RTF_UP | RTF_GATEWAY; - error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; + error = af_inet_ioctl(SIOCADDRT, &rt); if (error) { - VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); } return error; } @@ -2236,7 +2516,7 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, *netdev_name = NULL; stream = fopen(fn, "r"); if (stream == NULL) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); return errno; } @@ -2248,12 +2528,11 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, int refcnt, metric, mtu; unsigned int flags, use, window, irtt; - if (sscanf(line, - "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 - " %d %u %u\n", - iface, &dest, &gateway, &flags, &refcnt, - &use, &metric, &mask, &mtu, &window, &irtt) != 11) { - + if (!ovs_scan(line, + "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 + " %d %u %u\n", + iface, &dest, &gateway, &flags, &refcnt, + &use, &metric, &mask, &mtu, &window, &irtt)) { VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", fn, ln, line); continue; @@ -2291,6 +2570,7 @@ netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error = 0; + ovs_mutex_lock(&netdev->mutex); if (!(netdev->cache_valid & VALID_DRVINFO)) { struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; @@ -2310,6 +2590,8 @@ netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) smap_add(smap, "driver_version", netdev->drvinfo.version); smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); } + ovs_mutex_unlock(&netdev->mutex); + return error; } @@ -2343,12 +2625,13 @@ netdev_linux_arp_lookup(const struct netdev *netdev, r.arp_flags = 0; ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); COVERAGE_INC(netdev_arp_lookup); - retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; + retval = af_inet_ioctl(SIOCGARP, &r); if (!retval) { memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); } else if (retval != ENXIO) { VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", - netdev_get_name(netdev), IP_ARGS(ip), strerror(retval)); + netdev_get_name(netdev), IP_ARGS(ip), + ovs_strerror(retval)); } return retval; } @@ -2363,6 +2646,9 @@ nd_to_iff_flags(enum netdev_flags nd) if (nd & NETDEV_PROMISC) { iff |= IFF_PROMISC; } + if (nd & NETDEV_LOOPBACK) { + iff |= IFF_LOOPBACK; + } return iff; } @@ -2376,14 +2662,17 @@ iff_to_nd_flags(int iff) if (iff & IFF_PROMISC) { nd |= NETDEV_PROMISC; } + if (iff & IFF_LOOPBACK) { + nd |= NETDEV_LOOPBACK; + } return nd; } static int -netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, - enum netdev_flags on, enum netdev_flags *old_flagsp) +update_flags(struct netdev_linux *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) + OVS_REQUIRES(netdev->mutex) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); int old_flags, new_flags; int error = 0; @@ -2391,35 +2680,44 @@ netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, *old_flagsp = iff_to_nd_flags(old_flags); new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); if (new_flags != old_flags) { - error = set_flags(netdev_get_name(netdev_), new_flags); - get_flags(netdev_, &netdev->ifi_flags); + error = set_flags(netdev_get_name(&netdev->up), new_flags); + get_flags(&netdev->up, &netdev->ifi_flags); } + return error; } -static unsigned int -netdev_linux_change_seq(const struct netdev *netdev) +static int +netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) { - return netdev_linux_cast(netdev)->change_seq; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = update_flags(netdev, off, on, old_flagsp); + ovs_mutex_unlock(&netdev->mutex); + + return error; } -#define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \ +#define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, SET_STATS, \ GET_FEATURES, GET_STATUS) \ { \ NAME, \ \ - netdev_linux_init, \ + NULL, \ netdev_linux_run, \ netdev_linux_wait, \ \ - CREATE, \ - netdev_linux_destroy, \ + netdev_linux_alloc, \ + CONSTRUCT, \ + netdev_linux_destruct, \ + netdev_linux_dealloc, \ NULL, /* get_config */ \ NULL, /* set_config */ \ NULL, /* get_tunnel_config */ \ \ - netdev_linux_rx_open, \ - \ netdev_linux_send, \ netdev_linux_send_wait, \ \ @@ -2446,7 +2744,9 @@ netdev_linux_change_seq(const struct netdev *netdev) netdev_linux_set_queue, \ netdev_linux_delete_queue, \ netdev_linux_get_queue_stats, \ - netdev_linux_dump_queues, \ + netdev_linux_queue_dump_start, \ + netdev_linux_queue_dump_next, \ + netdev_linux_queue_dump_done, \ netdev_linux_dump_queue_stats, \ \ netdev_linux_get_in4, \ @@ -2459,13 +2759,19 @@ netdev_linux_change_seq(const struct netdev *netdev) \ netdev_linux_update_flags, \ \ - netdev_linux_change_seq \ + netdev_linux_rxq_alloc, \ + netdev_linux_rxq_construct, \ + netdev_linux_rxq_destruct, \ + netdev_linux_rxq_dealloc, \ + netdev_linux_rxq_recv, \ + netdev_linux_rxq_wait, \ + netdev_linux_rxq_drain, \ } const struct netdev_class netdev_linux_class = NETDEV_LINUX_CLASS( "system", - netdev_linux_create, + netdev_linux_construct, netdev_linux_get_stats, NULL, /* set_stats */ netdev_linux_get_features, @@ -2474,7 +2780,7 @@ const struct netdev_class netdev_linux_class = const struct netdev_class netdev_tap_class = NETDEV_LINUX_CLASS( "tap", - netdev_linux_create_tap, + netdev_linux_construct_tap, netdev_tap_get_stats, NULL, /* set_stats */ netdev_linux_get_features, @@ -2483,18 +2789,11 @@ const struct netdev_class netdev_tap_class = const struct netdev_class netdev_internal_class = NETDEV_LINUX_CLASS( "internal", - netdev_linux_create, + netdev_linux_construct, netdev_internal_get_stats, netdev_internal_set_stats, NULL, /* get_features */ netdev_internal_get_status); - -static const struct netdev_rx_class netdev_rx_linux_class = { - netdev_rx_linux_destroy, - netdev_rx_linux_recv, - netdev_rx_linux_wait, - netdev_rx_linux_drain, -}; /* HTB traffic control class. */ @@ -2581,7 +2880,7 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, int error; int mtu; - error = netdev_get_mtu(netdev, &mtu); + error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); if (error) { VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", netdev_get_name(netdev)); @@ -2617,7 +2916,7 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, tc_get_major(handle), tc_get_minor(handle), tc_get_major(parent), tc_get_minor(parent), class->min_rate, class->max_rate, - class->burst, class->priority, strerror(error)); + class->burst, class->priority, ovs_strerror(error)); } return error; } @@ -2677,9 +2976,10 @@ htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, } static void -htb_parse_qdisc_details__(struct netdev *netdev, +htb_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, struct htb_class *hc) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); const char *max_rate_s; max_rate_s = smap_get(details, "max-rate"); @@ -2687,7 +2987,8 @@ htb_parse_qdisc_details__(struct netdev *netdev, if (!hc->max_rate) { enum netdev_features current; - netdev_get_features(netdev, ¤t, NULL, NULL, NULL); + netdev_linux_read_features(netdev); + current = !netdev->get_features_error ? netdev->current : 0; hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; } hc->min_rate = hc->max_rate; @@ -2706,7 +3007,7 @@ htb_parse_class_details__(struct netdev *netdev, const char *priority_s = smap_get(details, "priority"); int mtu, error; - error = netdev_get_mtu(netdev, &mtu); + error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); if (error) { VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", netdev_get_name(netdev)); @@ -2801,6 +3102,7 @@ htb_update_queue__(struct netdev *netdev, unsigned int queue_id, hcp = xmalloc(sizeof *hcp); queue = &hcp->tc_queue; queue->queue_id = queue_id; + queue->created = time_msec(); hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); } @@ -2814,7 +3116,7 @@ static int htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) { struct ofpbuf msg; - struct nl_dump dump; + struct queue_dump_state state; struct htb_class hc; /* Get qdisc options. */ @@ -2823,17 +3125,17 @@ htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) htb_install__(netdev, hc.max_rate); /* Get queues. */ - if (!start_queue_dump(netdev, &dump)) { + if (!start_queue_dump(netdev, &state)) { return ENODEV; } - while (nl_dump_next(&dump, &msg)) { + while (nl_dump_next(&state.dump, &msg, &state.buf)) { unsigned int queue_id; if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { htb_update_queue__(netdev, queue_id, &hc); } } - nl_dump_done(&dump); + finish_queue_dump(&state); return 0; } @@ -3034,6 +3336,7 @@ hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, hcp = xmalloc(sizeof *hcp); queue = &hcp->tc_queue; queue->queue_id = queue_id; + queue->created = time_msec(); hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); } @@ -3152,9 +3455,10 @@ hfsc_query_class__(const struct netdev *netdev, unsigned int handle, } static void -hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details, +hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, struct hfsc_class *class) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); uint32_t max_rate; const char *max_rate_s; @@ -3164,7 +3468,8 @@ hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details, if (!max_rate) { enum netdev_features current; - netdev_get_features(netdev, ¤t, NULL, NULL, NULL); + netdev_linux_read_features(netdev); + current = !netdev->get_features_error ? netdev->current : 0; max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; } @@ -3277,7 +3582,7 @@ hfsc_setup_class__(struct netdev *netdev, unsigned int handle, netdev_get_name(netdev), tc_get_major(handle), tc_get_minor(handle), tc_get_major(parent), tc_get_minor(parent), - class->min_rate, class->max_rate, strerror(error)); + class->min_rate, class->max_rate, ovs_strerror(error)); } return error; @@ -3311,18 +3616,18 @@ static int hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) { struct ofpbuf msg; - struct nl_dump dump; + struct queue_dump_state state; struct hfsc_class hc; hc.max_rate = 0; hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); hfsc_install__(netdev, hc.max_rate); - if (!start_queue_dump(netdev, &dump)) { + if (!start_queue_dump(netdev, &state)) { return ENODEV; } - while (nl_dump_next(&dump, &msg)) { + while (nl_dump_next(&state.dump, &msg, &state.buf)) { unsigned int queue_id; if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { @@ -3330,7 +3635,7 @@ hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) } } - nl_dump_done(&dump); + finish_queue_dump(&state); return 0; } @@ -3628,7 +3933,7 @@ tc_make_request(const struct netdev *netdev, int type, unsigned int flags, static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) { - int error = nl_sock_transact(rtnl_sock, request, replyp); + int error = nl_transact(NETLINK_ROUTE, request, replyp); ofpbuf_uninit(request); return error; } @@ -3773,30 +4078,35 @@ read_psched(void) * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion * [6] 2.6.34 from kernel.org on KVM */ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; static const char fn[] = "/proc/net/psched"; unsigned int a, b, c, d; FILE *stream; + if (!ovsthread_once_start(&once)) { + return; + } + ticks_per_s = 1.0; buffer_hz = 100; stream = fopen(fn, "r"); if (!stream) { - VLOG_WARN("%s: open failed: %s", fn, strerror(errno)); - return; + VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); + goto exit; } if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { VLOG_WARN("%s: read failed", fn); fclose(stream); - return; + goto exit; } VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); fclose(stream); if (!a || !c) { VLOG_WARN("%s: invalid scheduler parameters", fn); - return; + goto exit; } ticks_per_s = (double) a * c / b; @@ -3807,6 +4117,9 @@ read_psched(void) fn, a, b, c, d); } VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); + +exit: + ovsthread_once_done(&once); } /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a @@ -3814,9 +4127,7 @@ read_psched(void) static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) { - if (!buffer_hz) { - read_psched(); - } + read_psched(); return (rate * ticks) / ticks_per_s; } @@ -3825,9 +4136,7 @@ tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size) { - if (!buffer_hz) { - read_psched(); - } + read_psched(); return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; } @@ -3836,9 +4145,7 @@ tc_bytes_to_ticks(unsigned int rate, unsigned int size) static unsigned int tc_buffer_per_jiffy(unsigned int rate) { - if (!buffer_hz) { - read_psched(); - } + read_psched(); return rate / buffer_hz; } @@ -3986,7 +4293,7 @@ tc_query_class(const struct netdev *netdev, netdev_get_name(netdev), tc_get_major(handle), tc_get_minor(handle), tc_get_major(parent), tc_get_minor(parent), - strerror(error)); + ovs_strerror(error)); } return error; } @@ -4011,7 +4318,7 @@ tc_delete_class(const struct netdev *netdev, unsigned int handle) VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", netdev_get_name(netdev), tc_get_major(handle), tc_get_minor(handle), - strerror(error)); + ovs_strerror(error)); } return error; } @@ -4111,7 +4418,7 @@ tc_query_qdisc(const struct netdev *netdev_) } else { /* Who knows? Maybe the device got deleted. */ VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", - netdev_get_name(netdev_), strerror(error)); + netdev_get_name(netdev_), ovs_strerror(error)); ops = &tc_ops_other; } @@ -4194,14 +4501,6 @@ tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) /* Linux-only functions declared in netdev-linux.h */ -/* Returns a fd for an AF_INET socket or a negative errno value. */ -int -netdev_linux_get_af_inet_sock(void) -{ - int error = netdev_linux_init(); - return error ? -error : af_inet_sock; -} - /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If * 'enable' is true, the bit is set. Otherwise, it is cleared. */ int @@ -4281,110 +4580,41 @@ netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, } static int -get_stats_via_netlink(int ifindex, struct netdev_stats *stats) +get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats) { - /* Policy for RTNLGRP_LINK messages. - * - * There are *many* more fields in these messages, but currently we only - * care about these fields. */ - static const struct nl_policy rtnlgrp_link_policy[] = { - [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false }, - [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, - .min_len = sizeof(struct rtnl_link_stats) }, - }; - struct ofpbuf request; struct ofpbuf *reply; - struct ifinfomsg *ifi; - struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; int error; ofpbuf_init(&request, 0); - nl_msg_put_nlmsghdr(&request, 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); + nl_msg_put_nlmsghdr(&request, + sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), + RTM_GETLINK, NLM_F_REQUEST); + ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); + nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_)); + error = nl_transact(NETLINK_ROUTE, &request, &reply); ofpbuf_uninit(&request); if (error) { return error; } - 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"); - ofpbuf_delete(reply); - return EPROTO; + if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) { + const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS); + if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) { + netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a)); + error = 0; + } else { + VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); + error = EPROTO; + } + } else { + VLOG_WARN_RL(&rl, "short RTM_GETLINK reply"); + error = EPROTO; } - netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS])); ofpbuf_delete(reply); - - return 0; -} - -static int -get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) -{ - static const char fn[] = "/proc/net/dev"; - char line[1024]; - FILE *stream; - int ln; - - stream = fopen(fn, "r"); - if (!stream) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); - return errno; - } - - ln = 0; - while (fgets(line, sizeof line, stream)) { - if (++ln >= 3) { - char devname[16]; -#define X64 "%"SCNu64 - if (sscanf(line, - " %15[^:]:" - X64 X64 X64 X64 X64 X64 X64 "%*u" - X64 X64 X64 X64 X64 X64 X64 "%*u", - devname, - &stats->rx_bytes, - &stats->rx_packets, - &stats->rx_errors, - &stats->rx_dropped, - &stats->rx_fifo_errors, - &stats->rx_frame_errors, - &stats->multicast, - &stats->tx_bytes, - &stats->tx_packets, - &stats->tx_errors, - &stats->tx_dropped, - &stats->tx_fifo_errors, - &stats->collisions, - &stats->tx_carrier_errors) != 15) { - VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); - } else if (!strcmp(devname, netdev_name)) { - stats->rx_length_errors = UINT64_MAX; - stats->rx_over_errors = UINT64_MAX; - stats->rx_crc_errors = UINT64_MAX; - stats->rx_missed_errors = UINT64_MAX; - stats->tx_aborted_errors = UINT64_MAX; - stats->tx_heartbeat_errors = UINT64_MAX; - stats->tx_window_errors = UINT64_MAX; - fclose(stream); - return 0; - } - } - } - VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); - fclose(stream); - return ENODEV; + return error; } static int @@ -4394,8 +4624,7 @@ get_flags(const struct netdev *dev, unsigned int *flags) int error; *flags = 0; - error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS, - "SIOCGIFFLAGS"); + error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); if (!error) { *flags = ifr.ifr_flags; } @@ -4408,20 +4637,23 @@ set_flags(const char *name, unsigned int flags) struct ifreq ifr; ifr.ifr_flags = flags; - return netdev_linux_do_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); + return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); } static int do_get_ifindex(const char *netdev_name) { struct ifreq ifr; + int error; ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); COVERAGE_INC(netdev_get_ifindex); - if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { + + error = af_inet_ioctl(SIOCGIFINDEX, &ifr); + if (error) { VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", - netdev_name, strerror(errno)); - return -errno; + netdev_name, ovs_strerror(error)); + return -error; } return ifr.ifr_ifindex; } @@ -4453,18 +4685,20 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) { struct ifreq ifr; int hwaddr_family; + int error; memset(&ifr, 0, sizeof ifr); ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); COVERAGE_INC(netdev_get_hwaddr); - if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { + error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); + if (error) { /* ENODEV probably means that a vif disappeared asynchronously and * hasn't been removed from the database yet, so reduce the log level * to INFO for that case. */ - VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR, + VLOG(error == ENODEV ? VLL_INFO : VLL_ERR, "ioctl(SIOCGIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; + netdev_name, ovs_strerror(error)); + return error; } hwaddr_family = ifr.ifr_hwaddr.sa_family; if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { @@ -4480,18 +4714,19 @@ set_etheraddr(const char *netdev_name, const uint8_t mac[ETH_ADDR_LEN]) { struct ifreq ifr; + int error; memset(&ifr, 0, sizeof ifr); ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); COVERAGE_INC(netdev_set_hwaddr); - if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { + error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); + if (error) { VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; + netdev_name, ovs_strerror(error)); } - return 0; + return error; } static int @@ -4499,37 +4734,24 @@ netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, int cmd, const char *cmd_name) { struct ifreq ifr; + int error; memset(&ifr, 0, sizeof ifr); ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); ifr.ifr_data = (caddr_t) ecmd; ecmd->cmd = cmd; - if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { - return 0; - } else { - if (errno != EOPNOTSUPP) { + error = af_inet_ioctl(SIOCETHTOOL, &ifr); + if (error) { + if (error != EOPNOTSUPP) { VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " - "failed: %s", cmd_name, name, strerror(errno)); + "failed: %s", cmd_name, name, ovs_strerror(error)); } else { /* The device doesn't support this operation. That's pretty * common, so there's no point in logging anything. */ } - return errno; } -} - -static int -netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, - const char *cmd_name) -{ - ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name); - if (ioctl(af_inet_sock, cmd, ifr) == -1) { - VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, - strerror(errno)); - return errno; - } - return 0; + return error; } static int @@ -4540,9 +4762,10 @@ netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, int error; ifr.ifr_addr.sa_family = AF_INET; - error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); + error = af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); if (!error) { - const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; + const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, + &ifr.ifr_addr); *ip = sin->sin_addr; } return error; @@ -4552,9 +4775,10 @@ netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, static int af_packet_sock(void) { - static int sock = INT_MIN; + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static int sock; - if (sock == INT_MIN) { + if (ovsthread_once_start(&once)) { sock = socket(AF_PACKET, SOCK_RAW, 0); if (sock >= 0) { int error = set_nonblocking(sock); @@ -4564,8 +4788,10 @@ af_packet_sock(void) } } else { sock = -errno; - VLOG_ERR("failed to create packet socket: %s", strerror(errno)); + VLOG_ERR("failed to create packet socket: %s", + ovs_strerror(errno)); } + ovsthread_once_done(&once); } return sock;