From: Ben Pfaff Date: Tue, 22 Sep 2009 17:17:44 +0000 (-0700) Subject: Merge citrix branch into master. X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=576e26d7b47f4e53116ef0b5f035d260f426d37b;p=openvswitch Merge citrix branch into master. --- 576e26d7b47f4e53116ef0b5f035d260f426d37b diff --cc configure.ac index d5b5938d,41ff7bea..400ea786 --- a/configure.ac +++ b/configure.ac @@@ -12,8 -12,8 +12,8 @@@ # See the License for the specific language governing permissions and # limitations under the License. -AC_PREREQ(2.60) +AC_PREREQ(2.63) - AC_INIT(openvswitch, 0.90.3, bugs@openvswitch.org) + AC_INIT(openvswitch, 0.90.5, ovs-bugs@openvswitch.org) NX_BUILDNR AC_CONFIG_SRCDIR([datapath/datapath.c]) AC_CONFIG_MACRO_DIR([m4]) diff --cc lib/netdev-linux.c index 2faffa34,00000000..5abf6e16 mode 100644,000000..100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@@ -1,1733 -1,0 +1,1736 @@@ +/* + * Copyright (c) 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "coverage.h" +#include "dynamic-string.h" +#include "fatal-signal.h" +#include "netdev-provider.h" +#include "netlink.h" +#include "ofpbuf.h" +#include "openflow/openflow.h" +#include "packets.h" +#include "poll-loop.h" +#include "rtnetlink.h" +#include "socket-util.h" +#include "shash.h" +#include "svec.h" + +#define THIS_MODULE VLM_netdev_linux +#include "vlog.h" + +/* These were introduced in Linux 2.6.14, so they might be missing if we have + * old headers. */ +#ifndef ADVERTISED_Pause +#define ADVERTISED_Pause (1 << 13) +#endif +#ifndef ADVERTISED_Asym_Pause +#define ADVERTISED_Asym_Pause (1 << 14) +#endif + +struct netdev_linux { + struct netdev netdev; + + /* File descriptors. For ordinary network devices, the two fds below are + * the same; for tap devices, they differ. */ + int netdev_fd; /* Network device. */ + int tap_fd; /* TAP character device, if any, otherwise the + * network device. */ + + struct netdev_linux_cache *cache; +}; + +enum { + VALID_IFINDEX = 1 << 0, + VALID_ETHERADDR = 1 << 1, + VALID_IN4 = 1 << 2, + VALID_IN6 = 1 << 3, + VALID_MTU = 1 << 4, + VALID_CARRIER = 1 << 5, + VALID_IS_INTERNAL = 1 << 6 +}; + +/* Cached network device information. */ +struct netdev_linux_cache { + struct shash_node *shash_node; + unsigned int valid; + int ref_cnt; + + int ifindex; + uint8_t etheraddr[ETH_ADDR_LEN]; + struct in_addr address, netmask; + struct in6_addr in6; + int mtu; + int carrier; + bool is_internal; +}; + +static struct shash cache_map = SHASH_INITIALIZER(&cache_map); +static struct rtnetlink_notifier netdev_linux_cache_notifier; + +/* An AF_INET socket (used for ioctl operations). */ +static int af_inet_sock = -1; + +struct netdev_linux_notifier { + struct netdev_notifier notifier; + struct list node; +}; + +static struct shash netdev_linux_notifiers = + SHASH_INITIALIZER(&netdev_linux_notifiers); +static struct rtnetlink_notifier netdev_linux_poll_notifier; + +/* This is set pretty low because we probably won't learn anything from the + * additional log messages. */ +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + +static int netdev_linux_do_ethtool(struct netdev *, struct ethtool_cmd *, + int cmd, const char *cmd_name); +static int netdev_linux_do_ioctl(const struct netdev *, struct ifreq *, + int cmd, const char *cmd_name); +static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, + int cmd, const char *cmd_name); +static int get_flags(const struct netdev *, int *flagsp); +static int set_flags(struct netdev *, int flags); +static int do_get_ifindex(const char *netdev_name); +static int get_ifindex(const struct netdev *, int *ifindexp); +static int do_set_addr(struct netdev *netdev, + int ioctl_nr, const char *ioctl_name, + struct in_addr addr); +static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); +static int set_etheraddr(const char *netdev_name, int hwaddr_family, + const uint8_t[ETH_ADDR_LEN]); +static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); +static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); + +static struct netdev_linux * +netdev_linux_cast(const struct netdev *netdev) +{ + netdev_assert_class(netdev, &netdev_linux_class); + return CONTAINER_OF(netdev, struct netdev_linux, netdev); +} + +static int +netdev_linux_init(void) +{ + static int status = -1; + if (status < 0) { + af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); + status = af_inet_sock >= 0 ? 0 : errno; + if (status) { + VLOG_ERR("failed to create inet socket: %s", strerror(status)); + } + } + return status; +} + +static void +netdev_linux_run(void) +{ + rtnetlink_notifier_run(); +} + +static void +netdev_linux_wait(void) +{ + rtnetlink_notifier_wait(); +} + +static void +netdev_linux_cache_cb(const struct rtnetlink_change *change, + void *aux UNUSED) +{ + struct netdev_linux_cache *cache; + if (change) { + cache = shash_find_data(&cache_map, change->ifname); + if (cache) { + cache->valid = 0; + } + } else { + struct shash_node *node; + SHASH_FOR_EACH (node, &cache_map) { + cache = node->data; + cache->valid = 0; + } + } +} + +static int +netdev_linux_open(const char *name, char *suffix, int ethertype, + struct netdev **netdevp) +{ + struct netdev_linux *netdev; + enum netdev_flags flags; + int error; + + /* Allocate network device. */ + netdev = xcalloc(1, sizeof *netdev); + netdev_init(&netdev->netdev, suffix, &netdev_linux_class); + netdev->netdev_fd = -1; + netdev->tap_fd = -1; + netdev->cache = shash_find_data(&cache_map, suffix); + if (!netdev->cache) { + if (shash_is_empty(&cache_map)) { + int error = rtnetlink_notifier_register( + &netdev_linux_cache_notifier, netdev_linux_cache_cb, NULL); + if (error) { + netdev_close(&netdev->netdev); + return error; + } + } + netdev->cache = xmalloc(sizeof *netdev->cache); + netdev->cache->shash_node = shash_add(&cache_map, suffix, + netdev->cache); + netdev->cache->valid = 0; + netdev->cache->ref_cnt = 0; + } + netdev->cache->ref_cnt++; + + if (!strncmp(name, "tap:", 4)) { + static const char tap_dev[] = "/dev/net/tun"; + struct ifreq ifr; + + /* Open tap device. */ + netdev->tap_fd = open(tap_dev, O_RDWR); + if (netdev->tap_fd < 0) { + error = errno; + VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); + goto error; + } + + /* Create tap device. */ + ifr.ifr_flags = IFF_TAP | IFF_NO_PI; + error = netdev_linux_do_ioctl(&netdev->netdev, &ifr, + TUNSETIFF, "TUNSETIFF"); + if (error) { + goto error; + } + + /* Make non-blocking. */ + error = set_nonblocking(netdev->tap_fd); + if (error) { + goto error; + } + } + + error = netdev_get_flags(&netdev->netdev, &flags); + if (error == ENODEV) { + goto error; + } + + if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) { + struct sockaddr_ll sll; + int protocol; + int ifindex; + + /* Create file descriptor. */ + protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL + : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 + : ethertype); + netdev->netdev_fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); + if (netdev->netdev_fd < 0) { + error = errno; + goto error; + } + if (netdev->tap_fd < 0) { + netdev->tap_fd = netdev->netdev_fd; + } + + /* Set non-blocking mode. */ + error = set_nonblocking(netdev->netdev_fd); + if (error) { + goto error; + } + + /* Get ethernet device index. */ + error = get_ifindex(&netdev->netdev, &ifindex); + if (error) { + goto error; + } + + /* Bind to specific ethernet device. */ + memset(&sll, 0, sizeof sll); + sll.sll_family = AF_PACKET; + sll.sll_ifindex = ifindex; + if (bind(netdev->netdev_fd, + (struct sockaddr *) &sll, sizeof sll) < 0) { + error = errno; + VLOG_ERR("bind to %s failed: %s", suffix, strerror(error)); + goto error; + } + + /* Between the socket() and bind() calls above, the socket receives all + * packets of the requested type on all system interfaces. We do not + * want to receive that data, but there is no way to avoid it. So we + * must now drain out the receive queue. */ + error = drain_rcvbuf(netdev->netdev_fd); + if (error) { + goto error; + } + } + + *netdevp = &netdev->netdev; + return 0; + +error: + netdev_close(&netdev->netdev); + return error; +} + +/* Closes and destroys 'netdev'. */ +static void +netdev_linux_close(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (netdev->cache && !--netdev->cache->ref_cnt) { + shash_delete(&cache_map, netdev->cache->shash_node); + free(netdev->cache); + + if (shash_is_empty(&cache_map)) { + rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); + } + } + if (netdev->netdev_fd >= 0) { + close(netdev->netdev_fd); + } + if (netdev->tap_fd >= 0 && netdev->netdev_fd != netdev->tap_fd) { + close(netdev->tap_fd); + } + free(netdev); +} + +/* Initializes 'svec' with a list of the names of all known network devices. */ +static int +netdev_linux_enumerate(struct svec *svec) +{ + struct if_nameindex *names; + + names = if_nameindex(); + if (names) { + size_t i; + + for (i = 0; names[i].if_name != NULL; i++) { + svec_add(svec, names[i].if_name); + } + if_freenameindex(names); + return 0; + } else { + VLOG_WARN("could not obtain list of network device names: %s", + strerror(errno)); + return errno; + } +} + +static int +netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (netdev->tap_fd < 0) { + /* Device was opened with NETDEV_ETH_TYPE_NONE. */ + return EAGAIN; + } + + for (;;) { + ssize_t retval = read(netdev->tap_fd, data, size); + if (retval >= 0) { + return retval; + } else if (errno != EINTR) { + if (errno != EAGAIN) { + VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", + strerror(errno), netdev_get_name(netdev_)); + } + return errno; + } + } +} + +/* Registers with the poll loop to wake up from the next call to poll_block() + * when a packet is ready to be received with netdev_recv() on 'netdev'. */ +static void +netdev_linux_recv_wait(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (netdev->tap_fd >= 0) { + poll_fd_wait(netdev->tap_fd, POLLIN); + } +} + +/* Discards all packets waiting to be received from 'netdev'. */ +static int +netdev_linux_drain(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) { + return 0; + } else if (netdev->tap_fd != netdev->netdev_fd) { + struct ifreq ifr; + int error = netdev_linux_do_ioctl(netdev_, &ifr, + SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); + if (error) { + return error; + } + drain_fd(netdev->tap_fd, ifr.ifr_qlen); + return 0; + } else { + return drain_rcvbuf(netdev->netdev_fd); + } +} + +/* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive + * errno value. Returns EAGAIN without blocking if the packet cannot be queued + * immediately. Returns EMSGSIZE if a partial packet was transmitted or if + * the packet is too big or too small to transmit on the device. + * + * The caller retains ownership of 'buffer' in all cases. + * + * The kernel maintains a packet transmission queue, so the caller is not + * expected to do additional queuing of packets. */ +static int +netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. + */ + if (netdev->tap_fd < 0) { + return EPIPE; + } + + for (;;) { + ssize_t retval = write(netdev->tap_fd, data, size); + if (retval < 0) { + /* The Linux AF_PACKET implementation never blocks waiting for room + * for packets, instead returning ENOBUFS. Translate this into + * EAGAIN for the caller. */ + if (errno == ENOBUFS) { + return EAGAIN; + } else if (errno == EINTR) { + continue; + } else if (errno != EAGAIN) { + VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", + netdev_get_name(netdev_), strerror(errno)); + } + return errno; + } else if (retval != size) { + VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " + "%zu) on %s", retval, size, netdev_get_name(netdev_)); + return EMSGSIZE; + } else { + return 0; + } + } +} + +/* Registers with the poll loop to wake up from the next call to poll_block() + * when the packet transmission queue has sufficient room to transmit a packet + * with netdev_send(). + * + * The kernel maintains a packet transmission queue, so the client is not + * expected to do additional queuing of packets. Thus, this function is + * unlikely to ever be used. It is included for completeness. */ +static void +netdev_linux_send_wait(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) { + /* Nothing to do. */ + } else if (netdev->tap_fd == netdev->netdev_fd) { + poll_fd_wait(netdev->tap_fd, POLLOUT); + } else { + /* TAP device always accepts packets.*/ + poll_immediate_wake(); + } +} + +/* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, + * otherwise a positive errno value. */ +static int +netdev_linux_set_etheraddr(struct netdev *netdev_, + const uint8_t mac[ETH_ADDR_LEN]) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac); + if (!error) { + memcpy(netdev->cache->etheraddr, mac, ETH_ADDR_LEN); + } + return error; +} + +/* Returns a pointer to 'netdev''s MAC address. The caller must not modify or + * free the returned buffer. */ +static int +netdev_linux_get_etheraddr(const struct netdev *netdev_, + uint8_t mac[ETH_ADDR_LEN]) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache->valid & VALID_ETHERADDR)) { + int error = get_etheraddr(netdev_get_name(netdev_), + netdev->cache->etheraddr); + if (error) { + return error; + } + netdev->cache->valid |= VALID_ETHERADDR; + } + memcpy(mac, netdev->cache->etheraddr, ETH_ADDR_LEN); + return 0; +} + +/* Returns the maximum size of transmitted (and received) packets on 'netdev', + * in bytes, not including the hardware header; thus, this is typically 1500 + * bytes for Ethernet devices. */ +static int +netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache->valid & VALID_MTU)) { + struct ifreq ifr; + int error; + + error = netdev_linux_do_ioctl(netdev_, &ifr, SIOCGIFMTU, "SIOCGIFMTU"); + if (error) { + return error; + } + netdev->cache->mtu = ifr.ifr_mtu; + netdev->cache->valid |= VALID_MTU; + } + *mtup = netdev->cache->mtu; + return 0; +} + +static int +netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error = 0; + char *fn = NULL; + int fd = -1; + + if (!(netdev->cache->valid & VALID_CARRIER)) { + char line[8]; + int retval; + + fn = xasprintf("/sys/class/net/%s/carrier", netdev_get_name(netdev_)); + fd = open(fn, O_RDONLY); + if (fd < 0) { + error = errno; + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); + goto exit; + } + + retval = read(fd, line, sizeof line); + if (retval < 0) { + error = errno; + if (error == EINVAL) { + /* This is the normal return value when we try to check carrier + * if the network device is not up. */ + } else { + VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error)); + } + goto exit; + } else if (retval == 0) { + error = EPROTO; + VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); + goto exit; + } + + if (line[0] != '0' && line[0] != '1') { + error = EPROTO; + VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", + fn, line[0]); + goto exit; + } + netdev->cache->carrier = line[0] != '0'; + netdev->cache->valid |= VALID_CARRIER; + } + *carrier = netdev->cache->carrier; + error = 0; + +exit: + if (fd >= 0) { + close(fd); + } + free(fn); + return error; +} + +/* Check whether we can we use RTM_GETLINK to get network device statistics. + * In pre-2.6.19 kernels, this was only available if wireless extensions were + * enabled. */ +static bool +check_for_working_netlink_stats(void) +{ + /* Decide on the netdev_get_stats() implementation to use. Netlink is + * preferable, so if that works, we'll use it. */ + int ifindex = do_get_ifindex("lo"); + if (ifindex < 0) { + VLOG_WARN("failed to get ifindex for lo, " + "obtaining netdev stats from proc"); + return false; + } else { + struct netdev_stats stats; + int error = get_stats_via_netlink(ifindex, &stats); + if (!error) { + VLOG_DBG("obtaining netdev stats via rtnetlink"); + return true; + } else { + VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " + "via proc (you are probably running a pre-2.6.19 " + "kernel)", strerror(error)); + return false; + } + } +} + +/* Retrieves current device stats for 'netdev'. + * + * XXX All of the members of struct netdev_stats are 64 bits wide, but on + * 32-bit architectures the Linux network stats are only 32 bits. */ +static int +netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + static int use_netlink_stats = -1; + int error; + struct netdev_stats raw_stats; + struct netdev_stats *collect_stats = stats; + + COVERAGE_INC(netdev_get_stats); + + if (!(netdev->cache->valid & VALID_IS_INTERNAL)) { + netdev->cache->is_internal = (netdev->tap_fd != -1); + + if (!netdev->cache->is_internal) { + struct ethtool_drvinfo drvinfo; + + memset(&drvinfo, 0, sizeof drvinfo); + error = netdev_linux_do_ethtool(&netdev->netdev, + (struct ethtool_cmd *)&drvinfo, + ETHTOOL_GDRVINFO, + "ETHTOOL_GDRVINFO"); + + if (!error) { + netdev->cache->is_internal = !strcmp(drvinfo.driver, + "openvswitch"); + } + } + + netdev->cache->valid |= VALID_IS_INTERNAL; + } + + if (netdev->cache->is_internal) { + collect_stats = &raw_stats; + } + + if (use_netlink_stats < 0) { + use_netlink_stats = check_for_working_netlink_stats(); + } + if (use_netlink_stats) { + int ifindex; + + error = get_ifindex(&netdev->netdev, &ifindex); + if (!error) { + error = get_stats_via_netlink(ifindex, collect_stats); + } + } else { + error = get_stats_via_proc(netdev->netdev.name, collect_stats); + } + + /* If this port is an internal port then the transmit and receive stats + * will appear to be swapped relative to the other ports since we are the + * one sending the data, not a remote computer. For consistency, we swap + * them back here. */ + if (netdev->cache->is_internal) { + stats->rx_packets = raw_stats.tx_packets; + stats->tx_packets = raw_stats.rx_packets; + stats->rx_bytes = raw_stats.tx_bytes; + stats->tx_bytes = raw_stats.rx_bytes; + stats->rx_errors = raw_stats.tx_errors; + stats->tx_errors = raw_stats.rx_errors; + stats->rx_dropped = raw_stats.tx_dropped; + stats->tx_dropped = raw_stats.rx_dropped; + stats->multicast = raw_stats.multicast; + stats->collisions = raw_stats.collisions; + stats->rx_length_errors = 0; + stats->rx_over_errors = 0; + stats->rx_crc_errors = 0; + stats->rx_frame_errors = 0; + stats->rx_fifo_errors = 0; + stats->rx_missed_errors = 0; + stats->tx_aborted_errors = 0; + stats->tx_carrier_errors = 0; + stats->tx_fifo_errors = 0; + stats->tx_heartbeat_errors = 0; + stats->tx_window_errors = 0; + } + + return error; +} + +/* Stores the features supported by 'netdev' into each of '*current', + * '*advertised', '*supported', and '*peer' that are non-null. Each value is a + * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if + * successful, otherwise a positive errno value. On failure, all of the + * passed-in values are set to 0. */ +static int +netdev_linux_get_features(struct netdev *netdev, + uint32_t *current, uint32_t *advertised, + uint32_t *supported, uint32_t *peer) +{ + struct ethtool_cmd ecmd; + int error; + + memset(&ecmd, 0, sizeof ecmd); + error = netdev_linux_do_ethtool(netdev, &ecmd, + ETHTOOL_GSET, "ETHTOOL_GSET"); + if (error) { + return error; + } + + /* Supported features. */ + *supported = 0; + if (ecmd.supported & SUPPORTED_10baseT_Half) { + *supported |= OFPPF_10MB_HD; + } + if (ecmd.supported & SUPPORTED_10baseT_Full) { + *supported |= OFPPF_10MB_FD; + } + if (ecmd.supported & SUPPORTED_100baseT_Half) { + *supported |= OFPPF_100MB_HD; + } + if (ecmd.supported & SUPPORTED_100baseT_Full) { + *supported |= OFPPF_100MB_FD; + } + if (ecmd.supported & SUPPORTED_1000baseT_Half) { + *supported |= OFPPF_1GB_HD; + } + if (ecmd.supported & SUPPORTED_1000baseT_Full) { + *supported |= OFPPF_1GB_FD; + } + if (ecmd.supported & SUPPORTED_10000baseT_Full) { + *supported |= OFPPF_10GB_FD; + } + if (ecmd.supported & SUPPORTED_TP) { + *supported |= OFPPF_COPPER; + } + if (ecmd.supported & SUPPORTED_FIBRE) { + *supported |= OFPPF_FIBER; + } + if (ecmd.supported & SUPPORTED_Autoneg) { + *supported |= OFPPF_AUTONEG; + } + if (ecmd.supported & SUPPORTED_Pause) { + *supported |= OFPPF_PAUSE; + } + if (ecmd.supported & SUPPORTED_Asym_Pause) { + *supported |= OFPPF_PAUSE_ASYM; + } + + /* Advertised features. */ + *advertised = 0; + if (ecmd.advertising & ADVERTISED_10baseT_Half) { + *advertised |= OFPPF_10MB_HD; + } + if (ecmd.advertising & ADVERTISED_10baseT_Full) { + *advertised |= OFPPF_10MB_FD; + } + if (ecmd.advertising & ADVERTISED_100baseT_Half) { + *advertised |= OFPPF_100MB_HD; + } + if (ecmd.advertising & ADVERTISED_100baseT_Full) { + *advertised |= OFPPF_100MB_FD; + } + if (ecmd.advertising & ADVERTISED_1000baseT_Half) { + *advertised |= OFPPF_1GB_HD; + } + if (ecmd.advertising & ADVERTISED_1000baseT_Full) { + *advertised |= OFPPF_1GB_FD; + } + if (ecmd.advertising & ADVERTISED_10000baseT_Full) { + *advertised |= OFPPF_10GB_FD; + } + if (ecmd.advertising & ADVERTISED_TP) { + *advertised |= OFPPF_COPPER; + } + if (ecmd.advertising & ADVERTISED_FIBRE) { + *advertised |= OFPPF_FIBER; + } + if (ecmd.advertising & ADVERTISED_Autoneg) { + *advertised |= OFPPF_AUTONEG; + } + if (ecmd.advertising & ADVERTISED_Pause) { + *advertised |= OFPPF_PAUSE; + } + if (ecmd.advertising & ADVERTISED_Asym_Pause) { + *advertised |= OFPPF_PAUSE_ASYM; + } + + /* Current settings. */ + if (ecmd.speed == SPEED_10) { + *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD; + } else if (ecmd.speed == SPEED_100) { + *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD; + } else if (ecmd.speed == SPEED_1000) { + *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD; + } else if (ecmd.speed == SPEED_10000) { + *current = OFPPF_10GB_FD; + } else { + *current = 0; + } + + if (ecmd.port == PORT_TP) { + *current |= OFPPF_COPPER; + } else if (ecmd.port == PORT_FIBRE) { + *current |= OFPPF_FIBER; + } + + if (ecmd.autoneg) { + *current |= OFPPF_AUTONEG; + } + + /* Peer advertisements. */ + *peer = 0; /* XXX */ + + return 0; +} + +/* Set the features advertised by 'netdev' to 'advertise'. */ +static int +netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) +{ + struct ethtool_cmd ecmd; + int error; + + memset(&ecmd, 0, sizeof ecmd); + error = netdev_linux_do_ethtool(netdev, &ecmd, + ETHTOOL_GSET, "ETHTOOL_GSET"); + if (error) { + return error; + } + + ecmd.advertising = 0; + if (advertise & OFPPF_10MB_HD) { + ecmd.advertising |= ADVERTISED_10baseT_Half; + } + if (advertise & OFPPF_10MB_FD) { + ecmd.advertising |= ADVERTISED_10baseT_Full; + } + if (advertise & OFPPF_100MB_HD) { + ecmd.advertising |= ADVERTISED_100baseT_Half; + } + if (advertise & OFPPF_100MB_FD) { + ecmd.advertising |= ADVERTISED_100baseT_Full; + } + if (advertise & OFPPF_1GB_HD) { + ecmd.advertising |= ADVERTISED_1000baseT_Half; + } + if (advertise & OFPPF_1GB_FD) { + ecmd.advertising |= ADVERTISED_1000baseT_Full; + } + if (advertise & OFPPF_10GB_FD) { + ecmd.advertising |= ADVERTISED_10000baseT_Full; + } + if (advertise & OFPPF_COPPER) { + ecmd.advertising |= ADVERTISED_TP; + } + if (advertise & OFPPF_FIBER) { + ecmd.advertising |= ADVERTISED_FIBRE; + } + if (advertise & OFPPF_AUTONEG) { + ecmd.advertising |= ADVERTISED_Autoneg; + } + if (advertise & OFPPF_PAUSE) { + ecmd.advertising |= ADVERTISED_Pause; + } + if (advertise & OFPPF_PAUSE_ASYM) { + ecmd.advertising |= ADVERTISED_Asym_Pause; + } + return netdev_linux_do_ethtool(netdev, &ecmd, + ETHTOOL_SSET, "ETHTOOL_SSET"); +} + +/* If 'netdev_name' is the name of a VLAN network device (e.g. one created with + * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device + * and returns 0. Otherwise returns a errno value (specifically ENOENT if + * 'netdev_name' is the name of a network device that is not a VLAN device) and + * sets '*vlan_vid' to -1. */ +static int +netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) +{ + const char *netdev_name = netdev_get_name(netdev); + struct ds line = DS_EMPTY_INITIALIZER; + FILE *stream = NULL; + int error; + char *fn; + + COVERAGE_INC(netdev_get_vlan_vid); + fn = xasprintf("/proc/net/vlan/%s", netdev_name); + stream = fopen(fn, "r"); + if (!stream) { + error = errno; + goto done; + } + + if (ds_get_line(&line, stream)) { + if (ferror(stream)) { + error = errno; + VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno)); + } else { + error = EPROTO; + VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn); + } + goto done; + } + + if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) { + error = EPROTO; + VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"", + fn, ds_cstr(&line)); + goto done; + } + + error = 0; + +done: + free(fn); + if (stream) { + fclose(stream); + } + ds_destroy(&line); + if (error) { + *vlan_vid = -1; + } + return error; +} + +#define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress" +#define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1" +/* We redirect stderr to /dev/null because we often want to remove all + * traffic control configuration on a port so its in a known state. If + * this done when there is no such configuration, tc complains, so we just + * always ignore it. + */ +#define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null" + +/* Attempts to set input rate limiting (policing) policy. */ +static int +netdev_linux_set_policing(struct netdev *netdev, + uint32_t kbits_rate, uint32_t kbits_burst) +{ + const char *netdev_name = netdev_get_name(netdev); + char command[1024]; + + COVERAGE_INC(netdev_set_policing); + if (kbits_rate) { + if (!kbits_burst) { + /* Default to 10 kilobits if not specified. */ + kbits_burst = 10; + } + + /* xxx This should be more careful about only adding if it + * xxx actually exists, as opposed to always deleting it. */ + snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name); + if (system(command) == -1) { + VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name); + } + + snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name); + if (system(command) != 0) { + VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name); + return -1; + } + + snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name, + kbits_rate, kbits_burst); + if (system(command) != 0) { + VLOG_WARN_RL(&rl, "%s: problem configuring policing", + netdev_name); + return -1; + } + } else { + snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name); + if (system(command) == -1) { + VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name); + } + } + + return 0; +} + +static int +netdev_linux_get_in4(const struct netdev *netdev_, + struct in_addr *address, struct in_addr *netmask) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache->valid & VALID_IN4)) { + int error; + + error = netdev_linux_get_ipv4(netdev_, &netdev->cache->address, + SIOCGIFADDR, "SIOCGIFADDR"); + if (error) { + return error; + } + + error = netdev_linux_get_ipv4(netdev_, &netdev->cache->netmask, + SIOCGIFNETMASK, "SIOCGIFNETMASK"); + if (error) { + return error; + } + + netdev->cache->valid |= VALID_IN4; + } + *address = netdev->cache->address; + *netmask = netdev->cache->netmask; + return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; +} + +static int +netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, + struct in_addr netmask) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); + if (!error) { + netdev->cache->valid |= VALID_IN4; + netdev->cache->address = address; + netdev->cache->netmask = netmask; + if (address.s_addr != INADDR_ANY) { + error = do_set_addr(netdev_, SIOCSIFNETMASK, + "SIOCSIFNETMASK", netmask); + } + } + return error; +} + +static bool +parse_if_inet6_line(const char *line, + struct in6_addr *in6, char ifname[16 + 1]) +{ + uint8_t *s6 = in6->s6_addr; +#define X8 "%2"SCNx8 + return sscanf(line, + " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 + "%*x %*x %*x %*x %16s\n", + &s6[0], &s6[1], &s6[2], &s6[3], + &s6[4], &s6[5], &s6[6], &s6[7], + &s6[8], &s6[9], &s6[10], &s6[11], + &s6[12], &s6[13], &s6[14], &s6[15], + ifname) == 17; +} + +/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if + * 'in6' is non-null) and returns true. Otherwise, returns false. */ +static int +netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache->valid & VALID_IN6)) { + FILE *file; + char line[128]; + + netdev->cache->in6 = in6addr_any; + + file = fopen("/proc/net/if_inet6", "r"); + if (file != NULL) { + const char *name = netdev_get_name(netdev_); + while (fgets(line, sizeof line, file)) { + struct in6_addr in6; + char ifname[16 + 1]; + if (parse_if_inet6_line(line, &in6, ifname) + && !strcmp(name, ifname)) + { + netdev->cache->in6 = in6; + break; + } + } + fclose(file); + } + netdev->cache->valid |= VALID_IN6; + } + *in6 = netdev->cache->in6; + return 0; +} + +static void +make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) +{ + struct sockaddr_in sin; + memset(&sin, 0, sizeof sin); + sin.sin_family = AF_INET; + sin.sin_addr = addr; + sin.sin_port = 0; + + memset(sa, 0, sizeof *sa); + memcpy(sa, &sin, sizeof sin); +} + +static int +do_set_addr(struct netdev *netdev, + int ioctl_nr, const char *ioctl_name, struct in_addr addr) +{ + struct ifreq ifr; + strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); + make_in4_sockaddr(&ifr.ifr_addr, addr); + return netdev_linux_do_ioctl(netdev, &ifr, ioctl_nr, ioctl_name); +} + +/* Adds 'router' as a default IP gateway. */ +static int +netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router) +{ + struct in_addr any = { INADDR_ANY }; + struct rtentry rt; + int error; + + memset(&rt, 0, sizeof rt); + make_in4_sockaddr(&rt.rt_dst, any); + make_in4_sockaddr(&rt.rt_gateway, router); + make_in4_sockaddr(&rt.rt_genmask, any); + rt.rt_flags = RTF_UP | RTF_GATEWAY; + COVERAGE_INC(netdev_add_router); + error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; + if (error) { + VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + } + return error; +} + +static int +netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, + char **netdev_name) +{ + static const char fn[] = "/proc/net/route"; + FILE *stream; + char line[256]; + int ln; + + *netdev_name = NULL; + stream = fopen(fn, "r"); + if (stream == NULL) { + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); + return errno; + } + + ln = 0; + while (fgets(line, sizeof line, stream)) { + if (++ln >= 2) { + char iface[17]; + uint32_t dest, gateway, mask; + int refcnt, metric, mtu; + unsigned int flags, use, window, irtt; + + if (sscanf(line, + "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 + " %d %u %u\n", + iface, &dest, &gateway, &flags, &refcnt, + &use, &metric, &mask, &mtu, &window, &irtt) != 11) { + + VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", + fn, ln, line); + continue; + } + if (!(flags & RTF_UP)) { + /* Skip routes that aren't up. */ + continue; + } + + /* The output of 'dest', 'mask', and 'gateway' were given in + * network byte order, so we don't need need any endian + * conversions here. */ + if ((dest & mask) == (host->s_addr & mask)) { + if (!gateway) { + /* The host is directly reachable. */ + next_hop->s_addr = 0; + } else { + /* To reach the host, we must go through a gateway. */ + next_hop->s_addr = gateway; + } + *netdev_name = xstrdup(iface); + fclose(stream); + return 0; + } + } + } + + fclose(stream); + return ENXIO; +} + +/* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be + * successfully retrieved, it stores the corresponding MAC address in 'mac' and + * returns 0. Otherwise, it returns a positive errno value; in particular, + * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ +static int +netdev_linux_arp_lookup(const struct netdev *netdev, + uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) +{ + struct arpreq r; + struct sockaddr_in *pa; + int retval; + + memset(&r, 0, sizeof r); + pa = (struct sockaddr_in *) &r.arp_pa; + pa->sin_family = AF_INET; + pa->sin_addr.s_addr = ip; + pa->sin_port = 0; + r.arp_ha.sa_family = ARPHRD_ETHER; + r.arp_flags = 0; + strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev); + COVERAGE_INC(netdev_arp_lookup); + retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; + if (!retval) { + memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); + } else if (retval != ENXIO) { + VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", + netdev->name, IP_ARGS(&ip), strerror(retval)); + } + return retval; +} + +static int +nd_to_iff_flags(enum netdev_flags nd) +{ + int iff = 0; + if (nd & NETDEV_UP) { + iff |= IFF_UP; + } + if (nd & NETDEV_PROMISC) { + iff |= IFF_PROMISC; + } + return iff; +} + +static int +iff_to_nd_flags(int iff) +{ + enum netdev_flags nd = 0; + if (iff & IFF_UP) { + nd |= NETDEV_UP; + } + if (iff & IFF_PROMISC) { + nd |= NETDEV_PROMISC; + } + return nd; +} + +static int +netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) +{ + int old_flags, new_flags; + int error; + + error = get_flags(netdev, &old_flags); + if (!error) { + *old_flagsp = iff_to_nd_flags(old_flags); + new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); + if (new_flags != old_flags) { + error = set_flags(netdev, new_flags); + } + } + return error; +} + +static void +poll_notify(struct list *list) +{ + struct netdev_linux_notifier *notifier; + LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) { + struct netdev_notifier *n = ¬ifier->notifier; + n->cb(n); + } +} + +static void +netdev_linux_poll_cb(const struct rtnetlink_change *change, + void *aux UNUSED) +{ + if (change) { + struct list *list = shash_find_data(&netdev_linux_notifiers, + change->ifname); + if (list) { + poll_notify(list); + } + } else { + struct shash_node *node; + SHASH_FOR_EACH (node, &netdev_linux_notifiers) { + poll_notify(node->data); + } + } +} + +static int +netdev_linux_poll_add(struct netdev *netdev, + void (*cb)(struct netdev_notifier *), void *aux, + struct netdev_notifier **notifierp) +{ + const char *netdev_name = netdev_get_name(netdev); + struct netdev_linux_notifier *notifier; + struct list *list; + + if (shash_is_empty(&netdev_linux_notifiers)) { + int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier, + netdev_linux_poll_cb, NULL); + if (error) { + return error; + } + } + + list = shash_find_data(&netdev_linux_notifiers, netdev_name); + if (!list) { + list = xmalloc(sizeof *list); + list_init(list); + shash_add(&netdev_linux_notifiers, netdev_name, list); + } + + notifier = xmalloc(sizeof *notifier); + netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); + list_push_back(list, ¬ifier->node); + *notifierp = ¬ifier->notifier; + return 0; +} + +static void +netdev_linux_poll_remove(struct netdev_notifier *notifier_) +{ + struct netdev_linux_notifier *notifier = + CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier); + struct list *list; + + /* Remove 'notifier' from its list. */ + list = list_remove(¬ifier->node); + if (list_is_empty(list)) { + /* The list is now empty. Remove it from the hash and free it. */ + const char *netdev_name = netdev_get_name(notifier->notifier.netdev); + shash_delete(&netdev_linux_notifiers, + shash_find(&netdev_linux_notifiers, netdev_name)); + free(list); + } + free(notifier); + + /* If that was the last notifier, unregister. */ + if (shash_is_empty(&netdev_linux_notifiers)) { + rtnetlink_notifier_unregister(&netdev_linux_poll_notifier); + } +} + +const struct netdev_class netdev_linux_class = { + "", /* prefix */ + "linux", /* name */ + + netdev_linux_init, + netdev_linux_run, + netdev_linux_wait, + + netdev_linux_open, + netdev_linux_close, + + netdev_linux_enumerate, + + netdev_linux_recv, + netdev_linux_recv_wait, + netdev_linux_drain, + + netdev_linux_send, + netdev_linux_send_wait, + + netdev_linux_set_etheraddr, + netdev_linux_get_etheraddr, + netdev_linux_get_mtu, + netdev_linux_get_carrier, + netdev_linux_get_stats, + + netdev_linux_get_features, + netdev_linux_set_advertisements, + netdev_linux_get_vlan_vid, + netdev_linux_set_policing, + + netdev_linux_get_in4, + netdev_linux_set_in4, + netdev_linux_get_in6, + netdev_linux_add_router, + netdev_linux_get_next_hop, + netdev_linux_arp_lookup, + + netdev_linux_update_flags, + + netdev_linux_poll_add, + netdev_linux_poll_remove, +}; + +const struct netdev_class netdev_tap_class = { + "tap", /* prefix */ + "tap", /* name */ + + netdev_linux_init, + NULL, /* run */ + NULL, /* wait */ + + netdev_linux_open, + netdev_linux_close, + + netdev_linux_enumerate, + + netdev_linux_recv, + netdev_linux_recv_wait, + netdev_linux_drain, + + netdev_linux_send, + netdev_linux_send_wait, + + netdev_linux_set_etheraddr, + netdev_linux_get_etheraddr, + netdev_linux_get_mtu, + netdev_linux_get_carrier, + netdev_linux_get_stats, + + netdev_linux_get_features, + netdev_linux_set_advertisements, + netdev_linux_get_vlan_vid, + netdev_linux_set_policing, + + netdev_linux_get_in4, + netdev_linux_set_in4, + netdev_linux_get_in6, + netdev_linux_add_router, + netdev_linux_get_next_hop, + netdev_linux_arp_lookup, + + netdev_linux_update_flags, + + netdev_linux_poll_add, + netdev_linux_poll_remove, +}; + +static int +get_stats_via_netlink(int ifindex, struct netdev_stats *stats) +{ + /* 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) }, + }; + + + static struct nl_sock *rtnl_sock; + struct ofpbuf request; + struct ofpbuf *reply; + struct ifinfomsg *ifi; + const struct rtnl_link_stats *rtnl_stats; + struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; + int error; + + if (!rtnl_sock) { + error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); + if (error) { + VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", + strerror(error)); + return error; + } + } + + ofpbuf_init(&request, 0); + nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi, + RTM_GETLINK, NLM_F_REQUEST); + ifi = ofpbuf_put_zeros(&request, sizeof *ifi); + ifi->ifi_family = PF_UNSPEC; + ifi->ifi_index = ifindex; + error = nl_sock_transact(rtnl_sock, &request, &reply); + ofpbuf_uninit(&request); + if (error) { + return error; + } + + if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), + rtnlgrp_link_policy, + attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { + ofpbuf_delete(reply); + return EPROTO; + } + + if (!attrs[IFLA_STATS]) { + VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); ++ ofpbuf_delete(reply); + return EPROTO; + } + + rtnl_stats = nl_attr_get(attrs[IFLA_STATS]); + stats->rx_packets = rtnl_stats->rx_packets; + stats->tx_packets = rtnl_stats->tx_packets; + stats->rx_bytes = rtnl_stats->rx_bytes; + stats->tx_bytes = rtnl_stats->tx_bytes; + stats->rx_errors = rtnl_stats->rx_errors; + stats->tx_errors = rtnl_stats->tx_errors; + stats->rx_dropped = rtnl_stats->rx_dropped; + stats->tx_dropped = rtnl_stats->tx_dropped; + stats->multicast = rtnl_stats->multicast; + stats->collisions = rtnl_stats->collisions; + stats->rx_length_errors = rtnl_stats->rx_length_errors; + stats->rx_over_errors = rtnl_stats->rx_over_errors; + stats->rx_crc_errors = rtnl_stats->rx_crc_errors; + stats->rx_frame_errors = rtnl_stats->rx_frame_errors; + stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors; + stats->rx_missed_errors = rtnl_stats->rx_missed_errors; + stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors; + stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors; + stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors; + stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors; + stats->tx_window_errors = rtnl_stats->tx_window_errors; + ++ 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; +} + +static int +get_flags(const struct netdev *netdev, int *flags) +{ + struct ifreq ifr; + int error; + + error = netdev_linux_do_ioctl(netdev, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); + *flags = ifr.ifr_flags; + return error; +} + +static int +set_flags(struct netdev *netdev, int flags) +{ + struct ifreq ifr; + + ifr.ifr_flags = flags; + return netdev_linux_do_ioctl(netdev, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); +} + +static int +do_get_ifindex(const char *netdev_name) +{ + struct ifreq ifr; + + strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + COVERAGE_INC(netdev_get_ifindex); + if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { + VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", + netdev_name, strerror(errno)); + return -errno; + } + return ifr.ifr_ifindex; +} + +static int +get_ifindex(const struct netdev *netdev_, int *ifindexp) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + *ifindexp = 0; + if (!(netdev->cache->valid & VALID_IFINDEX)) { + int ifindex = do_get_ifindex(netdev_get_name(netdev_)); + if (ifindex < 0) { + return -ifindex; + } + netdev->cache->valid |= VALID_IFINDEX; + netdev->cache->ifindex = ifindex; + } + *ifindexp = netdev->cache->ifindex; + return 0; +} + +static int +get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) +{ + struct ifreq ifr; + int hwaddr_family; + + memset(&ifr, 0, sizeof ifr); + strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + COVERAGE_INC(netdev_get_hwaddr); + if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { + VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s", + netdev_name, strerror(errno)); + return errno; + } + hwaddr_family = ifr.ifr_hwaddr.sa_family; + if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { + VLOG_WARN("%s device has unknown hardware address family %d", + netdev_name, hwaddr_family); + } + memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); + return 0; +} + +static int +set_etheraddr(const char *netdev_name, int hwaddr_family, + const uint8_t mac[ETH_ADDR_LEN]) +{ + struct ifreq ifr; + + memset(&ifr, 0, sizeof ifr); + strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + ifr.ifr_hwaddr.sa_family = hwaddr_family; + memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); + COVERAGE_INC(netdev_set_hwaddr); + if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { + VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", + netdev_name, strerror(errno)); + return errno; + } + return 0; +} + +static int +netdev_linux_do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd, + int cmd, const char *cmd_name) +{ + struct ifreq ifr; + + memset(&ifr, 0, sizeof ifr); + strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name); + ifr.ifr_data = (caddr_t) ecmd; + + ecmd->cmd = cmd; + COVERAGE_INC(netdev_ethtool); + if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { + return 0; + } else { + if (errno != EOPNOTSUPP) { + VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " + "failed: %s", cmd_name, netdev->name, + strerror(errno)); + } else { + /* The device doesn't support this operation. That's pretty + * common, so there's no point in logging anything. */ + } + return errno; + } +} + +static int +netdev_linux_do_ioctl(const struct netdev *netdev, struct ifreq *ifr, + int cmd, const char *cmd_name) +{ + strncpy(ifr->ifr_name, netdev_get_name(netdev), sizeof ifr->ifr_name); + if (ioctl(af_inet_sock, cmd, ifr) == -1) { + VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", + netdev_get_name(netdev), cmd_name, strerror(errno)); + return errno; + } + return 0; +} + +static int +netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, + int cmd, const char *cmd_name) +{ + struct ifreq ifr; + int error; + + ifr.ifr_addr.sa_family = AF_INET; + error = netdev_linux_do_ioctl(netdev, &ifr, cmd, cmd_name); + if (!error) { + const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; + *ip = sin->sin_addr; + } + return error; +} diff --cc ofproto/fail-open.c index 60890d49,00000000..48f70694 mode 100644,000000..100644 --- a/ofproto/fail-open.c +++ b/ofproto/fail-open.c @@@ -1,141 -1,0 +1,243 @@@ +/* + * Copyright (c) 2008, 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "fail-open.h" +#include +#include +#include "flow.h" +#include "mac-learning.h" +#include "odp-util.h" ++#include "ofpbuf.h" +#include "ofproto.h" ++#include "pktbuf.h" ++#include "poll-loop.h" +#include "rconn.h" +#include "status.h" +#include "timeval.h" ++#include "vconn.h" + +#define THIS_MODULE VLM_fail_open +#include "vlog.h" + ++/* ++ * Fail-open mode. ++ * ++ * In fail-open mode, the switch detects when the controller cannot be ++ * contacted or when the controller is dropping switch connections because the ++ * switch does not pass its admission control policy. In those situations the ++ * switch sets up flows itself using the "normal" action. ++ * ++ * There is a little subtlety to implementation, to properly handle the case ++ * where the controller allows switch connections but drops them a few seconds ++ * later for admission control reasons. Because of this case, we don't want to ++ * just stop setting up flows when we connect to the controller: if we did, ++ * then new flow setup and existing flows would stop during the duration of ++ * connection to the controller, and thus the whole network would go down for ++ * that period of time. ++ * ++ * So, instead, we add some special caseswhen we are connected to a controller, ++ * but not yet sure that it has admitted us: ++ * ++ * - We set up flows immediately ourselves, but simultaneously send out an ++ * OFPT_PACKET_IN to the controller. We put a special bogus buffer-id in ++ * these OFPT_PACKET_IN messages so that duplicate packets don't get sent ++ * out to the network when the controller replies. ++ * ++ * - We also send out OFPT_PACKET_IN messages for totally bogus packets ++ * every so often, in case no real new flows are arriving in the network. ++ * ++ * - We don't flush the flow table at the time we connect, because this ++ * could cause network stuttering in a switch with lots of flows or very ++ * high-bandwidth flows by suddenly throwing lots of packets down to ++ * userspace. ++ */ ++ +struct fail_open { + struct ofproto *ofproto; + struct rconn *controller; + int trigger_duration; + int last_disconn_secs; + struct status_category *ss_cat; ++ long long int next_bogus_packet_in; ++ struct rconn_packet_counter *bogus_packet_counter; +}; + - /* Causes the switch to enter or leave fail-open mode, if appropriate. */ - void - fail_open_run(struct fail_open *fo) ++/* Returns true if 'fo' should be in fail-open mode, otherwise false. */ ++static inline bool ++should_fail_open(const struct fail_open *fo) +{ - int disconn_secs = rconn_failure_duration(fo->controller); - bool open = disconn_secs >= fo->trigger_duration; - if (open != (fo->last_disconn_secs != 0)) { - if (!open) { - flow_t flow; ++ return rconn_failure_duration(fo->controller) >= fo->trigger_duration; ++} ++ ++/* Returns true if 'fo' is currently in fail-open mode, otherwise false. */ ++bool ++fail_open_is_active(const struct fail_open *fo) ++{ ++ return fo->last_disconn_secs != 0; ++} + - VLOG_WARN("No longer in fail-open mode"); - fo->last_disconn_secs = 0; ++static void ++send_bogus_packet_in(struct fail_open *fo) ++{ ++ uint8_t mac[ETH_ADDR_LEN]; ++ struct ofpbuf *opi; ++ struct ofpbuf b; + - memset(&flow, 0, sizeof flow); - ofproto_delete_flow(fo->ofproto, &flow, OFPFW_ALL, 70000); - } else { ++ /* Compose ofp_packet_in. */ ++ ofpbuf_init(&b, 128); ++ eth_addr_random(mac); ++ compose_benign_packet(&b, "Open vSwitch Controller Probe", 0xa033, mac); ++ opi = make_packet_in(pktbuf_get_null(), OFPP_LOCAL, OFPR_NO_MATCH, &b, 64); ++ ofpbuf_uninit(&b); ++ ++ /* Send. */ ++ rconn_send_with_limit(fo->controller, opi, fo->bogus_packet_counter, 1); ++} ++ ++/* Enter fail-open mode if we should be in it. Handle reconnecting to a ++ * controller from fail-open mode. */ ++void ++fail_open_run(struct fail_open *fo) ++{ ++ /* Enter fail-open mode if 'fo' is not in it but should be. */ ++ if (should_fail_open(fo)) { ++ int disconn_secs = rconn_failure_duration(fo->controller); ++ if (!fail_open_is_active(fo)) { + VLOG_WARN("Could not connect to controller (or switch failed " + "controller's post-connection admission control " + "policy) for %d seconds, failing open", disconn_secs); + fo->last_disconn_secs = disconn_secs; + + /* Flush all OpenFlow and datapath flows. We will set up our + * fail-open rule from fail_open_flushed() when + * ofproto_flush_flows() calls back to us. */ + ofproto_flush_flows(fo->ofproto); ++ } else if (disconn_secs > fo->last_disconn_secs + 60) { ++ VLOG_INFO("Still in fail-open mode after %d seconds disconnected " ++ "from controller", disconn_secs); ++ fo->last_disconn_secs = disconn_secs; + } - } else if (open && disconn_secs > fo->last_disconn_secs + 60) { - VLOG_INFO("Still in fail-open mode after %d seconds disconnected " - "from controller", disconn_secs); - fo->last_disconn_secs = disconn_secs; + } ++ ++ /* Schedule a bogus packet-in if we're connected and in fail-open. */ ++ if (fail_open_is_active(fo)) { ++ if (rconn_is_connected(fo->controller)) { ++ bool expired = time_msec() >= fo->next_bogus_packet_in; ++ if (expired) { ++ send_bogus_packet_in(fo); ++ } ++ if (expired || fo->next_bogus_packet_in == LLONG_MAX) { ++ fo->next_bogus_packet_in = time_msec() + 2000; ++ } ++ } else { ++ fo->next_bogus_packet_in = LLONG_MAX; ++ } ++ } ++ +} + ++/* If 'fo' is currently in fail-open mode and its rconn has connected to the ++ * controller, exits fail open mode. */ +void - fail_open_wait(struct fail_open *fo UNUSED) ++fail_open_maybe_recover(struct fail_open *fo) +{ - /* Nothing to do. */ ++ if (fail_open_is_active(fo) && rconn_is_admitted(fo->controller)) { ++ flow_t flow; ++ ++ VLOG_WARN("No longer in fail-open mode"); ++ fo->last_disconn_secs = 0; ++ fo->next_bogus_packet_in = LLONG_MAX; ++ ++ memset(&flow, 0, sizeof flow); ++ ofproto_delete_flow(fo->ofproto, &flow, OFPFW_ALL, FAIL_OPEN_PRIORITY); ++ } ++} ++ ++void ++fail_open_wait(struct fail_open *fo) ++{ ++ if (fo->next_bogus_packet_in != LLONG_MAX) { ++ poll_timer_wait(fo->next_bogus_packet_in - time_msec()); ++ } +} + +void +fail_open_flushed(struct fail_open *fo) +{ + int disconn_secs = rconn_failure_duration(fo->controller); + bool open = disconn_secs >= fo->trigger_duration; + if (open) { + union ofp_action action; + flow_t flow; + + /* Set up a flow that matches every packet and directs them to + * OFPP_NORMAL. */ + memset(&action, 0, sizeof action); + action.type = htons(OFPAT_OUTPUT); + action.output.len = htons(sizeof action); + action.output.port = htons(OFPP_NORMAL); + memset(&flow, 0, sizeof flow); - ofproto_add_flow(fo->ofproto, &flow, OFPFW_ALL, 70000, ++ ofproto_add_flow(fo->ofproto, &flow, OFPFW_ALL, FAIL_OPEN_PRIORITY, + &action, 1, 0); + } +} + +static void +fail_open_status_cb(struct status_reply *sr, void *fo_) +{ + struct fail_open *fo = fo_; + int cur_duration = rconn_failure_duration(fo->controller); + + status_reply_put(sr, "trigger-duration=%d", fo->trigger_duration); + status_reply_put(sr, "current-duration=%d", cur_duration); + status_reply_put(sr, "triggered=%s", + cur_duration >= fo->trigger_duration ? "true" : "false"); +} + +struct fail_open * +fail_open_create(struct ofproto *ofproto, + int trigger_duration, struct switch_status *switch_status, + struct rconn *controller) +{ + struct fail_open *fo = xmalloc(sizeof *fo); + fo->ofproto = ofproto; + fo->controller = controller; + fo->trigger_duration = trigger_duration; + fo->last_disconn_secs = 0; + fo->ss_cat = switch_status_register(switch_status, "fail-open", + fail_open_status_cb, fo); ++ fo->next_bogus_packet_in = LLONG_MAX; ++ fo->bogus_packet_counter = rconn_packet_counter_create(); + return fo; +} + +void +fail_open_set_trigger_duration(struct fail_open *fo, int trigger_duration) +{ + fo->trigger_duration = trigger_duration; +} + +void +fail_open_destroy(struct fail_open *fo) +{ + if (fo) { + /* We don't own fo->controller. */ + switch_status_unregister(fo->ss_cat); ++ rconn_packet_counter_destroy(fo->bogus_packet_counter); + free(fo); + } +} diff --cc ofproto/fail-open.h index c0ada2eb,00000000..900d587e mode 100644,000000..100644 --- a/ofproto/fail-open.h +++ b/ofproto/fail-open.h @@@ -1,38 -1,0 +1,46 @@@ +/* + * Copyright (c) 2008, 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef FAIL_OPEN_H +#define FAIL_OPEN_H 1 + +#include +#include +#include "flow.h" + +struct fail_open; +struct ofproto; +struct rconn; +struct switch_status; + ++/* Priority of the rule added by the fail-open subsystem when a switch enters ++ * fail-open mode. This priority value uniquely identifies a fail-open flow ++ * (OpenFlow priorities max out at 65535 and nothing else in Open vSwitch ++ * creates flows with this priority). */ ++#define FAIL_OPEN_PRIORITY 70000 ++ +struct fail_open *fail_open_create(struct ofproto *, int trigger_duration, + struct switch_status *, + struct rconn *controller); +void fail_open_set_trigger_duration(struct fail_open *, int trigger_duration); +void fail_open_destroy(struct fail_open *); +void fail_open_wait(struct fail_open *); ++bool fail_open_is_active(const struct fail_open *); +void fail_open_run(struct fail_open *); ++void fail_open_maybe_recover(struct fail_open *); +void fail_open_flushed(struct fail_open *); + +#endif /* fail-open.h */ diff --cc ofproto/in-band.c index 35ea534c,00000000..2b362bc0 mode 100644,000000..100644 --- a/ofproto/in-band.c +++ b/ofproto/in-band.c @@@ -1,502 -1,0 +1,652 @@@ +/* + * Copyright (c) 2008, 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "in-band.h" +#include +#include +#include +#include +#include +#include +#include "dhcp.h" +#include "dpif.h" +#include "flow.h" +#include "mac-learning.h" +#include "netdev.h" +#include "odp-util.h" +#include "ofp-print.h" +#include "ofproto.h" +#include "ofpbuf.h" +#include "openflow/openflow.h" +#include "openvswitch/datapath-protocol.h" +#include "packets.h" +#include "poll-loop.h" +#include "rconn.h" +#include "status.h" +#include "timeval.h" +#include "vconn.h" + +#define THIS_MODULE VLM_in_band +#include "vlog.h" + ++/* In-band control allows a single network to be used for OpenFlow ++ * traffic and other data traffic. Refer to ovs-vswitchd.conf(5) and ++ * secchan(8) for a description of configuring in-band control. ++ * ++ * This comment is an attempt to describe how in-band control works at a ++ * wire- and implementation-level. Correctly implementing in-band ++ * control has proven difficult due to its many subtleties, and has thus ++ * gone through many iterations. Please read through and understand the ++ * reasoning behind the chosen rules before making modifications. ++ * ++ * In Open vSwitch, in-band control is implemented as "hidden" flows (in ++ * that they are not visible through OpenFlow) and at a higher priority ++ * than wildcarded flows can be setup by the controller. This is done ++ * so that the controller cannot interfere with them and possibly break ++ * connectivity with its switches. It is possible to see all flows, ++ * including in-band ones, with the ovs-appctl "bridge/dump-flows" ++ * command. ++ * ++ * The following rules are always enabled with the "normal" action by a ++ * switch with in-band control: ++ * ++ * a. DHCP requests sent from the local port. ++ * b. ARP replies to the local port's MAC address. ++ * c. ARP requests from the local port's MAC address. ++ * d. ARP replies to the remote side's MAC address. Note that the ++ * remote side is either the controller or the gateway to reach ++ * the controller. ++ * e. ARP requests from the remote side's MAC address. Note that ++ * like (d), the MAC is either for the controller or gateway. ++ * f. ARP replies containing the controller's IP address as a target. ++ * g. ARP requests containing the controller's IP address as a source. ++ * h. OpenFlow (6633/tcp) traffic to the controller's IP. ++ * i. OpenFlow (6633/tcp) traffic from the controller's IP. ++ * ++ * The goal of these rules is to be as narrow as possible to allow a ++ * switch to join a network and be able to communicate with a ++ * controller. As mentioned earlier, these rules have higher priority ++ * than the controller's rules, so if they are too broad, they may ++ * prevent the controller from implementing its policy. As such, ++ * in-band actively monitors some aspects of flow and packet processing ++ * so that the rules can be made more precise. ++ * ++ * In-band control monitors attempts to add flows into the datapath that ++ * could interfere with its duties. The datapath only allows exact ++ * match entries, so in-band control is able to be very precise about ++ * the flows it prevents. Flows that miss in the datapath are sent to ++ * userspace to be processed, so preventing these flows from being ++ * cached in the "fast path" does not affect correctness. The only type ++ * of flow that is currently prevented is one that would prevent DHCP ++ * replies from being seen by the local port. For example, a rule that ++ * forwarded all DHCP traffic to the controller would not be allowed, ++ * but one that forwarded to all ports (including the local port) would. ++ * ++ * As mentioned earlier, packets that miss in the datapath are sent to ++ * the userspace for processing. The userspace has its own flow table, ++ * the "classifier", so in-band checks whether any special processing ++ * is needed before the classifier is consulted. If a packet is a DHCP ++ * response to a request from the local port, the packet is forwarded to ++ * the local port, regardless of the flow table. Note that this requires ++ * L7 processing of DHCP replies to determine whether the 'chaddr' field ++ * matches the MAC address of the local port. ++ * ++ * It is interesting to note that for an L3-based in-band control ++ * mechanism, the majority of rules are devoted to ARP traffic. At first ++ * glance, some of these rules appear redundant. However, each serves an ++ * important role. First, in order to determine the MAC address of the ++ * remote side (controller or gateway) for other ARP rules, we must allow ++ * ARP traffic for our local port with rules (b) and (c). If we are ++ * between a switch and its connection to the controller, we have to ++ * allow the other switch's ARP traffic to through. This is done with ++ * rules (d) and (e), since we do not know the addresses of the other ++ * switches a priori, but do know the controller's or gateway's. Finally, ++ * if the controller is running in a local guest VM that is not reached ++ * through the local port, the switch that is connected to the VM must ++ * allow ARP traffic based on the controller's IP address, since it will ++ * not know the MAC address of the local port that is sending the traffic ++ * or the MAC address of the controller in the guest VM. ++ * ++ * With a few notable exceptions below, in-band should work in most ++ * network setups. The following are considered "supported' in the ++ * current implementation: ++ * ++ * - Locally Connected. The switch and controller are on the same ++ * subnet. This uses rules (a), (b), (c), (h), and (i). ++ * ++ * - Reached through Gateway. The switch and controller are on ++ * different subnets and must go through a gateway. This uses ++ * rules (a), (b), (c), (h), and (i). ++ * ++ * - Between Switch and Controller. This switch is between another ++ * switch and the controller, and we want to allow the other ++ * switch's traffic through. This uses rules (d), (e), (h), and ++ * (i). It uses (b) and (c) indirectly in order to know the MAC ++ * address for rules (d) and (e). Note that DHCP for the other ++ * switch will not work unless the controller explicitly lets this ++ * switch pass the traffic. ++ * ++ * - Between Switch and Gateway. This switch is between another ++ * switch and the gateway, and we want to allow the other switch's ++ * traffic through. This uses the same rules and logic as the ++ * "Between Switch and Controller" configuration described earlier. ++ * ++ * - Controller on Local VM. The controller is a guest VM on the ++ * system running in-band control. This uses rules (a), (b), (c), ++ * (h), and (i). ++ * ++ * - Controller on Local VM with Different Networks. The controller ++ * is a guest VM on the system running in-band control, but the ++ * local port is not used to connect to the controller. For ++ * example, an IP address is configured on eth0 of the switch. The ++ * controller's VM is connected through eth1 of the switch, but an ++ * IP address has not been configured for that port on the switch. ++ * As such, the switch will use eth0 to connect to the controller, ++ * and eth1's rules about the local port will not work. In the ++ * example, the switch attached to eth0 would use rules (a), (b), ++ * (c), (h), and (i) on eth0. The switch attached to eth1 would use ++ * rules (f), (g), (h), and (i). ++ * ++ * The following are explicitly *not* supported by in-band control: ++ * ++ * - Specify Controller by Name. Currently, the controller must be ++ * identified by IP address. A naive approach would be to permit ++ * all DNS traffic. Unfortunately, this would prevent the ++ * controller from defining any policy over DNS. Since switches ++ * that are located behind us need to connect to the controller, ++ * in-band cannot simply add a rule that allows DNS traffic from ++ * the local port. The "correct" way to support this is to parse ++ * DNS requests to allow all traffic related to a request for the ++ * controller's name through. Due to the potential security ++ * problems and amount of processing, we decided to hold off for ++ * the time-being. ++ * ++ * - Multiple Controllers. There is nothing intrinsic in the high- ++ * level design that prevents using multiple (known) controllers, ++ * however, the current implementation's data structures assume ++ * only one. ++ * ++ * - Differing Controllers for Switches. All switches must know ++ * the L3 addresses for all the controllers that other switches ++ * may use, since rules need to be setup to allow traffic related ++ * to those controllers through. See rules (f), (g), (h), and (i). ++ * ++ * - Differing Routes for Switches. In order for the switch to ++ * allow other switches to connect to a controller through a ++ * gateway, it allows the gateway's traffic through with rules (d) ++ * and (e). If the routes to the controller differ for the two ++ * switches, we will not know the MAC address of the alternate ++ * gateway. ++ */ ++ +#define IB_BASE_PRIORITY 18181800 + +enum { - IBR_FROM_LOCAL_DHCP, /* From local port, DHCP. */ - IBR_TO_LOCAL_ARP, /* To local port, ARP. */ - IBR_FROM_LOCAL_ARP, /* From local port, ARP. */ - IBR_TO_REMOTE_ARP, /* To remote MAC, ARP. */ - IBR_FROM_REMOTE_ARP, /* From remote MAC, ARP. */ - IBR_TO_CTL_ARP, /* To controller IP, ARP. */ - IBR_FROM_CTL_ARP, /* From controller IP, ARP. */ - IBR_TO_CTL_OFP, /* To controller, OpenFlow port. */ - IBR_FROM_CTL_OFP, /* From controller, OpenFlow port. */ ++ IBR_FROM_LOCAL_DHCP, /* (a) From local port, DHCP. */ ++ IBR_TO_LOCAL_ARP, /* (b) To local port, ARP. */ ++ IBR_FROM_LOCAL_ARP, /* (c) From local port, ARP. */ ++ IBR_TO_REMOTE_ARP, /* (d) To remote MAC, ARP. */ ++ IBR_FROM_REMOTE_ARP, /* (e) From remote MAC, ARP. */ ++ IBR_TO_CTL_ARP, /* (f) To controller IP, ARP. */ ++ IBR_FROM_CTL_ARP, /* (g) From controller IP, ARP. */ ++ IBR_TO_CTL_OFP, /* (h) To controller, OpenFlow port. */ ++ IBR_FROM_CTL_OFP, /* (i) From controller, OpenFlow port. */ +#if OFP_TCP_PORT != OFP_SSL_PORT +#error Need to support separate TCP and SSL flows. +#endif + N_IB_RULES +}; + +struct ib_rule { + bool installed; + flow_t flow; + uint32_t wildcards; + unsigned int priority; +}; + +struct in_band { + struct ofproto *ofproto; + struct rconn *controller; + struct status_category *ss_cat; + + /* Keep track of local port's information. */ + uint8_t local_mac[ETH_ADDR_LEN]; /* Current MAC. */ + struct netdev *local_netdev; /* Local port's network device. */ + time_t next_local_refresh; + + /* Keep track of controller and next hop's information. */ + uint32_t controller_ip; /* Controller IP, 0 if unknown. */ + uint8_t remote_mac[ETH_ADDR_LEN]; /* Remote MAC. */ + struct netdev *remote_netdev; + uint8_t last_remote_mac[ETH_ADDR_LEN]; /* Previous remote MAC. */ + time_t next_remote_refresh; + + /* Rules that we set up. */ + struct ib_rule rules[N_IB_RULES]; +}; + +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60); + +static const uint8_t * +get_remote_mac(struct in_band *ib) +{ + int retval; + bool have_mac; + struct in_addr c_in4; /* Controller's IP address. */ + struct in_addr r_in4; /* Next hop IP address. */ + char *next_hop_dev; + time_t now = time_now(); + + if (now >= ib->next_remote_refresh) { + /* Find the next-hop IP address. */ + c_in4.s_addr = ib->controller_ip; + memset(ib->remote_mac, 0, sizeof ib->remote_mac); + retval = netdev_get_next_hop(ib->local_netdev, + &c_in4, &r_in4, &next_hop_dev); + if (retval) { + VLOG_WARN("cannot find route for controller ("IP_FMT"): %s", + IP_ARGS(&ib->controller_ip), strerror(retval)); + ib->next_remote_refresh = now + 1; + return NULL; + } + if (!r_in4.s_addr) { + r_in4.s_addr = c_in4.s_addr; + } + + /* Get the next-hop IP and network device. */ + if (!ib->remote_netdev + || strcmp(netdev_get_name(ib->remote_netdev), next_hop_dev)) + { + netdev_close(ib->remote_netdev); + retval = netdev_open(next_hop_dev, NETDEV_ETH_TYPE_NONE, + &ib->remote_netdev); + if (retval) { + VLOG_WARN_RL(&rl, "cannot open netdev %s (next hop " + "to controller "IP_FMT"): %s", + next_hop_dev, IP_ARGS(&ib->controller_ip), + strerror(retval)); + ib->next_remote_refresh = now + 1; + return NULL; + } + } + + /* Look up the MAC address of the next-hop IP address. */ + retval = netdev_arp_lookup(ib->remote_netdev, r_in4.s_addr, + ib->remote_mac); + if (retval) { + VLOG_DBG_RL(&rl, "cannot look up remote MAC address ("IP_FMT"): %s", + IP_ARGS(&r_in4.s_addr), strerror(retval)); + } + have_mac = !eth_addr_is_zero(ib->remote_mac); + free(next_hop_dev); + if (have_mac + && !eth_addr_equals(ib->last_remote_mac, ib->remote_mac)) { + VLOG_DBG("remote MAC address changed from "ETH_ADDR_FMT" to " + ETH_ADDR_FMT, + ETH_ADDR_ARGS(ib->last_remote_mac), + ETH_ADDR_ARGS(ib->remote_mac)); + memcpy(ib->last_remote_mac, ib->remote_mac, ETH_ADDR_LEN); + } + + /* Schedule next refresh. + * + * If we have an IP address but not a MAC address, then refresh + * quickly, since we probably will get a MAC address soon (via ARP). + * Otherwise, we can afford to wait a little while. */ + ib->next_remote_refresh + = now + (!ib->controller_ip || have_mac ? 10 : 1); + } + + return !eth_addr_is_zero(ib->remote_mac) ? ib->remote_mac : NULL; +} + +static const uint8_t * +get_local_mac(struct in_band *ib) +{ + time_t now = time_now(); + if (now >= ib->next_local_refresh) { + uint8_t ea[ETH_ADDR_LEN]; + if (ib->local_netdev && !netdev_get_etheraddr(ib->local_netdev, ea)) { + memcpy(ib->local_mac, ea, ETH_ADDR_LEN); + } + ib->next_local_refresh = now + 1; + } + return !eth_addr_is_zero(ib->local_mac) ? ib->local_mac : NULL; +} + +static void +in_band_status_cb(struct status_reply *sr, void *in_band_) +{ + struct in_band *in_band = in_band_; + + if (!eth_addr_is_zero(in_band->local_mac)) { + status_reply_put(sr, "local-mac="ETH_ADDR_FMT, + ETH_ADDR_ARGS(in_band->local_mac)); + } + + if (!eth_addr_is_zero(in_band->remote_mac)) { + status_reply_put(sr, "remote-mac="ETH_ADDR_FMT, + ETH_ADDR_ARGS(in_band->remote_mac)); + } +} + +static void +drop_flow(struct in_band *in_band, int rule_idx) +{ + struct ib_rule *rule = &in_band->rules[rule_idx]; + + if (rule->installed) { + rule->installed = false; + ofproto_delete_flow(in_band->ofproto, &rule->flow, rule->wildcards, + rule->priority); + } +} + +/* out_port and fixed_fields are assumed never to change. */ +static void +setup_flow(struct in_band *in_band, int rule_idx, const flow_t *flow, + uint32_t fixed_fields, uint16_t out_port) +{ + struct ib_rule *rule = &in_band->rules[rule_idx]; + + if (!rule->installed || memcmp(flow, &rule->flow, sizeof *flow)) { + union ofp_action action; + + drop_flow(in_band, rule_idx); + + rule->installed = true; + rule->flow = *flow; + rule->wildcards = OFPFW_ALL & ~fixed_fields; + rule->priority = IB_BASE_PRIORITY + (N_IB_RULES - rule_idx); + + action.type = htons(OFPAT_OUTPUT); + action.output.len = htons(sizeof action); + action.output.port = htons(out_port); + action.output.max_len = htons(0); + ofproto_add_flow(in_band->ofproto, &rule->flow, rule->wildcards, + rule->priority, &action, 1, 0); + } +} + +/* Returns true if 'packet' should be sent to the local port regardless + * of the flow table. */ +bool +in_band_msg_in_hook(struct in_band *in_band, const flow_t *flow, + const struct ofpbuf *packet) +{ + if (!in_band) { + return false; + } + + /* Regardless of how the flow table is configured, we want to be + * able to see replies to our DHCP requests. */ + if (flow->dl_type == htons(ETH_TYPE_IP) + && flow->nw_proto == IP_TYPE_UDP + && flow->tp_src == htons(DHCP_SERVER_PORT) + && flow->tp_dst == htons(DHCP_CLIENT_PORT) + && packet->l7) { + struct dhcp_header *dhcp; + const uint8_t *local_mac; + + dhcp = ofpbuf_at(packet, (char *)packet->l7 - (char *)packet->data, + sizeof *dhcp); + if (!dhcp) { + return false; + } + + local_mac = get_local_mac(in_band); + if (eth_addr_equals(dhcp->chaddr, local_mac)) { + return true; + } + } + + return false; +} + +/* Returns true if the rule that would match 'flow' with 'actions' is + * allowed to be set up in the datapath. */ +bool +in_band_rule_check(struct in_band *in_band, const flow_t *flow, + const struct odp_actions *actions) +{ + if (!in_band) { + return true; + } + + /* Don't allow flows that would prevent DHCP replies from being seen + * by the local port. */ + if (flow->dl_type == htons(ETH_TYPE_IP) + && flow->nw_proto == IP_TYPE_UDP + && flow->tp_src == htons(DHCP_SERVER_PORT) + && flow->tp_dst == htons(DHCP_CLIENT_PORT)) { + int i; + + for (i=0; in_actions; i++) { + if (actions->actions[i].output.type == ODPAT_OUTPUT + && actions->actions[i].output.port == ODPP_LOCAL) { + return true; + } + } + return false; + } + + return true; +} + +void +in_band_run(struct in_band *in_band) +{ + time_t now = time_now(); + uint32_t controller_ip; + const uint8_t *remote_mac; + const uint8_t *local_mac; + flow_t flow; + + if (now < in_band->next_remote_refresh + && now < in_band->next_local_refresh) { + return; + } + + controller_ip = rconn_get_remote_ip(in_band->controller); + if (in_band->controller_ip && controller_ip != in_band->controller_ip) { + VLOG_DBG("controller IP address changed from "IP_FMT" to "IP_FMT, + IP_ARGS(&in_band->controller_ip), + IP_ARGS(&controller_ip)); + } + in_band->controller_ip = controller_ip; + + remote_mac = get_remote_mac(in_band); + local_mac = get_local_mac(in_band); + + if (local_mac) { + /* Allow DHCP requests to be sent from the local port. */ + memset(&flow, 0, sizeof flow); + flow.in_port = ODPP_LOCAL; + flow.dl_type = htons(ETH_TYPE_IP); + memcpy(flow.dl_src, local_mac, ETH_ADDR_LEN); + flow.nw_proto = IP_TYPE_UDP; + flow.tp_src = htons(DHCP_CLIENT_PORT); + flow.tp_dst = htons(DHCP_SERVER_PORT); + setup_flow(in_band, IBR_FROM_LOCAL_DHCP, &flow, + (OFPFW_IN_PORT | OFPFW_DL_TYPE | OFPFW_DL_SRC + | OFPFW_NW_PROTO | OFPFW_TP_SRC | OFPFW_TP_DST), + OFPP_NORMAL); + + /* Allow the connection's interface to receive directed ARP traffic. */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_ARP); + memcpy(flow.dl_dst, local_mac, ETH_ADDR_LEN); + flow.nw_proto = ARP_OP_REPLY; + setup_flow(in_band, IBR_TO_LOCAL_ARP, &flow, + (OFPFW_DL_TYPE | OFPFW_DL_DST | OFPFW_NW_PROTO), + OFPP_NORMAL); + + /* Allow the connection's interface to be the source of ARP traffic. */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_ARP); + memcpy(flow.dl_src, local_mac, ETH_ADDR_LEN); + flow.nw_proto = ARP_OP_REQUEST; + setup_flow(in_band, IBR_FROM_LOCAL_ARP, &flow, + (OFPFW_DL_TYPE | OFPFW_DL_SRC | OFPFW_NW_PROTO), + OFPP_NORMAL); + } else { + drop_flow(in_band, IBR_TO_LOCAL_ARP); + drop_flow(in_band, IBR_FROM_LOCAL_ARP); + } + + if (remote_mac) { + /* Allow ARP replies to the remote side's MAC. */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_ARP); + memcpy(flow.dl_dst, remote_mac, ETH_ADDR_LEN); + flow.nw_proto = ARP_OP_REPLY; + setup_flow(in_band, IBR_TO_REMOTE_ARP, &flow, + (OFPFW_DL_TYPE | OFPFW_DL_DST | OFPFW_NW_PROTO), + OFPP_NORMAL); + + /* Allow ARP requests from the remote side's MAC. */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_ARP); + memcpy(flow.dl_src, remote_mac, ETH_ADDR_LEN); + flow.nw_proto = ARP_OP_REQUEST; + setup_flow(in_band, IBR_FROM_REMOTE_ARP, &flow, + (OFPFW_DL_TYPE | OFPFW_DL_SRC | OFPFW_NW_PROTO), + OFPP_NORMAL); + } else { + drop_flow(in_band, IBR_TO_REMOTE_ARP); + drop_flow(in_band, IBR_FROM_REMOTE_ARP); + } + + if (controller_ip) { + /* Allow ARP replies to the controller's IP. */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_ARP); + flow.nw_proto = ARP_OP_REPLY; + flow.nw_dst = controller_ip; + setup_flow(in_band, IBR_TO_CTL_ARP, &flow, + (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_DST_MASK), + OFPP_NORMAL); + + /* Allow ARP requests from the controller's IP. */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_ARP); + flow.nw_proto = ARP_OP_REQUEST; + flow.nw_src = controller_ip; + setup_flow(in_band, IBR_FROM_CTL_ARP, &flow, + (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_SRC_MASK), + OFPP_NORMAL); + + /* OpenFlow traffic to or from the controller. + * + * (A given field's value is completely ignored if it is wildcarded, + * which is why we can get away with using a single 'flow' in each + * case here.) */ + memset(&flow, 0, sizeof flow); + flow.dl_type = htons(ETH_TYPE_IP); + flow.nw_proto = IP_TYPE_TCP; + flow.nw_src = controller_ip; + flow.nw_dst = controller_ip; + flow.tp_src = htons(OFP_TCP_PORT); + flow.tp_dst = htons(OFP_TCP_PORT); + setup_flow(in_band, IBR_TO_CTL_OFP, &flow, + (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_DST_MASK + | OFPFW_TP_DST), OFPP_NORMAL); + setup_flow(in_band, IBR_FROM_CTL_OFP, &flow, + (OFPFW_DL_TYPE | OFPFW_NW_PROTO | OFPFW_NW_SRC_MASK + | OFPFW_TP_SRC), OFPP_NORMAL); + } else { + drop_flow(in_band, IBR_TO_CTL_ARP); + drop_flow(in_band, IBR_FROM_CTL_ARP); + drop_flow(in_band, IBR_TO_CTL_OFP); + drop_flow(in_band, IBR_FROM_CTL_OFP); + } +} + +void +in_band_wait(struct in_band *in_band) +{ + time_t now = time_now(); + time_t wakeup + = MIN(in_band->next_remote_refresh, in_band->next_local_refresh); + if (wakeup > now) { + poll_timer_wait((wakeup - now) * 1000); + } else { + poll_immediate_wake(); + } +} + +void +in_band_flushed(struct in_band *in_band) +{ + int i; + + for (i = 0; i < N_IB_RULES; i++) { + in_band->rules[i].installed = false; + } +} + +int +in_band_create(struct ofproto *ofproto, struct dpif *dpif, + struct switch_status *ss, struct rconn *controller, + struct in_band **in_bandp) +{ + struct in_band *in_band; + char local_name[IF_NAMESIZE]; + struct netdev *local_netdev; + int error; + + error = dpif_port_get_name(dpif, ODPP_LOCAL, + local_name, sizeof local_name); + if (error) { + VLOG_ERR("failed to initialize in-band control: cannot get name " + "of datapath local port (%s)", strerror(error)); + return error; + } + + error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &local_netdev); + if (error) { + VLOG_ERR("failed to initialize in-band control: cannot open " + "datapath local port %s (%s)", local_name, strerror(error)); + return error; + } + + in_band = xcalloc(1, sizeof *in_band); + in_band->ofproto = ofproto; + in_band->controller = controller; + in_band->ss_cat = switch_status_register(ss, "in-band", + in_band_status_cb, in_band); + in_band->local_netdev = local_netdev; + in_band->next_local_refresh = TIME_MIN; + in_band->remote_netdev = NULL; + in_band->next_remote_refresh = TIME_MIN; + + *in_bandp = in_band; + + return 0; +} + +void +in_band_destroy(struct in_band *in_band) +{ + if (in_band) { + switch_status_unregister(in_band->ss_cat); + netdev_close(in_band->local_netdev); + netdev_close(in_band->remote_netdev); + /* We don't own the rconn. */ + } +} + diff --cc ofproto/ofproto.c index 7650068e,00000000..37033660 mode 100644,000000..100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@@ -1,3406 -1,0 +1,3428 @@@ +/* + * Copyright (c) 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "ofproto.h" +#include +#include +#include +#include +#include +#include +#include "classifier.h" +#include "coverage.h" +#include "discovery.h" +#include "dpif.h" +#include "dynamic-string.h" +#include "executer.h" +#include "fail-open.h" +#include "in-band.h" +#include "mac-learning.h" +#include "netdev.h" +#include "netflow.h" +#include "odp-util.h" +#include "ofp-print.h" +#include "ofpbuf.h" +#include "openflow/nicira-ext.h" +#include "openflow/openflow.h" +#include "openflow/openflow-mgmt.h" +#include "openvswitch/datapath-protocol.h" +#include "packets.h" +#include "pinsched.h" +#include "pktbuf.h" +#include "poll-loop.h" +#include "port-array.h" +#include "rconn.h" +#include "shash.h" +#include "status.h" +#include "stp.h" +#include "svec.h" +#include "tag.h" +#include "timeval.h" +#include "unixctl.h" +#include "vconn.h" +#include "vconn-ssl.h" +#include "xtoxll.h" + +#define THIS_MODULE VLM_ofproto +#include "vlog.h" + +enum { + DP_GROUP_FLOOD = 0, + DP_GROUP_ALL = 1 +}; + +enum { + TABLEID_HASH = 0, + TABLEID_CLASSIFIER = 1 +}; + +struct ofport { + struct netdev *netdev; + struct ofp_phy_port opp; /* In host byte order. */ +}; + +static void ofport_free(struct ofport *); +static void hton_ofp_phy_port(struct ofp_phy_port *); + +static int xlate_actions(const union ofp_action *in, size_t n_in, + const flow_t *flow, struct ofproto *ofproto, + const struct ofpbuf *packet, + struct odp_actions *out, tag_type *tags, + bool *may_setup_flow); + +struct rule { + struct cls_rule cr; + + uint16_t idle_timeout; /* In seconds from time of last use. */ + uint16_t hard_timeout; /* In seconds from time of creation. */ + long long int used; /* Last-used time (0 if never used). */ + long long int created; /* Creation time. */ + uint64_t packet_count; /* Number of packets received. */ + uint64_t byte_count; /* Number of bytes received. */ + uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */ + uint8_t tcp_flags; /* Bitwise-OR of all TCP flags seen. */ + uint8_t ip_tos; /* Last-seen IP type-of-service. */ + tag_type tags; /* Tags (set only by hooks). */ + + /* If 'super' is non-NULL, this rule is a subrule, that is, it is an + * exact-match rule (having cr.wc.wildcards of 0) generated from the + * wildcard rule 'super'. In this case, 'list' is an element of the + * super-rule's list. + * + * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of + * a list of subrules. A super-rule with no wildcards (where + * cr.wc.wildcards is 0) will never have any subrules. */ + struct rule *super; + struct list list; + + /* OpenFlow actions. + * + * A subrule has no actions (it uses the super-rule's actions). */ + int n_actions; + union ofp_action *actions; + + /* Datapath actions. + * + * A super-rule with wildcard fields never has ODP actions (since the + * datapath only supports exact-match flows). */ + bool installed; /* Installed in datapath? */ + bool may_install; /* True ordinarily; false if actions must + * be reassessed for every packet. */ + int n_odp_actions; + union odp_action *odp_actions; +}; + +static inline bool +rule_is_hidden(const struct rule *rule) +{ + /* Subrules are merely an implementation detail, so hide them from the + * controller. */ + if (rule->super != NULL) { + return true; + } + + /* Rules with priority higher than UINT16_MAX are set up by ofproto itself + * (e.g. by in-band control) and are intentionally hidden from the + * controller. */ + if (rule->cr.priority > UINT16_MAX) { + return true; + } + + return false; +} + +static struct rule *rule_create(struct rule *super, const union ofp_action *, + size_t n_actions, uint16_t idle_timeout, + uint16_t hard_timeout); +static void rule_free(struct rule *); +static void rule_destroy(struct ofproto *, struct rule *); +static struct rule *rule_from_cls_rule(const struct cls_rule *); +static void rule_insert(struct ofproto *, struct rule *, + struct ofpbuf *packet, uint16_t in_port); +static void rule_remove(struct ofproto *, struct rule *); +static bool rule_make_actions(struct ofproto *, struct rule *, + const struct ofpbuf *packet); +static void rule_install(struct ofproto *, struct rule *, + struct rule *displaced_rule); +static void rule_uninstall(struct ofproto *, struct rule *); +static void rule_post_uninstall(struct ofproto *, struct rule *); + +struct ofconn { + struct list node; + struct rconn *rconn; + struct pktbuf *pktbuf; + bool send_flow_exp; + int miss_send_len; + + struct rconn_packet_counter *packet_in_counter; + + /* Number of OpenFlow messages queued as replies to OpenFlow requests, and + * the maximum number before we stop reading OpenFlow requests. */ +#define OFCONN_REPLY_MAX 100 + struct rconn_packet_counter *reply_counter; +}; + +static struct ofconn *ofconn_create(struct ofproto *, struct rconn *); +static void ofconn_destroy(struct ofconn *, struct ofproto *); +static void ofconn_run(struct ofconn *, struct ofproto *); +static void ofconn_wait(struct ofconn *); +static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, + struct rconn_packet_counter *counter); + +struct ofproto { + /* Settings. */ + uint64_t datapath_id; /* Datapath ID. */ + uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ + uint64_t mgmt_id; /* Management channel identifier. */ + char *manufacturer; /* Manufacturer. */ + char *hardware; /* Hardware. */ + char *software; /* Software version. */ + char *serial; /* Serial number. */ + + /* Datapath. */ + struct dpif *dpif; + struct netdev_monitor *netdev_monitor; + struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is + * OFP port nr. */ + struct shash port_by_name; + uint32_t max_ports; + + /* Configuration. */ + struct switch_status *switch_status; + struct status_category *ss_cat; + struct in_band *in_band; + struct discovery *discovery; + struct fail_open *fail_open; + struct pinsched *miss_sched, *action_sched; + struct executer *executer; + struct netflow *netflow; + + /* Flow table. */ + struct classifier cls; + bool need_revalidate; + long long int next_expiration; + struct tag_set revalidate_set; + + /* OpenFlow connections. */ + struct list all_conns; + struct ofconn *controller; + struct pvconn **listeners; + size_t n_listeners; + struct pvconn **snoops; + size_t n_snoops; + + /* Hooks for ovs-vswitchd. */ + const struct ofhooks *ofhooks; + void *aux; + + /* Used by default ofhooks. */ + struct mac_learning *ml; +}; + +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + +static const struct ofhooks default_ofhooks; + +static uint64_t pick_datapath_id(const struct ofproto *); +static uint64_t pick_fallback_dpid(void); +static void send_packet_in_miss(struct ofpbuf *, void *ofproto); +static void send_packet_in_action(struct ofpbuf *, void *ofproto); +static void update_used(struct ofproto *); +static void update_stats(struct rule *, const struct odp_flow_stats *); +static void expire_rule(struct cls_rule *, void *ofproto); +static bool revalidate_rule(struct ofproto *p, struct rule *rule); +static void revalidate_cb(struct cls_rule *rule_, void *p_); + +static void handle_odp_msg(struct ofproto *, struct ofpbuf *); + +static void handle_openflow(struct ofconn *, struct ofproto *, + struct ofpbuf *); + +static void refresh_port_group(struct ofproto *, unsigned int group); +static void update_port(struct ofproto *, const char *devname); +static int init_ports(struct ofproto *); +static void reinit_ports(struct ofproto *); + +int +ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, + struct ofproto **ofprotop) +{ + struct odp_stats stats; + struct ofproto *p; + struct dpif *dpif; + int error; + + *ofprotop = NULL; + + /* Connect to datapath and start listening for messages. */ + error = dpif_open(datapath, &dpif); + if (error) { + VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); + return error; + } + error = dpif_get_dp_stats(dpif, &stats); + if (error) { + VLOG_ERR("failed to obtain stats for datapath %s: %s", + datapath, strerror(error)); + dpif_close(dpif); + return error; + } + error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION); + if (error) { + VLOG_ERR("failed to listen on datapath %s: %s", + datapath, strerror(error)); + dpif_close(dpif); + return error; + } + dpif_flow_flush(dpif); + dpif_recv_purge(dpif); + + /* Initialize settings. */ + p = xcalloc(1, sizeof *p); + p->fallback_dpid = pick_fallback_dpid(); + p->datapath_id = p->fallback_dpid; + p->manufacturer = xstrdup("Nicira Networks, Inc."); + p->hardware = xstrdup("Reference Implementation"); + p->software = xstrdup(VERSION BUILDNR); + p->serial = xstrdup("None"); + + /* Initialize datapath. */ + p->dpif = dpif; + p->netdev_monitor = netdev_monitor_create(); + port_array_init(&p->ports); + shash_init(&p->port_by_name); + p->max_ports = stats.max_ports; + + /* Initialize submodules. */ + p->switch_status = switch_status_create(p); + p->in_band = NULL; + p->discovery = NULL; + p->fail_open = NULL; + p->miss_sched = p->action_sched = NULL; + p->executer = NULL; + p->netflow = NULL; + + /* Initialize flow table. */ + classifier_init(&p->cls); + p->need_revalidate = false; + p->next_expiration = time_msec() + 1000; + tag_set_init(&p->revalidate_set); + + /* Initialize OpenFlow connections. */ + list_init(&p->all_conns); + p->controller = ofconn_create(p, rconn_create(5, 8)); + p->controller->pktbuf = pktbuf_create(); + p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; + p->listeners = NULL; + p->n_listeners = 0; + p->snoops = NULL; + p->n_snoops = 0; + + /* Initialize hooks. */ + if (ofhooks) { + p->ofhooks = ofhooks; + p->aux = aux; + p->ml = NULL; + } else { + p->ofhooks = &default_ofhooks; + p->aux = p; + p->ml = mac_learning_create(); + } + + /* Register switch status category. */ + p->ss_cat = switch_status_register(p->switch_status, "remote", + rconn_status_cb, p->controller->rconn); + + /* Almost done... */ + error = init_ports(p); + if (error) { + ofproto_destroy(p); + return error; + } + + /* Pick final datapath ID. */ + p->datapath_id = pick_datapath_id(p); + VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id); + + *ofprotop = p; + return 0; +} + +void +ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) +{ + uint64_t old_dpid = p->datapath_id; + p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); + if (p->datapath_id != old_dpid) { + VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id); + rconn_reconnect(p->controller->rconn); + } +} + +void +ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id) +{ + p->mgmt_id = mgmt_id; +} + +void +ofproto_set_probe_interval(struct ofproto *p, int probe_interval) +{ + probe_interval = probe_interval ? MAX(probe_interval, 5) : 0; + rconn_set_probe_interval(p->controller->rconn, probe_interval); + if (p->fail_open) { + int trigger_duration = probe_interval ? probe_interval * 3 : 15; + fail_open_set_trigger_duration(p->fail_open, trigger_duration); + } +} + +void +ofproto_set_max_backoff(struct ofproto *p, int max_backoff) +{ + rconn_set_max_backoff(p->controller->rconn, max_backoff); +} + +void +ofproto_set_desc(struct ofproto *p, + const char *manufacturer, const char *hardware, + const char *software, const char *serial) +{ + if (manufacturer) { + free(p->manufacturer); + p->manufacturer = xstrdup(manufacturer); + } + if (hardware) { + free(p->hardware); + p->hardware = xstrdup(hardware); + } + if (software) { + free(p->software); + p->software = xstrdup(software); + } + if (serial) { + free(p->serial); + p->serial = xstrdup(serial); + } +} + +int +ofproto_set_in_band(struct ofproto *p, bool in_band) +{ + if (in_band != (p->in_band != NULL)) { + if (in_band) { + return in_band_create(p, p->dpif, p->switch_status, + p->controller->rconn, &p->in_band); + } else { + ofproto_set_discovery(p, false, NULL, true); + in_band_destroy(p->in_band); + p->in_band = NULL; + } + rconn_reconnect(p->controller->rconn); + } + return 0; +} + +int +ofproto_set_discovery(struct ofproto *p, bool discovery, + const char *re, bool update_resolv_conf) +{ + if (discovery != (p->discovery != NULL)) { + if (discovery) { + int error = ofproto_set_in_band(p, true); + if (error) { + return error; + } + error = discovery_create(re, update_resolv_conf, + p->dpif, p->switch_status, + &p->discovery); + if (error) { + return error; + } + } else { + discovery_destroy(p->discovery); + p->discovery = NULL; + } + rconn_disconnect(p->controller->rconn); + } else if (discovery) { + discovery_set_update_resolv_conf(p->discovery, update_resolv_conf); + return discovery_set_accept_controller_re(p->discovery, re); + } + return 0; +} + +int +ofproto_set_controller(struct ofproto *ofproto, const char *controller) +{ + if (ofproto->discovery) { + return EINVAL; + } else if (controller) { + if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) { + return rconn_connect(ofproto->controller->rconn, controller); + } else { + return 0; + } + } else { + rconn_disconnect(ofproto->controller->rconn); + return 0; + } +} + +static int +set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, + const struct svec *svec) +{ + struct pvconn **pvconns = *pvconnsp; + size_t n_pvconns = *n_pvconnsp; + int retval = 0; + size_t i; + + for (i = 0; i < n_pvconns; i++) { + pvconn_close(pvconns[i]); + } + free(pvconns); + + pvconns = xmalloc(svec->n * sizeof *pvconns); + n_pvconns = 0; + for (i = 0; i < svec->n; i++) { + const char *name = svec->names[i]; + struct pvconn *pvconn; + int error; + + error = pvconn_open(name, &pvconn); + if (!error) { + pvconns[n_pvconns++] = pvconn; + } else { + VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); + if (!retval) { + retval = error; + } + } + } + + *pvconnsp = pvconns; + *n_pvconnsp = n_pvconns; + + return retval; +} + +int +ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners) +{ + return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners); +} + +int +ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) +{ + return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); +} + +int +ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors, + uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface) +{ + if (collectors && collectors->n) { + if (!ofproto->netflow) { + ofproto->netflow = netflow_create(); + } + netflow_set_engine(ofproto->netflow, engine_type, engine_id, + add_id_to_iface); + return netflow_set_collectors(ofproto->netflow, collectors); + } else { + netflow_destroy(ofproto->netflow); + ofproto->netflow = NULL; + return 0; + } +} + +void +ofproto_set_failure(struct ofproto *ofproto, bool fail_open) +{ + if (fail_open) { + struct rconn *rconn = ofproto->controller->rconn; + int trigger_duration = rconn_get_probe_interval(rconn) * 3; + if (!ofproto->fail_open) { + ofproto->fail_open = fail_open_create(ofproto, trigger_duration, + ofproto->switch_status, + rconn); + } else { + fail_open_set_trigger_duration(ofproto->fail_open, + trigger_duration); + } + } else { + fail_open_destroy(ofproto->fail_open); + ofproto->fail_open = NULL; + } +} + +void +ofproto_set_rate_limit(struct ofproto *ofproto, + int rate_limit, int burst_limit) +{ + if (rate_limit > 0) { + if (!ofproto->miss_sched) { + ofproto->miss_sched = pinsched_create(rate_limit, burst_limit, + ofproto->switch_status); + ofproto->action_sched = pinsched_create(rate_limit, burst_limit, + NULL); + } else { + pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit); + pinsched_set_limits(ofproto->action_sched, + rate_limit, burst_limit); + } + } else { + pinsched_destroy(ofproto->miss_sched); + ofproto->miss_sched = NULL; + pinsched_destroy(ofproto->action_sched); + ofproto->action_sched = NULL; + } +} + +int +ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp) +{ + /* XXX */ + if (enable_stp) { + VLOG_WARN("STP is not yet implemented"); + return EINVAL; + } else { + return 0; + } +} + +int +ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl, + const char *command_dir) +{ + if (command_acl) { + if (!ofproto->executer) { + return executer_create(command_acl, command_dir, + &ofproto->executer); + } else { + executer_set_acl(ofproto->executer, command_acl, command_dir); + } + } else { + executer_destroy(ofproto->executer); + ofproto->executer = NULL; + } + return 0; +} + +uint64_t +ofproto_get_datapath_id(const struct ofproto *ofproto) +{ + return ofproto->datapath_id; +} + +uint64_t +ofproto_get_mgmt_id(const struct ofproto *ofproto) +{ + return ofproto->mgmt_id; +} + +int +ofproto_get_probe_interval(const struct ofproto *ofproto) +{ + return rconn_get_probe_interval(ofproto->controller->rconn); +} + +int +ofproto_get_max_backoff(const struct ofproto *ofproto) +{ + return rconn_get_max_backoff(ofproto->controller->rconn); +} + +bool +ofproto_get_in_band(const struct ofproto *ofproto) +{ + return ofproto->in_band != NULL; +} + +bool +ofproto_get_discovery(const struct ofproto *ofproto) +{ + return ofproto->discovery != NULL; +} + +const char * +ofproto_get_controller(const struct ofproto *ofproto) +{ + return rconn_get_name(ofproto->controller->rconn); +} + +void +ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners) +{ + size_t i; + + for (i = 0; i < ofproto->n_listeners; i++) { + svec_add(listeners, pvconn_get_name(ofproto->listeners[i])); + } +} + +void +ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) +{ + size_t i; + + for (i = 0; i < ofproto->n_snoops; i++) { + svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); + } +} + +void +ofproto_destroy(struct ofproto *p) +{ + struct ofconn *ofconn, *next_ofconn; + struct ofport *ofport; + unsigned int port_no; + size_t i; + + if (!p) { + return; + } + + ofproto_flush_flows(p); + classifier_destroy(&p->cls); + + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, + &p->all_conns) { + ofconn_destroy(ofconn, p); + } + + dpif_close(p->dpif); + netdev_monitor_destroy(p->netdev_monitor); + PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { + ofport_free(ofport); + } + shash_destroy(&p->port_by_name); + + switch_status_destroy(p->switch_status); + in_band_destroy(p->in_band); + discovery_destroy(p->discovery); + fail_open_destroy(p->fail_open); + pinsched_destroy(p->miss_sched); + pinsched_destroy(p->action_sched); + executer_destroy(p->executer); + netflow_destroy(p->netflow); + + switch_status_unregister(p->ss_cat); + + for (i = 0; i < p->n_listeners; i++) { + pvconn_close(p->listeners[i]); + } + free(p->listeners); + + for (i = 0; i < p->n_snoops; i++) { + pvconn_close(p->snoops[i]); + } + free(p->snoops); + + mac_learning_destroy(p->ml); + + free(p); +} + +int +ofproto_run(struct ofproto *p) +{ + int error = ofproto_run1(p); + if (!error) { + error = ofproto_run2(p, false); + } + return error; +} + +static void +process_port_change(struct ofproto *ofproto, int error, char *devname) +{ + if (error == ENOBUFS) { + reinit_ports(ofproto); + } else if (!error) { + update_port(ofproto, devname); + free(devname); + } +} + +int +ofproto_run1(struct ofproto *p) +{ + struct ofconn *ofconn, *next_ofconn; + char *devname; + int error; + int i; + + for (i = 0; i < 50; i++) { + struct ofpbuf *buf; + int error; + + error = dpif_recv(p->dpif, &buf); + if (error) { + if (error == ENODEV) { + /* Someone destroyed the datapath behind our back. The caller + * better destroy us and give up, because we're just going to + * spin from here on out. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally", + dpif_name(p->dpif)); + return ENODEV; + } + break; + } + + handle_odp_msg(p, buf); + } + + while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { + process_port_change(p, error, devname); + } + while ((error = netdev_monitor_poll(p->netdev_monitor, + &devname)) != EAGAIN) { + process_port_change(p, error, devname); + } + + if (p->in_band) { + in_band_run(p->in_band); + } + if (p->discovery) { + char *controller_name; + if (rconn_is_connectivity_questionable(p->controller->rconn)) { + discovery_question_connectivity(p->discovery); + } + if (discovery_run(p->discovery, &controller_name)) { + if (controller_name) { + rconn_connect(p->controller->rconn, controller_name); + } else { + rconn_disconnect(p->controller->rconn); + } + } + } - if (p->fail_open) { - fail_open_run(p->fail_open); - } + pinsched_run(p->miss_sched, send_packet_in_miss, p); + pinsched_run(p->action_sched, send_packet_in_action, p); + if (p->executer) { + executer_run(p->executer); + } + + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, + &p->all_conns) { + ofconn_run(ofconn, p); + } + ++ /* Fail-open maintenance. Do this after processing the ofconns since ++ * fail-open checks the status of the controller rconn. */ ++ if (p->fail_open) { ++ fail_open_run(p->fail_open); ++ } ++ + for (i = 0; i < p->n_listeners; i++) { + struct vconn *vconn; + int retval; + + retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn); + if (!retval) { + ofconn_create(p, rconn_new_from_vconn("passive", vconn)); + } else if (retval != EAGAIN) { + VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); + } + } + + for (i = 0; i < p->n_snoops; i++) { + struct vconn *vconn; + int retval; + + retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); + if (!retval) { + rconn_add_monitor(p->controller->rconn, vconn); + } else if (retval != EAGAIN) { + VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); + } + } + + if (time_msec() >= p->next_expiration) { + COVERAGE_INC(ofproto_expiration); + p->next_expiration = time_msec() + 1000; + update_used(p); + + classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); + + /* Let the hook know that we're at a stable point: all outstanding data + * in existing flows has been accounted to the account_cb. Thus, the + * hook can now reasonably do operations that depend on having accurate + * flow volume accounting (currently, that's just bond rebalancing). */ + if (p->ofhooks->account_checkpoint_cb) { + p->ofhooks->account_checkpoint_cb(p->aux); + } + } + + if (p->netflow) { + netflow_run(p->netflow); + } + + return 0; +} + +struct revalidate_cbdata { + struct ofproto *ofproto; + bool revalidate_all; /* Revalidate all exact-match rules? */ + bool revalidate_subrules; /* Revalidate all exact-match subrules? */ + struct tag_set revalidate_set; /* Set of tags to revalidate. */ +}; + +int +ofproto_run2(struct ofproto *p, bool revalidate_all) +{ + if (p->need_revalidate || revalidate_all + || !tag_set_is_empty(&p->revalidate_set)) { + struct revalidate_cbdata cbdata; + cbdata.ofproto = p; + cbdata.revalidate_all = revalidate_all; + cbdata.revalidate_subrules = p->need_revalidate; + cbdata.revalidate_set = p->revalidate_set; + tag_set_init(&p->revalidate_set); + COVERAGE_INC(ofproto_revalidate); + classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); + p->need_revalidate = false; + } + + return 0; +} + +void +ofproto_wait(struct ofproto *p) +{ + struct ofconn *ofconn; + size_t i; + + dpif_recv_wait(p->dpif); + dpif_port_poll_wait(p->dpif); + netdev_monitor_poll_wait(p->netdev_monitor); + LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + ofconn_wait(ofconn); + } + if (p->in_band) { + in_band_wait(p->in_band); + } + if (p->discovery) { + discovery_wait(p->discovery); + } + if (p->fail_open) { + fail_open_wait(p->fail_open); + } + pinsched_wait(p->miss_sched); + pinsched_wait(p->action_sched); + if (p->executer) { + executer_wait(p->executer); + } + if (!tag_set_is_empty(&p->revalidate_set)) { + poll_immediate_wake(); + } + if (p->need_revalidate) { + /* Shouldn't happen, but if it does just go around again. */ + VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); + poll_immediate_wake(); + } else if (p->next_expiration != LLONG_MAX) { + poll_timer_wait(p->next_expiration - time_msec()); + } + for (i = 0; i < p->n_listeners; i++) { + pvconn_wait(p->listeners[i]); + } + for (i = 0; i < p->n_snoops; i++) { + pvconn_wait(p->snoops[i]); + } +} + +void +ofproto_revalidate(struct ofproto *ofproto, tag_type tag) +{ + tag_set_add(&ofproto->revalidate_set, tag); +} + +struct tag_set * +ofproto_get_revalidate_set(struct ofproto *ofproto) +{ + return &ofproto->revalidate_set; +} + +bool +ofproto_is_alive(const struct ofproto *p) +{ + return p->discovery || rconn_is_alive(p->controller->rconn); +} + +int +ofproto_send_packet(struct ofproto *p, const flow_t *flow, + const union ofp_action *actions, size_t n_actions, + const struct ofpbuf *packet) +{ + struct odp_actions odp_actions; + int error; + + error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, + NULL, NULL); + if (error) { + return error; + } + + /* XXX Should we translate the dpif_execute() errno value into an OpenFlow + * error code? */ + dpif_execute(p->dpif, flow->in_port, odp_actions.actions, + odp_actions.n_actions, packet); + return 0; +} + +void +ofproto_add_flow(struct ofproto *p, + const flow_t *flow, uint32_t wildcards, unsigned int priority, + const union ofp_action *actions, size_t n_actions, + int idle_timeout) +{ + struct rule *rule; + rule = rule_create(NULL, actions, n_actions, + idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0); + cls_rule_from_flow(&rule->cr, flow, wildcards, priority); + rule_insert(p, rule, NULL, 0); +} + +void +ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, + uint32_t wildcards, unsigned int priority) +{ + struct rule *rule; + + rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, + flow, wildcards, + priority)); + if (rule) { + rule_remove(ofproto, rule); + } +} + +static void +destroy_rule(struct cls_rule *rule_, void *ofproto_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct ofproto *ofproto = ofproto_; + + /* Mark the flow as not installed, even though it might really be + * installed, so that rule_remove() doesn't bother trying to uninstall it. + * There is no point in uninstalling it individually since we are about to + * blow away all the flows with dpif_flow_flush(). */ + rule->installed = false; + + rule_remove(ofproto, rule); +} + +void +ofproto_flush_flows(struct ofproto *ofproto) +{ + COVERAGE_INC(ofproto_flush); + classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); + dpif_flow_flush(ofproto->dpif); + if (ofproto->in_band) { + in_band_flushed(ofproto->in_band); + } + if (ofproto->fail_open) { + fail_open_flushed(ofproto->fail_open); + } +} + +static void +reinit_ports(struct ofproto *p) +{ + struct svec devnames; + struct ofport *ofport; + unsigned int port_no; + struct odp_port *odp_ports; + size_t n_odp_ports; + size_t i; + + svec_init(&devnames); + PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { + svec_add (&devnames, (char *) ofport->opp.name); + } + dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); + for (i = 0; i < n_odp_ports; i++) { + svec_add (&devnames, odp_ports[i].devname); + } + free(odp_ports); + + svec_sort_unique(&devnames); + for (i = 0; i < devnames.n; i++) { + update_port(p, devnames.names[i]); + } + svec_destroy(&devnames); +} + +static void +refresh_port_group(struct ofproto *p, unsigned int group) +{ + uint16_t *ports; + size_t n_ports; + struct ofport *port; + unsigned int port_no; + + assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); + + ports = xmalloc(port_array_count(&p->ports) * sizeof *ports); + n_ports = 0; + PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { + ports[n_ports++] = port_no; + } + } + dpif_port_group_set(p->dpif, group, ports, n_ports); + free(ports); +} + +static void +refresh_port_groups(struct ofproto *p) +{ + refresh_port_group(p, DP_GROUP_FLOOD); + refresh_port_group(p, DP_GROUP_ALL); +} + +static struct ofport * +make_ofport(const struct odp_port *odp_port) +{ + enum netdev_flags flags; + struct ofport *ofport; + struct netdev *netdev; + bool carrier; + int error; + + error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev); + if (error) { + VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " + "cannot be opened (%s)", + odp_port->devname, odp_port->port, + odp_port->devname, strerror(error)); + return NULL; + } + + ofport = xmalloc(sizeof *ofport); + ofport->netdev = netdev; + ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); + netdev_get_etheraddr(netdev, ofport->opp.hw_addr); + memcpy(ofport->opp.name, odp_port->devname, + MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); + ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; + + netdev_get_flags(netdev, &flags); + ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; + + netdev_get_carrier(netdev, &carrier); + ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; + + netdev_get_features(netdev, + &ofport->opp.curr, &ofport->opp.advertised, + &ofport->opp.supported, &ofport->opp.peer); + return ofport; +} + +static bool +ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) +{ + if (port_array_get(&p->ports, odp_port->port)) { + VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", + odp_port->port); + return true; + } else if (shash_find(&p->port_by_name, odp_port->devname)) { + VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", + odp_port->devname); + return true; + } else { + return false; + } +} + +static int +ofport_equal(const struct ofport *a_, const struct ofport *b_) +{ + const struct ofp_phy_port *a = &a_->opp; + const struct ofp_phy_port *b = &b_->opp; + + BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ + return (a->port_no == b->port_no + && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) + && !strcmp((char *) a->name, (char *) b->name) + && a->state == b->state + && a->config == b->config + && a->curr == b->curr + && a->advertised == b->advertised + && a->supported == b->supported + && a->peer == b->peer); +} + +static void +send_port_status(struct ofproto *p, const struct ofport *ofport, + uint8_t reason) +{ + /* XXX Should limit the number of queued port status change messages. */ + struct ofconn *ofconn; + LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + struct ofp_port_status *ops; + struct ofpbuf *b; + + ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); + ops->reason = reason; + ops->desc = ofport->opp; + hton_ofp_phy_port(&ops->desc); + queue_tx(b, ofconn, NULL); + } + if (p->ofhooks->port_changed_cb) { + p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); + } +} + +static void +ofport_install(struct ofproto *p, struct ofport *ofport) +{ + netdev_monitor_add(p->netdev_monitor, ofport->netdev); + port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), + ofport); + shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport); +} + +static void +ofport_remove(struct ofproto *p, struct ofport *ofport) +{ + netdev_monitor_remove(p->netdev_monitor, ofport->netdev); + port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL); + shash_delete(&p->port_by_name, + shash_find(&p->port_by_name, (char *) ofport->opp.name)); +} + +static void +ofport_free(struct ofport *ofport) +{ + if (ofport) { + netdev_close(ofport->netdev); + free(ofport); + } +} + +static void +update_port(struct ofproto *p, const char *devname) +{ + struct odp_port odp_port; + struct ofport *old_ofport; + struct ofport *new_ofport; + int error; + + COVERAGE_INC(ofproto_update_port); + + /* Query the datapath for port information. */ + error = dpif_port_query_by_name(p->dpif, devname, &odp_port); + + /* Find the old ofport. */ + old_ofport = shash_find_data(&p->port_by_name, devname); + if (!error) { + if (!old_ofport) { + /* There's no port named 'devname' but there might be a port with + * the same port number. This could happen if a port is deleted + * and then a new one added in its place very quickly, or if a port + * is renamed. In the former case we want to send an OFPPR_DELETE + * and an OFPPR_ADD, and in the latter case we want to send a + * single OFPPR_MODIFY. We can distinguish the cases by comparing + * the old port's ifindex against the new port, or perhaps less + * reliably but more portably by comparing the old port's MAC + * against the new port's MAC. However, this code isn't that smart + * and always sends an OFPPR_MODIFY (XXX). */ + old_ofport = port_array_get(&p->ports, odp_port.port); + } + } else if (error != ENOENT && error != ENODEV) { + VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " + "%s", strerror(error)); + return; + } + + /* Create a new ofport. */ + new_ofport = !error ? make_ofport(&odp_port) : NULL; + + /* Eliminate a few pathological cases. */ + if (!old_ofport && !new_ofport) { + return; + } else if (old_ofport && new_ofport) { + /* Most of the 'config' bits are OpenFlow soft state, but + * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow + * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and + * leaves the other bits 0.) */ + new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; + + if (ofport_equal(old_ofport, new_ofport)) { + /* False alarm--no change. */ + ofport_free(new_ofport); + return; + } + } + + /* Now deal with the normal cases. */ + if (old_ofport) { + ofport_remove(p, old_ofport); + } + if (new_ofport) { + ofport_install(p, new_ofport); + } + send_port_status(p, new_ofport ? new_ofport : old_ofport, + (!old_ofport ? OFPPR_ADD + : !new_ofport ? OFPPR_DELETE + : OFPPR_MODIFY)); + ofport_free(old_ofport); + + /* Update port groups. */ + refresh_port_groups(p); +} + +static int +init_ports(struct ofproto *p) +{ + struct odp_port *ports; + size_t n_ports; + size_t i; + int error; + + error = dpif_port_list(p->dpif, &ports, &n_ports); + if (error) { + return error; + } + + for (i = 0; i < n_ports; i++) { + const struct odp_port *odp_port = &ports[i]; + if (!ofport_conflicts(p, odp_port)) { + struct ofport *ofport = make_ofport(odp_port); + if (ofport) { + ofport_install(p, ofport); + } + } + } + free(ports); + refresh_port_groups(p); + return 0; +} + +static struct ofconn * +ofconn_create(struct ofproto *p, struct rconn *rconn) +{ + struct ofconn *ofconn = xmalloc(sizeof *ofconn); + list_push_back(&p->all_conns, &ofconn->node); + ofconn->rconn = rconn; + ofconn->pktbuf = NULL; + ofconn->send_flow_exp = false; + ofconn->miss_send_len = 0; + ofconn->packet_in_counter = rconn_packet_counter_create (); + ofconn->reply_counter = rconn_packet_counter_create (); + return ofconn; +} + +static void +ofconn_destroy(struct ofconn *ofconn, struct ofproto *p) +{ + if (p->executer) { + executer_rconn_closing(p->executer, ofconn->rconn); + } + + list_remove(&ofconn->node); + rconn_destroy(ofconn->rconn); + rconn_packet_counter_destroy(ofconn->packet_in_counter); + rconn_packet_counter_destroy(ofconn->reply_counter); + pktbuf_destroy(ofconn->pktbuf); + free(ofconn); +} + +static void +ofconn_run(struct ofconn *ofconn, struct ofproto *p) +{ + int iteration; + + rconn_run(ofconn->rconn); + + if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { + /* Limit the number of iterations to prevent other tasks from + * starving. */ + for (iteration = 0; iteration < 50; iteration++) { + struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); + if (!of_msg) { + break; + } ++ if (p->fail_open) { ++ fail_open_maybe_recover(p->fail_open); ++ } + handle_openflow(ofconn, p, of_msg); + ofpbuf_delete(of_msg); + } + } + + if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) { + ofconn_destroy(ofconn, p); + } +} + +static void +ofconn_wait(struct ofconn *ofconn) +{ + rconn_run_wait(ofconn->rconn); + if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { + rconn_recv_wait(ofconn->rconn); + } else { + COVERAGE_INC(ofproto_ofconn_stuck); + } +} + +/* Caller is responsible for initializing the 'cr' member of the returned + * rule. */ +static struct rule * +rule_create(struct rule *super, + const union ofp_action *actions, size_t n_actions, + uint16_t idle_timeout, uint16_t hard_timeout) +{ + struct rule *rule = xcalloc(1, sizeof *rule); + rule->idle_timeout = idle_timeout; + rule->hard_timeout = hard_timeout; + rule->used = rule->created = time_msec(); + rule->super = super; + if (super) { + list_push_back(&super->list, &rule->list); + } else { + list_init(&rule->list); + } + rule->n_actions = n_actions; + rule->actions = xmemdup(actions, n_actions * sizeof *actions); + return rule; +} + +static struct rule * +rule_from_cls_rule(const struct cls_rule *cls_rule) +{ + return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; +} + +static void +rule_free(struct rule *rule) +{ + free(rule->actions); + free(rule->odp_actions); + free(rule); +} + +/* Destroys 'rule'. If 'rule' is a subrule, also removes it from its + * super-rule's list of subrules. If 'rule' is a super-rule, also iterates + * through all of its subrules and revalidates them, destroying any that no + * longer has a super-rule (which is probably all of them). + * + * Before calling this function, the caller must make have removed 'rule' from + * the classifier. If 'rule' is an exact-match rule, the caller is also + * responsible for ensuring that it has been uninstalled from the datapath. */ +static void +rule_destroy(struct ofproto *ofproto, struct rule *rule) +{ + if (!rule->super) { + struct rule *subrule, *next; + LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { + revalidate_rule(ofproto, subrule); + } + } else { + list_remove(&rule->list); + } + rule_free(rule); +} + +static bool +rule_has_out_port(const struct rule *rule, uint16_t out_port) +{ + const union ofp_action *oa; + struct actions_iterator i; + + if (out_port == htons(OFPP_NONE)) { + return true; + } + for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; + oa = actions_next(&i)) { + if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) { + return true; + } + } + return false; +} + +/* Executes the actions indicated by 'rule' on 'packet', which is in flow + * 'flow' and is considered to have arrived on ODP port 'in_port'. + * + * The flow that 'packet' actually contains does not need to actually match + * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, + * the packet and byte counters for 'rule' will be credited for the packet sent + * out whether or not the packet actually matches 'rule'. + * + * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, + * the caller must already have accurately composed ODP actions for it given + * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if + * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this + * function will compose a set of ODP actions based on 'rule''s OpenFlow + * actions and apply them to 'packet'. */ +static void +rule_execute(struct ofproto *ofproto, struct rule *rule, + struct ofpbuf *packet, const flow_t *flow) +{ + const union odp_action *actions; + size_t n_actions; + struct odp_actions a; + + /* Grab or compose the ODP actions. + * + * The special case for an exact-match 'rule' where 'flow' is not the + * rule's flow is important to avoid, e.g., sending a packet out its input + * port simply because the ODP actions were composed for the wrong + * scenario. */ + if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { + struct rule *super = rule->super ? rule->super : rule; + if (xlate_actions(super->actions, super->n_actions, flow, ofproto, + packet, &a, NULL, 0)) { + return; + } + actions = a.actions; + n_actions = a.n_actions; + } else { + actions = rule->odp_actions; + n_actions = rule->n_odp_actions; + } + + /* Execute the ODP actions. */ + if (!dpif_execute(ofproto->dpif, flow->in_port, + actions, n_actions, packet)) { + struct odp_flow_stats stats; + flow_extract_stats(flow, packet, &stats); + update_stats(rule, &stats); + rule->used = time_msec(); + } +} + +static void +rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, + uint16_t in_port) +{ + struct rule *displaced_rule; + + /* Insert the rule in the classifier. */ + displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); + if (!rule->cr.wc.wildcards) { + rule_make_actions(p, rule, packet); + } + + /* Send the packet and credit it to the rule. */ + if (packet) { + flow_t flow; + flow_extract(packet, in_port, &flow); + rule_execute(p, rule, packet, &flow); + } + + /* Install the rule in the datapath only after sending the packet, to + * avoid packet reordering. */ + if (rule->cr.wc.wildcards) { + COVERAGE_INC(ofproto_add_wc_flow); + p->need_revalidate = true; + } else { + rule_install(p, rule, displaced_rule); + } + + /* Free the rule that was displaced, if any. */ + if (displaced_rule) { + rule_destroy(p, displaced_rule); + } +} + +static struct rule * +rule_create_subrule(struct ofproto *ofproto, struct rule *rule, + const flow_t *flow) +{ + struct rule *subrule = rule_create(rule, NULL, 0, + rule->idle_timeout, rule->hard_timeout); + COVERAGE_INC(ofproto_subrule_create); + cls_rule_from_flow(&subrule->cr, flow, 0, + (rule->cr.priority <= UINT16_MAX ? UINT16_MAX + : rule->cr.priority)); + classifier_insert_exact(&ofproto->cls, &subrule->cr); + + return subrule; +} + +static void +rule_remove(struct ofproto *ofproto, struct rule *rule) +{ + if (rule->cr.wc.wildcards) { + COVERAGE_INC(ofproto_del_wc_flow); + ofproto->need_revalidate = true; + } else { + rule_uninstall(ofproto, rule); + } + classifier_remove(&ofproto->cls, &rule->cr); + rule_destroy(ofproto, rule); +} + +/* Returns true if the actions changed, false otherwise. */ +static bool +rule_make_actions(struct ofproto *p, struct rule *rule, + const struct ofpbuf *packet) +{ + const struct rule *super; + struct odp_actions a; + size_t actions_len; + + assert(!rule->cr.wc.wildcards); + + super = rule->super ? rule->super : rule; + rule->tags = 0; + xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, + packet, &a, &rule->tags, &rule->may_install); + + actions_len = a.n_actions * sizeof *a.actions; + if (rule->n_odp_actions != a.n_actions + || memcmp(rule->odp_actions, a.actions, actions_len)) { + COVERAGE_INC(ofproto_odp_unchanged); + free(rule->odp_actions); + rule->n_odp_actions = a.n_actions; + rule->odp_actions = xmemdup(a.actions, actions_len); + return true; + } else { + return false; + } +} + +static int +do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, + struct odp_flow_put *put) +{ + memset(&put->flow.stats, 0, sizeof put->flow.stats); + put->flow.key = rule->cr.flow; + put->flow.actions = rule->odp_actions; + put->flow.n_actions = rule->n_odp_actions; + put->flags = flags; + return dpif_flow_put(ofproto->dpif, put); +} + +static void +rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) +{ + assert(!rule->cr.wc.wildcards); + + if (rule->may_install) { + struct odp_flow_put put; + if (!do_put_flow(p, rule, + ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, + &put)) { + rule->installed = true; + if (displaced_rule) { + update_stats(rule, &put.flow.stats); + rule_post_uninstall(p, displaced_rule); + } + } + } else if (displaced_rule) { + rule_uninstall(p, displaced_rule); + } +} + +static void +rule_reinstall(struct ofproto *ofproto, struct rule *rule) +{ + if (rule->installed) { + struct odp_flow_put put; + COVERAGE_INC(ofproto_dp_missed); + do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); + } else { + rule_install(ofproto, rule, NULL); + } +} + +static void +rule_update_actions(struct ofproto *ofproto, struct rule *rule) +{ + bool actions_changed = rule_make_actions(ofproto, rule, NULL); + if (rule->may_install) { + if (rule->installed) { + if (actions_changed) { + /* XXX should really do rule_post_uninstall() for the *old* set + * of actions, and distinguish the old stats from the new. */ + struct odp_flow_put put; + do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); + } + } else { + rule_install(ofproto, rule, NULL); + } + } else { + rule_uninstall(ofproto, rule); + } +} + +static void +rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) +{ + uint64_t total_bytes = rule->byte_count + extra_bytes; + + if (ofproto->ofhooks->account_flow_cb + && total_bytes > rule->accounted_bytes) + { + ofproto->ofhooks->account_flow_cb( + &rule->cr.flow, rule->odp_actions, rule->n_odp_actions, + total_bytes - rule->accounted_bytes, ofproto->aux); + rule->accounted_bytes = total_bytes; + } +} + +static void +rule_uninstall(struct ofproto *p, struct rule *rule) +{ + assert(!rule->cr.wc.wildcards); + if (rule->installed) { + struct odp_flow odp_flow; + + odp_flow.key = rule->cr.flow; + odp_flow.actions = NULL; + odp_flow.n_actions = 0; + if (!dpif_flow_del(p->dpif, &odp_flow)) { + update_stats(rule, &odp_flow.stats); + } + rule->installed = false; + + rule_post_uninstall(p, rule); + } +} + +static void +rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) +{ + struct rule *super = rule->super; + + rule_account(ofproto, rule, 0); + if (ofproto->netflow && rule->byte_count) { + struct ofexpired expired; + expired.flow = rule->cr.flow; + expired.packet_count = rule->packet_count; + expired.byte_count = rule->byte_count; + expired.used = rule->used; + expired.created = rule->created; + expired.tcp_flags = rule->tcp_flags; + expired.ip_tos = rule->ip_tos; + netflow_expire(ofproto->netflow, &expired); + } + if (super) { + super->packet_count += rule->packet_count; + super->byte_count += rule->byte_count; + super->tcp_flags |= rule->tcp_flags; + if (rule->packet_count) { + super->ip_tos = rule->ip_tos; + } + } + + /* Reset counters to prevent double counting if the rule ever gets + * reinstalled. */ + rule->packet_count = 0; + rule->byte_count = 0; + rule->accounted_bytes = 0; + rule->tcp_flags = 0; + rule->ip_tos = 0; +} + +static void +queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, + struct rconn_packet_counter *counter) +{ + update_openflow_length(msg); + if (rconn_send(ofconn->rconn, msg, counter)) { + ofpbuf_delete(msg); + } +} + +static void +send_error(const struct ofconn *ofconn, const struct ofp_header *oh, + int error, const void *data, size_t len) +{ + struct ofpbuf *buf; + struct ofp_error_msg *oem; + + if (!(error >> 16)) { + VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", + error); + return; + } + + COVERAGE_INC(ofproto_error); + oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, + oh ? oh->xid : 0, &buf); + oem->type = htons((unsigned int) error >> 16); + oem->code = htons(error & 0xffff); + memcpy(oem->data, data, len); + queue_tx(buf, ofconn, ofconn->reply_counter); +} + +static void +send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, + int error) +{ + size_t oh_length = ntohs(oh->length); + send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); +} + +static void +hton_ofp_phy_port(struct ofp_phy_port *opp) +{ + opp->port_no = htons(opp->port_no); + opp->config = htonl(opp->config); + opp->state = htonl(opp->state); + opp->curr = htonl(opp->curr); + opp->advertised = htonl(opp->advertised); + opp->supported = htonl(opp->supported); + opp->peer = htonl(opp->peer); +} + +static int +handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) +{ + struct ofp_header *rq = oh; + queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); + return 0; +} + +static int +handle_features_request(struct ofproto *p, struct ofconn *ofconn, + struct ofp_header *oh) +{ + struct ofp_switch_features *osf; + struct ofpbuf *buf; + unsigned int port_no; + struct ofport *port; + + osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); + osf->datapath_id = htonll(p->datapath_id); + osf->n_buffers = htonl(pktbuf_capacity()); + osf->n_tables = 2; + osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | + OFPC_PORT_STATS | OFPC_MULTI_PHY_TX); + osf->actions = htonl((1u << OFPAT_OUTPUT) | + (1u << OFPAT_SET_VLAN_VID) | + (1u << OFPAT_SET_VLAN_PCP) | + (1u << OFPAT_STRIP_VLAN) | + (1u << OFPAT_SET_DL_SRC) | + (1u << OFPAT_SET_DL_DST) | + (1u << OFPAT_SET_NW_SRC) | + (1u << OFPAT_SET_NW_DST) | + (1u << OFPAT_SET_TP_SRC) | + (1u << OFPAT_SET_TP_DST)); + + PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); + } + + queue_tx(buf, ofconn, ofconn->reply_counter); + return 0; +} + +static int +handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, + struct ofp_header *oh) +{ + struct ofpbuf *buf; + struct ofp_switch_config *osc; + uint16_t flags; + bool drop_frags; + + /* Figure out flags. */ + dpif_get_drop_frags(p->dpif, &drop_frags); + flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; + if (ofconn->send_flow_exp) { + flags |= OFPC_SEND_FLOW_EXP; + } + + /* Send reply. */ + osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); + osc->flags = htons(flags); + osc->miss_send_len = htons(ofconn->miss_send_len); + queue_tx(buf, ofconn, ofconn->reply_counter); + + return 0; +} + +static int +handle_set_config(struct ofproto *p, struct ofconn *ofconn, + struct ofp_switch_config *osc) +{ + uint16_t flags; + int error; + + error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); + if (error) { + return error; + } + flags = ntohs(osc->flags); + + ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0; + + if (ofconn == p->controller) { + switch (flags & OFPC_FRAG_MASK) { + case OFPC_FRAG_NORMAL: + dpif_set_drop_frags(p->dpif, false); + break; + case OFPC_FRAG_DROP: + dpif_set_drop_frags(p->dpif, true); + break; + default: + VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", + osc->flags); + break; + } + } + + if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) { + if (ntohs(osc->miss_send_len) != 0) { + ofconn->pktbuf = pktbuf_create(); + } else { + pktbuf_destroy(ofconn->pktbuf); + } + } + + ofconn->miss_send_len = ntohs(osc->miss_send_len); + + return 0; +} + +static void +add_output_group_action(struct odp_actions *actions, uint16_t group) +{ + odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; +} + +static void +add_controller_action(struct odp_actions *actions, + const struct ofp_action_output *oao) +{ + union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); + a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX; +} + +struct action_xlate_ctx { + /* Input. */ + const flow_t *flow; /* Flow to which these actions correspond. */ + int recurse; /* Recursion level, via xlate_table_action. */ + struct ofproto *ofproto; + const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a + * null pointer if we are revalidating + * without a packet to refer to. */ + + /* Output. */ + struct odp_actions *out; /* Datapath actions. */ + tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ + bool may_setup_flow; /* True ordinarily; false if the actions must + * be reassessed for every packet. */ +}; + +static void do_xlate_actions(const union ofp_action *in, size_t n_in, + struct action_xlate_ctx *ctx); + +static void +add_output_action(struct action_xlate_ctx *ctx, uint16_t port) +{ + const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); + if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) { + odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; + } +} + +static struct rule * +lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) +{ + struct rule *rule; + rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); + + /* The rule we found might not be valid, since we could be in need of + * revalidation. If it is not valid, don't return it. */ + if (rule + && rule->super + && ofproto->need_revalidate + && !revalidate_rule(ofproto, rule)) { + COVERAGE_INC(ofproto_invalidated); + return NULL; + } + + return rule; +} + +static void +xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) +{ + if (!ctx->recurse) { + struct rule *rule; + flow_t flow; + + flow = *ctx->flow; + flow.in_port = in_port; + + rule = lookup_valid_rule(ctx->ofproto, &flow); + if (rule) { + if (rule->super) { + rule = rule->super; + } + + ctx->recurse++; + do_xlate_actions(rule->actions, rule->n_actions, ctx); + ctx->recurse--; + } + } +} + +static void +xlate_output_action(struct action_xlate_ctx *ctx, + const struct ofp_action_output *oao) +{ + uint16_t odp_port; + + switch (ntohs(oao->port)) { + case OFPP_IN_PORT: + add_output_action(ctx, ctx->flow->in_port); + break; + case OFPP_TABLE: + xlate_table_action(ctx, ctx->flow->in_port); + break; + case OFPP_NORMAL: + if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet, + ctx->out, ctx->tags, + ctx->ofproto->aux)) { + COVERAGE_INC(ofproto_uninstallable); + ctx->may_setup_flow = false; + } + break; + case OFPP_FLOOD: + add_output_group_action(ctx->out, DP_GROUP_FLOOD); + break; + case OFPP_ALL: + add_output_group_action(ctx->out, DP_GROUP_ALL); + break; + case OFPP_CONTROLLER: + add_controller_action(ctx->out, oao); + break; + case OFPP_LOCAL: + add_output_action(ctx, ODPP_LOCAL); + break; + default: + odp_port = ofp_port_to_odp_port(ntohs(oao->port)); + if (odp_port != ctx->flow->in_port) { + add_output_action(ctx, odp_port); + } + break; + } +} + +static void +xlate_nicira_action(struct action_xlate_ctx *ctx, + const struct nx_action_header *nah) +{ + const struct nx_action_resubmit *nar; + int subtype = ntohs(nah->subtype); + + assert(nah->vendor == htonl(NX_VENDOR_ID)); + switch (subtype) { + case NXAST_RESUBMIT: + nar = (const struct nx_action_resubmit *) nah; + xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); + break; + + default: + VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); + break; + } +} + +static void +do_xlate_actions(const union ofp_action *in, size_t n_in, + struct action_xlate_ctx *ctx) +{ + struct actions_iterator iter; + const union ofp_action *ia; + const struct ofport *port; + + port = port_array_get(&ctx->ofproto->ports, ctx->flow->in_port); + if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && + port->opp.config & (eth_addr_equals(ctx->flow->dl_dst, stp_eth_addr) + ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { + /* Drop this flow. */ + return; + } + + for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { + uint16_t type = ntohs(ia->type); + union odp_action *oa; + + switch (type) { + case OFPAT_OUTPUT: + xlate_output_action(ctx, &ia->output); + break; + + case OFPAT_SET_VLAN_VID: + oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); + oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; + break; + + case OFPAT_SET_VLAN_PCP: + oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); + oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; + break; + + case OFPAT_STRIP_VLAN: + odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); + break; + + case OFPAT_SET_DL_SRC: + oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); + memcpy(oa->dl_addr.dl_addr, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + break; + + case OFPAT_SET_DL_DST: + oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); + memcpy(oa->dl_addr.dl_addr, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + break; + + case OFPAT_SET_NW_SRC: + oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); + oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_TP_SRC: + oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); + oa->tp_port.tp_port = ia->tp_port.tp_port; + break; + + case OFPAT_VENDOR: + xlate_nicira_action(ctx, (const struct nx_action_header *) ia); + break; + + default: + VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); + break; + } + } +} + +static int +xlate_actions(const union ofp_action *in, size_t n_in, + const flow_t *flow, struct ofproto *ofproto, + const struct ofpbuf *packet, + struct odp_actions *out, tag_type *tags, bool *may_setup_flow) +{ + tag_type no_tags = 0; + struct action_xlate_ctx ctx; + COVERAGE_INC(ofproto_ofp2odp); + odp_actions_init(out); + ctx.flow = flow; + ctx.recurse = 0; + ctx.ofproto = ofproto; + ctx.packet = packet; + ctx.out = out; + ctx.tags = tags ? tags : &no_tags; + ctx.may_setup_flow = true; + do_xlate_actions(in, n_in, &ctx); + + /* Check with in-band control to see if we're allowed to setup this + * flow. */ + if (!in_band_rule_check(ofproto->in_band, flow, out)) { + ctx.may_setup_flow = false; + } + + if (may_setup_flow) { + *may_setup_flow = ctx.may_setup_flow; + } + if (odp_actions_overflow(out)) { + odp_actions_init(out); + return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); + } + return 0; +} + +static int +handle_packet_out(struct ofproto *p, struct ofconn *ofconn, + struct ofp_header *oh) +{ + struct ofp_packet_out *opo; + struct ofpbuf payload, *buffer; + struct odp_actions actions; + int n_actions; + uint16_t in_port; + flow_t flow; + int error; + + error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); + if (error) { + return error; + } + opo = (struct ofp_packet_out *) oh; + + COVERAGE_INC(ofproto_packet_out); + if (opo->buffer_id != htonl(UINT32_MAX)) { + error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), + &buffer, &in_port); - if (error) { ++ if (error || !buffer) { + return error; + } + payload = *buffer; + } else { + buffer = NULL; + } + + flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); + error = xlate_actions((const union ofp_action *) opo->actions, n_actions, + &flow, p, &payload, &actions, NULL, NULL); + if (error) { + return error; + } + + dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, + &payload); + ofpbuf_delete(buffer); + + return 0; +} + +static void +update_port_config(struct ofproto *p, struct ofport *port, + uint32_t config, uint32_t mask) +{ + mask &= config ^ port->opp.config; + if (mask & OFPPC_PORT_DOWN) { + if (config & OFPPC_PORT_DOWN) { + netdev_turn_flags_off(port->netdev, NETDEV_UP, true); + } else { + netdev_turn_flags_on(port->netdev, NETDEV_UP, true); + } + } +#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) + if (mask & REVALIDATE_BITS) { + COVERAGE_INC(ofproto_costly_flags); + port->opp.config ^= mask & REVALIDATE_BITS; + p->need_revalidate = true; + } +#undef REVALIDATE_BITS + if (mask & OFPPC_NO_FLOOD) { + port->opp.config ^= OFPPC_NO_FLOOD; + refresh_port_group(p, DP_GROUP_FLOOD); + } + if (mask & OFPPC_NO_PACKET_IN) { + port->opp.config ^= OFPPC_NO_PACKET_IN; + } +} + +static int +handle_port_mod(struct ofproto *p, struct ofp_header *oh) +{ + const struct ofp_port_mod *opm; + struct ofport *port; + int error; + + error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); + if (error) { + return error; + } + opm = (struct ofp_port_mod *) oh; + + port = port_array_get(&p->ports, + ofp_port_to_odp_port(ntohs(opm->port_no))); + if (!port) { + return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); + } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { + return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); + } else { + update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); + if (opm->advertise) { + netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); + } + } + return 0; +} + +static struct ofpbuf * +make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) +{ + struct ofp_stats_reply *osr; + struct ofpbuf *msg; + + msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); + osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); + osr->type = type; + osr->flags = htons(0); + return msg; +} + +static struct ofpbuf * +start_stats_reply(const struct ofp_stats_request *request, size_t body_len) +{ + return make_stats_reply(request->header.xid, request->type, body_len); +} + +static void * +append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) +{ + struct ofpbuf *msg = *msgp; + assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); + if (nbytes + msg->size > UINT16_MAX) { + struct ofp_stats_reply *reply = msg->data; + reply->flags = htons(OFPSF_REPLY_MORE); + *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes); + queue_tx(msg, ofconn, ofconn->reply_counter); + } + return ofpbuf_put_uninit(*msgp, nbytes); +} + +static int +handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, + struct ofp_stats_request *request) +{ + struct ofp_desc_stats *ods; + struct ofpbuf *msg; + + msg = start_stats_reply(request, sizeof *ods); + ods = append_stats_reply(sizeof *ods, ofconn, &msg); + strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc); + strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc); + strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc); + strncpy(ods->serial_num, p->serial, sizeof ods->serial_num); + queue_tx(msg, ofconn, ofconn->reply_counter); + + return 0; +} + +static void +count_subrules(struct cls_rule *cls_rule, void *n_subrules_) +{ + struct rule *rule = rule_from_cls_rule(cls_rule); + int *n_subrules = n_subrules_; + + if (rule->super) { + (*n_subrules)++; + } +} + +static int +handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, + struct ofp_stats_request *request) +{ + struct ofp_table_stats *ots; + struct ofpbuf *msg; + struct odp_stats dpstats; + int n_exact, n_subrules, n_wild; + + msg = start_stats_reply(request, sizeof *ots * 2); + + /* Count rules of various kinds. */ + n_subrules = 0; + classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); + n_exact = classifier_count_exact(&p->cls) - n_subrules; + n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); + + /* Hash table. */ + dpif_get_dp_stats(p->dpif, &dpstats); + ots = append_stats_reply(sizeof *ots, ofconn, &msg); + memset(ots, 0, sizeof *ots); + ots->table_id = TABLEID_HASH; + strcpy(ots->name, "hash"); + ots->wildcards = htonl(0); + ots->max_entries = htonl(dpstats.max_capacity); + ots->active_count = htonl(n_exact); + ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + + dpstats.n_missed); + ots->matched_count = htonll(dpstats.n_hit); /* XXX */ + + /* Classifier table. */ + ots = append_stats_reply(sizeof *ots, ofconn, &msg); + memset(ots, 0, sizeof *ots); + ots->table_id = TABLEID_CLASSIFIER; + strcpy(ots->name, "classifier"); + ots->wildcards = htonl(OFPFW_ALL); + ots->max_entries = htonl(65536); + ots->active_count = htonl(n_wild); + ots->lookup_count = htonll(0); /* XXX */ + ots->matched_count = htonll(0); /* XXX */ + + queue_tx(msg, ofconn, ofconn->reply_counter); + return 0; +} + +static int +handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, + struct ofp_stats_request *request) +{ + struct ofp_port_stats *ops; + struct ofpbuf *msg; + struct ofport *port; + unsigned int port_no; + + msg = start_stats_reply(request, sizeof *ops * 16); + PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + struct netdev_stats stats; + + /* Intentionally ignore return value, since errors will set 'stats' to + * all-1s, which is correct for OpenFlow, and netdev_get_stats() will + * log errors. */ + netdev_get_stats(port->netdev, &stats); + + ops = append_stats_reply(sizeof *ops, ofconn, &msg); + ops->port_no = htons(odp_port_to_ofp_port(port_no)); + memset(ops->pad, 0, sizeof ops->pad); + ops->rx_packets = htonll(stats.rx_packets); + ops->tx_packets = htonll(stats.tx_packets); + ops->rx_bytes = htonll(stats.rx_bytes); + ops->tx_bytes = htonll(stats.tx_bytes); + ops->rx_dropped = htonll(stats.rx_dropped); + ops->tx_dropped = htonll(stats.tx_dropped); + ops->rx_errors = htonll(stats.rx_errors); + ops->tx_errors = htonll(stats.tx_errors); + ops->rx_frame_err = htonll(stats.rx_frame_errors); + ops->rx_over_err = htonll(stats.rx_over_errors); + ops->rx_crc_err = htonll(stats.rx_crc_errors); + ops->collisions = htonll(stats.collisions); + } + + queue_tx(msg, ofconn, ofconn->reply_counter); + return 0; +} + +struct flow_stats_cbdata { + struct ofproto *ofproto; + struct ofconn *ofconn; + uint16_t out_port; + struct ofpbuf *msg; +}; + +static void +query_stats(struct ofproto *p, struct rule *rule, + uint64_t *packet_countp, uint64_t *byte_countp) +{ + uint64_t packet_count, byte_count; + struct rule *subrule; + struct odp_flow *odp_flows; + size_t n_odp_flows; + + n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; + odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows); + if (rule->cr.wc.wildcards) { + size_t i = 0; + LIST_FOR_EACH (subrule, struct rule, list, &rule->list) { + odp_flows[i++].key = subrule->cr.flow; + } + } else { + odp_flows[0].key = rule->cr.flow; + } + + packet_count = rule->packet_count; + byte_count = rule->byte_count; + if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { + size_t i; + for (i = 0; i < n_odp_flows; i++) { + struct odp_flow *odp_flow = &odp_flows[i]; + packet_count += odp_flow->stats.n_packets; + byte_count += odp_flow->stats.n_bytes; + } + } + free(odp_flows); + + *packet_countp = packet_count; + *byte_countp = byte_count; +} + +static void +flow_stats_cb(struct cls_rule *rule_, void *cbdata_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct flow_stats_cbdata *cbdata = cbdata_; + struct ofp_flow_stats *ofs; + uint64_t packet_count, byte_count; + size_t act_len, len; + + if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { + return; + } + + act_len = sizeof *rule->actions * rule->n_actions; + len = offsetof(struct ofp_flow_stats, actions) + act_len; + + query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); + + ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); + ofs->length = htons(len); + ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; + ofs->pad = 0; + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match); + ofs->duration = htonl((time_msec() - rule->created) / 1000); + ofs->priority = htons(rule->cr.priority); + ofs->idle_timeout = htons(rule->idle_timeout); + ofs->hard_timeout = htons(rule->hard_timeout); + memset(ofs->pad2, 0, sizeof ofs->pad2); + ofs->packet_count = htonll(packet_count); + ofs->byte_count = htonll(byte_count); + memcpy(ofs->actions, rule->actions, act_len); +} + +static int +table_id_to_include(uint8_t table_id) +{ + return (table_id == TABLEID_HASH ? CLS_INC_EXACT + : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD + : table_id == 0xff ? CLS_INC_ALL + : 0); +} + +static int +handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, + const struct ofp_stats_request *osr, + size_t arg_size) +{ + struct ofp_flow_stats_request *fsr; + struct flow_stats_cbdata cbdata; + struct cls_rule target; + + if (arg_size != sizeof *fsr) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + } + fsr = (struct ofp_flow_stats_request *) osr->body; + + COVERAGE_INC(ofproto_flows_req); + cbdata.ofproto = p; + cbdata.ofconn = ofconn; + cbdata.out_port = fsr->out_port; + cbdata.msg = start_stats_reply(osr, 1024); + cls_rule_from_match(&target, &fsr->match, 0); + classifier_for_each_match(&p->cls, &target, + table_id_to_include(fsr->table_id), + flow_stats_cb, &cbdata); + queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); + return 0; +} + +struct flow_stats_ds_cbdata { + struct ofproto *ofproto; + struct ds *results; +}; + +static void +flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct flow_stats_ds_cbdata *cbdata = cbdata_; + struct ds *results = cbdata->results; + struct ofp_match match; + uint64_t packet_count, byte_count; + size_t act_len = sizeof *rule->actions * rule->n_actions; + + /* Don't report on subrules. */ + if (rule->super != NULL) { + return; + } + + query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); + flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match); + + ds_put_format(results, "duration=%llds, ", + (time_msec() - rule->created) / 1000); + ds_put_format(results, "priority=%u, ", rule->cr.priority); + ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); + ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); + ofp_print_match(results, &match, true); + ofp_print_actions(results, &rule->actions->header, act_len); + ds_put_cstr(results, "\n"); +} + +/* Adds a pretty-printed description of all flows to 'results', including + * those marked hidden by secchan (e.g., by in-band control). */ +void +ofproto_get_all_flows(struct ofproto *p, struct ds *results) +{ + struct ofp_match match; + struct cls_rule target; + struct flow_stats_ds_cbdata cbdata; + + memset(&match, 0, sizeof match); + match.wildcards = htonl(OFPFW_ALL); + + cbdata.ofproto = p; + cbdata.results = results; + + cls_rule_from_match(&target, &match, 0); + classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, + flow_stats_ds_cb, &cbdata); +} + +struct aggregate_stats_cbdata { + struct ofproto *ofproto; + uint16_t out_port; + uint64_t packet_count; + uint64_t byte_count; + uint32_t n_flows; +}; + +static void +aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct aggregate_stats_cbdata *cbdata = cbdata_; + uint64_t packet_count, byte_count; + + if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { + return; + } + + query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); + + cbdata->packet_count += packet_count; + cbdata->byte_count += byte_count; + cbdata->n_flows++; +} + +static int +handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, + const struct ofp_stats_request *osr, + size_t arg_size) +{ + struct ofp_aggregate_stats_request *asr; + struct ofp_aggregate_stats_reply *reply; + struct aggregate_stats_cbdata cbdata; + struct cls_rule target; + struct ofpbuf *msg; + + if (arg_size != sizeof *asr) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + } + asr = (struct ofp_aggregate_stats_request *) osr->body; + + COVERAGE_INC(ofproto_agg_request); + cbdata.ofproto = p; + cbdata.out_port = asr->out_port; + cbdata.packet_count = 0; + cbdata.byte_count = 0; + cbdata.n_flows = 0; + cls_rule_from_match(&target, &asr->match, 0); + classifier_for_each_match(&p->cls, &target, + table_id_to_include(asr->table_id), + aggregate_stats_cb, &cbdata); + + msg = start_stats_reply(osr, sizeof *reply); + reply = append_stats_reply(sizeof *reply, ofconn, &msg); + reply->flow_count = htonl(cbdata.n_flows); + reply->packet_count = htonll(cbdata.packet_count); + reply->byte_count = htonll(cbdata.byte_count); + queue_tx(msg, ofconn, ofconn->reply_counter); + return 0; +} + +static int +handle_stats_request(struct ofproto *p, struct ofconn *ofconn, + struct ofp_header *oh) +{ + struct ofp_stats_request *osr; + size_t arg_size; + int error; + + error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, + 1, &arg_size); + if (error) { + return error; + } + osr = (struct ofp_stats_request *) oh; + + switch (ntohs(osr->type)) { + case OFPST_DESC: + return handle_desc_stats_request(p, ofconn, osr); + + case OFPST_FLOW: + return handle_flow_stats_request(p, ofconn, osr, arg_size); + + case OFPST_AGGREGATE: + return handle_aggregate_stats_request(p, ofconn, osr, arg_size); + + case OFPST_TABLE: + return handle_table_stats_request(p, ofconn, osr); + + case OFPST_PORT: + return handle_port_stats_request(p, ofconn, osr); + + case OFPST_VENDOR: + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); + + default: + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + } +} + +static long long int +msec_from_nsec(uint64_t sec, uint32_t nsec) +{ + return !sec ? 0 : sec * 1000 + nsec / 1000000; +} + +static void +update_time(struct rule *rule, const struct odp_flow_stats *stats) +{ + long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); + if (used > rule->used) { + rule->used = used; + } +} + +static void +update_stats(struct rule *rule, const struct odp_flow_stats *stats) +{ + update_time(rule, stats); + rule->packet_count += stats->n_packets; + rule->byte_count += stats->n_bytes; + rule->tcp_flags |= stats->tcp_flags; + if (stats->n_packets) { + rule->ip_tos = stats->ip_tos; + } +} + +static int +add_flow(struct ofproto *p, struct ofconn *ofconn, + struct ofp_flow_mod *ofm, size_t n_actions) +{ + struct ofpbuf *packet; + struct rule *rule; + uint16_t in_port; + int error; + + rule = rule_create(NULL, (const union ofp_action *) ofm->actions, + n_actions, ntohs(ofm->idle_timeout), + ntohs(ofm->hard_timeout)); + cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority)); + + packet = NULL; + error = 0; + if (ofm->buffer_id != htonl(UINT32_MAX)) { + error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), + &packet, &in_port); + } + + rule_insert(p, rule, packet, in_port); + ofpbuf_delete(packet); + return error; +} + +static int +modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, + size_t n_actions, uint16_t command, struct rule *rule) +{ + if (rule_is_hidden(rule)) { + return 0; + } + + if (command == OFPFC_DELETE) { + rule_remove(p, rule); + } else { + size_t actions_len = n_actions * sizeof *rule->actions; + + if (n_actions == rule->n_actions + && !memcmp(ofm->actions, rule->actions, actions_len)) + { + return 0; + } + + free(rule->actions); + rule->actions = xmemdup(ofm->actions, actions_len); + rule->n_actions = n_actions; + + if (rule->cr.wc.wildcards) { + COVERAGE_INC(ofproto_mod_wc_flow); + p->need_revalidate = true; + } else { + rule_update_actions(p, rule); + } + } + + return 0; +} + +static int +modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm, + size_t n_actions, uint16_t command) +{ + struct rule *rule; + uint32_t wildcards; + flow_t flow; + + flow_from_match(&flow, &wildcards, &ofm->match); + rule = rule_from_cls_rule(classifier_find_rule_exactly( + &p->cls, &flow, wildcards, + ntohs(ofm->priority))); + + if (rule) { + if (command == OFPFC_DELETE + && ofm->out_port != htons(OFPP_NONE) + && !rule_has_out_port(rule, ofm->out_port)) { + return 0; + } + + modify_flow(p, ofm, n_actions, command, rule); + } + return 0; +} + +struct modify_flows_cbdata { + struct ofproto *ofproto; + const struct ofp_flow_mod *ofm; + uint16_t out_port; + size_t n_actions; + uint16_t command; +}; + +static void +modify_flows_cb(struct cls_rule *rule_, void *cbdata_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct modify_flows_cbdata *cbdata = cbdata_; + + if (cbdata->out_port != htons(OFPP_NONE) + && !rule_has_out_port(rule, cbdata->out_port)) { + return; + } + + modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, + cbdata->command, rule); +} + +static int +modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm, + size_t n_actions, uint16_t command) +{ + struct modify_flows_cbdata cbdata; + struct cls_rule target; + + cbdata.ofproto = p; + cbdata.ofm = ofm; + cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port + : htons(OFPP_NONE)); + cbdata.n_actions = n_actions; + cbdata.command = command; + + cls_rule_from_match(&target, &ofm->match, 0); + + classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, + modify_flows_cb, &cbdata); + return 0; +} + +static int +handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, + struct ofp_flow_mod *ofm) +{ + size_t n_actions; + int error; + + error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, + sizeof *ofm->actions, &n_actions); + if (error) { + return error; + } + + normalize_match(&ofm->match); + if (!ofm->match.wildcards) { + ofm->priority = htons(UINT16_MAX); + } + + error = validate_actions((const union ofp_action *) ofm->actions, + n_actions, p->max_ports); + if (error) { + return error; + } + + switch (ntohs(ofm->command)) { + case OFPFC_ADD: + return add_flow(p, ofconn, ofm, n_actions); + + case OFPFC_MODIFY: + return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY); + + case OFPFC_MODIFY_STRICT: + return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY); + + case OFPFC_DELETE: + return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE); + + case OFPFC_DELETE_STRICT: + return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE); + + default: + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); + } +} + +static void +send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid) +{ + struct ofmp_capability_reply *ocr; + struct ofpbuf *b; + char capabilities[] = "com.nicira.mgmt.manager=false\n"; + + ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b); + ocr->header.header.vendor = htonl(NX_VENDOR_ID); + ocr->header.header.subtype = htonl(NXT_MGMT); + ocr->header.type = htons(OFMPT_CAPABILITY_REPLY); + + ocr->format = htonl(OFMPCOF_SIMPLE); + ocr->mgmt_id = htonll(p->mgmt_id); + + ofpbuf_put(b, capabilities, strlen(capabilities)); + + queue_tx(b, ofconn, ofconn->reply_counter); +} + +static int +handle_ofmp(struct ofproto *p, struct ofconn *ofconn, + struct ofmp_header *ofmph) +{ + size_t msg_len = ntohs(ofmph->header.header.length); + if (msg_len < sizeof(*ofmph)) { + VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + } + + if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) { + struct ofmp_capability_request *ofmpcr; + + if (msg_len < sizeof(struct ofmp_capability_request)) { + VLOG_WARN_RL(&rl, "dropping short capability request: %d\n", + msg_len); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + } + + ofmpcr = (struct ofmp_capability_request *)ofmph; + if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) { + /* xxx Find a better type than bad subtype */ + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); + } + + send_capability_reply(p, ofconn, ofmph->header.header.xid); + return 0; + } else { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); + } +} + +static int +handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) +{ + struct ofp_vendor_header *ovh = msg; + struct nicira_header *nh; + + if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + } + if (ovh->vendor != htonl(NX_VENDOR_ID)) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); + } + if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + } + + nh = msg; + switch (ntohl(nh->subtype)) { + case NXT_STATUS_REQUEST: + return switch_status_handle_request(p->switch_status, ofconn->rconn, + msg); + + case NXT_ACT_SET_CONFIG: + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */ + + case NXT_ACT_GET_CONFIG: + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */ + + case NXT_COMMAND_REQUEST: + if (p->executer) { + return executer_handle_request(p->executer, ofconn->rconn, msg); + } + break; + + case NXT_MGMT: + return handle_ofmp(p, ofconn, msg); + } + + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); +} + +static void +handle_openflow(struct ofconn *ofconn, struct ofproto *p, + struct ofpbuf *ofp_msg) +{ + struct ofp_header *oh = ofp_msg->data; + int error; + + COVERAGE_INC(ofproto_recv_openflow); + switch (oh->type) { + case OFPT_ECHO_REQUEST: + error = handle_echo_request(ofconn, oh); + break; + + case OFPT_ECHO_REPLY: + error = 0; + break; + + case OFPT_FEATURES_REQUEST: + error = handle_features_request(p, ofconn, oh); + break; + + case OFPT_GET_CONFIG_REQUEST: + error = handle_get_config_request(p, ofconn, oh); + break; + + case OFPT_SET_CONFIG: + error = handle_set_config(p, ofconn, ofp_msg->data); + break; + + case OFPT_PACKET_OUT: + error = handle_packet_out(p, ofconn, ofp_msg->data); + break; + + case OFPT_PORT_MOD: + error = handle_port_mod(p, oh); + break; + + case OFPT_FLOW_MOD: + error = handle_flow_mod(p, ofconn, ofp_msg->data); + break; + + case OFPT_STATS_REQUEST: + error = handle_stats_request(p, ofconn, oh); + break; + + case OFPT_VENDOR: + error = handle_vendor(p, ofconn, ofp_msg->data); + break; + + default: + if (VLOG_IS_WARN_ENABLED()) { + char *s = ofp_to_string(oh, ntohs(oh->length), 2); + VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); + free(s); + } + error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); + break; + } + + if (error) { + send_error_oh(ofconn, ofp_msg->data, error); + } +} + +static void +handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) +{ + struct odp_msg *msg = packet->data; + uint16_t in_port = odp_port_to_ofp_port(msg->port); + struct rule *rule; + struct ofpbuf payload; + flow_t flow; + + /* Handle controller actions. */ + if (msg->type == _ODPL_ACTION_NR) { + COVERAGE_INC(ofproto_ctlr_action); + pinsched_send(p->action_sched, in_port, packet, + send_packet_in_action, p); + return; + } + + payload.data = msg + 1; + payload.size = msg->length - sizeof *msg; + flow_extract(&payload, msg->port, &flow); + + /* Check with in-band control to see if this packet should be sent + * to the local port regardless of the flow table. */ + if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { + union odp_action action; + + memset(&action, 0, sizeof(action)); + action.output.type = ODPAT_OUTPUT; + action.output.port = ODPP_LOCAL; + dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); + } + + rule = lookup_valid_rule(p, &flow); + if (!rule) { + /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ + struct ofport *port = port_array_get(&p->ports, msg->port); + if (port) { + if (port->opp.config & OFPPC_NO_PACKET_IN) { + COVERAGE_INC(ofproto_no_packet_in); + /* XXX install 'drop' flow entry */ + ofpbuf_delete(packet); + return; + } + } else { + VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); + } + + COVERAGE_INC(ofproto_packet_in); + pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); + return; + } + + if (rule->cr.wc.wildcards) { + rule = rule_create_subrule(p, rule, &flow); + rule_make_actions(p, rule, packet); + } else { + if (!rule->may_install) { + /* The rule is not installable, that is, we need to process every + * packet, so process the current packet and set its actions into + * 'subrule'. */ + rule_make_actions(p, rule, packet); + } else { + /* XXX revalidate rule if it needs it */ + } + } + + rule_execute(p, rule, &payload, &flow); + rule_reinstall(p, rule); - ofpbuf_delete(packet); ++ ++ if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY ++ && rconn_is_connected(p->controller->rconn)) { ++ /* ++ * Extra-special case for fail-open mode. ++ * ++ * We are in fail-open mode and the packet matched the fail-open rule, ++ * but we are connected to a controller too. We should send the packet ++ * up to the controller in the hope that it will try to set up a flow ++ * and thereby allow us to exit fail-open. ++ * ++ * See the top-level comment in fail-open.c for more information. ++ */ ++ pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); ++ } else { ++ ofpbuf_delete(packet); ++ } +} + +static void +revalidate_cb(struct cls_rule *sub_, void *cbdata_) +{ + struct rule *sub = rule_from_cls_rule(sub_); + struct revalidate_cbdata *cbdata = cbdata_; + + if (cbdata->revalidate_all + || (cbdata->revalidate_subrules && sub->super) + || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { + revalidate_rule(cbdata->ofproto, sub); + } +} + +static bool +revalidate_rule(struct ofproto *p, struct rule *rule) +{ + const flow_t *flow = &rule->cr.flow; + + COVERAGE_INC(ofproto_revalidate_rule); + if (rule->super) { + struct rule *super; + super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); + if (!super) { + rule_remove(p, rule); + return false; + } else if (super != rule->super) { + COVERAGE_INC(ofproto_revalidate_moved); + list_remove(&rule->list); + list_push_back(&super->list, &rule->list); + rule->super = super; + rule->hard_timeout = super->hard_timeout; + rule->idle_timeout = super->idle_timeout; + rule->created = super->created; + rule->used = 0; + } + } + + rule_update_actions(p, rule); + return true; +} + +static struct ofpbuf * +compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason) +{ + struct ofp_flow_expired *ofe; + struct ofpbuf *buf; + + ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf); + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match); + ofe->priority = htons(rule->cr.priority); + ofe->reason = reason; + ofe->duration = (now - rule->created) / 1000; + ofe->packet_count = rule->packet_count; + ofe->byte_count = rule->byte_count; + + return buf; +} + +static void +send_flow_exp(struct ofproto *p, struct rule *rule, + long long int now, uint8_t reason) +{ + struct ofconn *ofconn; + struct ofconn *prev; + struct ofpbuf *buf = NULL; + + /* We limit the maximum number of queued flow expirations it by accounting + * them under the counter for replies. That works because preventing + * OpenFlow requests from being processed also prevents new flows from + * being added (and expiring). (It also prevents processing OpenFlow + * requests that would not add new flows, so it is imperfect.) */ + + prev = NULL; + LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) { + if (prev) { - queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter); ++ queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); + } else { + buf = compose_flow_exp(rule, now, reason); + } + prev = ofconn; + } + } + if (prev) { - queue_tx(buf, prev, ofconn->reply_counter); ++ queue_tx(buf, prev, prev->reply_counter); + } +} + +static void +uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) +{ + assert(rule->installed); + assert(!rule->cr.wc.wildcards); + + if (rule->super) { + rule_remove(ofproto, rule); + } else { + rule_uninstall(ofproto, rule); + } +} + +static void +expire_rule(struct cls_rule *cls_rule, void *p_) +{ + struct ofproto *p = p_; + struct rule *rule = rule_from_cls_rule(cls_rule); + long long int hard_expire, idle_expire, expire, now; + + hard_expire = (rule->hard_timeout + ? rule->created + rule->hard_timeout * 1000 + : LLONG_MAX); + idle_expire = (rule->idle_timeout + && (rule->super || list_is_empty(&rule->list)) + ? rule->used + rule->idle_timeout * 1000 + : LLONG_MAX); + expire = MIN(hard_expire, idle_expire); + if (expire == LLONG_MAX) { + if (rule->installed && time_msec() >= rule->used + 5000) { + uninstall_idle_flow(p, rule); + } + return; + } + + now = time_msec(); + if (now < expire) { + if (rule->installed && now >= rule->used + 5000) { + uninstall_idle_flow(p, rule); + } + return; + } + + COVERAGE_INC(ofproto_expired); + if (rule->cr.wc.wildcards) { + /* Update stats. (This code will be a no-op if the rule expired + * due to an idle timeout, because in that case the rule has no + * subrules left.) */ + struct rule *subrule, *next; + LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { + rule_remove(p, subrule); + } + } + + send_flow_exp(p, rule, now, + (now >= hard_expire + ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT)); + rule_remove(p, rule); +} + +static void +update_used(struct ofproto *p) +{ + struct odp_flow *flows; + size_t n_flows; + size_t i; + int error; + + error = dpif_flow_list_all(p->dpif, &flows, &n_flows); + if (error) { + return; + } + + for (i = 0; i < n_flows; i++) { + struct odp_flow *f = &flows[i]; + struct rule *rule; + + rule = rule_from_cls_rule( + classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); + if (!rule || !rule->installed) { + COVERAGE_INC(ofproto_unexpected_rule); + dpif_flow_del(p->dpif, f); + continue; + } + + update_time(rule, &f->stats); + rule_account(p, rule, f->stats.n_bytes); + } + free(flows); +} + +static void +do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id, + const struct ofpbuf *packet, int send_len) +{ - struct ofp_packet_in *opi; - struct ofpbuf payload, *buf; - struct odp_msg *msg; ++ struct odp_msg *msg = packet->data; ++ struct ofpbuf payload; ++ struct ofpbuf *opi; ++ uint8_t reason; + - msg = packet->data; ++ /* Extract packet payload from 'msg'. */ + payload.data = msg + 1; + payload.size = msg->length - sizeof *msg; + - send_len = MIN(send_len, payload.size); - buf = ofpbuf_new(sizeof *opi + send_len); - opi = put_openflow_xid(offsetof(struct ofp_packet_in, data), - OFPT_PACKET_IN, 0, buf); - opi->buffer_id = htonl(buffer_id); - opi->total_len = htons(payload.size); - opi->in_port = htons(odp_port_to_ofp_port(msg->port)); - opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH; - ofpbuf_put(buf, payload.data, MIN(send_len, payload.size)); - update_openflow_length(buf); - rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100); ++ /* Construct ofp_packet_in message. */ ++ reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH; ++ opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason, ++ &payload, send_len); ++ ++ /* Send. */ ++ rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100); +} + +static void +send_packet_in_action(struct ofpbuf *packet, void *p_) +{ + struct ofproto *p = p_; + struct ofconn *ofconn; + struct odp_msg *msg; + + msg = packet->data; + LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + if (ofconn == p->controller || ofconn->miss_send_len) { + do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg); + } + } + ofpbuf_delete(packet); +} + +static void +send_packet_in_miss(struct ofpbuf *packet, void *p_) +{ + struct ofproto *p = p_; ++ bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open); + struct ofconn *ofconn; + struct ofpbuf payload; + struct odp_msg *msg; + + msg = packet->data; + payload.data = msg + 1; + payload.size = msg->length - sizeof *msg; + LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + if (ofconn->miss_send_len) { - uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload, - msg->port); ++ struct pktbuf *pb = ofconn->pktbuf; ++ uint32_t buffer_id = (in_fail_open ++ ? pktbuf_get_null() ++ : pktbuf_save(pb, &payload, msg->port)); + int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len + : UINT32_MAX); + do_send_packet_in(ofconn, buffer_id, packet, send_len); + } + } + ofpbuf_delete(packet); +} + +static uint64_t +pick_datapath_id(const struct ofproto *ofproto) +{ + const struct ofport *port; + + port = port_array_get(&ofproto->ports, ODPP_LOCAL); + if (port) { + uint8_t ea[ETH_ADDR_LEN]; + int error; + + error = netdev_get_etheraddr(port->netdev, ea); + if (!error) { + return eth_addr_to_uint64(ea); + } + VLOG_WARN("could not get MAC address for %s (%s)", + netdev_get_name(port->netdev), strerror(error)); + } + return ofproto->fallback_dpid; +} + +static uint64_t +pick_fallback_dpid(void) +{ + uint8_t ea[ETH_ADDR_LEN]; + eth_addr_random(ea); + ea[0] = 0x00; /* Set Nicira OUI. */ + ea[1] = 0x23; + ea[2] = 0x20; + return eth_addr_to_uint64(ea); +} + +static bool +default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, + struct odp_actions *actions, tag_type *tags, + void *ofproto_) +{ + struct ofproto *ofproto = ofproto_; + int out_port; + + /* Drop frames for reserved multicast addresses. */ + if (eth_addr_is_reserved(flow->dl_dst)) { + return true; + } + + /* Learn source MAC (but don't try to learn from revalidation). */ + if (packet != NULL) { + tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, + 0, flow->in_port); + if (rev_tag) { + /* The log messages here could actually be useful in debugging, + * so keep the rate limit relatively high. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); + VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, + ETH_ADDR_ARGS(flow->dl_src), flow->in_port); + ofproto_revalidate(ofproto, rev_tag); + } + } + + /* Determine output port. */ + out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags); + if (out_port < 0) { + add_output_group_action(actions, DP_GROUP_FLOOD); + } else if (out_port != flow->in_port) { + odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; + } else { + /* Drop. */ + } + + return true; +} + +static const struct ofhooks default_ofhooks = { + NULL, + default_normal_ofhook_cb, + NULL, + NULL +}; diff --cc ofproto/pktbuf.c index b4198a8f,00000000..450cc3b6 mode 100644,000000..100644 --- a/ofproto/pktbuf.c +++ b/ofproto/pktbuf.c @@@ -1,150 -1,0 +1,209 @@@ +/* + * Copyright (c) 2008, 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include "pktbuf.h" +#include +#include +#include "coverage.h" +#include "ofpbuf.h" +#include "timeval.h" +#include "util.h" +#include "vconn.h" + +#define THIS_MODULE VLM_pktbuf +#include "vlog.h" + +/* Buffers are identified by a 32-bit opaque ID. We divide the ID + * into a buffer number (low bits) and a cookie (high bits). The buffer number + * is an index into an array of buffers. The cookie distinguishes between + * different packets that have occupied a single buffer. Thus, the more + * buffers we have, the lower-quality the cookie... */ +#define PKTBUF_BITS 8 +#define PKTBUF_MASK (PKTBUF_CNT - 1) +#define PKTBUF_CNT (1u << PKTBUF_BITS) + +#define COOKIE_BITS (32 - PKTBUF_BITS) +#define COOKIE_MAX ((1u << COOKIE_BITS) - 1) + +#define OVERWRITE_MSECS 5000 + +struct packet { + struct ofpbuf *buffer; + uint32_t cookie; + long long int timeout; + uint16_t in_port; +}; + +struct pktbuf { + struct packet packets[PKTBUF_CNT]; + unsigned int buffer_idx; ++ unsigned int null_idx; +}; + +int +pktbuf_capacity(void) +{ + return PKTBUF_CNT; +} + +struct pktbuf * +pktbuf_create(void) +{ + return xcalloc(1, sizeof *pktbuf_create()); +} + +void +pktbuf_destroy(struct pktbuf *pb) +{ + if (pb) { + size_t i; + + for (i = 0; i < PKTBUF_CNT; i++) { + ofpbuf_delete(pb->packets[i].buffer); + } + free(pb); + } +} + ++static unsigned int ++make_id(unsigned int buffer_idx, unsigned int cookie) ++{ ++ return buffer_idx | (cookie << PKTBUF_BITS); ++} ++ ++/* Attempts to allocate an OpenFlow packet buffer id within 'pb'. The packet ++ * buffer will store a copy of 'buffer' and the port number 'in_port', which ++ * should be the datapath port number on which 'buffer' was received. ++ * ++ * If successful, returns the packet buffer id (a number other than ++ * UINT32_MAX). pktbuf_retrieve() can later be used to retrieve the buffer and ++ * its input port number (buffers do expire after a time, so this is not ++ * guaranteed to be true forever). On failure, returns UINT32_MAX. ++ * ++ * The caller retains ownership of 'buffer'. */ +uint32_t +pktbuf_save(struct pktbuf *pb, struct ofpbuf *buffer, uint16_t in_port) +{ + struct packet *p = &pb->packets[pb->buffer_idx]; + pb->buffer_idx = (pb->buffer_idx + 1) & PKTBUF_MASK; + if (p->buffer) { + if (time_msec() < p->timeout) { + return UINT32_MAX; + } + ofpbuf_delete(p->buffer); + } + + /* Don't use maximum cookie value since all-1-bits ID is special. */ + if (++p->cookie >= COOKIE_MAX) { + p->cookie = 0; + } + p->buffer = ofpbuf_clone(buffer); + p->timeout = time_msec() + OVERWRITE_MSECS; + p->in_port = in_port; - return (p - pb->packets) | (p->cookie << PKTBUF_BITS); ++ return make_id(p - pb->packets, p->cookie); ++} ++ ++/* ++ * Allocates and returns a "null" packet buffer id. The returned packet buffer ++ * id is considered valid by pktbuf_retrieve(), but it is not associated with ++ * actual buffered data. ++ * ++ * This function is always successful. ++ * ++ * This is useful in one special case: with the current OpenFlow design, the ++ * "fail-open" code cannot always know whether a connection to a controller is ++ * actually valid until it receives a OFPT_PACKET_OUT or OFPT_FLOW_MOD request, ++ * but at that point the packet in question has already been forwarded (since ++ * we are still in "fail-open" mode). If the packet was buffered in the usual ++ * way, then the OFPT_PACKET_OUT or OFPT_FLOW_MOD would cause a duplicate ++ * packet in the network. Null packet buffer ids identify such a packet that ++ * has already been forwarded, so that Open vSwitch can quietly ignore the ++ * request to re-send it. (After that happens, the switch exits fail-open ++ * mode.) ++ * ++ * See the top-level comment in fail-open.c for an overview. ++ */ ++uint32_t ++pktbuf_get_null(void) ++{ ++ return make_id(0, COOKIE_MAX); +} + ++/* Attempts to retrieve a saved packet with the given 'id' from 'pb'. Returns ++ * 0 if successful, otherwise an OpenFlow error code constructed with ++ * ofp_mkerr(). ++ * ++ * On success, ordinarily stores the buffered packet in '*bufferp' and the ++ * datapath port number on which the packet was received in '*in_port'. The ++ * caller becomes responsible for freeing the buffer. However, if 'id' ++ * identifies a "null" packet buffer (created with pktbuf_get_null()), stores ++ * NULL in '*bufferp' and -1 in '*in_port'. ++ * ++ * On failure, stores NULL in in '*bufferp' and -1 in '*in_port'. */ +int +pktbuf_retrieve(struct pktbuf *pb, uint32_t id, struct ofpbuf **bufferp, + uint16_t *in_port) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 20); + struct packet *p; + int error; + + if (!pb) { + VLOG_WARN_RL(&rl, "attempt to send buffered packet via connection " + "without buffers"); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_COOKIE); + } + + p = &pb->packets[id & PKTBUF_MASK]; + if (p->cookie == id >> PKTBUF_BITS) { + struct ofpbuf *buffer = p->buffer; + if (buffer) { + *bufferp = buffer; + *in_port = p->in_port; + p->buffer = NULL; + COVERAGE_INC(pktbuf_retrieved); + return 0; + } else { + COVERAGE_INC(pktbuf_reuse_error); + VLOG_WARN_RL(&rl, "attempt to reuse buffer %08"PRIx32, id); + error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BUFFER_EMPTY); + } - } else { ++ } else if (id >> PKTBUF_BITS != COOKIE_MAX) { + COVERAGE_INC(pktbuf_bad_cookie); + VLOG_WARN_RL(&rl, "cookie mismatch: %08"PRIx32" != %08"PRIx32, + id, (id & PKTBUF_MASK) | (p->cookie << PKTBUF_BITS)); + error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_COOKIE); ++ } else { ++ COVERAGE_INC(pktbuf_null_cookie); ++ VLOG_INFO_RL(&rl, "Received null cookie %08"PRIx32" (this is normal " ++ "if the switch was recently in fail-open mode)", id); ++ error = 0; + } + *bufferp = NULL; + *in_port = -1; + return error; +} + +void +pktbuf_discard(struct pktbuf *pb, uint32_t id) +{ + struct packet *p = &pb->packets[id & PKTBUF_MASK]; + if (p->cookie == id >> PKTBUF_BITS) { + ofpbuf_delete(p->buffer); + p->buffer = NULL; + } +} diff --cc ofproto/pktbuf.h index b27b7490,00000000..67f4973c mode 100644,000000..100644 --- a/ofproto/pktbuf.h +++ b/ofproto/pktbuf.h @@@ -1,34 -1,0 +1,35 @@@ +/* + * Copyright (c) 2008, 2009 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef PKTBUF_H +#define PKTBUF_H 1 + +#include + +struct pktbuf; +struct ofpbuf; + +int pktbuf_capacity(void); + +struct pktbuf *pktbuf_create(void); +void pktbuf_destroy(struct pktbuf *); +uint32_t pktbuf_save(struct pktbuf *, struct ofpbuf *buffer, uint16_t in_port); ++uint32_t pktbuf_get_null(void); +int pktbuf_retrieve(struct pktbuf *, uint32_t id, struct ofpbuf **bufferp, + uint16_t *in_port); +void pktbuf_discard(struct pktbuf *, uint32_t id); + +#endif /* pktbuf.h */