path: root/app/src/tap.c

#include <common.h>
#include <pcap/pcap.h>
#include <rte_atomic.h>
#include <rte_epoll.h>
#include <rte_ether.h>
#include <rte_malloc.h>
#include <rte_mbuf.h>
#include <sys/epoll.h>
#include <sys/socket.h>
#include <sys/types.h>

#include <assert.h>
#include <fcntl.h>
#include <linux/if.h>
#include <linux/if_tun.h>
#include <marsio.h>
#include <mrapp.h>
#include <mrb_define.h>
#include <netinet/in.h>
#include <sys/ioctl.h>
#include <tap.h>
#include <unistd.h>

static int tap_ioctl(int fd, unsigned long request, struct ifreq * ifr, int set)
{
    short req_flags = ifr->ifr_flags;
    switch (request)
    {
    case SIOCSIFFLAGS:
        if (ioctl(fd, SIOCGIFFLAGS, ifr) < 0)
            goto error;
        if (set)
            ifr->ifr_flags |= req_flags;
        else
            ifr->ifr_flags &= ~req_flags;
        break;

    case SIOCGIFFLAGS:
    case SIOCGIFHWADDR:
    case SIOCSIFHWADDR:
    case SIOCSIFMTU:
        break;
    default:
        return -EINVAL;
    }

    if (ioctl(fd, request, ifr) < 0)
        goto error;

    return 0;

error:
    MR_ERROR("%s: ioctl(%lu) failed with error: %s", ifr->ifr_name, request, strerror(errno));
    return -errno;
}

struct tap_device
{
    struct mr_instance * ref_mr_instance;
    struct mr_vdev * ref_vdev;

    int tap_fd;
    struct rte_epoll_event epoll_event;

    rte_atomic64_t stat_write_pkts;
    rte_atomic64_t stat_write_pktlen;
    rte_atomic64_t stat_write_drops;

    rte_atomic64_t stat_read_pkts;
    rte_atomic64_t stat_read_pktlen;
    rte_atomic64_t stat_read_drops;
};

static int tap_resp_dev_filter(struct vdev * vdev_desc, struct rte_mbuf * mbuf)
{
    struct mrb_metadata * mrb_meta = (struct mrb_metadata *)mrbuf_cz_data(mbuf, MR_NODE_CTRLZONE_ID);
    struct pkt_parser_result * parser_result = &mrb_meta->pkt_parser_result;
    const struct rte_ether_hdr * ether_hdr = rte_pktmbuf_mtod(mbuf, const struct rte_ether_hdr *);

    /* not local's mac addr or broadcast packet, ignore it */
    if (rte_is_broadcast_ether_addr(&ether_hdr->dst_addr) == 0 &&
        rte_is_same_ether_addr(&ether_hdr->dst_addr, &vdev_desc->ether_addr) == 0)
    {
        return 0;
    }

    /* for arp, rarp and lldp, only check the dest's mac address */
    if (ether_hdr->ether_type == htons(RTE_ETHER_TYPE_ARP) && vdev_desc->representor_config.redirect_local_arp > 0)
    {
        return 1;
    }

    if (ether_hdr->ether_type == htons(RTE_ETHER_TYPE_RARP) && vdev_desc->representor_config.redirect_local_rarp > 0)
    {
        return 1;
    }

    if (ether_hdr->ether_type == htons(RTE_ETHER_TYPE_LLDP) && vdev_desc->representor_config.redirect_local_lldp > 0)
    {
        return 1;
    }

    /* allow layers are ETHER->IPv4->TCP, ETHER->IPv4->UDP, the ipv6 is not supported for now. */
    static const uint16_t exp_ipv4_tcp[] = {
        LAYER_TYPE_ID_ETHER,
        LAYER_TYPE_ID_IPV4,
        LAYER_TYPE_ID_TCP,
    };

    static const uint16_t exp_ipv4_udp[] = {
        LAYER_TYPE_ID_ETHER,
        LAYER_TYPE_ID_IPV4,
        LAYER_TYPE_ID_UDP,
    };
    static const uint16_t exp_ipv4_others[] = {
        LAYER_TYPE_ID_ETHER,
        LAYER_TYPE_ID_IPV4,
    };

    if (complex_layer_type_expect(parser_result, exp_ipv4_tcp, RTE_DIM(exp_ipv4_tcp)) == 0 ||
        complex_layer_type_expect(parser_result, exp_ipv4_udp, RTE_DIM(exp_ipv4_udp)) == 0 ||
        complex_layer_type_expect(parser_result, exp_ipv4_others, RTE_DIM(exp_ipv4_others)) == 0)
    {
        return 1;
    }
    else
    {
        return 0;
    }

    assert(false);
}

static int tap_device_tx(struct tap_device * tap_dev, struct rte_mbuf * mbuf)
{
    const char * pkt_ptr = rte_pktmbuf_mtod(mbuf, const char *);
    unsigned int pkt_len = rte_pktmbuf_data_len(mbuf);

    if (pkt_ptr == NULL || pkt_len == 0)
    {
        rte_atomic64_add(&tap_dev->stat_write_drops, 1);
        goto err;
    }

    ssize_t len = write(tap_dev->tap_fd, pkt_ptr, pkt_len);
    if (unlikely(len < 0))
    {
        rte_atomic64_add(&tap_dev->stat_write_drops, 1);
        goto err;
    }

    rte_atomic64_add(&tap_dev->stat_write_pkts, 1);
    rte_atomic64_add(&tap_dev->stat_write_pktlen, pkt_len);

    return 0;

err:
    return -1;
}

static int tap_device_rx(struct tap_device * tap_dev, unsigned int queue_id, marsio_buff_t * buffs[],
                         unsigned int nr_buffs)
{
    char buff[ETH_MAX_MTU];
    ssize_t sz_buff = read(tap_dev->tap_fd, buff, sizeof(buff));
    if (sz_buff == -1 && (errno == EWOULDBLOCK || errno == EAGAIN))
    {
        return 0;
    }
    else if (unlikely(sz_buff < 0))
    {
        return -1;
    }

    int ret = marsio_buff_malloc_global(tap_dev->ref_mr_instance, buffs, 1, MARSIO_SOCKET_ID_ANY, MARSIO_LCORE_ID_ANY);
    if (unlikely(ret < 0))
    {
        return -2;
    }

    struct rte_mbuf * mbuf = (struct rte_mbuf *)buffs[0];
    assert(mbuf != NULL);

    char * mbuf_data_ptr = rte_pktmbuf_append(mbuf, sz_buff);
    if (unlikely(mbuf_data_ptr == NULL))
    {
        marsio_buff_free(tap_dev->ref_mr_instance, buffs, 1, MARSIO_SOCKET_ID_ANY, MARSIO_LCORE_ID_ANY);
        return -3;
    }

    rte_memcpy(mbuf_data_ptr, buff, sz_buff);
    return 1;
}

static struct tap_device * tap_device_create(struct mr_instance * mr_instance, const char * name,
                                             struct rte_ether_addr * hwaddr)
{
    struct tap_device * tap_dev_inst = rte_zmalloc(NULL, sizeof(struct tap_device), 0);
    MR_VERIFY_MALLOC(tap_dev_inst);

    tap_dev_inst->ref_mr_instance = mr_instance;
    int ioctl_sock = -1;
    int tap_fd = -1;

    tap_fd = open("/dev/net/tun", O_RDWR);
    if (tap_fd < 0)
    {
        MR_ERROR("open /dev/net/tun failed: %s", strerror(errno));
        goto errout;
    }

    ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
    if (ioctl_sock < 0)
    {
        MR_ERROR("unable to get a socket for management: %s", strerror(errno));
        goto errout;
    }

    struct ifreq ifr = {.ifr_flags = IFF_TAP | IFF_NO_PI};
    snprintf(ifr.ifr_name, IFNAMSIZ, "%s", name);

    int ret = ioctl(tap_fd, TUNSETIFF, (void *)&ifr);
    if (ret < 0)
    {
        MR_ERROR("tap %s tap_fd(tap_fd = %d) ioctl failed: %s", name, tap_fd, strerror(errno));
        goto errout;
    }

    /* set the tap device in up and running status */
    struct ifreq link_up_ifr = {.ifr_flags = IFF_UP | IFF_RUNNING};
    snprintf(link_up_ifr.ifr_name, IFNAMSIZ, "%s", name);

    ret = tap_ioctl(ioctl_sock, SIOCSIFFLAGS, &link_up_ifr, 1);
    if (ret < 0)
    {
        MR_ERROR("tap device %s link up failed.", name);
        goto errout;
    }

    /* clear the ifr, and use it as mac addr setup */
    ret = tap_ioctl(ioctl_sock, SIOCGIFHWADDR, &ifr, 0);
    if (ret < 0)
    {
        MR_ERROR("tap device %s mac address get failed.", name);
        goto errout;
    }

    ifr.ifr_hwaddr.sa_family = AF_LOCAL;
    rte_memcpy(ifr.ifr_hwaddr.sa_data, hwaddr, RTE_ETHER_ADDR_LEN);
    ret = tap_ioctl(ioctl_sock, SIOCSIFHWADDR, &ifr, 1);
    if (ret < 0)
    {
        MR_ERROR("tap device %s mac address set failed.", name);
        goto errout;
    }

    MR_INFO("tap device %s for created.", name);
    tap_dev_inst->tap_fd = tap_fd;

    close(ioctl_sock);
    ioctl_sock = -1;

    /* first time called, need to create evfd */
    if (mr_instance->tap_resp_epfd == 0)
    {
        mr_instance->tap_resp_epfd = epoll_create1(EPOLL_CLOEXEC);
    }

    /* prepare add the device handle to instance, and join to epoll fd */
    struct rte_epoll_event epoll_event = {
        .epdata.event = EPOLLIN,
        .epdata.data = (void *)tap_dev_inst,
    };

    tap_dev_inst->epoll_event = epoll_event;
    ret = rte_epoll_ctl(mr_instance->tap_resp_epfd, EPOLL_CTL_ADD, tap_fd, &tap_dev_inst->epoll_event);
    if (ret < 0)
    {
        MR_ERROR("failed at add tap_fd %d to epoll_fd for tap representor %s.", tap_fd, name);
        goto errout;
    }

    /* add to the tap resp sets */
    mr_instance->tap_resp_devices[mr_instance->nr_tap_resp_devices] = tap_dev_inst;
    mr_instance->nr_tap_resp_devices++;
    return tap_dev_inst;

errout:
    if (tap_dev_inst != NULL)
    {
        rte_free(tap_dev_inst);
    }

    if (ioctl_sock >= 0)
    {
        close(ioctl_sock);
    }

    if (tap_fd >= 0)
    {
        close(tap_fd);
    }

    return NULL;
}

static void * tap_representor_poll_thread_entry(void * arg)
{
#define TAP_RESP_EVENTS_MAX 8
    struct mr_instance * mr_instance = (struct mr_instance *)arg;
    struct rte_epoll_event epoll_events[TAP_RESP_EVENTS_MAX] = {};
    int ret = 0;

    while (1)
    {
        int n = rte_epoll_wait(mr_instance->tap_resp_epfd, epoll_events, TAP_RESP_EVENTS_MAX, -1);
        if (unlikely(n < 0))
        {
            MR_ERROR("rte_epoll_wait returned error %d, tap_resp poll thread terminated.", errno);
            goto errout;
        }

        /* handle the read event, read the packet then redirect to shmdev queues */
        for (int i = 0; i < n; i++)
        {
            struct tap_device * tap_dev = (struct tap_device *)(epoll_events[i].epdata.data);
            struct mr_vdev * vdev = tap_dev->ref_vdev;

            marsio_buff_t * buff[1];
            ret = tap_device_rx(tap_dev, 0, buff, RTE_DIM(buff));

            if (ret <= 0)
            {
                continue;
            }

            ret = mrapp_packet_fast_send_burst(vdev->vdi, 0, (struct rte_mbuf **)buff, RTE_DIM(buff));
            if (unlikely(ret < 0))
            {
                marsio_buff_free(mr_instance, buff, RTE_DIM(buff), MARSIO_SOCKET_ID_ANY, MARSIO_LCORE_ID_ANY);
            }
        }
    }

errout:
    return (void *)NULL;
}

int tap_representor_entry(struct mr_vdev * vdev, unsigned int qid, struct rte_mbuf * buffs[], unsigned int nr_buffs)
{
    static rte_spinlock_t tap_dev_write_lock = {0};

    for (unsigned int i = 0; i < nr_buffs; i++)
    {
        if (tap_resp_dev_filter(vdev->vdi->vdev, buffs[i]) <= 0)
        {
            continue;
        }

        /* spinlock at here */
        rte_spinlock_lock(&tap_dev_write_lock);
        tap_device_tx(vdev->tap_representor, buffs[i]);
        rte_spinlock_unlock(&tap_dev_write_lock);
    }

    return 0;
}

int tap_representor_init(struct mr_instance * mr_instance, struct mr_vdev * vdev)
{
    struct vdev * _vdev = vdev->vdi->vdev;
    struct tap_device * tap_dev = tap_device_create(mr_instance, vdev->devsym, &_vdev->ether_addr);
    if (unlikely(tap_dev == NULL))
    {
        MR_ERROR("failed at create tap representor for %s, ignore it.", vdev->devsym);
        return -1;
    }

    tap_dev->ref_mr_instance = mr_instance;
    tap_dev->ref_vdev = vdev;
    vdev->tap_representor = tap_dev;

    /* create the tap representor poll thread at first time.
     * this thread is use for recv the packet from the tap, and forward to the rings */
    if (mr_instance->pid_tap_resp_poll == 0)
    {
        int ret = pthread_create(&mr_instance->pid_tap_resp_poll, NULL, tap_representor_poll_thread_entry,
                                 (void *)mr_instance);
        if (unlikely(ret < 0))
        {
            MR_ERROR("failed at creating thread for tap representor poll routine: %s", strerror(errno));
            return -2;
        }
    }

    MR_INFO("tap representor for %s created. ", vdev->devsym);
    return 0;
}