Consolidate network devices

[chiangkd]

This PR refactors the existing TAP-based network device implementation without affecting the current functionality. It separates network client handling to support multiple backends, allowing the use of the -netdev flag to specify the client type, defaulting to TAP if not provided. Support for other backends is yet to be implemented.

Additionally, packet transfer functions are now designed to be specific to each network client, making the system more modular and extensible for future development.

Verification of existing functionality (in privileged mode):

1. Open TAP device and run simulator

sudo make check

2. Configure the IP address for tap0 and enable it on the host side:

sudo ip addr add 192.168.10.1/24 dev tap0
sudo ip link set tap0 up

3. Enable eth0 and set its IP address:

ip l set eth0 up
ip a add 192.168.10.2/24 dev eth0

[jserv]

Enhance CI pipeline to automate the network specific tests.

See vwifi for reference.

[jserv]

Can you hide the implementation details as possible?

[chiangkd]

I use an X-Macro to define supported network devices:

#define SUPPORTED_DEVICES   \
        _(tap)              \
        _(user)

Previously, NET_DEV_DRIVER_MAX was used to denote the total number of network device drivers implemented. To simplify the design, this constant has been removed. Instead, the find_net_dev_idx function dynamically determines the range of the lookup table based on its content.

The final NULL ensures the array is properly terminated, allowing for safe iteration.

static const char *netdev_driver_lookup[] = {
#define _(dev) [NET_DEV_DRIVER_##dev] = #dev,
    SUPPORTED_DEVICES
#undef _
        NULL,
};

...

static int find_net_dev_idx(const char *net_type, const char **netlookup)
{
    if (!net_type)
        return -1;
    int i = 0;
    while (netlookup[i]) {
        if (!strcmp(net_type, netlookup[i])) {
            return i;
        }
        i++;
    }
    return -1;
}

[jserv]

Think of the use of opaque pointer.

[chiangkd]

I try to encapsulate the implementation details of netdev_t as follow:

typedef struct {
    char *name;
    net_dev_driver_t type;
    void *device_op;
} netdev_t;

The device_op pointer is allocated and defined during the netdev_init process, which is called by virtio_net_init. Also read/write operations in virtio-net depend on the underlying peer device specified by netdev_t.

static ssize_t handle_write(netdev_t *netdev,
                            virtio_net_queue_t *queue,
                            struct iovec *iovs_cursor,
                            size_t niovs)
{
    ssize_t plen = 0;
#define _(dev) NET_DEV_DRIVER_##dev
    switch (netdev->type) {
    case _(tap):
        net_tap_options_t *tap = (net_tap_options_t *) netdev->device_op;
        plen = writev(tap->tap_fd, iovs_cursor, niovs);
        if (plen < 0 && (errno == EWOULDBLOCK || errno == EAGAIN)) {
            queue->fd_ready = false;
            return -1;
        }
        if (plen < 0) {
            plen = 0;
            fprintf(stderr, "[VNET] could not write packet: %s\n",
                    strerror(errno));
        }
        break;
        ...
    }
#undef _
    return plen;
}

[jserv]

Thank @chiangkd for contributing!