nbd-proxy.c

/* Copyright 2018 IBM Corp.
 *
 * Author: Jeremy Kerr <jk@ozlabs.org>
 *
 * 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.
 */

#define _GNU_SOURCE

#include "config.h"

#include <dirent.h>
#include <err.h>
#include <errno.h>
#include <fcntl.h>
#include <getopt.h>
#include <json.h>
#include <libudev.h>
#include <limits.h>
#include <signal.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/poll.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/un.h>
#include <sys/wait.h>
#include <unistd.h>

struct config
{
    char* name;
    bool is_default;
    char* nbd_device;
    struct json_object* metadata;
};

struct ctx
{
    int sock;
    int sock_client;
    int signal_pipe[2];
    char* sock_path;
    pid_t nbd_client_pid;
    pid_t state_hook_pid;
    int nbd_timeout;
    dev_t nbd_devno;
    uint8_t* buf;
    size_t bufsize;
    struct config* configs;
    int n_configs;
    struct config* default_config;
    struct config* config;
    struct udev* udev;
    struct udev_monitor* monitor;
};

static const char* conf_path = SYSCONFDIR "/nbd-proxy/config.json";
static const char* state_hook_path = SYSCONFDIR "/nbd-proxy/state";
static const char* sockpath_tmpl = RUNSTATEDIR "/nbd.%d.sock";

static const size_t bufsize = 0x20000;
static const int nbd_timeout_default = 30;

static int open_nbd_socket(struct ctx* ctx)
{
    struct sockaddr_un addr;
    char* path;
    int sd, rc;

    rc = asprintf(&path, sockpath_tmpl, getpid());
    if (rc < 0)
        return -1;

    sd = socket(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0);
    if (sd < 0)
    {
        warn("can't create socket");
        goto err_free;
    }

    rc = fchmod(sd, S_IRUSR | S_IWUSR);
    if (rc)
    {
        warn("can't set permissions on socket");
        goto err_close;
    }

    addr.sun_family = AF_UNIX;
    strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);

    rc = bind(sd, (struct sockaddr*)&addr, sizeof(addr));
    if (rc)
    {
        warn("can't bind to path %s", path);
        goto err_close;
    }

    rc = listen(sd, 1);
    if (rc)
    {
        warn("can't listen on socket %s", path);
        goto err_unlink;
    }

    ctx->sock = sd;
    ctx->sock_path = path;
    return 0;

err_unlink:
    unlink(path);
err_close:
    close(sd);
err_free:
    free(path);
    return -1;
}

static int start_nbd_client(struct ctx* ctx)
{
    pid_t pid;

    pid = fork();
    if (pid < 0)
    {
        warn("can't create client process");
        return -1;
    }

    /* child process: run nbd-client in non-fork mode */
    if (pid == 0)
    {
        char timeout_str[10];
        int fd;

        snprintf(timeout_str, sizeof(timeout_str), "%d", ctx->nbd_timeout);

        fd = open("/dev/null", O_RDWR | O_CLOEXEC);
        if (fd < 0)
            err(EXIT_FAILURE, "can't open /dev/null");

        dup2(fd, STDIN_FILENO);
        dup2(fd, STDOUT_FILENO);
        close(fd);

        execlp("nbd-client", "nbd-client", "-u", ctx->sock_path, "-n", "-L",
               "-t", timeout_str, ctx->config->nbd_device, NULL);
        err(EXIT_FAILURE, "can't start ndb client");
    }

    ctx->nbd_client_pid = pid;
    return 0;
}

static void stop_nbd_client(struct ctx* ctx)
{
    int rc;

    if (!ctx->nbd_client_pid)
        return;

    rc = kill(ctx->nbd_client_pid, SIGTERM);
    if (rc)
        return;

    waitpid(ctx->nbd_client_pid, NULL, 0);
    ctx->nbd_client_pid = 0;
}

static int copy_fd(struct ctx* ctx, int fd_in, int fd_out)
{
#ifdef HAVE_SPLICE
    int rc;

    rc = splice(fd_in, NULL, fd_out, NULL, ctx->bufsize, 0);
    if (rc < 0)
        warn("splice");

    return rc;
#else
    size_t len, pos;
    ssize_t rc;

    for (;;)
    {
        errno = 0;
        rc = read(fd_in, ctx->buf, ctx->bufsize);
        if (rc < 0)
        {
            if (errno == EINTR)
                continue;
            warn("read failure");
            return -1;
        }
        if (rc == 0)
            return 0;
        break;
    }

    len = rc;

    for (pos = 0; pos < len;)
    {
        errno = 0;
        rc = write(fd_out, ctx->buf + pos, len - pos);
        if (rc < 0)
        {
            if (errno == EINTR)
                continue;
            warn("write failure");
            return -1;
        }
        if (rc == 0)
            break;
        pos += rc;
    }

    return pos;
#endif
}

static int signal_pipe_fd = -1;

static void signal_handler(int signal)
{
    int rc;

    rc = write(signal_pipe_fd, &signal, sizeof(signal));

    /* not a lot we can do here but exit... */
    if (rc != sizeof(signal))
        exit(EXIT_FAILURE);
}

static int setup_signals(struct ctx* ctx)
{
    struct sigaction sa;
    int rc;

    rc = pipe2(ctx->signal_pipe, O_CLOEXEC);
    if (rc)
    {
        warn("cant setup signal pipe");
        return -1;
    }

    signal_pipe_fd = ctx->signal_pipe[1];

    memset(&sa, 0, sizeof(sa));
    sa.sa_handler = signal_handler;

    sigaction(SIGINT, &sa, NULL);
    sigaction(SIGTERM, &sa, NULL);
    sigaction(SIGCHLD, &sa, NULL);

    return 0;
}

static void cleanup_signals(struct ctx* ctx)
{
    struct sigaction sa;
    memset(&sa, 0, sizeof(sa));
    sa.sa_handler = SIG_DFL;

    sigaction(SIGINT, &sa, NULL);
    sigaction(SIGTERM, &sa, NULL);
    sigaction(SIGCHLD, &sa, NULL);

    close(ctx->signal_pipe[0]);
    close(ctx->signal_pipe[1]);
}

static void process_sigchld(struct ctx* ctx, bool* exit)
{
    int status;
    pid_t pid;

    for (;;)
    {
        pid = waitpid(-1, &status, WNOHANG);
        if (pid == 0)
            break;

        if (pid == ctx->nbd_client_pid)
        {
            warnx("nbd client stopped (%s: %d); exiting",
                  WIFEXITED(status) ? "rc" : "sig",
                  WIFEXITED(status) ? WEXITSTATUS(status) : WTERMSIG(status));
            ctx->nbd_client_pid = 0;
            *exit = true;
        }
        else if (pid == ctx->state_hook_pid)
        {
            if (!WIFEXITED(status) || WEXITSTATUS(status))
            {
                warnx("state hook failed (%s: %d); exiting",
                      WIFEXITED(status) ? "rc" : "sig",
                      WIFEXITED(status) ? WEXITSTATUS(status)
                                        : WTERMSIG(status));
                *exit = true;
            }
            ctx->state_hook_pid = 0;
        }
    }
}

static int process_signal_pipe(struct ctx* ctx, bool* exit)
{
    int buf, rc;

    rc = read(ctx->signal_pipe[0], &buf, sizeof(buf));
    if (rc != sizeof(buf))
        return -1;

    *exit = false;

    switch (buf)
    {
        case SIGCHLD:
            process_sigchld(ctx, exit);
            break;
        case SIGINT:
        case SIGTERM:
            *exit = true;
            break;
    }

    return 0;
}

static int wait_for_nbd_client(struct ctx* ctx)
{
    struct pollfd pollfds[2];
    int rc;

    pollfds[0].fd = ctx->sock;
    pollfds[0].events = POLLIN;
    pollfds[1].fd = ctx->signal_pipe[0];
    pollfds[1].events = POLLIN;

    for (;;)
    {
        errno = 0;
        rc = poll(pollfds, 2, -1);
        if (rc < 0)
        {
            if (errno == EINTR)
                continue;
            warn("poll failed");
            return -1;
        }

        if (pollfds[0].revents)
        {
            rc = accept4(ctx->sock, NULL, NULL, SOCK_CLOEXEC);
            if (rc < 0)
            {
                warn("can't create connection");
                return -1;
            }
            ctx->sock_client = rc;
            break;
        }

        if (pollfds[1].revents)
        {
            bool exit;
            rc = process_signal_pipe(ctx, &exit);
            if (rc || exit)
                return -1;
        }
    }

    return 0;
}

static int run_state_hook(struct ctx* ctx, const char* action, bool wait)
{
    int status, rc, fd;
    pid_t pid;

    /* if the hook isn't present or executable, that's not necessarily
     * an error condition */
    if (access(state_hook_path, X_OK))
        return 0;

    pid = fork();
    if (pid < 0)
    {
        warn("can't fork to execute hook %s", state_hook_path);
        return -1;
    }

    if (!pid)
    {
        const char* argv0;

        argv0 = strchr(state_hook_path, '/');
        if (!argv0)
            argv0 = state_hook_path;

        fd = open("/dev/null", O_RDWR | O_CLOEXEC);
        if (fd < 0)
            exit(EXIT_FAILURE);

        dup2(fd, STDIN_FILENO);
        dup2(fd, STDOUT_FILENO);
        execl(state_hook_path, argv0, action, ctx->config->name, NULL);
        exit(EXIT_FAILURE);
    }

    if (!wait)
    {
        ctx->state_hook_pid = pid;
        return 0;
    }

    rc = waitpid(pid, &status, 0);
    if (rc < 0)
    {
        warn("wait");
        return -1;
    }

    if (!WIFEXITED(status) || WEXITSTATUS(status) != 0)
    {
        warnx("hook %s failed", state_hook_path);
        return -1;
    }

    return 0;
}

static int udev_init(struct ctx* ctx)
{
    int rc;

    ctx->udev = udev_new();
    if (!ctx->udev)
    {
        warn("can't create udev object");
        return -1;
    }

    ctx->monitor = udev_monitor_new_from_netlink(ctx->udev, "kernel");
    if (!ctx->monitor)
    {
        warn("can't create udev monitor");
        goto out_unref_udev;
    }

    rc = udev_monitor_filter_add_match_subsystem_devtype(ctx->monitor, "block",
                                                         "disk");
    if (rc)
    {
        warn("can't create udev monitor filter");
        goto out_unref_monitor;
    }

    rc = udev_monitor_enable_receiving(ctx->monitor);
    if (rc)
    {
        warn("can't start udev monitor");
        goto out_unref_monitor;
    }

    return 0;

out_unref_monitor:
    udev_monitor_unref(ctx->monitor);
out_unref_udev:
    udev_unref(ctx->udev);
    return -1;
}

static void udev_free(struct ctx* ctx)
{
    udev_monitor_unref(ctx->monitor);
    udev_unref(ctx->udev);
}

/* Check for the change event on our nbd device, signifying that the kernel
 * has finished initialising the block device. Once we see the event, we run
 * the "start" state hook, and close the udev monitor.
 *
 * Returns:
 *   0 if no processing was performed
 *  -1 on state hook error (and the nbd session should be closed)
 */
static int udev_process(struct ctx* ctx)
{
    struct udev_device* dev;
    bool action_is_change;
    dev_t devno;
    int rc;

    dev = udev_monitor_receive_device(ctx->monitor);
    if (!dev)
        return 0;

    devno = udev_device_get_devnum(dev);
    action_is_change = !strcmp(udev_device_get_action(dev), "change");
    udev_device_unref(dev);

    if (devno != ctx->nbd_devno)
        return 0;

    if (!action_is_change)
        return 0;

    udev_monitor_unref(ctx->monitor);
    udev_unref(ctx->udev);
    ctx->monitor = NULL;
    ctx->udev = NULL;

    rc = run_state_hook(ctx, "start", false);

    return rc;
}

static int run_proxy(struct ctx* ctx)
{
    struct pollfd pollfds[4];
    bool exit = false;
    int rc, n_fd;

    /* main proxy: forward data between stdio & socket */
    pollfds[0].fd = ctx->sock_client;
    pollfds[0].events = POLLIN;
    pollfds[1].fd = STDIN_FILENO;
    pollfds[1].events = POLLIN;
    pollfds[2].fd = ctx->signal_pipe[0];
    pollfds[2].events = POLLIN;
    pollfds[3].fd = udev_monitor_get_fd(ctx->monitor);
    pollfds[3].events = POLLIN;

    n_fd = 4;

    for (;;)
    {
        errno = 0;
        rc = poll(pollfds, n_fd, -1);
        if (rc < 0)
        {
            if (errno == EINTR)
                continue;
            warn("poll failed");
            break;
        }

        if (pollfds[0].revents)
        {
            rc = copy_fd(ctx, ctx->sock_client, STDOUT_FILENO);
            if (rc <= 0)
                break;
        }

        if (pollfds[1].revents)
        {
            rc = copy_fd(ctx, STDIN_FILENO, ctx->sock_client);
            if (rc <= 0)
                break;
        }

        if (pollfds[2].revents)
        {
            rc = process_signal_pipe(ctx, &exit);
            if (rc || exit)
                break;
        }

        if (pollfds[3].revents)
        {
            rc = udev_process(ctx);
            if (rc)
                break;

            /* udev_process may have closed the udev connection,
             * in which case we can stop polling on its fd */
            if (!ctx->udev)
            {
                pollfds[3].fd = 0;
                pollfds[3].revents = 0;
                n_fd = 3;
            }
        }
    }

    return rc ? -1 : 0;
}

static void print_metadata(struct ctx* ctx)
{
    struct json_object* md;
    int i;

    md = json_object_new_object();

    for (i = 0; i < ctx->n_configs; i++)
    {
        struct config* config = &ctx->configs[i];
        json_object_object_add(md, config->name,
                               json_object_get(config->metadata));
    }

    puts(json_object_get_string(md));

    json_object_put(md);
}

static void config_free_one(struct config* config)
{
    if (config->metadata)
        json_object_put(config->metadata);
    free(config->nbd_device);
    free(config->name);
}

static int config_parse_one(struct config* config, const char* name,
                            json_object* obj)
{
    struct json_object *tmp, *meta;
    json_bool jrc;

    jrc = json_object_object_get_ex(obj, "nbd-device", &tmp);
    if (!jrc)
    {
        warnx("config %s doesn't specify a nbd-device", name);
        return -1;
    }

    if (!json_object_is_type(tmp, json_type_string))
    {
        warnx("config %s has invalid nbd-device", name);
        return -1;
    }

    config->nbd_device = strdup(json_object_get_string(tmp));
    config->name = strdup(name);

    jrc = json_object_object_get_ex(obj, "default", &tmp);
    config->is_default = jrc && json_object_get_boolean(tmp);

    jrc = json_object_object_get_ex(obj, "metadata", &meta);
    if (jrc && json_object_is_type(meta, json_type_object))
        config->metadata = json_object_get(meta);
    else
        config->metadata = NULL;

    return 0;
}

static void config_free(struct ctx* ctx)
{
    int i;

    for (i = 0; i < ctx->n_configs; i++)
        config_free_one(&ctx->configs[i]);

    free(ctx->configs);
    ctx->n_configs = 0;
}

static int config_init(struct ctx* ctx)
{
    struct json_object *obj, *tmp;
    json_bool jrc;
    int i, rc;

    /* apply defaults */
    ctx->nbd_timeout = nbd_timeout_default;

    obj = json_object_from_file(conf_path);
    if (!obj)
    {
        warnx("can't read configuration from %s\n", conf_path);
        return -1;
    }

    /* global configuration */
    jrc = json_object_object_get_ex(obj, "timeout", &tmp);
    if (jrc)
    {
        errno = 0;
        ctx->nbd_timeout = json_object_get_int(tmp);
        if (ctx->nbd_timeout == 0 && errno)
        {
            warnx("can't parse timeout value");
            goto err_free;
        }
    }

    /* per-config configuration */
    jrc = json_object_object_get_ex(obj, "configurations", &tmp);
    if (!jrc)
    {
        warnx("no configurations specified");
        goto err_free;
    }

    if (!json_object_is_type(tmp, json_type_object))
    {
        warnx("invalid configurations format");
        goto err_free;
    }

    ctx->n_configs = json_object_object_length(tmp);
    ctx->configs = calloc(ctx->n_configs, sizeof(*ctx->configs));

    i = 0;
    json_object_object_foreach(tmp, name, config_json)
    {
        struct config* config = &ctx->configs[i];

        rc = config_parse_one(config, name, config_json);
        if (rc)
            goto err_free;

        if (config->is_default)
        {
            if (ctx->default_config)
            {
                warn("multiple configs flagged as default");
                goto err_free;
            }
            ctx->default_config = config;
        }
        i++;
    }

    json_object_put(obj);

    if (ctx->n_configs == 1)
        ctx->default_config = &ctx->configs[0];

    return 0;

err_free:
    warnx("failed to load config from %s", conf_path);
    json_object_put(obj);
    return -1;
}

static int config_select(struct ctx* ctx, const char* name)
{
    struct config* config;
    struct stat statbuf;
    int i, rc;

    config = NULL;

    if (!name)
    {
        /* no config specified: use the default */
        if (!ctx->default_config)
        {
            warnx("no config specified, and no default");
            return -1;
        }
        config = ctx->default_config;
    }
    else
    {
        /* find a matching config... */
        for (i = 0; i < ctx->n_configs; i++)
        {
            if (!strcmp(ctx->configs[i].name, name))
            {
                config = &ctx->configs[i];
                break;
            }
        }

        if (!config)
        {
            warnx("no such configuration '%s'", name);
            return -1;
        }
    }

    /* check that the device exists */
    rc = stat(config->nbd_device, &statbuf);
    if (rc)
    {
        warn("can't stat nbd device %s", config->nbd_device);
        return -1;
    }

    if (!S_ISBLK(statbuf.st_mode))
    {
        warn("specified nbd path %s isn't a block device", config->nbd_device);
        return -1;
    }

    /* ... and apply it */
    ctx->config = config;
    ctx->nbd_devno = statbuf.st_rdev;

    return 0;
}

static const struct option options[] = {
    {.name = "help", .val = 'h'},
    {.name = "metadata", .val = 'm'},
    {0},
};

enum action
{
    ACTION_PROXY,
    ACTION_METADATA,
};

static void print_usage(const char* progname)
{
    fprintf(stderr, "usage:\n");
    fprintf(stderr, "\t%s [configuration]\n", progname);
    fprintf(stderr, "\t%s --metadata\n", progname);
}

int main(int argc, char** argv)
{
    enum action action = ACTION_PROXY;
    const char* config_name;
    struct ctx _ctx, *ctx;
    int rc;

    config_name = NULL;

    for (;;)
    {
        int c = getopt_long(argc, argv, "h", options, NULL);
        if (c == -1)
            break;

        switch (c)
        {
            case 'm':
                action = ACTION_METADATA;
                break;
            case 'h':
            case '?':
                print_usage(argv[0]);
                return c == 'h' ? EXIT_SUCCESS : EXIT_FAILURE;
        }
    }

    if (optind < argc)
        config_name = argv[optind];

    ctx = &_ctx;
    memset(ctx, 0, sizeof(*ctx));
    ctx->bufsize = bufsize;
    ctx->buf = malloc(ctx->bufsize);

    rc = config_init(ctx);
    if (rc)
        goto out_free;

    if (action == ACTION_METADATA)
    {
        print_metadata(ctx);
        goto out_free;
    }

    rc = config_select(ctx, config_name);
    if (rc)
        goto out_free;

    rc = open_nbd_socket(ctx);
    if (rc)
        goto out_free;

    rc = setup_signals(ctx);
    if (rc)
        goto out_close;

    rc = start_nbd_client(ctx);
    if (rc)
        goto out_stop_client;

    rc = wait_for_nbd_client(ctx);
    if (rc)
        goto out_stop_client;

    rc = udev_init(ctx);
    if (rc)
        goto out_stop_client;

    rc = run_proxy(ctx);

    if (ctx->udev)
        udev_free(ctx);

    run_state_hook(ctx, "stop", true);

out_stop_client:
    /* we cleanup signals before stopping the client, because we
     * no longer care about SIGCHLD from the stopping nbd-client
     * process. stop_nbd_client will be a no-op if the client hasn't
     * been started. */
    cleanup_signals(ctx);

    stop_nbd_client(ctx);
    close(ctx->sock_client);

out_close:
    if (ctx->sock_path)
    {
        unlink(ctx->sock_path);
        free(ctx->sock_path);
    }
    close(ctx->sock);
out_free:
    config_free(ctx);
    free(ctx->buf);
    return rc ? EXIT_FAILURE : EXIT_SUCCESS;
}