niceSOFT/systemd: The systemd System and Service Manager

src/modules-load/modules-load.c
467 строк15 KB
Lennart Poettering
cpu-set-util: introduce cpus_online().
14 май 2026, 09:12
14 май 2026, 09:1294ef901
/* SPDX-License-Identifier: LGPL-2.1-or-later */

#include <limits.h>
#include <pthread.h>
#include <signal.h>
#include <sys/socket.h>
#include <unistd.h>

#include "alloc-util.h"
#include "assert.h"
#include "build.h"
#include "conf-files.h"
#include "constants.h"
#include "cpu-set-util.h"
#include "errno-util.h"
#include "fd-util.h"
#include "fileio.h"
#include "format-table.h"
#include "log.h"
#include "macro.h"
#include "main-func.h"
#include "module-util.h"
#include "options.h"
#include "ordered-set.h"
#include "parse-util.h"
#include "pretty-print.h"
#include "proc-cmdline.h"
#include "string-util.h"
#include "strv.h"

#define MODULE_NAME_MAX_LEN (4096UL)

static char **arg_proc_cmdline_modules = NULL;
static const char conf_file_dirs[] = CONF_PATHS_NULSTR("modules-load.d");

STATIC_DESTRUCTOR_REGISTER(arg_proc_cmdline_modules, strv_freep);

static int modules_list_append_suffix(OrderedSet **module_set, char *modp) {
        /* take possession of string no matter what */
        _cleanup_free_ char *mod = modp;
        int r;

        assert(module_set);
        assert(mod);

        if (strlen(mod) > MODULE_NAME_MAX_LEN)
                return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Module name max length exceeded (%lu): %s",
                                       MODULE_NAME_MAX_LEN, mod);

        /* kmod will do it anyway later, so replace now dashes with
           underscores to detect duplications due to different spelling. */
        string_replace_char(mod, '-', '_');

        r = ordered_set_ensure_put(module_set, &string_hash_ops_free, mod);
        if (r == -EEXIST)
                return 0;
        if (r < 0)
                return log_error_errno(r, "Failed to enqueue module '%s': %m", mod);

        TAKE_PTR(mod);

        return 0;
}

static int modules_list_append_dup(OrderedSet **module_set, const char *module) {
        _cleanup_free_ char *m = NULL;

        assert(module);

        if (strdup_to(&m, module) < 0)
                return log_oom();

        return modules_list_append_suffix(module_set, TAKE_PTR(m));
}

static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
        int r;

        if (proc_cmdline_key_streq(key, "modules_load")) {

                if (proc_cmdline_value_missing(key, value))
                        return 0;

                r = strv_split_and_extend(&arg_proc_cmdline_modules, value, ",", /* filter_duplicates= */ true);
                if (r < 0)
                        return log_error_errno(r, "Failed to parse modules_load= kernel command line option: %m");
        }

        return 0;
}

static int apply_file(FILE *f, const char *filename, OrderedSet **module_set) {
        int ret = 0, r;

        assert(f);

        log_debug("apply: %s", filename);
        for (;;) {
                _cleanup_free_ char *line = NULL;

                r = read_stripped_line(f, LONG_LINE_MAX, &line);
                if (r < 0)
                        return log_error_errno(r, "Failed to read file '%s': %m", filename);
                if (r == 0)
                        break;

                if (isempty(line))
                        continue;
                if (strchr(COMMENTS, *line))
                        continue;

                RET_GATHER(ret, modules_list_append_suffix(module_set, TAKE_PTR(line)));
        }

        return ret;
}

static int apply_file_from_path(const char *path, OrderedSet **module_set) {
        _cleanup_fclose_ FILE *f = NULL;
        _cleanup_free_ char *pp = NULL;
        int r;

        assert(path);

        r = search_and_fopen_nulstr(path, "re", NULL, conf_file_dirs, &f, &pp);
        if (r < 0)
                return log_error_errno(r, "Failed to open %s: %m", path);

        return apply_file(f, pp, module_set);
}

static int apply_conf_file(ConfFile *c, OrderedSet **module_set) {
        _cleanup_fclose_ FILE *f = NULL;

        f = fopen(FORMAT_PROC_FD_PATH(c->fd), "re");
        if (!f) {
                if (errno == ENOENT)
                        return 0;

                return log_error_errno(errno, "Failed to open %s: %m", c->original_path);
        }

        return apply_file(f, c->original_path, module_set);
}

static int do_direct_probe(OrderedSet *module_set) {
        _cleanup_(kmod_unrefp) struct kmod_ctx *ctx = NULL;
        char *module;
        int ret = 0, r;

        r = module_setup_context(&ctx);
        if (r < 0)
                return log_error_errno(r, "Failed to initialize libkmod context: %m");

        ORDERED_SET_FOREACH(module, module_set) {
                r = module_load_and_warn(ctx, module, /* verbose= */ true);
                if (r != -ENOENT)
                        RET_GATHER(ret, r);
        }

        return ret;
}

static int enqueue_module_to_load(int sock, const char *module) {
        ssize_t bytes;

        assert(sock >= 0);
        assert(module);

        bytes = send(sock, module, strlen(module), /* flags= */ 0);
        if (bytes < 0)
                return log_error_errno(errno, "Failed to send '%s' to thread pool: %m", module);

        return 0;
}

static int dequeue_module_to_load(int sock, char *buffer, size_t buffer_len) {
        ssize_t bytes;

        assert(sock >= 0);
        assert(buffer);

        /* Dequeue one module to be loaded from the socket pair. In case no more
         * modules are present, recv() will return 0. */
        bytes = recv(sock, buffer, buffer_len, /* flags= */ 0);
        if (bytes == 0)
                return 0;
        if (bytes < 0)
                return negative_errno();
        if ((size_t)bytes == buffer_len)
                return -E2BIG;

        buffer[bytes] = '\0';

        return 1;
}

static int run_prober(int sock) {
        _cleanup_(kmod_unrefp) struct kmod_ctx *ctx = NULL;
        char buffer[MODULE_NAME_MAX_LEN + 1];
        int ret = 0, r;

        r = module_setup_context(&ctx);
        if (r < 0)
                return log_error_errno(r, "Failed to initialize libkmod context: %m");

        for (;;) {
                r = dequeue_module_to_load(sock, buffer, sizeof(buffer));
                if (ERRNO_IS_NEG_TRANSIENT(r))
                        continue;
                if (r == -E2BIG) {
                        log_warning_errno(SYNTHETIC_ERRNO(E2BIG), "Dequeued module name too long, proceeding anyway");
                        continue;
                }
                if (r < 0)
                        return log_error_errno(r, "Failed to receive from module queue: %m");
                if (r == 0) {
                        log_debug("No more queued modules, terminate thread");
                        break;
                }

                r = module_load_and_warn(ctx, buffer, /* verbose= */ true);
                if (r != -ENOENT)
                        RET_GATHER(ret, r);
        }

        return ret;
}

static void *prober_thread(void *arg) {
        int sock = PTR_TO_FD(arg);
        return INT_TO_PTR(run_prober(sock));
}

static int create_worker_threads(size_t n_threads, void *arg, pthread_t **ret_threads, size_t *ret_n_threads) {
        _cleanup_free_ pthread_t *new_threads = NULL;
        size_t created_threads;
        sigset_t ss, saved_ss;
        int r;

        assert(ret_threads);
        assert(ret_n_threads);

        if (n_threads == 0) {
                *ret_threads = NULL;
                *ret_n_threads = 0;
                return 0;
        }

        /* Create worker threads with masked signals */
        new_threads = new(pthread_t, n_threads);
        if (!new_threads)
                return log_oom();

        /* No signals in worker threads. */
        assert_se(sigfillset(&ss) >= 0);
        r = pthread_sigmask(SIG_BLOCK, &ss, &saved_ss);
        if (r != 0)
                return log_error_errno(r, "Failed to mask signals for workers: %m");

        for (created_threads = 0; created_threads < n_threads; ++created_threads) {
                r = pthread_create(&new_threads[created_threads], /* attr= */ NULL, prober_thread, arg);
                if (r != 0) {
                        log_error_errno(r, "Failed to create worker thread %zu: %m", created_threads);
                        break;
                }
        }

        /* Restore the signal mask */
        r = pthread_sigmask(SIG_SETMASK, &saved_ss, NULL);
        if (r != 0)
                log_warning_errno(r, "Failed to restore signal mask, ignoring: %m");

        *ret_threads = TAKE_PTR(new_threads);
        *ret_n_threads = created_threads;

        return 0;
}

static int destroy_worker_threads(pthread_t **threads, size_t n_threads) {
        int ret = 0, r;

        assert(threads);
        assert(n_threads == 0 || *threads);

        for (size_t i = 0; i < n_threads; ++i) {
                void *p;
                r = pthread_join((*threads)[i], &p);
                if (r != 0)
                        RET_GATHER(ret, log_warning_errno(r, "Failed to join worker thread, proceeding anyway: %m"));
                else
                        RET_GATHER(ret, PTR_TO_INT(p));
        }

        *threads = mfree(*threads);

        return ret;
}

/* Determine number of workers, either from env or from online CPUs */
static unsigned determine_num_worker_threads(unsigned n_modules) {
        unsigned n_threads = UINT_MAX;
        int r;

        if (n_modules == 0)
                return 0;

        const char *e = secure_getenv("SYSTEMD_MODULES_LOAD_NUM_THREADS");
        if (e) {
                r = safe_atou(e, &n_threads);
                if (r < 0)
                        log_debug("Invalid value in $SYSTEMD_MODULES_LOAD_NUM_THREADS, ignoring: %s", e);
        }

        if (n_threads == UINT_MAX) {
                /* By default, use a number of worker threads equal the number of online CPUs,
                 * but clamp it to avoid a probing storm on machines with many CPUs. */
                unsigned n_cpus;
                r = cpus_online(&n_cpus);
                if (r < 0) {
                        log_warning_errno(r, "Failed to get number of online CPUs, ignoring: %m");
                        n_threads = 1;
                } else
                        n_threads = CLAMP(n_cpus, 1U, 16U);
        }

        /* There's no reason to spawn more threads than the modules that need to be loaded */
        n_threads = CLAMP(n_threads, 1U, n_modules);

        /* One of the probe threads is the main process */
        return n_threads - 1U;
}

static int help(void) {
        _cleanup_free_ char *link = NULL;
        _cleanup_(table_unrefp) Table *options = NULL;
        int r;

        if (terminal_urlify_man("systemd-modules-load.service", "8", &link) < 0)
                return log_oom();

        r = option_parser_get_help_table(&options);
        if (r < 0)
                return r;

        printf("%s [OPTIONS...] [CONFIGURATION FILE...]\n\n"
               "Loads statically configured kernel modules.\n\n",
               program_invocation_short_name);

        r = table_print_or_warn(options);
        if (r < 0)
                return r;

        printf("\nSee the %s for details.\n", link);
        return 0;
}

static int parse_argv(int argc, char *argv[], char ***ret_args) {
        assert(argc >= 0);
        assert(argv);
        assert(ret_args);

        OptionParser opts = { argc, argv };

        FOREACH_OPTION_OR_RETURN(c, &opts)
                switch (c) {

                OPTION_COMMON_HELP:
                        return help();

                OPTION_COMMON_VERSION:
                        return version();
                }

        *ret_args = option_parser_get_args(&opts);
        return 1;
}

static int run(int argc, char *argv[]) {
        /* A OrderedSet instead of a plain Set is used here to make debug easier:
         * in case a race condition is observed during module probing, the threading
         * can be disabled through the SYSTEMD_MODULES_LOAD_NUM_THREADS environment
         * variable and the modules reordered at will by the user that is debugging it.
         * In that case, the probing order would be the same in which the modules
         * appear inside the modules-load.d files (this wouldn't be true with a Set). */
        _cleanup_ordered_set_free_ OrderedSet *module_set = NULL;
        _cleanup_close_pair_ int pair[2] = EBADF_PAIR;
        _cleanup_free_ pthread_t *threads = NULL;
        size_t n_threads = 0;
        char *module;
        int ret = 0, r;

        char **args = NULL;
        r = parse_argv(argc, argv, &args);
        if (r <= 0)
                return r;

        log_setup();

        umask(0022);

        r = proc_cmdline_parse(parse_proc_cmdline_item, /* userdata= */ NULL, PROC_CMDLINE_STRIP_RD_PREFIX);
        if (r < 0)
                log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");

        if (!strv_isempty(args)) {
                STRV_FOREACH(i, args) {
                        r = apply_file_from_path(*i, &module_set);
                        if (r < 0)
                                RET_GATHER(ret, r);
                }
        } else {
                ConfFile **files = NULL;
                size_t n_files = 0;

                CLEANUP_ARRAY(files, n_files, conf_file_free_array);

                STRV_FOREACH(i, arg_proc_cmdline_modules)
                        RET_GATHER(ret, modules_list_append_dup(&module_set, *i));

                r = conf_files_list_nulstr_full(".conf", /* root= */ NULL,
                                                CONF_FILES_REGULAR | CONF_FILES_FILTER_MASKED | CONF_FILES_WARN,
                                                conf_file_dirs, &files, &n_files);
                if (r < 0)
                        RET_GATHER(ret, log_error_errno(r, "Failed to enumerate modules-load.d files: %m"));
                else
                        FOREACH_ARRAY(cf, files, n_files)
                                RET_GATHER(ret, apply_conf_file(*cf, &module_set));
        }

        n_threads = determine_num_worker_threads((size_t) ordered_set_size(module_set));

        /* If no additional thread is required, there is no need to create the
         * thread pool or the mean to communicate with its members. */
        if (n_threads == 0) {
                log_debug("Single-threaded probe");
                return RET_GATHER(ret, do_direct_probe(module_set));
        }

        /* Create a socketpair for communication with probe workers */
        r = RET_NERRNO(socketpair(AF_UNIX, SOCK_SEQPACKET | SOCK_CLOEXEC, /* protocol= */ 0, pair));
        if (r < 0)
                return log_error_errno(r, "Failed to create socket pair: %m");

        /* Create threads, which will then wait for modules to probe. */
        log_info("Using %zu probe threads", n_threads + 1);
        r = create_worker_threads(n_threads, FD_TO_PTR(pair[1]), &threads, &n_threads);
        if (r < 0)
                log_warning_errno(r, "Failed to create probe threads, continuing as single-threaded probe");

        /* Send modules to be probed */
        ORDERED_SET_FOREACH(module, module_set)
                RET_GATHER(ret, enqueue_module_to_load(pair[0], module));

        /* Close one end of the socketpair; workers will run until the queue is empty */
        pair[0] = safe_close(pair[0]);

        /* Run the prober function also in original thread */
        RET_GATHER(ret, run_prober(pair[1]));

        /* Wait for all threads (if any) to finish and gather errors */
        RET_GATHER(ret, destroy_worker_threads(&threads, n_threads));

        return ret;
}

DEFINE_MAIN_FUNCTION(run);