X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcore%2Fexecute.c;h=4b0767aed3151946f63653bc973503fe744e0d75;hp=ce8b9bcb8b9b212937471b2b1ace5d8eaf44b44e;hb=8fa6cbe1a9f3f3353224116772619a4d3eb26f88;hpb=417116f23432073162ebfcb286a7800846482eed diff --git a/src/core/execute.c b/src/core/execute.c index ce8b9bcb8..4b0767aed 100644 --- a/src/core/execute.c +++ b/src/core/execute.c @@ -83,6 +83,8 @@ #include "af-list.h" #include "mkdir.h" #include "apparmor-util.h" +#include "bus-kernel.h" +#include "label.h" #ifdef HAVE_SECCOMP #include "seccomp-util.h" @@ -333,7 +335,7 @@ static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty i == EXEC_INPUT_TTY_FAIL, i == EXEC_INPUT_TTY_FORCE, false, - (usec_t) -1); + USEC_INFINITY); if (fd < 0) return fd; @@ -561,7 +563,7 @@ static int restore_confirm_stdio(int *saved_stdin, static int ask_for_confirmation(char *response, char **argv) { int saved_stdout = -1, saved_stdin = -1, r; - char *line; + _cleanup_free_ char *line = NULL; r = setup_confirm_stdio(&saved_stdin, &saved_stdout); if (r < 0) @@ -571,8 +573,7 @@ static int ask_for_confirmation(char *response, char **argv) { if (!line) return -ENOMEM; - r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line); - free(line); + r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line); restore_confirm_stdio(&saved_stdin, &saved_stdout); @@ -939,7 +940,7 @@ static void rename_process_from_path(const char *path) { #ifdef HAVE_SECCOMP -static int apply_seccomp(ExecContext *c) { +static int apply_seccomp(const ExecContext *c) { uint32_t negative_action, action; scmp_filter_ctx *seccomp; Iterator i; @@ -988,7 +989,7 @@ finish: return r; } -static int apply_address_families(ExecContext *c) { +static int apply_address_families(const ExecContext *c) { scmp_filter_ctx *seccomp; Iterator i; int r; @@ -1139,7 +1140,7 @@ static void do_idle_pipe_dance(int idle_pipe[4]) { } static int build_environment( - ExecContext *c, + const ExecContext *c, unsigned n_fds, usec_t watchdog_usec, const char *home, @@ -1224,572 +1225,605 @@ static int build_environment( return 0; } -int exec_spawn(ExecCommand *command, - char **argv, - ExecContext *context, - int fds[], unsigned n_fds, - char **environment, - bool apply_permissions, - bool apply_chroot, - bool apply_tty_stdin, - bool confirm_spawn, - CGroupControllerMask cgroup_supported, - const char *cgroup_path, - const char *runtime_prefix, - const char *unit_id, - usec_t watchdog_usec, - int idle_pipe[4], - ExecRuntime *runtime, - pid_t *ret) { - - _cleanup_strv_free_ char **files_env = NULL; - int socket_fd; - char *line; - pid_t pid; - int r; +static int exec_child(ExecCommand *command, + const ExecContext *context, + const ExecParameters *params, + ExecRuntime *runtime, + char **argv, + int socket_fd, + int *fds, unsigned n_fds, + char **files_env, + int *error) { + + _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL; + const char *username = NULL, *home = NULL, *shell = NULL; + unsigned n_dont_close = 0; + int dont_close[n_fds + 4]; + uid_t uid = (uid_t) -1; + gid_t gid = (gid_t) -1; + int i, err; assert(command); assert(context); - assert(ret); - assert(fds || n_fds <= 0); + assert(params); + assert(error); - if (context->std_input == EXEC_INPUT_SOCKET || - context->std_output == EXEC_OUTPUT_SOCKET || - context->std_error == EXEC_OUTPUT_SOCKET) { + rename_process_from_path(command->path); - if (n_fds != 1) - return -EINVAL; + /* We reset exactly these signals, since they are the + * only ones we set to SIG_IGN in the main daemon. All + * others we leave untouched because we set them to + * SIG_DFL or a valid handler initially, both of which + * will be demoted to SIG_DFL. */ + default_signals(SIGNALS_CRASH_HANDLER, + SIGNALS_IGNORE, -1); - socket_fd = fds[0]; + if (context->ignore_sigpipe) + ignore_signals(SIGPIPE, -1); - fds = NULL; - n_fds = 0; - } else - socket_fd = -1; - - r = exec_context_load_environment(context, &files_env); - if (r < 0) { - log_struct_unit(LOG_ERR, - unit_id, - "MESSAGE=Failed to load environment files: %s", strerror(-r), - "ERRNO=%d", -r, - NULL); - return r; + err = reset_signal_mask(); + if (err < 0) { + *error = EXIT_SIGNAL_MASK; + return err; } - if (!argv) - argv = command->argv; + if (params->idle_pipe) + do_idle_pipe_dance(params->idle_pipe); - line = exec_command_line(argv); - if (!line) - return log_oom(); + /* Close sockets very early to make sure we don't + * block init reexecution because it cannot bind its + * sockets */ + log_forget_fds(); - log_struct_unit(LOG_DEBUG, - unit_id, - "EXECUTABLE=%s", command->path, - "MESSAGE=About to execute: %s", line, - NULL); - free(line); + if (socket_fd >= 0) + dont_close[n_dont_close++] = socket_fd; + if (n_fds > 0) { + memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds); + n_dont_close += n_fds; + } + if (params->bus_endpoint_fd >= 0) + dont_close[n_dont_close++] = params->bus_endpoint_fd; + if (runtime) { + if (runtime->netns_storage_socket[0] >= 0) + dont_close[n_dont_close++] = runtime->netns_storage_socket[0]; + if (runtime->netns_storage_socket[1] >= 0) + dont_close[n_dont_close++] = runtime->netns_storage_socket[1]; + } - pid = fork(); - if (pid < 0) - return -errno; + err = close_all_fds(dont_close, n_dont_close); + if (err < 0) { + *error = EXIT_FDS; + return err; + } - if (pid == 0) { - _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL; - const char *username = NULL, *home = NULL, *shell = NULL; - unsigned n_dont_close = 0; - int dont_close[n_fds + 3]; - uid_t uid = (uid_t) -1; - gid_t gid = (gid_t) -1; - sigset_t ss; - int i, err; - - /* child */ - - rename_process_from_path(command->path); - - /* We reset exactly these signals, since they are the - * only ones we set to SIG_IGN in the main daemon. All - * others we leave untouched because we set them to - * SIG_DFL or a valid handler initially, both of which - * will be demoted to SIG_DFL. */ - default_signals(SIGNALS_CRASH_HANDLER, - SIGNALS_IGNORE, -1); - - if (context->ignore_sigpipe) - ignore_signals(SIGPIPE, -1); - - assert_se(sigemptyset(&ss) == 0); - if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) { - err = -errno; - r = EXIT_SIGNAL_MASK; - goto fail_child; + if (!context->same_pgrp) + if (setsid() < 0) { + *error = EXIT_SETSID; + return -errno; } - if (idle_pipe) - do_idle_pipe_dance(idle_pipe); + exec_context_tty_reset(context); + + if (params->confirm_spawn) { + char response; + + err = ask_for_confirmation(&response, argv); + if (err == -ETIMEDOUT) + write_confirm_message("Confirmation question timed out, assuming positive response.\n"); + else if (err < 0) + write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err)); + else if (response == 's') { + write_confirm_message("Skipping execution.\n"); + *error = EXIT_CONFIRM; + return -ECANCELED; + } else if (response == 'n') { + write_confirm_message("Failing execution.\n"); + *error = 0; + return 0; + } + } - /* Close sockets very early to make sure we don't - * block init reexecution because it cannot bind its - * sockets */ - log_forget_fds(); + /* If a socket is connected to STDIN/STDOUT/STDERR, we + * must sure to drop O_NONBLOCK */ + if (socket_fd >= 0) + fd_nonblock(socket_fd, false); - if (socket_fd >= 0) - dont_close[n_dont_close++] = socket_fd; - if (n_fds > 0) { - memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds); - n_dont_close += n_fds; - } - if (runtime) { - if (runtime->netns_storage_socket[0] >= 0) - dont_close[n_dont_close++] = runtime->netns_storage_socket[0]; - if (runtime->netns_storage_socket[1] >= 0) - dont_close[n_dont_close++] = runtime->netns_storage_socket[1]; - } + err = setup_input(context, socket_fd, params->apply_tty_stdin); + if (err < 0) { + *error = EXIT_STDIN; + return err; + } + + err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin); + if (err < 0) { + *error = EXIT_STDOUT; + return err; + } + + err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin); + if (err < 0) { + *error = EXIT_STDERR; + return err; + } - err = close_all_fds(dont_close, n_dont_close); + if (params->cgroup_path) { + err = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0); if (err < 0) { - r = EXIT_FDS; - goto fail_child; + *error = EXIT_CGROUP; + return err; } + } - if (!context->same_pgrp) - if (setsid() < 0) { - err = -errno; - r = EXIT_SETSID; - goto fail_child; - } + if (context->oom_score_adjust_set) { + char t[16]; - exec_context_tty_reset(context); + snprintf(t, sizeof(t), "%i", context->oom_score_adjust); + char_array_0(t); - if (confirm_spawn) { - char response; - - err = ask_for_confirmation(&response, argv); - if (err == -ETIMEDOUT) - write_confirm_message("Confirmation question timed out, assuming positive response.\n"); - else if (err < 0) - write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err)); - else if (response == 's') { - write_confirm_message("Skipping execution.\n"); - err = -ECANCELED; - r = EXIT_CONFIRM; - goto fail_child; - } else if (response == 'n') { - write_confirm_message("Failing execution.\n"); - err = r = 0; - goto fail_child; - } + if (write_string_file("/proc/self/oom_score_adj", t) < 0) { + *error = EXIT_OOM_ADJUST; + return -errno; } + } - /* If a socket is connected to STDIN/STDOUT/STDERR, we - * must sure to drop O_NONBLOCK */ - if (socket_fd >= 0) - fd_nonblock(socket_fd, false); + if (context->nice_set) + if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) { + *error = EXIT_NICE; + return -errno; + } - err = setup_input(context, socket_fd, apply_tty_stdin); + if (context->cpu_sched_set) { + struct sched_param param = { + .sched_priority = context->cpu_sched_priority, + }; + + err = sched_setscheduler(0, + context->cpu_sched_policy | + (context->cpu_sched_reset_on_fork ? + SCHED_RESET_ON_FORK : 0), + ¶m); if (err < 0) { - r = EXIT_STDIN; - goto fail_child; + *error = EXIT_SETSCHEDULER; + return -errno; } + } - err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin); - if (err < 0) { - r = EXIT_STDOUT; - goto fail_child; + if (context->cpuset) + if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) { + *error = EXIT_CPUAFFINITY; + return -errno; } - err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin); - if (err < 0) { - r = EXIT_STDERR; - goto fail_child; + if (context->ioprio_set) + if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) { + *error = EXIT_IOPRIO; + return -errno; } - if (cgroup_path) { - err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } + if (context->timer_slack_nsec != NSEC_INFINITY) + if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) { + *error = EXIT_TIMERSLACK; + return -errno; } - if (context->oom_score_adjust_set) { - char t[16]; + if (context->personality != 0xffffffffUL) + if (personality(context->personality) < 0) { + *error = EXIT_PERSONALITY; + return -errno; + } - snprintf(t, sizeof(t), "%i", context->oom_score_adjust); - char_array_0(t); + if (context->utmp_id) + utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path); - if (write_string_file("/proc/self/oom_score_adj", t) < 0) { - err = -errno; - r = EXIT_OOM_ADJUST; - goto fail_child; - } + if (context->user) { + username = context->user; + err = get_user_creds(&username, &uid, &gid, &home, &shell); + if (err < 0) { + *error = EXIT_USER; + return err; } - if (context->nice_set) - if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) { - err = -errno; - r = EXIT_NICE; - goto fail_child; + if (is_terminal_input(context->std_input)) { + err = chown_terminal(STDIN_FILENO, uid); + if (err < 0) { + *error = EXIT_STDIN; + return err; } + } + } - if (context->cpu_sched_set) { - struct sched_param param = { - .sched_priority = context->cpu_sched_priority, - }; +#ifdef ENABLE_KDBUS + if (params->bus_endpoint_fd >= 0 && context->bus_endpoint) { + uid_t ep_uid = (uid == (uid_t) -1) ? 0 : uid; - r = sched_setscheduler(0, - context->cpu_sched_policy | - (context->cpu_sched_reset_on_fork ? - SCHED_RESET_ON_FORK : 0), - ¶m); - if (r < 0) { - err = -errno; - r = EXIT_SETSCHEDULER; - goto fail_child; - } + err = bus_kernel_set_endpoint_policy(params->bus_endpoint_fd, ep_uid, context->bus_endpoint); + if (err < 0) { + *error = EXIT_BUS_ENDPOINT; + return err; } + } +#endif - if (context->cpuset) - if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) { - err = -errno; - r = EXIT_CPUAFFINITY; - goto fail_child; - } +#ifdef HAVE_PAM + if (params->cgroup_path && context->user && context->pam_name) { + err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid); + if (err < 0) { + *error = EXIT_CGROUP; + return err; + } - if (context->ioprio_set) - if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) { - err = -errno; - r = EXIT_IOPRIO; - goto fail_child; - } - if (context->timer_slack_nsec != (nsec_t) -1) - if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) { - err = -errno; - r = EXIT_TIMERSLACK; - goto fail_child; - } + err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid); + if (err < 0) { + *error = EXIT_CGROUP; + return err; + } + } +#endif - if (context->personality != 0xffffffffUL) - if (personality(context->personality) < 0) { - err = -errno; - r = EXIT_PERSONALITY; - goto fail_child; - } + if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) { + char **rt; - if (context->utmp_id) - utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path); + STRV_FOREACH(rt, context->runtime_directory) { + _cleanup_free_ char *p; - if (context->user) { - username = context->user; - err = get_user_creds(&username, &uid, &gid, &home, &shell); - if (err < 0) { - r = EXIT_USER; - goto fail_child; + p = strjoin(params->runtime_prefix, "/", *rt, NULL); + if (!p) { + *error = EXIT_RUNTIME_DIRECTORY; + return -ENOMEM; } - if (is_terminal_input(context->std_input)) { - err = chown_terminal(STDIN_FILENO, uid); - if (err < 0) { - r = EXIT_STDIN; - goto fail_child; - } + err = mkdir_safe(p, context->runtime_directory_mode, uid, gid); + if (err < 0) { + *error = EXIT_RUNTIME_DIRECTORY; + return err; } } + } -#ifdef HAVE_PAM - if (cgroup_path && context->user && context->pam_name) { - err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } + if (params->apply_permissions) { + err = enforce_groups(context, username, gid); + if (err < 0) { + *error = EXIT_GROUP; + return err; + } + } + umask(context->umask); - err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } +#ifdef HAVE_PAM + if (params->apply_permissions && context->pam_name && username) { + err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds); + if (err < 0) { + *error = EXIT_PAM; + return err; } + } #endif - if (!strv_isempty(context->runtime_directory) && runtime_prefix) { - char **rt; + if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) { + err = setup_netns(runtime->netns_storage_socket); + if (err < 0) { + *error = EXIT_NETWORK; + return err; + } + } - STRV_FOREACH(rt, context->runtime_directory) { - _cleanup_free_ char *p; + if (!strv_isempty(context->read_write_dirs) || + !strv_isempty(context->read_only_dirs) || + !strv_isempty(context->inaccessible_dirs) || + context->mount_flags != 0 || + (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) || + params->bus_endpoint_path || + context->private_devices || + context->protect_system != PROTECT_SYSTEM_NO || + context->protect_home != PROTECT_HOME_NO) { + + char *tmp = NULL, *var = NULL; + + /* The runtime struct only contains the parent + * of the private /tmp, which is + * non-accessible to world users. Inside of it + * there's a /tmp that is sticky, and that's + * the one we want to use here. */ + + if (context->private_tmp && runtime) { + if (runtime->tmp_dir) + tmp = strappenda(runtime->tmp_dir, "/tmp"); + if (runtime->var_tmp_dir) + var = strappenda(runtime->var_tmp_dir, "/tmp"); + } - p = strjoin(runtime_prefix, "/", *rt, NULL); - if (!p) { - r = EXIT_RUNTIME_DIRECTORY; - err = -ENOMEM; - goto fail_child; - } + err = setup_namespace( + context->read_write_dirs, + context->read_only_dirs, + context->inaccessible_dirs, + tmp, + var, + params->bus_endpoint_path, + context->private_devices, + context->protect_home, + context->protect_system, + context->mount_flags); + + if (err == -EPERM) + log_warning_unit(params->unit_id, "Failed to set up file system namespace due to lack of privileges. Execution sandbox will not be in effect: %s", strerror(-err)); + else if (err < 0) { + *error = EXIT_NAMESPACE; + return err; + } + } - err = mkdir_safe(p, context->runtime_directory_mode, uid, gid); - if (err < 0) { - r = EXIT_RUNTIME_DIRECTORY; - goto fail_child; - } + if (params->apply_chroot) { + if (context->root_directory) + if (chroot(context->root_directory) < 0) { + *error = EXIT_CHROOT; + return -errno; } + + if (chdir(context->working_directory ? context->working_directory : "/") < 0) { + *error = EXIT_CHDIR; + return -errno; } + } else { + _cleanup_free_ char *d = NULL; - if (apply_permissions) { - err = enforce_groups(context, username, gid); - if (err < 0) { - r = EXIT_GROUP; - goto fail_child; - } + if (asprintf(&d, "%s/%s", + context->root_directory ? context->root_directory : "", + context->working_directory ? context->working_directory : "") < 0) { + *error = EXIT_MEMORY; + return -ENOMEM; } - umask(context->umask); + if (chdir(d) < 0) { + *error = EXIT_CHDIR; + return -errno; + } + } -#ifdef HAVE_PAM - if (apply_permissions && context->pam_name && username) { - err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds); - if (err < 0) { - r = EXIT_PAM; - goto fail_child; + /* We repeat the fd closing here, to make sure that + * nothing is leaked from the PAM modules. Note that + * we are more aggressive this time since socket_fd + * and the netns fds we don't need anymore. The custom + * endpoint fd was needed to upload the policy and can + * now be closed as well. */ + err = close_all_fds(fds, n_fds); + if (err >= 0) + err = shift_fds(fds, n_fds); + if (err >= 0) + err = flags_fds(fds, n_fds, context->non_blocking); + if (err < 0) { + *error = EXIT_FDS; + return err; + } + + if (params->apply_permissions) { + + for (i = 0; i < _RLIMIT_MAX; i++) { + if (!context->rlimit[i]) + continue; + + if (setrlimit_closest(i, context->rlimit[i]) < 0) { + *error = EXIT_LIMITS; + return -errno; } } -#endif - if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) { - err = setup_netns(runtime->netns_storage_socket); + + if (context->capability_bounding_set_drop) { + err = capability_bounding_set_drop(context->capability_bounding_set_drop, false); if (err < 0) { - r = EXIT_NETWORK; - goto fail_child; + *error = EXIT_CAPABILITIES; + return err; } } - if (!strv_isempty(context->read_write_dirs) || - !strv_isempty(context->read_only_dirs) || - !strv_isempty(context->inaccessible_dirs) || - context->mount_flags != 0 || - (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) || - context->private_devices || - context->read_only_system || - context->protected_home != PROTECTED_HOME_NO) { - - char *tmp = NULL, *var = NULL; - - /* The runtime struct only contains the parent - * of the private /tmp, which is - * non-accessible to world users. Inside of it - * there's a /tmp that is sticky, and that's - * the one we want to use here. */ - - if (context->private_tmp && runtime) { - if (runtime->tmp_dir) - tmp = strappenda(runtime->tmp_dir, "/tmp"); - if (runtime->var_tmp_dir) - var = strappenda(runtime->var_tmp_dir, "/tmp"); - } - - err = setup_namespace( - context->read_write_dirs, - context->read_only_dirs, - context->inaccessible_dirs, - tmp, - var, - context->private_devices, - context->protected_home, - context->read_only_system, - context->mount_flags); + if (context->user) { + err = enforce_user(context, uid); if (err < 0) { - r = EXIT_NAMESPACE; - goto fail_child; + *error = EXIT_USER; + return err; } } - if (apply_chroot) { - if (context->root_directory) - if (chroot(context->root_directory) < 0) { - err = -errno; - r = EXIT_CHROOT; - goto fail_child; - } + /* PR_GET_SECUREBITS is not privileged, while + * PR_SET_SECUREBITS is. So to suppress + * potential EPERMs we'll try not to call + * PR_SET_SECUREBITS unless necessary. */ + if (prctl(PR_GET_SECUREBITS) != context->secure_bits) + if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) { + *error = EXIT_SECUREBITS; + return -errno; + } - if (chdir(context->working_directory ? context->working_directory : "/") < 0) { - err = -errno; - r = EXIT_CHDIR; - goto fail_child; + if (context->capabilities) + if (cap_set_proc(context->capabilities) < 0) { + *error = EXIT_CAPABILITIES; + return -errno; } - } else { - _cleanup_free_ char *d = NULL; - - if (asprintf(&d, "%s/%s", - context->root_directory ? context->root_directory : "", - context->working_directory ? context->working_directory : "") < 0) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; + + if (context->no_new_privileges) + if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { + *error = EXIT_NO_NEW_PRIVILEGES; + return -errno; } - if (chdir(d) < 0) { - err = -errno; - r = EXIT_CHDIR; - goto fail_child; +#ifdef HAVE_SECCOMP + if (context->address_families_whitelist || + !set_isempty(context->address_families)) { + err = apply_address_families(context); + if (err < 0) { + *error = EXIT_ADDRESS_FAMILIES; + return err; } } - /* We repeat the fd closing here, to make sure that - * nothing is leaked from the PAM modules */ - err = close_all_fds(fds, n_fds); - if (err >= 0) - err = shift_fds(fds, n_fds); - if (err >= 0) - err = flags_fds(fds, n_fds, context->non_blocking); - if (err < 0) { - r = EXIT_FDS; - goto fail_child; + if (context->syscall_whitelist || + !set_isempty(context->syscall_filter) || + !set_isempty(context->syscall_archs)) { + err = apply_seccomp(context); + if (err < 0) { + *error = EXIT_SECCOMP; + return err; + } } +#endif - if (apply_permissions) { - - for (i = 0; i < _RLIMIT_MAX; i++) { - if (!context->rlimit[i]) - continue; - - if (setrlimit_closest(i, context->rlimit[i]) < 0) { - err = -errno; - r = EXIT_LIMITS; - goto fail_child; +#ifdef HAVE_SELINUX + if (use_selinux()) { + if (context->selinux_context) { + err = setexeccon(context->selinux_context); + if (err < 0 && !context->selinux_context_ignore) { + *error = EXIT_SELINUX_CONTEXT; + return err; } } - if (context->capability_bounding_set_drop) { - err = capability_bounding_set_drop(context->capability_bounding_set_drop, false); + if (params->selinux_context_net && socket_fd >= 0) { + _cleanup_free_ char *label = NULL; + + err = label_get_child_mls_label(socket_fd, command->path, &label); if (err < 0) { - r = EXIT_CAPABILITIES; - goto fail_child; + *error = EXIT_SELINUX_CONTEXT; + return err; } - } - if (context->user) { - err = enforce_user(context, uid); + err = setexeccon(label); if (err < 0) { - r = EXIT_USER; - goto fail_child; + *error = EXIT_SELINUX_CONTEXT; + return err; } } + } +#endif - /* PR_GET_SECUREBITS is not privileged, while - * PR_SET_SECUREBITS is. So to suppress - * potential EPERMs we'll try not to call - * PR_SET_SECUREBITS unless necessary. */ - if (prctl(PR_GET_SECUREBITS) != context->secure_bits) - if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) { - err = -errno; - r = EXIT_SECUREBITS; - goto fail_child; - } +#ifdef HAVE_APPARMOR + if (context->apparmor_profile && use_apparmor()) { + err = aa_change_onexec(context->apparmor_profile); + if (err < 0 && !context->apparmor_profile_ignore) { + *error = EXIT_APPARMOR_PROFILE; + return -errno; + } + } +#endif + } - if (context->capabilities) - if (cap_set_proc(context->capabilities) < 0) { - err = -errno; - r = EXIT_CAPABILITIES; - goto fail_child; - } + err = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env); + if (err < 0) { + *error = EXIT_MEMORY; + return err; + } - if (context->no_new_privileges) - if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { - err = -errno; - r = EXIT_NO_NEW_PRIVILEGES; - goto fail_child; - } + final_env = strv_env_merge(5, + params->environment, + our_env, + context->environment, + files_env, + pam_env, + NULL); + if (!final_env) { + *error = EXIT_MEMORY; + return -ENOMEM; + } -#ifdef HAVE_SECCOMP - if (context->address_families_whitelist || - !set_isempty(context->address_families)) { - err = apply_address_families(context); - if (err < 0) { - r = EXIT_ADDRESS_FAMILIES; - goto fail_child; - } - } + final_argv = replace_env_argv(argv, final_env); + if (!final_argv) { + *error = EXIT_MEMORY; + return -ENOMEM; + } - if (context->syscall_whitelist || - !set_isempty(context->syscall_filter) || - !set_isempty(context->syscall_archs)) { - err = apply_seccomp(context); - if (err < 0) { - r = EXIT_SECCOMP; - goto fail_child; - } - } -#endif + final_env = strv_env_clean(final_env); -#ifdef HAVE_SELINUX - if (context->selinux_context && use_selinux()) { - err = setexeccon(context->selinux_context); - if (err < 0 && !context->selinux_context_ignore) { - r = EXIT_SELINUX_CONTEXT; - goto fail_child; - } - } -#endif + if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) { + _cleanup_free_ char *line; -#ifdef HAVE_APPARMOR - if (context->apparmor_profile && use_apparmor()) { - err = aa_change_onexec(context->apparmor_profile); - if (err < 0 && !context->apparmor_profile_ignore) { - r = EXIT_APPARMOR_PROFILE; - goto fail_child; - } - } -#endif + line = exec_command_line(final_argv); + if (line) { + log_open(); + log_struct_unit(LOG_DEBUG, + params->unit_id, + "EXECUTABLE=%s", command->path, + "MESSAGE=Executing: %s", line, + NULL); + log_close(); } + } + execve(command->path, final_argv, final_env); + *error = EXIT_EXEC; + return -errno; +} - err = build_environment(context, n_fds, watchdog_usec, home, username, shell, &our_env); - if (r < 0) { - r = EXIT_MEMORY; - goto fail_child; - } +int exec_spawn(ExecCommand *command, + const ExecContext *context, + const ExecParameters *params, + ExecRuntime *runtime, + pid_t *ret) { - final_env = strv_env_merge(5, - environment, - our_env, - context->environment, - files_env, - pam_env, - NULL); - if (!final_env) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + _cleanup_strv_free_ char **files_env = NULL; + int *fds = NULL; unsigned n_fds = 0; + char *line, **argv; + int socket_fd; + pid_t pid; + int err; - final_argv = replace_env_argv(argv, final_env); - if (!final_argv) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + assert(command); + assert(context); + assert(ret); + assert(params); + assert(params->fds || params->n_fds <= 0); - final_env = strv_env_clean(final_env); - - if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) { - line = exec_command_line(final_argv); - if (line) { - log_open(); - log_struct_unit(LOG_DEBUG, - unit_id, - "EXECUTABLE=%s", command->path, - "MESSAGE=Executing: %s", line, - NULL); - log_close(); - free(line); - line = NULL; - } - } - execve(command->path, final_argv, final_env); - err = -errno; - r = EXIT_EXEC; + if (context->std_input == EXEC_INPUT_SOCKET || + context->std_output == EXEC_OUTPUT_SOCKET || + context->std_error == EXEC_OUTPUT_SOCKET) { + + if (params->n_fds != 1) + return -EINVAL; + + socket_fd = params->fds[0]; + } else { + socket_fd = -1; + fds = params->fds; + n_fds = params->n_fds; + } + + err = exec_context_load_environment(context, &files_env); + if (err < 0) { + log_struct_unit(LOG_ERR, + params->unit_id, + "MESSAGE=Failed to load environment files: %s", strerror(-err), + "ERRNO=%d", -err, + NULL); + return err; + } + + argv = params->argv ?: command->argv; + + line = exec_command_line(argv); + if (!line) + return log_oom(); + + log_struct_unit(LOG_DEBUG, + params->unit_id, + "EXECUTABLE=%s", command->path, + "MESSAGE=About to execute: %s", line, + NULL); + free(line); + + pid = fork(); + if (pid < 0) + return -errno; - fail_child: + if (pid == 0) { + int r; + + err = exec_child(command, + context, + params, + runtime, + argv, + socket_fd, + fds, n_fds, + files_env, + &r); if (r != 0) { log_open(); log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED), @@ -1806,7 +1840,7 @@ int exec_spawn(ExecCommand *command, } log_struct_unit(LOG_DEBUG, - unit_id, + params->unit_id, "MESSAGE=Forked %s as "PID_FMT, command->path, pid, NULL); @@ -1816,8 +1850,8 @@ int exec_spawn(ExecCommand *command, * outside of the cgroup) and in the parent (so that we can be * sure that when we kill the cgroup the process will be * killed too). */ - if (cgroup_path) - cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid); + if (params->cgroup_path) + cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid); exec_status_start(&command->exec_status, pid); @@ -1834,7 +1868,7 @@ void exec_context_init(ExecContext *c) { c->syslog_priority = LOG_DAEMON|LOG_INFO; c->syslog_level_prefix = true; c->ignore_sigpipe = true; - c->timer_slack_nsec = (nsec_t) -1; + c->timer_slack_nsec = NSEC_INFINITY; c->personality = 0xffffffffUL; c->runtime_directory_mode = 0755; } @@ -1915,6 +1949,9 @@ void exec_context_done(ExecContext *c) { strv_free(c->runtime_directory); c->runtime_directory = NULL; + + bus_endpoint_free(c->bus_endpoint); + c->bus_endpoint = NULL; } int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) { @@ -2024,7 +2061,7 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { return -EINVAL; } for (n = 0; n < count; n++) { - k = load_env_file(pglob.gl_pathv[n], NULL, &p); + k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p); if (k < 0) { if (ignore) continue; @@ -2058,8 +2095,8 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { } static bool tty_may_match_dev_console(const char *tty) { - char *active = NULL, *console; - bool b; + _cleanup_free_ char *active = NULL; + char *console; if (startswith(tty, "/dev/")) tty += 5; @@ -2074,10 +2111,7 @@ static bool tty_may_match_dev_console(const char *tty) { return true; /* "tty0" means the active VC, so it may be the same sometimes */ - b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty)); - free(active); - - return b; + return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty)); } bool exec_context_may_touch_console(ExecContext *ec) { @@ -2114,8 +2148,8 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { "%sPrivateTmp: %s\n" "%sPrivateNetwork: %s\n" "%sPrivateDevices: %s\n" - "%sProtectedHome: %s\n" - "%sReadOnlySystem: %s\n" + "%sProtectHome: %s\n" + "%sProtectSystem: %s\n" "%sIgnoreSIGPIPE: %s\n", prefix, c->umask, prefix, c->working_directory ? c->working_directory : "/", @@ -2124,8 +2158,8 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { prefix, yes_no(c->private_tmp), prefix, yes_no(c->private_network), prefix, yes_no(c->private_devices), - prefix, protected_home_to_string(c->protected_home), - prefix, yes_no(c->read_only_system), + prefix, protect_home_to_string(c->protect_home), + prefix, protect_system_to_string(c->protect_system), prefix, yes_no(c->ignore_sigpipe)); STRV_FOREACH(e, c->environment) @@ -2181,7 +2215,7 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { fputs("\n", f); } - if (c->timer_slack_nsec != (nsec_t) -1) + if (c->timer_slack_nsec != NSEC_INFINITY) fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec); fprintf(f, @@ -2402,12 +2436,11 @@ void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) { assert(s); assert(f); - if (!prefix) - prefix = ""; - if (s->pid <= 0) return; + prefix = strempty(prefix); + fprintf(f, "%sPID: "PID_FMT"\n", prefix, s->pid); @@ -2467,37 +2500,27 @@ char *exec_command_line(char **argv) { } void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) { - char *p2; + _cleanup_free_ char *cmd = NULL; const char *prefix2; - char *cmd; - assert(c); assert(f); - if (!prefix) - prefix = ""; - p2 = strappend(prefix, "\t"); - prefix2 = p2 ? p2 : prefix; + prefix = strempty(prefix); + prefix2 = strappenda(prefix, "\t"); cmd = exec_command_line(c->argv); - fprintf(f, "%sCommand Line: %s\n", prefix, cmd ? cmd : strerror(ENOMEM)); - free(cmd); - exec_status_dump(&c->exec_status, f, prefix2); - - free(p2); } void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) { assert(f); - if (!prefix) - prefix = ""; + prefix = strempty(prefix); LIST_FOREACH(command, c, c) exec_command_dump(c, f, prefix); @@ -2546,6 +2569,29 @@ int exec_command_set(ExecCommand *c, const char *path, ...) { return 0; } +int exec_command_append(ExecCommand *c, const char *path, ...) { + _cleanup_strv_free_ char **l = NULL; + va_list ap; + int r; + + assert(c); + assert(path); + + va_start(ap, path); + l = strv_new_ap(path, ap); + va_end(ap); + + if (!l) + return -ENOMEM; + + r = strv_extend_strv(&c->argv, l); + if (r < 0) + return r; + + return 0; +} + + static int exec_runtime_allocate(ExecRuntime **rt) { if (*rt)