X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=src%2Fcore%2Fexecute.c;h=43b571e043ad0b7231311e748e68b9875e8c353e;hb=31a7eb86;hp=c0e8f9e01302677ea1fd89016ea0885ce530b90d;hpb=af6da548aa14c57da7f17b3a1f2211efdb811d19;p=elogind.git diff --git a/src/core/execute.c b/src/core/execute.c index c0e8f9e01..43b571e04 100644 --- a/src/core/execute.c +++ b/src/core/execute.c @@ -38,6 +38,9 @@ #include #include #include +#include +#include +#include #ifdef HAVE_PAM #include @@ -49,9 +52,9 @@ #include "capability.h" #include "util.h" #include "log.h" +#include "sd-messages.h" #include "ioprio.h" #include "securebits.h" -#include "cgroup.h" #include "namespace.h" #include "tcpwrap.h" #include "exit-status.h" @@ -60,8 +63,13 @@ #include "def.h" #include "loopback-setup.h" #include "path-util.h" +#include "syscall-list.h" +#include "env-util.h" +#include "fileio.h" +#include "unit.h" #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC) +#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC) /* This assumes there is a 'tty' group */ #define TTY_MODE 0620 @@ -137,7 +145,7 @@ static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) { return 0; } -static const char *tty_path(const ExecContext *context) { +_pure_ static const char *tty_path(const ExecContext *context) { assert(context); if (context->tty_path) @@ -159,6 +167,26 @@ void exec_context_tty_reset(const ExecContext *context) { vt_disallocate(context->tty_path); } +static bool is_terminal_output(ExecOutput o) { + return + o == EXEC_OUTPUT_TTY || + o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || + o == EXEC_OUTPUT_KMSG_AND_CONSOLE || + o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE; +} + +void exec_context_serialize(const ExecContext *context, Unit *u, FILE *f) { + assert(context); + assert(u); + assert(f); + + if (context->tmp_dir) + unit_serialize_item(u, f, "tmp-dir", context->tmp_dir); + + if (context->var_tmp_dir) + unit_serialize_item(u, f, "var-tmp-dir", context->var_tmp_dir); +} + static int open_null_as(int flags, int nfd) { int fd, r; @@ -178,7 +206,10 @@ static int open_null_as(int flags, int nfd) { static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) { int fd, r; - union sockaddr_union sa; + union sockaddr_union sa = { + .un.sun_family = AF_UNIX, + .un.sun_path = "/run/systemd/journal/stdout", + }; assert(context); assert(output < _EXEC_OUTPUT_MAX); @@ -189,10 +220,6 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons if (fd < 0) return -errno; - zero(sa); - sa.un.sun_family = AF_UNIX; - strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path)); - r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)); if (r < 0) { close_nointr_nofail(fd); @@ -218,7 +245,7 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons !!context->syslog_level_prefix, output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE, output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE, - output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE || output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE); + is_terminal_output(output)); if (fd != nfd) { r = dup2(fd, nfd) < 0 ? -errno : nfd; @@ -314,9 +341,10 @@ static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty } } -static int setup_output(const ExecContext *context, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) { +static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) { ExecOutput o; ExecInput i; + int r; assert(context); assert(ident); @@ -324,91 +352,55 @@ static int setup_output(const ExecContext *context, int socket_fd, const char *i i = fixup_input(context->std_input, socket_fd, apply_tty_stdin); o = fixup_output(context->std_output, socket_fd); - /* This expects the input is already set up */ + if (fileno == STDERR_FILENO) { + ExecOutput e; + e = fixup_output(context->std_error, socket_fd); - switch (o) { + /* This expects the input and output are already set up */ + + /* Don't change the stderr file descriptor if we inherit all + * the way and are not on a tty */ + if (e == EXEC_OUTPUT_INHERIT && + o == EXEC_OUTPUT_INHERIT && + i == EXEC_INPUT_NULL && + !is_terminal_input(context->std_input) && + getppid () != 1) + return fileno; + + /* Duplicate from stdout if possible */ + if (e == o || e == EXEC_OUTPUT_INHERIT) + return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno; - case EXEC_OUTPUT_INHERIT: + o = e; + } else if (o == EXEC_OUTPUT_INHERIT) { /* If input got downgraded, inherit the original value */ if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input)) - return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO); + return open_terminal_as(tty_path(context), O_WRONLY, fileno); /* If the input is connected to anything that's not a /dev/null, inherit that... */ if (i != EXEC_INPUT_NULL) - return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO; + return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno; /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */ if (getppid() != 1) - return STDOUT_FILENO; - - /* We need to open /dev/null here anew, to get the - * right access mode. So we fall through */ - - case EXEC_OUTPUT_NULL: - return open_null_as(O_WRONLY, STDOUT_FILENO); - - case EXEC_OUTPUT_TTY: - if (is_terminal_input(i)) - return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO; - - /* We don't reset the terminal if this is just about output */ - return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO); + return fileno; - case EXEC_OUTPUT_SYSLOG: - case EXEC_OUTPUT_SYSLOG_AND_CONSOLE: - case EXEC_OUTPUT_KMSG: - case EXEC_OUTPUT_KMSG_AND_CONSOLE: - case EXEC_OUTPUT_JOURNAL: - case EXEC_OUTPUT_JOURNAL_AND_CONSOLE: - return connect_logger_as(context, o, ident, unit_id, STDOUT_FILENO); - - case EXEC_OUTPUT_SOCKET: - assert(socket_fd >= 0); - return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO; - - default: - assert_not_reached("Unknown output type"); + /* We need to open /dev/null here anew, to get the right access mode. */ + return open_null_as(O_WRONLY, fileno); } -} - -static int setup_error(const ExecContext *context, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) { - ExecOutput o, e; - ExecInput i; - - assert(context); - assert(ident); - - i = fixup_input(context->std_input, socket_fd, apply_tty_stdin); - o = fixup_output(context->std_output, socket_fd); - e = fixup_output(context->std_error, socket_fd); - - /* This expects the input and output are already set up */ - - /* Don't change the stderr file descriptor if we inherit all - * the way and are not on a tty */ - if (e == EXEC_OUTPUT_INHERIT && - o == EXEC_OUTPUT_INHERIT && - i == EXEC_INPUT_NULL && - !is_terminal_input(context->std_input) && - getppid () != 1) - return STDERR_FILENO; - /* Duplicate from stdout if possible */ - if (e == o || e == EXEC_OUTPUT_INHERIT) - return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO; - - switch (e) { + switch (o) { case EXEC_OUTPUT_NULL: - return open_null_as(O_WRONLY, STDERR_FILENO); + return open_null_as(O_WRONLY, fileno); case EXEC_OUTPUT_TTY: if (is_terminal_input(i)) - return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO; + return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno; /* We don't reset the terminal if this is just about output */ - return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO); + return open_terminal_as(tty_path(context), O_WRONLY, fileno); case EXEC_OUTPUT_SYSLOG: case EXEC_OUTPUT_SYSLOG_AND_CONSOLE: @@ -416,11 +408,21 @@ static int setup_error(const ExecContext *context, int socket_fd, const char *id case EXEC_OUTPUT_KMSG_AND_CONSOLE: case EXEC_OUTPUT_JOURNAL: case EXEC_OUTPUT_JOURNAL_AND_CONSOLE: - return connect_logger_as(context, e, ident, unit_id, STDERR_FILENO); + r = connect_logger_as(context, o, ident, unit_id, fileno); + if (r < 0) { + log_struct_unit(LOG_CRIT, unit_id, + "MESSAGE=Failed to connect std%s of %s to the journal socket: %s", + fileno == STDOUT_FILENO ? "out" : "err", + unit_id, strerror(-r), + "ERRNO=%d", -r, + NULL); + r = open_null_as(O_WRONLY, fileno); + } + return r; case EXEC_OUTPUT_SOCKET: assert(socket_fd >= 0); - return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO; + return dup2(socket_fd, fileno) < 0 ? -errno : fileno; default: assert_not_reached("Unknown error type"); @@ -508,7 +510,7 @@ fail: return r; } -static int write_confirm_message(const char *format, ...) { +_printf_attr_(1, 2) static int write_confirm_message(const char *format, ...) { int fd; va_list ap; @@ -669,9 +671,9 @@ static int enforce_user(const ExecContext *context, uid_t uid) { /* First step: If we need to keep capabilities but * drop privileges we need to make sure we keep our - * caps, whiel we drop privileges. */ + * caps, while we drop privileges. */ if (uid != 0) { - int sb = context->secure_bits|SECURE_KEEP_CAPS; + int sb = context->secure_bits | 1<> 4] & (1 << (i & 31))) + n++; + + /* Second: build the filter program from a header the syscall + * matches and the footer */ + f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer))); + memcpy(f, header, sizeof(header)); + + for (i = 0, n = 0; i < syscall_max(); i++) + if (syscall_filter[i >> 4] & (1 << (i & 31))) { + struct sock_filter item[] = { + BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1), + BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW) + }; + + assert_cc(ELEMENTSOF(item) == 2); + + f[ELEMENTSOF(header) + 2*n] = item[0]; + f[ELEMENTSOF(header) + 2*n+1] = item[1]; + + n++; + } + + memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer)); + + /* Third: install the filter */ + prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n; + prog.filter = f; + if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0) + return -errno; + + return 0; +} + +static void do_idle_pipe_dance(int idle_pipe[4]) { + assert(idle_pipe); + + if (idle_pipe[1] >= 0) + close_nointr_nofail(idle_pipe[1]); + if (idle_pipe[2] >= 0) + close_nointr_nofail(idle_pipe[2]); + + if (idle_pipe[0] >= 0) { + int r; + + r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC); + + if (idle_pipe[3] >= 0 && r == 0 /* timeout */) { + /* Signal systemd that we are bored and want to continue. */ + write(idle_pipe[3], "x", 1); + + /* Wait for systemd to react to the signal above. */ + fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC); + } + + close_nointr_nofail(idle_pipe[0]); + + } + + if (idle_pipe[3] >= 0) + close_nointr_nofail(idle_pipe[3]); +} + int exec_spawn(ExecCommand *command, char **argv, - const ExecContext *context, + ExecContext *context, int fds[], unsigned n_fds, char **environment, bool apply_permissions, bool apply_chroot, bool apply_tty_stdin, bool confirm_spawn, - CGroupBonding *cgroup_bondings, - CGroupAttribute *cgroup_attributes, - const char *cgroup_suffix, + CGroupControllerMask cgroup_mask, + const char *cgroup_path, const char *unit_id, - int idle_pipe[2], + int idle_pipe[4], pid_t *ret) { + _cleanup_strv_free_ char **files_env = NULL; + int socket_fd; + char *line; pid_t pid; int r; - char *line; - int socket_fd; - char **files_env = NULL; assert(command); assert(context); @@ -965,8 +1047,13 @@ int exec_spawn(ExecCommand *command, } else socket_fd = -1; - if ((r = exec_context_load_environment(context, &files_env)) < 0) { - log_error("Failed to load environment files: %s", strerror(-r)); + 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; } @@ -974,34 +1061,35 @@ int exec_spawn(ExecCommand *command, argv = command->argv; line = exec_command_line(argv); - if (!line) { - r = -ENOMEM; - goto fail_parent; - } + if (!line) + return log_oom(); - log_debug("About to execute: %s", line); + log_struct_unit(LOG_DEBUG, + unit_id, + "EXECUTABLE=%s", command->path, + "MESSAGE=About to execute: %s", line, + NULL); free(line); - r = cgroup_bonding_realize_list(cgroup_bondings); - if (r < 0) - goto fail_parent; - - cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings); - - if ((pid = fork()) < 0) { - r = -errno; - goto fail_parent; + if (context->private_tmp && !context->tmp_dir && !context->var_tmp_dir) { + r = setup_tmpdirs(&context->tmp_dir, &context->var_tmp_dir); + if (r < 0) + return r; } + pid = fork(); + if (pid < 0) + return -errno; + if (pid == 0) { int i, err; sigset_t ss; const char *username = NULL, *home = NULL; uid_t uid = (uid_t) -1; gid_t gid = (gid_t) -1; - char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL; + _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, + **final_env = NULL, **final_argv = NULL; unsigned n_env = 0; - bool set_access = false; /* child */ @@ -1025,14 +1113,8 @@ int exec_spawn(ExecCommand *command, goto fail_child; } - if (idle_pipe) { - if (idle_pipe[1] >= 0) - close_nointr_nofail(idle_pipe[1]); - if (idle_pipe[0] >= 0) { - fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC); - close_nointr_nofail(idle_pipe[0]); - } - } + if (idle_pipe) + do_idle_pipe_dance(idle_pipe); /* Close sockets very early to make sure we don't * block init reexecution because it cannot bind its @@ -1102,20 +1184,20 @@ int exec_spawn(ExecCommand *command, goto fail_child; } - err = setup_output(context, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin); + err = setup_output(context, STDOUT_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin); if (err < 0) { r = EXIT_STDOUT; goto fail_child; } - err = setup_error(context, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin); + err = setup_output(context, STDERR_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin); if (err < 0) { r = EXIT_STDERR; goto fail_child; } - if (cgroup_bondings) { - err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix); + if (cgroup_path) { + err = cg_attach_with_mask(cgroup_mask, cgroup_path, 0); if (err < 0) { r = EXIT_CGROUP; goto fail_child; @@ -1128,7 +1210,7 @@ int exec_spawn(ExecCommand *command, snprintf(t, sizeof(t), "%i", context->oom_score_adjust); char_array_0(t); - if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) { + if (write_string_file("/proc/self/oom_score_adj", t) < 0) { err = -errno; r = EXIT_OOM_ADJUST; goto fail_child; @@ -1143,13 +1225,16 @@ int exec_spawn(ExecCommand *command, } if (context->cpu_sched_set) { - struct sched_param param; - - zero(param); - param.sched_priority = context->cpu_sched_priority; - - if (sched_setscheduler(0, context->cpu_sched_policy | - (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) { + struct sched_param param = { + .sched_priority = context->cpu_sched_priority, + }; + + 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; @@ -1182,7 +1267,7 @@ int exec_spawn(ExecCommand *command, if (context->user) { username = context->user; - err = get_user_creds(&username, &uid, &gid, &home); + err = get_user_creds(&username, &uid, &gid, &home, NULL); if (err < 0) { r = EXIT_USER; goto fail_child; @@ -1195,27 +1280,24 @@ int exec_spawn(ExecCommand *command, goto fail_child; } } + } - if (cgroup_bondings && context->control_group_modify) { - err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid); - if (err >= 0) - err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } - - set_access = true; +#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 (cgroup_bondings && !set_access && context->control_group_persistent >= 0) { - err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent); + + err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid); if (err < 0) { r = EXIT_CGROUP; goto fail_child; } } +#endif if (apply_permissions) { err = enforce_groups(context, username, gid); @@ -1228,7 +1310,7 @@ int exec_spawn(ExecCommand *command, umask(context->umask); #ifdef HAVE_PAM - if (context->pam_name && username) { + 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; @@ -1249,11 +1331,13 @@ int exec_spawn(ExecCommand *command, if (strv_length(context->read_write_dirs) > 0 || strv_length(context->read_only_dirs) > 0 || strv_length(context->inaccessible_dirs) > 0 || - context->mount_flags != MS_SHARED || + context->mount_flags != 0 || context->private_tmp) { err = setup_namespace(context->read_write_dirs, context->read_only_dirs, context->inaccessible_dirs, + context->tmp_dir, + context->var_tmp_dir, context->private_tmp, context->mount_flags); if (err < 0) { @@ -1276,8 +1360,7 @@ int exec_spawn(ExecCommand *command, goto fail_child; } } else { - - char *d; + _cleanup_free_ char *d = NULL; if (asprintf(&d, "%s/%s", context->root_directory ? context->root_directory : "", @@ -1289,12 +1372,9 @@ int exec_spawn(ExecCommand *command, if (chdir(d) < 0) { err = -errno; - free(d); r = EXIT_CHDIR; goto fail_child; } - - free(d); } /* We repeat the fd closing here, to make sure that @@ -1355,9 +1435,25 @@ int exec_spawn(ExecCommand *command, r = EXIT_CAPABILITIES; goto fail_child; } + + 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; + } + + if (context->syscall_filter) { + err = apply_seccomp(context->syscall_filter); + if (err < 0) { + r = EXIT_SECCOMP; + goto fail_child; + } + } } - if (!(our_env = new0(char*, 7))) { + our_env = new0(char*, 7); + if (!our_env) { err = -ENOMEM; r = EXIT_MEMORY; goto fail_child; @@ -1397,20 +1493,21 @@ int exec_spawn(ExecCommand *command, assert(n_env <= 7); - if (!(final_env = strv_env_merge( - 5, - environment, - our_env, - context->environment, - files_env, - pam_env, - NULL))) { + 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; } - if (!(final_argv = replace_env_argv(argv, final_env))) { + final_argv = replace_env_argv(argv, final_env); + if (!final_argv) { err = -ENOMEM; r = EXIT_MEMORY; goto fail_child; @@ -1418,6 +1515,20 @@ int exec_spawn(ExecCommand *command, 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; @@ -1425,41 +1536,37 @@ int exec_spawn(ExecCommand *command, fail_child: if (r != 0) { log_open(); - log_warning("Failed at step %s spawning %s: %s", - exit_status_to_string(r, EXIT_STATUS_SYSTEMD), - command->path, strerror(-err)); + log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED), + "EXECUTABLE=%s", command->path, + "MESSAGE=Failed at step %s spawning %s: %s", + exit_status_to_string(r, EXIT_STATUS_SYSTEMD), + command->path, strerror(-err), + "ERRNO=%d", -err, + NULL); + log_close(); } - strv_free(our_env); - strv_free(final_env); - strv_free(pam_env); - strv_free(files_env); - strv_free(final_argv); - _exit(r); } - strv_free(files_env); + log_struct_unit(LOG_DEBUG, + unit_id, + "MESSAGE=Forked %s as %lu", + command->path, (unsigned long) pid, + NULL); /* We add the new process to the cgroup both in the child (so * that we can be sure that no user code is ever executed * 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_bondings) - cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix); - - log_debug("Forked %s as %lu", command->path, (unsigned long) pid); + if (cgroup_path) + cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid); exec_status_start(&command->exec_status, pid); *ret = pid; return 0; - -fail_parent: - strv_free(files_env); - - return r; } void exec_context_init(ExecContext *c) { @@ -1470,15 +1577,39 @@ void exec_context_init(ExecContext *c) { c->cpu_sched_policy = SCHED_OTHER; c->syslog_priority = LOG_DAEMON|LOG_INFO; c->syslog_level_prefix = true; - c->mount_flags = MS_SHARED; - c->kill_signal = SIGTERM; - c->send_sigkill = true; - c->control_group_persistent = -1; c->ignore_sigpipe = true; c->timer_slack_nsec = (nsec_t) -1; } -void exec_context_done(ExecContext *c) { +void exec_context_tmp_dirs_done(ExecContext *c) { + char* dirs[] = {c->tmp_dir ? c->tmp_dir : c->var_tmp_dir, + c->tmp_dir ? c->var_tmp_dir : NULL, + NULL}; + char **dirp; + + for(dirp = dirs; *dirp; dirp++) { + char *dir; + int r; + + r = rm_rf_dangerous(*dirp, false, true, false); + dir = dirname(*dirp); + if (r < 0) + log_warning("Failed to remove content of temporary directory %s: %s", + dir, strerror(-r)); + else { + r = rmdir(dir); + if (r < 0) + log_warning("Failed to remove temporary directory %s: %s", + dir, strerror(-r)); + } + + free(*dirp); + } + + c->tmp_dir = c->var_tmp_dir = NULL; +} + +void exec_context_done(ExecContext *c, bool reloading_or_reexecuting) { unsigned l; assert(c); @@ -1539,6 +1670,12 @@ void exec_context_done(ExecContext *c) { free(c->utmp_id); c->utmp_id = NULL; + + free(c->syscall_filter); + c->syscall_filter = NULL; + + if (!reloading_or_reexecuting) + exec_context_tmp_dirs_done(c); } void exec_command_done(ExecCommand *c) { @@ -1588,6 +1725,8 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { int k; bool ignore = false; char **p; + _cleanup_globfree_ glob_t pglob = {}; + int count, n; fn = *i; @@ -1597,7 +1736,6 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { } if (!path_is_absolute(fn)) { - if (ignore) continue; @@ -1605,28 +1743,49 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { return -EINVAL; } - if ((k = load_env_file(fn, &p)) < 0) { - + /* Filename supports globbing, take all matching files */ + errno = 0; + if (glob(fn, 0, NULL, &pglob) != 0) { if (ignore) continue; strv_free(r); - return k; + return errno ? -errno : -EINVAL; } + count = pglob.gl_pathc; + if (count == 0) { + if (ignore) + continue; - if (r == NULL) - r = p; - else { - char **m; - - m = strv_env_merge(2, r, p); strv_free(r); - strv_free(p); - - if (!m) - return -ENOMEM; + return -EINVAL; + } + for (n = 0; n < count; n++) { + k = load_env_file(pglob.gl_pathv[n], NULL, &p); + if (k < 0) { + if (ignore) + continue; - r = m; + strv_free(r); + return k; + } + /* Log invalid environment variables with filename */ + if (p) + p = strv_env_clean_log(p, pglob.gl_pathv[n]); + + if (r == NULL) + r = p; + else { + char **m; + + m = strv_env_merge(2, r, p); + strv_free(r); + strv_free(p); + if (!m) + return -ENOMEM; + + r = m; + } } } @@ -1635,6 +1794,37 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { return 0; } +static bool tty_may_match_dev_console(const char *tty) { + char *active = NULL, *console; + bool b; + + if (startswith(tty, "/dev/")) + tty += 5; + + /* trivial identity? */ + if (streq(tty, "console")) + return true; + + console = resolve_dev_console(&active); + /* if we could not resolve, assume it may */ + if (!console) + 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; +} + +bool exec_context_may_touch_console(ExecContext *ec) { + return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate || + is_terminal_input(ec->std_input) || + is_terminal_output(ec->std_output) || + is_terminal_output(ec->std_error)) && + tty_may_match_dev_console(tty_path(ec)); +} + static void strv_fprintf(FILE *f, char **l) { char **g; @@ -1645,14 +1835,13 @@ static void strv_fprintf(FILE *f, char **l) { } void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { - char ** e; + char **e; unsigned i; assert(c); assert(f); - if (!prefix) - prefix = ""; + prefix = strempty(prefix); fprintf(f, "%sUMask: %04o\n" @@ -1660,17 +1849,15 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { "%sRootDirectory: %s\n" "%sNonBlocking: %s\n" "%sPrivateTmp: %s\n" - "%sControlGroupModify: %s\n" - "%sControlGroupPersistent: %s\n" - "%sPrivateNetwork: %s\n", + "%sPrivateNetwork: %s\n" + "%sIgnoreSIGPIPE: %s\n", prefix, c->umask, prefix, c->working_directory ? c->working_directory : "/", prefix, c->root_directory ? c->root_directory : "/", prefix, yes_no(c->non_blocking), prefix, yes_no(c->private_tmp), - prefix, yes_no(c->control_group_modify), - prefix, yes_no(c->control_group_persistent), - prefix, yes_no(c->private_network)); + prefix, yes_no(c->private_network), + prefix, yes_no(c->ignore_sigpipe)); STRV_FOREACH(e, c->environment) fprintf(f, "%sEnvironment: %s\n", prefix, *e); @@ -1697,21 +1884,37 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { if (c->rlimit[i]) fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max); - if (c->ioprio_set) + if (c->ioprio_set) { + char *class_str; + int r; + + r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str); + if (r < 0) + class_str = NULL; fprintf(f, "%sIOSchedulingClass: %s\n" "%sIOPriority: %i\n", - prefix, ioprio_class_to_string(IOPRIO_PRIO_CLASS(c->ioprio)), + prefix, strna(class_str), prefix, (int) IOPRIO_PRIO_DATA(c->ioprio)); + free(class_str); + } - if (c->cpu_sched_set) + if (c->cpu_sched_set) { + char *policy_str; + int r; + + r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str); + if (r < 0) + policy_str = NULL; fprintf(f, "%sCPUSchedulingPolicy: %s\n" "%sCPUSchedulingPriority: %i\n" "%sCPUSchedulingResetOnFork: %s\n", - prefix, sched_policy_to_string(c->cpu_sched_policy), + prefix, strna(policy_str), prefix, c->cpu_sched_priority, prefix, yes_no(c->cpu_sched_reset_on_fork)); + free(policy_str); + } if (c->cpuset) { fprintf(f, "%sCPUAffinity:", prefix); @@ -1746,12 +1949,26 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL || c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE || c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL || - c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) + c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) { + char *fac_str, *lvl_str; + int r; + + r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str); + if (r < 0) + fac_str = NULL; + + r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str); + if (r < 0) + lvl_str = NULL; + fprintf(f, "%sSyslogFacility: %s\n" "%sSyslogLevel: %s\n", - prefix, log_facility_unshifted_to_string(c->syslog_priority >> 3), - prefix, log_level_to_string(LOG_PRI(c->syslog_priority))); + prefix, strna(fac_str), + prefix, strna(lvl_str)); + free(lvl_str); + free(fac_str); + } if (c->capabilities) { char *t; @@ -1765,12 +1982,12 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { if (c->secure_bits) fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n", prefix, - (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "", - (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "", - (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "", - (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "", - (c->secure_bits & SECURE_NOROOT) ? " noroot" : "", - (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : ""); + (c->secure_bits & 1<secure_bits & 1<secure_bits & 1<secure_bits & 1<secure_bits & 1<secure_bits & 1<capability_bounding_set_drop) { unsigned long l; @@ -1821,16 +2038,6 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { fputs("\n", f); } - fprintf(f, - "%sKillMode: %s\n" - "%sKillSignal: SIG%s\n" - "%sSendSIGKILL: %s\n" - "%sIgnoreSIGPIPE: %s\n", - prefix, kill_mode_to_string(c->kill_mode), - prefix, signal_to_string(c->kill_signal), - prefix, yes_no(c->send_sigkill), - prefix, yes_no(c->ignore_sigpipe)); - if (c->utmp_id) fprintf(f, "%sUtmpIdentifier: %s\n", @@ -2038,19 +2245,3 @@ static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = { }; DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput); - -static const char* const kill_mode_table[_KILL_MODE_MAX] = { - [KILL_CONTROL_GROUP] = "control-group", - [KILL_PROCESS] = "process", - [KILL_NONE] = "none" -}; - -DEFINE_STRING_TABLE_LOOKUP(kill_mode, KillMode); - -static const char* const kill_who_table[_KILL_WHO_MAX] = { - [KILL_MAIN] = "main", - [KILL_CONTROL] = "control", - [KILL_ALL] = "all" -}; - -DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);