execute.c revision 6bf6e43e7e214a4bd03008a91a7fc77ce6934d65
383N/A/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
383N/A
383N/A/***
383N/A This file is part of systemd.
383N/A
383N/A Copyright 2010 Lennart Poettering
383N/A
383N/A systemd is free software; you can redistribute it and/or modify it
383N/A under the terms of the GNU Lesser General Public License as published by
383N/A the Free Software Foundation; either version 2.1 of the License, or
383N/A (at your option) any later version.
383N/A
383N/A systemd is distributed in the hope that it will be useful, but
383N/A WITHOUT ANY WARRANTY; without even the implied warranty of
383N/A MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
383N/A Lesser General Public License for more details.
383N/A
383N/A You should have received a copy of the GNU Lesser General Public License
383N/A along with systemd; If not, see <http://www.gnu.org/licenses/>.
383N/A***/
383N/A
383N/A#include <errno.h>
383N/A#include <fcntl.h>
383N/A#include <unistd.h>
383N/A#include <string.h>
383N/A#include <signal.h>
383N/A#include <sys/socket.h>
383N/A#include <sys/un.h>
383N/A#include <sys/prctl.h>
383N/A#include <sys/stat.h>
383N/A#include <grp.h>
383N/A#include <poll.h>
383N/A#include <glob.h>
383N/A#include <utmpx.h>
383N/A#include <sys/personality.h>
383N/A
383N/A#ifdef HAVE_PAM
383N/A#include <security/pam_appl.h>
383N/A#endif
383N/A
383N/A#ifdef HAVE_SELINUX
383N/A#include <selinux/selinux.h>
383N/A#endif
383N/A
383N/A#ifdef HAVE_SECCOMP
383N/A#include <seccomp.h>
383N/A#endif
383N/A
383N/A#ifdef HAVE_APPARMOR
383N/A#include <sys/apparmor.h>
383N/A#endif
383N/A
383N/A#include "sd-messages.h"
383N/A#include "rm-rf.h"
383N/A#include "strv.h"
383N/A#include "macro.h"
383N/A#include "capability.h"
383N/A#include "util.h"
383N/A#include "log.h"
383N/A#include "ioprio.h"
383N/A#include "securebits.h"
383N/A#include "namespace.h"
383N/A#include "exit-status.h"
383N/A#include "missing.h"
383N/A#include "utmp-wtmp.h"
383N/A#include "def.h"
383N/A#include "path-util.h"
383N/A#include "env-util.h"
383N/A#include "fileio.h"
383N/A#include "unit.h"
383N/A#include "async.h"
383N/A#include "selinux-util.h"
383N/A#include "errno-list.h"
383N/A#include "af-list.h"
383N/A#include "mkdir.h"
383N/A#include "smack-util.h"
383N/A#include "bus-endpoint.h"
383N/A#include "cap-list.h"
383N/A#include "formats-util.h"
383N/A#include "process-util.h"
383N/A#include "terminal-util.h"
383N/A#include "signal-util.h"
383N/A
383N/A#ifdef HAVE_APPARMOR
383N/A#include "apparmor-util.h"
383N/A#endif
383N/A
383N/A#ifdef HAVE_SECCOMP
383N/A#include "seccomp-util.h"
383N/A#endif
383N/A
383N/A#include "execute.h"
383N/A
383N/A#define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
383N/A#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
383N/A
383N/A/* This assumes there is a 'tty' group */
383N/A#define TTY_MODE 0620
383N/A
383N/A#define SNDBUF_SIZE (8*1024*1024)
383N/A
383N/Astatic int shift_fds(int fds[], unsigned n_fds) {
383N/A int start, restart_from;
383N/A
383N/A if (n_fds <= 0)
383N/A return 0;
383N/A
383N/A /* Modifies the fds array! (sorts it) */
383N/A
383N/A assert(fds);
383N/A
383N/A start = 0;
383N/A for (;;) {
383N/A int i;
383N/A
383N/A restart_from = -1;
383N/A
383N/A for (i = start; i < (int) n_fds; i++) {
383N/A int nfd;
383N/A
383N/A /* Already at right index? */
383N/A if (fds[i] == i+3)
383N/A continue;
383N/A
383N/A nfd = fcntl(fds[i], F_DUPFD, i + 3);
383N/A if (nfd < 0)
383N/A return -errno;
383N/A
383N/A safe_close(fds[i]);
383N/A fds[i] = nfd;
383N/A
383N/A /* Hmm, the fd we wanted isn't free? Then
383N/A * let's remember that and try again from here */
383N/A if (nfd != i+3 && restart_from < 0)
383N/A restart_from = i;
383N/A }
383N/A
383N/A if (restart_from < 0)
383N/A break;
383N/A
383N/A start = restart_from;
383N/A }
383N/A
383N/A return 0;
383N/A}
383N/A
383N/Astatic int flags_fds(const int fds[], unsigned n_fds, bool nonblock) {
383N/A unsigned i;
383N/A int r;
383N/A
383N/A if (n_fds <= 0)
383N/A return 0;
383N/A
383N/A assert(fds);
383N/A
383N/A /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
383N/A
383N/A for (i = 0; i < n_fds; i++) {
383N/A
383N/A r = fd_nonblock(fds[i], nonblock);
383N/A if (r < 0)
383N/A return r;
383N/A
383N/A /* We unconditionally drop FD_CLOEXEC from the fds,
383N/A * since after all we want to pass these fds to our
383N/A * children */
383N/A
383N/A r = fd_cloexec(fds[i], false);
383N/A if (r < 0)
383N/A return r;
383N/A }
383N/A
383N/A return 0;
383N/A}
383N/A
383N/A_pure_ static const char *tty_path(const ExecContext *context) {
383N/A assert(context);
383N/A
383N/A if (context->tty_path)
383N/A return context->tty_path;
383N/A
383N/A return "/dev/console";
383N/A}
383N/A
383N/Astatic void exec_context_tty_reset(const ExecContext *context) {
383N/A assert(context);
383N/A
383N/A if (context->tty_vhangup)
383N/A terminal_vhangup(tty_path(context));
383N/A
383N/A if (context->tty_reset)
383N/A reset_terminal(tty_path(context));
383N/A
383N/A if (context->tty_vt_disallocate && context->tty_path)
383N/A vt_disallocate(context->tty_path);
383N/A}
383N/A
383N/Astatic bool is_terminal_output(ExecOutput o) {
383N/A return
383N/A o == EXEC_OUTPUT_TTY ||
383N/A o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
383N/A o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
383N/A o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
383N/A}
383N/A
383N/Astatic int open_null_as(int flags, int nfd) {
383N/A int fd, r;
383N/A
383N/A assert(nfd >= 0);
383N/A
383N/A fd = open("/dev/null", flags|O_NOCTTY);
383N/A if (fd < 0)
383N/A return -errno;
383N/A
383N/A if (fd != nfd) {
383N/A r = dup2(fd, nfd) < 0 ? -errno : nfd;
383N/A safe_close(fd);
383N/A } else
383N/A r = nfd;
383N/A
383N/A return r;
383N/A}
383N/A
383N/Astatic int connect_journal_socket(int fd, uid_t uid, gid_t gid) {
383N/A union sockaddr_union sa = {
383N/A .un.sun_family = AF_UNIX,
383N/A .un.sun_path = "/run/systemd/journal/stdout",
383N/A };
383N/A uid_t olduid = UID_INVALID;
383N/A gid_t oldgid = GID_INVALID;
383N/A int r;
383N/A
383N/A if (gid != GID_INVALID) {
383N/A oldgid = getgid();
383N/A
383N/A r = setegid(gid);
383N/A if (r < 0)
383N/A return -errno;
383N/A }
383N/A
383N/A if (uid != UID_INVALID) {
383N/A olduid = getuid();
383N/A
383N/A r = seteuid(uid);
383N/A if (r < 0) {
383N/A r = -errno;
383N/A goto restore_gid;
383N/A }
383N/A }
383N/A
383N/A r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
383N/A if (r < 0)
383N/A r = -errno;
383N/A
383N/A /* If we fail to restore the uid or gid, things will likely
383N/A fail later on. This should only happen if an LSM interferes. */
383N/A
383N/A if (uid != UID_INVALID)
383N/A (void) seteuid(olduid);
383N/A
383N/A restore_gid:
383N/A if (gid != GID_INVALID)
383N/A (void) setegid(oldgid);
383N/A
383N/A return r;
383N/A}
383N/A
383N/Astatic int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd, uid_t uid, gid_t gid) {
383N/A int fd, r;
383N/A
383N/A assert(context);
383N/A assert(output < _EXEC_OUTPUT_MAX);
383N/A assert(ident);
383N/A assert(nfd >= 0);
383N/A
383N/A fd = socket(AF_UNIX, SOCK_STREAM, 0);
383N/A if (fd < 0)
383N/A return -errno;
383N/A
383N/A r = connect_journal_socket(fd, uid, gid);
383N/A if (r < 0)
383N/A return r;
383N/A
383N/A if (shutdown(fd, SHUT_RD) < 0) {
383N/A safe_close(fd);
383N/A return -errno;
383N/A }
383N/A
383N/A fd_inc_sndbuf(fd, SNDBUF_SIZE);
383N/A
383N/A dprintf(fd,
383N/A "%s\n"
383N/A "%s\n"
383N/A "%i\n"
383N/A "%i\n"
383N/A "%i\n"
383N/A "%i\n"
383N/A "%i\n",
383N/A context->syslog_identifier ? context->syslog_identifier : ident,
383N/A unit_id,
383N/A context->syslog_priority,
383N/A !!context->syslog_level_prefix,
383N/A output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
383N/A output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
383N/A is_terminal_output(output));
383N/A
383N/A if (fd != nfd) {
383N/A r = dup2(fd, nfd) < 0 ? -errno : nfd;
383N/A safe_close(fd);
383N/A } else
383N/A r = nfd;
383N/A
383N/A return r;
383N/A}
383N/Astatic int open_terminal_as(const char *path, mode_t mode, int nfd) {
383N/A int fd, r;
383N/A
383N/A assert(path);
383N/A assert(nfd >= 0);
383N/A
383N/A fd = open_terminal(path, mode | O_NOCTTY);
383N/A if (fd < 0)
383N/A return fd;
383N/A
383N/A if (fd != nfd) {
383N/A r = dup2(fd, nfd) < 0 ? -errno : nfd;
383N/A safe_close(fd);
383N/A } else
383N/A r = nfd;
383N/A
383N/A return r;
383N/A}
383N/A
383N/Astatic bool is_terminal_input(ExecInput i) {
383N/A return
383N/A i == EXEC_INPUT_TTY ||
383N/A i == EXEC_INPUT_TTY_FORCE ||
383N/A i == EXEC_INPUT_TTY_FAIL;
383N/A}
383N/A
383N/Astatic int fixup_input(ExecInput std_input, int socket_fd, bool apply_tty_stdin) {
383N/A
383N/A if (is_terminal_input(std_input) && !apply_tty_stdin)
383N/A return EXEC_INPUT_NULL;
383N/A
383N/A if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
383N/A return EXEC_INPUT_NULL;
383N/A
383N/A return std_input;
383N/A}
383N/A
383N/Astatic int fixup_output(ExecOutput std_output, int socket_fd) {
383N/A
383N/A if (std_output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
383N/A return EXEC_OUTPUT_INHERIT;
383N/A
383N/A return std_output;
383N/A}
383N/A
383N/Astatic int setup_input(const ExecContext *context, int socket_fd, bool apply_tty_stdin) {
383N/A ExecInput i;
383N/A
383N/A assert(context);
383N/A
383N/A i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
383N/A
383N/A switch (i) {
383N/A
383N/A case EXEC_INPUT_NULL:
383N/A return open_null_as(O_RDONLY, STDIN_FILENO);
383N/A
383N/A case EXEC_INPUT_TTY:
383N/A case EXEC_INPUT_TTY_FORCE:
383N/A case EXEC_INPUT_TTY_FAIL: {
383N/A int fd, r;
383N/A
383N/A fd = acquire_terminal(tty_path(context),
383N/A i == EXEC_INPUT_TTY_FAIL,
383N/A i == EXEC_INPUT_TTY_FORCE,
383N/A false,
383N/A USEC_INFINITY);
383N/A if (fd < 0)
383N/A return fd;
383N/A
383N/A if (fd != STDIN_FILENO) {
383N/A r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
383N/A safe_close(fd);
383N/A } else
383N/A r = STDIN_FILENO;
383N/A
383N/A return r;
383N/A }
383N/A
383N/A case EXEC_INPUT_SOCKET:
383N/A return dup2(socket_fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO;
383N/A
383N/A default:
383N/A assert_not_reached("Unknown input type");
383N/A }
383N/A}
383N/A
383N/Astatic int setup_output(Unit *unit, const ExecContext *context, int fileno, int socket_fd, const char *ident, bool apply_tty_stdin, uid_t uid, gid_t gid) {
383N/A ExecOutput o;
383N/A ExecInput i;
383N/A int r;
383N/A
383N/A assert(unit);
383N/A assert(context);
383N/A assert(ident);
383N/A
383N/A i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
383N/A o = fixup_output(context->std_output, socket_fd);
383N/A
383N/A if (fileno == STDERR_FILENO) {
383N/A ExecOutput e;
383N/A e = fixup_output(context->std_error, socket_fd);
383N/A
383N/A /* This expects the input and output are already set up */
383N/A
383N/A /* Don't change the stderr file descriptor if we inherit all
383N/A * the way and are not on a tty */
383N/A if (e == EXEC_OUTPUT_INHERIT &&
383N/A o == EXEC_OUTPUT_INHERIT &&
383N/A i == EXEC_INPUT_NULL &&
383N/A !is_terminal_input(context->std_input) &&
383N/A getppid () != 1)
383N/A return fileno;
383N/A
383N/A /* Duplicate from stdout if possible */
383N/A if (e == o || e == EXEC_OUTPUT_INHERIT)
383N/A return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
383N/A
383N/A o = e;
383N/A
383N/A } else if (o == EXEC_OUTPUT_INHERIT) {
383N/A /* If input got downgraded, inherit the original value */
383N/A if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
383N/A return open_terminal_as(tty_path(context), O_WRONLY, fileno);
383N/A
383N/A /* If the input is connected to anything that's not a /dev/null, inherit that... */
383N/A if (i != EXEC_INPUT_NULL)
383N/A return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
383N/A
383N/A /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
383N/A if (getppid() != 1)
383N/A return fileno;
383N/A
383N/A /* We need to open /dev/null here anew, to get the right access mode. */
383N/A return open_null_as(O_WRONLY, fileno);
383N/A }
383N/A
383N/A switch (o) {
383N/A
383N/A case EXEC_OUTPUT_NULL:
383N/A return open_null_as(O_WRONLY, fileno);
383N/A
383N/A case EXEC_OUTPUT_TTY:
383N/A if (is_terminal_input(i))
383N/A return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
383N/A
383N/A /* We don't reset the terminal if this is just about output */
383N/A return open_terminal_as(tty_path(context), O_WRONLY, fileno);
383N/A
383N/A case EXEC_OUTPUT_SYSLOG:
383N/A case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
383N/A case EXEC_OUTPUT_KMSG:
383N/A case EXEC_OUTPUT_KMSG_AND_CONSOLE:
383N/A case EXEC_OUTPUT_JOURNAL:
383N/A case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
383N/A r = connect_logger_as(context, o, ident, unit->id, fileno, uid, gid);
383N/A if (r < 0) {
383N/A log_unit_error_errno(unit, r, "Failed to connect %s to the journal socket, ignoring: %m", fileno == STDOUT_FILENO ? "stdout" : "stderr");
383N/A r = open_null_as(O_WRONLY, fileno);
383N/A }
383N/A return r;
383N/A
383N/A case EXEC_OUTPUT_SOCKET:
383N/A assert(socket_fd >= 0);
383N/A return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
383N/A
383N/A default:
383N/A assert_not_reached("Unknown error type");
383N/A }
383N/A}
383N/A
383N/Astatic int chown_terminal(int fd, uid_t uid) {
383N/A struct stat st;
383N/A
383N/A assert(fd >= 0);
383N/A
383N/A /* This might fail. What matters are the results. */
383N/A (void) fchown(fd, uid, -1);
383N/A (void) fchmod(fd, TTY_MODE);
383N/A
383N/A if (fstat(fd, &st) < 0)
383N/A return -errno;
383N/A
383N/A if (st.st_uid != uid || (st.st_mode & 0777) != TTY_MODE)
383N/A return -EPERM;
383N/A
383N/A return 0;
383N/A}
383N/A
383N/Astatic int setup_confirm_stdio(int *_saved_stdin,
383N/A int *_saved_stdout) {
383N/A int fd = -1, saved_stdin, saved_stdout = -1, r;
383N/A
383N/A assert(_saved_stdin);
383N/A assert(_saved_stdout);
383N/A
383N/A saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
383N/A if (saved_stdin < 0)
383N/A return -errno;
383N/A
383N/A saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
383N/A if (saved_stdout < 0) {
383N/A r = errno;
383N/A goto fail;
383N/A }
383N/A
383N/A fd = acquire_terminal(
383N/A "/dev/console",
383N/A false,
383N/A false,
383N/A false,
383N/A DEFAULT_CONFIRM_USEC);
383N/A if (fd < 0) {
383N/A r = fd;
383N/A goto fail;
383N/A }
383N/A
383N/A r = chown_terminal(fd, getuid());
383N/A if (r < 0)
383N/A goto fail;
383N/A
383N/A if (dup2(fd, STDIN_FILENO) < 0) {
383N/A r = -errno;
383N/A goto fail;
383N/A }
383N/A
383N/A if (dup2(fd, STDOUT_FILENO) < 0) {
383N/A r = -errno;
383N/A goto fail;
383N/A }
383N/A
383N/A if (fd >= 2)
383N/A safe_close(fd);
383N/A
383N/A *_saved_stdin = saved_stdin;
383N/A *_saved_stdout = saved_stdout;
383N/A
383N/A return 0;
383N/A
383N/Afail:
383N/A safe_close(saved_stdout);
383N/A safe_close(saved_stdin);
383N/A safe_close(fd);
383N/A
383N/A return r;
383N/A}
383N/A
383N/A_printf_(1, 2) static int write_confirm_message(const char *format, ...) {
383N/A _cleanup_close_ int fd = -1;
383N/A va_list ap;
383N/A
383N/A assert(format);
383N/A
383N/A fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
383N/A if (fd < 0)
383N/A return fd;
383N/A
383N/A va_start(ap, format);
383N/A vdprintf(fd, format, ap);
383N/A va_end(ap);
383N/A
383N/A return 0;
383N/A}
383N/A
383N/Astatic int restore_confirm_stdio(int *saved_stdin,
383N/A int *saved_stdout) {
383N/A
383N/A int r = 0;
383N/A
383N/A assert(saved_stdin);
383N/A assert(saved_stdout);
383N/A
383N/A release_terminal();
383N/A
383N/A if (*saved_stdin >= 0)
383N/A if (dup2(*saved_stdin, STDIN_FILENO) < 0)
383N/A r = -errno;
383N/A
383N/A if (*saved_stdout >= 0)
383N/A if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
383N/A r = -errno;
383N/A
383N/A safe_close(*saved_stdin);
383N/A safe_close(*saved_stdout);
383N/A
383N/A return r;
383N/A}
383N/A
383N/Astatic int ask_for_confirmation(char *response, char **argv) {
383N/A int saved_stdout = -1, saved_stdin = -1, r;
383N/A _cleanup_free_ char *line = NULL;
383N/A
383N/A r = setup_confirm_stdio(&saved_stdin, &saved_stdout);
383N/A if (r < 0)
383N/A return r;
383N/A
383N/A line = exec_command_line(argv);
383N/A if (!line)
383N/A return -ENOMEM;
383N/A
383N/A r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line);
383N/A
383N/A restore_confirm_stdio(&saved_stdin, &saved_stdout);
383N/A
383N/A return r;
383N/A}
383N/A
383N/Astatic int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
383N/A bool keep_groups = false;
383N/A int r;
383N/A
383N/A assert(context);
383N/A
383N/A /* Lookup and set GID and supplementary group list. Here too
383N/A * we avoid NSS lookups for gid=0. */
383N/A
383N/A if (context->group || username) {
383N/A /* First step, initialize groups from /etc/groups */
383N/A if (username && gid != 0) {
383N/A if (initgroups(username, gid) < 0)
383N/A return -errno;
383N/A
383N/A keep_groups = true;
383N/A }
383N/A
383N/A /* Second step, set our gids */
383N/A if (setresgid(gid, gid, gid) < 0)
383N/A return -errno;
383N/A }
383N/A
383N/A if (context->supplementary_groups) {
383N/A int ngroups_max, k;
383N/A gid_t *gids;
383N/A char **i;
383N/A
383N/A /* Final step, initialize any manually set supplementary groups */
383N/A assert_se((ngroups_max = (int) sysconf(_SC_NGROUPS_MAX)) > 0);
383N/A
383N/A if (!(gids = new(gid_t, ngroups_max)))
383N/A return -ENOMEM;
383N/A
383N/A if (keep_groups) {
383N/A k = getgroups(ngroups_max, gids);
383N/A if (k < 0) {
383N/A free(gids);
383N/A return -errno;
383N/A }
383N/A } else
383N/A k = 0;
383N/A
383N/A STRV_FOREACH(i, context->supplementary_groups) {
383N/A const char *g;
383N/A
383N/A if (k >= ngroups_max) {
383N/A free(gids);
383N/A return -E2BIG;
383N/A }
383N/A
383N/A g = *i;
383N/A r = get_group_creds(&g, gids+k);
383N/A if (r < 0) {
383N/A free(gids);
383N/A return r;
383N/A }
383N/A
383N/A k++;
383N/A }
383N/A
383N/A if (setgroups(k, gids) < 0) {
383N/A free(gids);
383N/A return -errno;
383N/A }
383N/A
383N/A free(gids);
383N/A }
383N/A
383N/A return 0;
383N/A}
383N/A
383N/Astatic int enforce_user(const ExecContext *context, uid_t uid) {
383N/A assert(context);
383N/A
383N/A /* Sets (but doesn't lookup) the uid and make sure we keep the
383N/A * capabilities while doing so. */
383N/A
383N/A if (context->capabilities) {
383N/A _cleanup_cap_free_ cap_t d = NULL;
383N/A static const cap_value_t bits[] = {
383N/A CAP_SETUID, /* Necessary so that we can run setresuid() below */
383N/A CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
383N/A };
383N/A
383N/A /* First step: If we need to keep capabilities but
383N/A * drop privileges we need to make sure we keep our
383N/A * caps, while we drop privileges. */
383N/A if (uid != 0) {
383N/A int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
383N/A
383N/A if (prctl(PR_GET_SECUREBITS) != sb)
383N/A if (prctl(PR_SET_SECUREBITS, sb) < 0)
383N/A return -errno;
383N/A }
383N/A
383N/A /* Second step: set the capabilities. This will reduce
383N/A * the capabilities to the minimum we need. */
383N/A
383N/A d = cap_dup(context->capabilities);
383N/A if (!d)
383N/A return -errno;
383N/A
383N/A if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
383N/A cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
383N/A return -errno;
383N/A
383N/A if (cap_set_proc(d) < 0)
383N/A return -errno;
383N/A }
383N/A
383N/A /* Third step: actually set the uids */
383N/A if (setresuid(uid, uid, uid) < 0)
383N/A return -errno;
383N/A
383N/A /* At this point we should have all necessary capabilities but
383N/A are otherwise a normal user. However, the caps might got
383N/A corrupted due to the setresuid() so we need clean them up
383N/A later. This is done outside of this call. */
383N/A
383N/A return 0;
383N/A}
383N/A
383N/A#ifdef HAVE_PAM
383N/A
383N/Astatic int null_conv(
383N/A int num_msg,
383N/A const struct pam_message **msg,
383N/A struct pam_response **resp,
383N/A void *appdata_ptr) {
383N/A
383N/A /* We don't support conversations */
383N/A
383N/A return PAM_CONV_ERR;
383N/A}
383N/A
383N/Astatic int setup_pam(
383N/A const char *name,
383N/A const char *user,
383N/A uid_t uid,
383N/A const char *tty,
383N/A char ***pam_env,
383N/A int fds[], unsigned n_fds) {
383N/A
383N/A static const struct pam_conv conv = {
383N/A .conv = null_conv,
383N/A .appdata_ptr = NULL
383N/A };
383N/A
383N/A pam_handle_t *handle = NULL;
383N/A sigset_t old_ss;
383N/A int pam_code = PAM_SUCCESS;
383N/A int err;
383N/A char **e = NULL;
383N/A bool close_session = false;
383N/A pid_t pam_pid = 0, parent_pid;
383N/A int flags = 0;
383N/A
383N/A assert(name);
383N/A assert(user);
383N/A assert(pam_env);
383N/A
383N/A /* We set up PAM in the parent process, then fork. The child
383N/A * will then stay around until killed via PR_GET_PDEATHSIG or
383N/A * systemd via the cgroup logic. It will then remove the PAM
383N/A * session again. The parent process will exec() the actual
383N/A * daemon. We do things this way to ensure that the main PID
383N/A * of the daemon is the one we initially fork()ed. */
383N/A
383N/A if (log_get_max_level() < LOG_DEBUG)
383N/A flags |= PAM_SILENT;
383N/A
383N/A pam_code = pam_start(name, user, &conv, &handle);
383N/A if (pam_code != PAM_SUCCESS) {
383N/A handle = NULL;
383N/A goto fail;
383N/A }
383N/A
383N/A if (tty) {
383N/A pam_code = pam_set_item(handle, PAM_TTY, tty);
383N/A if (pam_code != PAM_SUCCESS)
383N/A goto fail;
383N/A }
383N/A
383N/A pam_code = pam_acct_mgmt(handle, flags);
383N/A if (pam_code != PAM_SUCCESS)
383N/A goto fail;
383N/A
383N/A pam_code = pam_open_session(handle, flags);
383N/A if (pam_code != PAM_SUCCESS)
383N/A goto fail;
383N/A
383N/A close_session = true;
383N/A
383N/A e = pam_getenvlist(handle);
383N/A if (!e) {
383N/A pam_code = PAM_BUF_ERR;
383N/A goto fail;
383N/A }
383N/A
383N/A /* Block SIGTERM, so that we know that it won't get lost in
383N/A * the child */
383N/A
383N/A assert_se(sigprocmask_many(SIG_BLOCK, &old_ss, SIGTERM, -1) >= 0);
383N/A
383N/A parent_pid = getpid();
383N/A
383N/A pam_pid = fork();
383N/A if (pam_pid < 0)
383N/A goto fail;
383N/A
383N/A if (pam_pid == 0) {
383N/A int sig;
383N/A int r = EXIT_PAM;
383N/A
383N/A /* The child's job is to reset the PAM session on
383N/A * termination */
383N/A
383N/A /* This string must fit in 10 chars (i.e. the length
383N/A * of "/sbin/init"), to look pretty in /bin/ps */
383N/A rename_process("(sd-pam)");
383N/A
383N/A /* Make sure we don't keep open the passed fds in this
383N/A child. We assume that otherwise only those fds are
383N/A open here that have been opened by PAM. */
383N/A close_many(fds, n_fds);
383N/A
383N/A /* Drop privileges - we don't need any to pam_close_session
383N/A * and this will make PR_SET_PDEATHSIG work in most cases.
383N/A * If this fails, ignore the error - but expect sd-pam threads
383N/A * to fail to exit normally */
383N/A if (setresuid(uid, uid, uid) < 0)
383N/A log_error_errno(r, "Error: Failed to setresuid() in sd-pam: %m");
383N/A
383N/A (void) ignore_signals(SIGPIPE, -1);
383N/A
383N/A /* Wait until our parent died. This will only work if
383N/A * the above setresuid() succeeds, otherwise the kernel
383N/A * will not allow unprivileged parents kill their privileged
383N/A * children this way. We rely on the control groups kill logic
383N/A * to do the rest for us. */
383N/A if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
383N/A goto child_finish;
383N/A
383N/A /* Check if our parent process might already have
383N/A * died? */
383N/A if (getppid() == parent_pid) {
383N/A sigset_t ss;
383N/A
383N/A assert_se(sigemptyset(&ss) >= 0);
383N/A assert_se(sigaddset(&ss, SIGTERM) >= 0);
383N/A
383N/A for (;;) {
383N/A if (sigwait(&ss, &sig) < 0) {
383N/A if (errno == EINTR)
383N/A continue;
383N/A
383N/A goto child_finish;
383N/A }
383N/A
383N/A assert(sig == SIGTERM);
383N/A break;
383N/A }
383N/A }
383N/A
383N/A /* If our parent died we'll end the session */
383N/A if (getppid() != parent_pid) {
383N/A pam_code = pam_close_session(handle, flags);
383N/A if (pam_code != PAM_SUCCESS)
383N/A goto child_finish;
383N/A }
383N/A
383N/A r = 0;
383N/A
383N/A child_finish:
383N/A pam_end(handle, pam_code | flags);
383N/A _exit(r);
383N/A }
383N/A
383N/A /* If the child was forked off successfully it will do all the
383N/A * cleanups, so forget about the handle here. */
383N/A handle = NULL;
383N/A
383N/A /* Unblock SIGTERM again in the parent */
383N/A assert_se(sigprocmask(SIG_SETMASK, &old_ss, NULL) >= 0);
383N/A
383N/A /* We close the log explicitly here, since the PAM modules
383N/A * might have opened it, but we don't want this fd around. */
383N/A closelog();
383N/A
383N/A *pam_env = e;
383N/A e = NULL;
383N/A
383N/A return 0;
383N/A
383N/Afail:
383N/A if (pam_code != PAM_SUCCESS) {
383N/A log_error("PAM failed: %s", pam_strerror(handle, pam_code));
383N/A err = -EPERM; /* PAM errors do not map to errno */
383N/A } else {
383N/A err = log_error_errno(errno, "PAM failed: %m");
383N/A }
383N/A
383N/A if (handle) {
383N/A if (close_session)
383N/A pam_code = pam_close_session(handle, flags);
383N/A
383N/A pam_end(handle, pam_code | flags);
383N/A }
383N/A
383N/A strv_free(e);
383N/A
383N/A closelog();
383N/A
383N/A if (pam_pid > 1) {
383N/A kill(pam_pid, SIGTERM);
383N/A kill(pam_pid, SIGCONT);
383N/A }
383N/A
383N/A return err;
383N/A}
383N/A#endif
383N/A
383N/Astatic void rename_process_from_path(const char *path) {
383N/A char process_name[11];
383N/A const char *p;
383N/A size_t l;
383N/A
383N/A /* This resulting string must fit in 10 chars (i.e. the length
383N/A * of "/sbin/init") to look pretty in /bin/ps */
383N/A
383N/A p = basename(path);
383N/A if (isempty(p)) {
383N/A rename_process("(...)");
383N/A return;
383N/A }
383N/A
383N/A l = strlen(p);
383N/A if (l > 8) {
383N/A /* The end of the process name is usually more
383N/A * interesting, since the first bit might just be
383N/A * "systemd-" */
383N/A p = p + l - 8;
383N/A l = 8;
383N/A }
383N/A
383N/A process_name[0] = '(';
383N/A memcpy(process_name+1, p, l);
383N/A process_name[1+l] = ')';
383N/A process_name[1+l+1] = 0;
383N/A
383N/A rename_process(process_name);
383N/A}
383N/A
383N/A#ifdef HAVE_SECCOMP
383N/A
383N/Astatic int apply_seccomp(const ExecContext *c) {
383N/A uint32_t negative_action, action;
383N/A scmp_filter_ctx *seccomp;
383N/A Iterator i;
383N/A void *id;
383N/A int r;
383N/A
383N/A assert(c);
383N/A
383N/A negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
383N/A
383N/A seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
383N/A if (!seccomp)
383N/A return -ENOMEM;
383N/A
383N/A if (c->syscall_archs) {
383N/A
383N/A SET_FOREACH(id, c->syscall_archs, i) {
383N/A r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
383N/A if (r == -EEXIST)
383N/A continue;
383N/A if (r < 0)
383N/A goto finish;
383N/A }
383N/A
383N/A } else {
383N/A r = seccomp_add_secondary_archs(seccomp);
383N/A if (r < 0)
383N/A goto finish;
383N/A }
383N/A
383N/A action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
383N/A SET_FOREACH(id, c->syscall_filter, i) {
383N/A r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
383N/A if (r < 0)
383N/A goto finish;
383N/A }
383N/A
383N/A r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
383N/A if (r < 0)
383N/A goto finish;
383N/A
383N/A r = seccomp_load(seccomp);
383N/A
383N/Afinish:
383N/A seccomp_release(seccomp);
383N/A return r;
383N/A}
383N/A
383N/Astatic int apply_address_families(const ExecContext *c) {
383N/A scmp_filter_ctx *seccomp;
383N/A Iterator i;
383N/A int r;
383N/A
383N/A assert(c);
383N/A
383N/A seccomp = seccomp_init(SCMP_ACT_ALLOW);
383N/A if (!seccomp)
383N/A return -ENOMEM;
383N/A
383N/A r = seccomp_add_secondary_archs(seccomp);
383N/A if (r < 0)
383N/A goto finish;
383N/A
383N/A if (c->address_families_whitelist) {
383N/A int af, first = 0, last = 0;
383N/A void *afp;
383N/A
383N/A /* If this is a whitelist, we first block the address
383N/A * families that are out of range and then everything
383N/A * that is not in the set. First, we find the lowest
383N/A * and highest address family in the set. */
383N/A
383N/A SET_FOREACH(afp, c->address_families, i) {
383N/A af = PTR_TO_INT(afp);
383N/A
383N/A if (af <= 0 || af >= af_max())
383N/A continue;
383N/A
383N/A if (first == 0 || af < first)
383N/A first = af;
383N/A
383N/A if (last == 0 || af > last)
383N/A last = af;
383N/A }
383N/A
383N/A assert((first == 0) == (last == 0));
383N/A
383N/A if (first == 0) {
383N/A
383N/A /* No entries in the valid range, block everything */
383N/A r = seccomp_rule_add(
383N/A seccomp,
383N/A SCMP_ACT_ERRNO(EPROTONOSUPPORT),
383N/A SCMP_SYS(socket),
383N/A 0);
383N/A if (r < 0)
383N/A goto finish;
383N/A
383N/A } else {
383N/A
383N/A /* Block everything below the first entry */
383N/A r = seccomp_rule_add(
383N/A seccomp,
383N/A SCMP_ACT_ERRNO(EPROTONOSUPPORT),
383N/A SCMP_SYS(socket),
383N/A 1,
383N/A SCMP_A0(SCMP_CMP_LT, first));
383N/A if (r < 0)
383N/A goto finish;
383N/A
383N/A /* Block everything above the last entry */
383N/A r = seccomp_rule_add(
383N/A seccomp,
383N/A SCMP_ACT_ERRNO(EPROTONOSUPPORT),
383N/A SCMP_SYS(socket),
383N/A 1,
383N/A SCMP_A0(SCMP_CMP_GT, last));
383N/A if (r < 0)
383N/A goto finish;
383N/A
383N/A /* Block everything between the first and last
383N/A * entry */
383N/A for (af = 1; af < af_max(); af++) {
383N/A
383N/A if (set_contains(c->address_families, INT_TO_PTR(af)))
383N/A continue;
383N/A
383N/A r = seccomp_rule_add(
383N/A seccomp,
383N/A SCMP_ACT_ERRNO(EPROTONOSUPPORT),
383N/A SCMP_SYS(socket),
383N/A 1,
383N/A SCMP_A0(SCMP_CMP_EQ, af));
383N/A if (r < 0)
383N/A goto finish;
383N/A }
383N/A }
383N/A
383N/A } else {
383N/A void *af;
383N/A
383N/A /* If this is a blacklist, then generate one rule for
383N/A * each address family that are then combined in OR
383N/A * checks. */
383N/A
383N/A SET_FOREACH(af, c->address_families, i) {
383N/A
383N/A r = seccomp_rule_add(
383N/A seccomp,
383N/A SCMP_ACT_ERRNO(EPROTONOSUPPORT),
383N/A SCMP_SYS(socket),
383N/A 1,
383N/A SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af)));
383N/A if (r < 0)
383N/A goto finish;
383N/A }
383N/A }
383N/A
383N/A r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
383N/A if (r < 0)
383N/A goto finish;
383N/A
383N/A r = seccomp_load(seccomp);
383N/A
383N/Afinish:
383N/A seccomp_release(seccomp);
383N/A return r;
383N/A}
383N/A
383N/A#endif
383N/A
383N/Astatic void do_idle_pipe_dance(int idle_pipe[4]) {
assert(idle_pipe);
idle_pipe[1] = safe_close(idle_pipe[1]);
idle_pipe[2] = safe_close(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 */) {
ssize_t n;
/* Signal systemd that we are bored and want to continue. */
n = write(idle_pipe[3], "x", 1);
if (n > 0)
/* Wait for systemd to react to the signal above. */
fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
}
idle_pipe[0] = safe_close(idle_pipe[0]);
}
idle_pipe[3] = safe_close(idle_pipe[3]);
}
static int build_environment(
const ExecContext *c,
unsigned n_fds,
usec_t watchdog_usec,
const char *home,
const char *username,
const char *shell,
char ***ret) {
_cleanup_strv_free_ char **our_env = NULL;
unsigned n_env = 0;
char *x;
assert(c);
assert(ret);
our_env = new0(char*, 10);
if (!our_env)
return -ENOMEM;
if (n_fds > 0) {
if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
return -ENOMEM;
our_env[n_env++] = x;
if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
return -ENOMEM;
our_env[n_env++] = x;
}
if (watchdog_usec > 0) {
if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
return -ENOMEM;
our_env[n_env++] = x;
if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0)
return -ENOMEM;
our_env[n_env++] = x;
}
if (home) {
x = strappend("HOME=", home);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
}
if (username) {
x = strappend("LOGNAME=", username);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
x = strappend("USER=", username);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
}
if (shell) {
x = strappend("SHELL=", shell);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
}
if (is_terminal_input(c->std_input) ||
c->std_output == EXEC_OUTPUT_TTY ||
c->std_error == EXEC_OUTPUT_TTY ||
c->tty_path) {
x = strdup(default_term_for_tty(tty_path(c)));
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
}
our_env[n_env++] = NULL;
assert(n_env <= 10);
*ret = our_env;
our_env = NULL;
return 0;
}
static bool exec_needs_mount_namespace(
const ExecContext *context,
const ExecParameters *params,
ExecRuntime *runtime) {
assert(context);
assert(params);
if (!strv_isempty(context->read_write_dirs) ||
!strv_isempty(context->read_only_dirs) ||
!strv_isempty(context->inaccessible_dirs))
return true;
if (context->mount_flags != 0)
return true;
if (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir))
return true;
if (params->bus_endpoint_path)
return true;
if (context->private_devices ||
context->protect_system != PROTECT_SYSTEM_NO ||
context->protect_home != PROTECT_HOME_NO)
return true;
return false;
}
static int exec_child(
Unit *unit,
ExecCommand *command,
const ExecContext *context,
const ExecParameters *params,
ExecRuntime *runtime,
char **argv,
int socket_fd,
int *fds, unsigned n_fds,
char **files_env,
int *exit_status) {
_cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
_cleanup_free_ char *mac_selinux_context_net = NULL;
const char *username = NULL, *home = NULL, *shell = NULL;
unsigned n_dont_close = 0;
int dont_close[n_fds + 4];
uid_t uid = UID_INVALID;
gid_t gid = GID_INVALID;
int i, r;
bool needs_mount_namespace;
assert(unit);
assert(command);
assert(context);
assert(params);
assert(exit_status);
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. */
(void) default_signals(SIGNALS_CRASH_HANDLER,
SIGNALS_IGNORE, -1);
if (context->ignore_sigpipe)
(void) ignore_signals(SIGPIPE, -1);
r = reset_signal_mask();
if (r < 0) {
*exit_status = EXIT_SIGNAL_MASK;
return r;
}
if (params->idle_pipe)
do_idle_pipe_dance(params->idle_pipe);
/* Close sockets very early to make sure we don't
* block init reexecution because it cannot bind its
* sockets */
log_forget_fds();
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];
}
r = close_all_fds(dont_close, n_dont_close);
if (r < 0) {
*exit_status = EXIT_FDS;
return r;
}
if (!context->same_pgrp)
if (setsid() < 0) {
*exit_status = EXIT_SETSID;
return -errno;
}
exec_context_tty_reset(context);
if (params->confirm_spawn) {
char response;
r = ask_for_confirmation(&response, argv);
if (r == -ETIMEDOUT)
write_confirm_message("Confirmation question timed out, assuming positive response.\n");
else if (r < 0)
write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-r));
else if (response == 's') {
write_confirm_message("Skipping execution.\n");
*exit_status = EXIT_CONFIRM;
return -ECANCELED;
} else if (response == 'n') {
write_confirm_message("Failing execution.\n");
*exit_status = 0;
return 0;
}
}
if (context->user) {
username = context->user;
r = get_user_creds(&username, &uid, &gid, &home, &shell);
if (r < 0) {
*exit_status = EXIT_USER;
return r;
}
}
if (context->group) {
const char *g = context->group;
r = get_group_creds(&g, &gid);
if (r < 0) {
*exit_status = EXIT_GROUP;
return r;
}
}
/* 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);
r = setup_input(context, socket_fd, params->apply_tty_stdin);
if (r < 0) {
*exit_status = EXIT_STDIN;
return r;
}
r = setup_output(unit, context, STDOUT_FILENO, socket_fd, basename(command->path), params->apply_tty_stdin, uid, gid);
if (r < 0) {
*exit_status = EXIT_STDOUT;
return r;
}
r = setup_output(unit, context, STDERR_FILENO, socket_fd, basename(command->path), params->apply_tty_stdin, uid, gid);
if (r < 0) {
*exit_status = EXIT_STDERR;
return r;
}
if (params->cgroup_path) {
r = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0, NULL, NULL);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return r;
}
}
if (context->oom_score_adjust_set) {
char t[DECIMAL_STR_MAX(context->oom_score_adjust)];
/* When we can't make this change due to EPERM, then
* let's silently skip over it. User namespaces
* prohibit write access to this file, and we
* shouldn't trip up over that. */
sprintf(t, "%i", context->oom_score_adjust);
r = write_string_file("/proc/self/oom_score_adj", t, 0);
if (r == -EPERM || r == -EACCES) {
log_open();
log_unit_debug_errno(unit, r, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
log_close();
} else if (r < 0) {
*exit_status = EXIT_OOM_ADJUST;
return -errno;
}
}
if (context->nice_set)
if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
*exit_status = EXIT_NICE;
return -errno;
}
if (context->cpu_sched_set) {
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),
&param);
if (r < 0) {
*exit_status = EXIT_SETSCHEDULER;
return -errno;
}
}
if (context->cpuset)
if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) {
*exit_status = EXIT_CPUAFFINITY;
return -errno;
}
if (context->ioprio_set)
if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
*exit_status = EXIT_IOPRIO;
return -errno;
}
if (context->timer_slack_nsec != NSEC_INFINITY)
if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
*exit_status = EXIT_TIMERSLACK;
return -errno;
}
if (context->personality != PERSONALITY_INVALID)
if (personality(context->personality) < 0) {
*exit_status = EXIT_PERSONALITY;
return -errno;
}
if (context->utmp_id)
utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path,
context->utmp_mode == EXEC_UTMP_INIT ? INIT_PROCESS :
context->utmp_mode == EXEC_UTMP_LOGIN ? LOGIN_PROCESS :
USER_PROCESS,
username ? "root" : context->user);
if (context->user && is_terminal_input(context->std_input)) {
r = chown_terminal(STDIN_FILENO, uid);
if (r < 0) {
*exit_status = EXIT_STDIN;
return r;
}
}
if (params->bus_endpoint_fd >= 0 && context->bus_endpoint) {
uid_t ep_uid = (uid == UID_INVALID) ? 0 : uid;
r = bus_kernel_set_endpoint_policy(params->bus_endpoint_fd, ep_uid, context->bus_endpoint);
if (r < 0) {
*exit_status = EXIT_BUS_ENDPOINT;
return r;
}
}
/* If delegation is enabled we'll pass ownership of the cgroup
* (but only in systemd's own controller hierarchy!) to the
* user of the new process. */
if (params->cgroup_path && context->user && params->cgroup_delegate) {
r = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return r;
}
r = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid);
if (r < 0) {
*exit_status = EXIT_CGROUP;
return r;
}
}
if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) {
char **rt;
STRV_FOREACH(rt, context->runtime_directory) {
_cleanup_free_ char *p;
p = strjoin(params->runtime_prefix, "/", *rt, NULL);
if (!p) {
*exit_status = EXIT_RUNTIME_DIRECTORY;
return -ENOMEM;
}
r = mkdir_p_label(p, context->runtime_directory_mode);
if (r < 0) {
*exit_status = EXIT_RUNTIME_DIRECTORY;
return r;
}
r = chmod_and_chown(p, context->runtime_directory_mode, uid, gid);
if (r < 0) {
*exit_status = EXIT_RUNTIME_DIRECTORY;
return r;
}
}
}
if (params->apply_permissions) {
r = enforce_groups(context, username, gid);
if (r < 0) {
*exit_status = EXIT_GROUP;
return r;
}
}
umask(context->umask);
#ifdef HAVE_SMACK
if (params->apply_permissions) {
if (context->smack_process_label) {
r = mac_smack_apply_pid(0, context->smack_process_label);
if (r < 0) {
*exit_status = EXIT_SMACK_PROCESS_LABEL;
return r;
}
}
#ifdef SMACK_DEFAULT_PROCESS_LABEL
else {
_cleanup_free_ char *exec_label = NULL;
r = mac_smack_read(command->path, SMACK_ATTR_EXEC, &exec_label);
if (r < 0 && r != -ENODATA && r != -EOPNOTSUPP) {
*exit_status = EXIT_SMACK_PROCESS_LABEL;
return r;
}
r = mac_smack_apply_pid(0, exec_label ? : SMACK_DEFAULT_PROCESS_LABEL);
if (r < 0) {
*exit_status = EXIT_SMACK_PROCESS_LABEL;
return r;
}
}
}
#endif
#endif
#ifdef HAVE_PAM
if (params->apply_permissions && context->pam_name && username) {
r = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds);
if (r < 0) {
*exit_status = EXIT_PAM;
return r;
}
}
#endif
if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
r = setup_netns(runtime->netns_storage_socket);
if (r < 0) {
*exit_status = EXIT_NETWORK;
return r;
}
}
needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime);
if (needs_mount_namespace) {
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 = strjoina(runtime->tmp_dir, "/tmp");
if (runtime->var_tmp_dir)
var = strjoina(runtime->var_tmp_dir, "/tmp");
}
r = setup_namespace(
params->apply_chroot ? context->root_directory : NULL,
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 we couldn't set up the namespace this is
* probably due to a missing capability. In this case,
* silently proceeed. */
if (r == -EPERM || r == -EACCES) {
log_open();
log_unit_debug_errno(unit, r, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
log_close();
} else if (r < 0) {
*exit_status = EXIT_NAMESPACE;
return r;
}
}
if (params->apply_chroot) {
if (!needs_mount_namespace && context->root_directory)
if (chroot(context->root_directory) < 0) {
*exit_status = EXIT_CHROOT;
return -errno;
}
if (chdir(context->working_directory ?: "/") < 0 &&
!context->working_directory_missing_ok) {
*exit_status = EXIT_CHDIR;
return -errno;
}
} else {
_cleanup_free_ char *d = NULL;
if (asprintf(&d, "%s/%s",
context->root_directory ?: "",
context->working_directory ?: "") < 0) {
*exit_status = EXIT_MEMORY;
return -ENOMEM;
}
if (chdir(d) < 0 &&
!context->working_directory_missing_ok) {
*exit_status = EXIT_CHDIR;
return -errno;
}
}
#ifdef HAVE_SELINUX
if (params->apply_permissions && mac_selinux_use() && params->selinux_context_net && socket_fd >= 0) {
r = mac_selinux_get_child_mls_label(socket_fd, command->path, context->selinux_context, &mac_selinux_context_net);
if (r < 0) {
*exit_status = EXIT_SELINUX_CONTEXT;
return r;
}
}
#endif
/* 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. */
r = close_all_fds(fds, n_fds);
if (r >= 0)
r = shift_fds(fds, n_fds);
if (r >= 0)
r = flags_fds(fds, n_fds, context->non_blocking);
if (r < 0) {
*exit_status = EXIT_FDS;
return r;
}
if (params->apply_permissions) {
for (i = 0; i < _RLIMIT_MAX; i++) {
if (!context->rlimit[i])
continue;
if (setrlimit_closest(i, context->rlimit[i]) < 0) {
*exit_status = EXIT_LIMITS;
return -errno;
}
}
if (context->capability_bounding_set_drop) {
r = capability_bounding_set_drop(context->capability_bounding_set_drop, false);
if (r < 0) {
*exit_status = EXIT_CAPABILITIES;
return r;
}
}
if (context->user) {
r = enforce_user(context, uid);
if (r < 0) {
*exit_status = EXIT_USER;
return r;
}
}
/* 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) {
*exit_status = EXIT_SECUREBITS;
return -errno;
}
if (context->capabilities)
if (cap_set_proc(context->capabilities) < 0) {
*exit_status = EXIT_CAPABILITIES;
return -errno;
}
if (context->no_new_privileges)
if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
*exit_status = EXIT_NO_NEW_PRIVILEGES;
return -errno;
}
#ifdef HAVE_SECCOMP
if (context->address_families_whitelist ||
!set_isempty(context->address_families)) {
r = apply_address_families(context);
if (r < 0) {
*exit_status = EXIT_ADDRESS_FAMILIES;
return r;
}
}
if (context->syscall_whitelist ||
!set_isempty(context->syscall_filter) ||
!set_isempty(context->syscall_archs)) {
r = apply_seccomp(context);
if (r < 0) {
*exit_status = EXIT_SECCOMP;
return r;
}
}
#endif
#ifdef HAVE_SELINUX
if (mac_selinux_use()) {
char *exec_context = mac_selinux_context_net ?: context->selinux_context;
if (exec_context) {
r = setexeccon(exec_context);
if (r < 0) {
*exit_status = EXIT_SELINUX_CONTEXT;
return r;
}
}
}
#endif
#ifdef HAVE_APPARMOR
if (context->apparmor_profile && mac_apparmor_use()) {
r = aa_change_onexec(context->apparmor_profile);
if (r < 0 && !context->apparmor_profile_ignore) {
*exit_status = EXIT_APPARMOR_PROFILE;
return -errno;
}
}
#endif
}
r = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env);
if (r < 0) {
*exit_status = EXIT_MEMORY;
return r;
}
final_env = strv_env_merge(5,
params->environment,
our_env,
context->environment,
files_env,
pam_env,
NULL);
if (!final_env) {
*exit_status = EXIT_MEMORY;
return -ENOMEM;
}
final_argv = replace_env_argv(argv, final_env);
if (!final_argv) {
*exit_status = EXIT_MEMORY;
return -ENOMEM;
}
final_env = strv_env_clean(final_env);
if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
_cleanup_free_ char *line;
line = exec_command_line(final_argv);
if (line) {
log_open();
log_struct(LOG_DEBUG,
LOG_UNIT_ID(unit),
"EXECUTABLE=%s", command->path,
LOG_UNIT_MESSAGE(unit, "Executing: %s", line),
NULL);
log_close();
}
}
execve(command->path, final_argv, final_env);
*exit_status = EXIT_EXEC;
return -errno;
}
int exec_spawn(Unit *unit,
ExecCommand *command,
const ExecContext *context,
const ExecParameters *params,
ExecRuntime *runtime,
pid_t *ret) {
_cleanup_strv_free_ char **files_env = NULL;
int *fds = NULL; unsigned n_fds = 0;
_cleanup_free_ char *line = NULL;
int socket_fd, r;
char **argv;
pid_t pid;
assert(unit);
assert(command);
assert(context);
assert(ret);
assert(params);
assert(params->fds || params->n_fds <= 0);
if (context->std_input == EXEC_INPUT_SOCKET ||
context->std_output == EXEC_OUTPUT_SOCKET ||
context->std_error == EXEC_OUTPUT_SOCKET) {
if (params->n_fds != 1) {
log_unit_error(unit, "Got more than one socket.");
return -EINVAL;
}
socket_fd = params->fds[0];
} else {
socket_fd = -1;
fds = params->fds;
n_fds = params->n_fds;
}
r = exec_context_load_environment(unit, context, &files_env);
if (r < 0)
return log_unit_error_errno(unit, r, "Failed to load environment files: %m");
argv = params->argv ?: command->argv;
line = exec_command_line(argv);
if (!line)
return log_oom();
log_struct(LOG_DEBUG,
LOG_UNIT_ID(unit),
LOG_UNIT_MESSAGE(unit, "About to execute: %s", line),
"EXECUTABLE=%s", command->path,
NULL);
pid = fork();
if (pid < 0)
return log_unit_error_errno(unit, r, "Failed to fork: %m");
if (pid == 0) {
int exit_status;
r = exec_child(unit,
command,
context,
params,
runtime,
argv,
socket_fd,
fds, n_fds,
files_env,
&exit_status);
if (r < 0) {
log_open();
log_struct_errno(LOG_ERR, r,
LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED),
LOG_UNIT_ID(unit),
LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m",
exit_status_to_string(exit_status, EXIT_STATUS_SYSTEMD),
command->path),
"EXECUTABLE=%s", command->path,
NULL);
}
_exit(exit_status);
}
log_unit_debug(unit, "Forked %s as "PID_FMT, command->path, pid);
/* 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 (params->cgroup_path)
(void) cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid);
exec_status_start(&command->exec_status, pid);
*ret = pid;
return 0;
}
void exec_context_init(ExecContext *c) {
assert(c);
c->umask = 0022;
c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
c->cpu_sched_policy = SCHED_OTHER;
c->syslog_priority = LOG_DAEMON|LOG_INFO;
c->syslog_level_prefix = true;
c->ignore_sigpipe = true;
c->timer_slack_nsec = NSEC_INFINITY;
c->personality = PERSONALITY_INVALID;
c->runtime_directory_mode = 0755;
}
void exec_context_done(ExecContext *c) {
unsigned l;
assert(c);
c->environment = strv_free(c->environment);
c->environment_files = strv_free(c->environment_files);
for (l = 0; l < ELEMENTSOF(c->rlimit); l++)
c->rlimit[l] = mfree(c->rlimit[l]);
c->working_directory = mfree(c->working_directory);
c->root_directory = mfree(c->root_directory);
c->tty_path = mfree(c->tty_path);
c->syslog_identifier = mfree(c->syslog_identifier);
c->user = mfree(c->user);
c->group = mfree(c->group);
c->supplementary_groups = strv_free(c->supplementary_groups);
c->pam_name = mfree(c->pam_name);
if (c->capabilities) {
cap_free(c->capabilities);
c->capabilities = NULL;
}
c->read_only_dirs = strv_free(c->read_only_dirs);
c->read_write_dirs = strv_free(c->read_write_dirs);
c->inaccessible_dirs = strv_free(c->inaccessible_dirs);
if (c->cpuset)
CPU_FREE(c->cpuset);
c->utmp_id = mfree(c->utmp_id);
c->selinux_context = mfree(c->selinux_context);
c->apparmor_profile = mfree(c->apparmor_profile);
c->syscall_filter = set_free(c->syscall_filter);
c->syscall_archs = set_free(c->syscall_archs);
c->address_families = set_free(c->address_families);
c->runtime_directory = strv_free(c->runtime_directory);
bus_endpoint_free(c->bus_endpoint);
c->bus_endpoint = NULL;
}
int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) {
char **i;
assert(c);
if (!runtime_prefix)
return 0;
STRV_FOREACH(i, c->runtime_directory) {
_cleanup_free_ char *p;
p = strjoin(runtime_prefix, "/", *i, NULL);
if (!p)
return -ENOMEM;
/* We execute this synchronously, since we need to be
* sure this is gone when we start the service
* next. */
(void) rm_rf(p, REMOVE_ROOT);
}
return 0;
}
void exec_command_done(ExecCommand *c) {
assert(c);
c->path = mfree(c->path);
c->argv = strv_free(c->argv);
}
void exec_command_done_array(ExecCommand *c, unsigned n) {
unsigned i;
for (i = 0; i < n; i++)
exec_command_done(c+i);
}
ExecCommand* exec_command_free_list(ExecCommand *c) {
ExecCommand *i;
while ((i = c)) {
LIST_REMOVE(command, c, i);
exec_command_done(i);
free(i);
}
return NULL;
}
void exec_command_free_array(ExecCommand **c, unsigned n) {
unsigned i;
for (i = 0; i < n; i++)
c[i] = exec_command_free_list(c[i]);
}
typedef struct InvalidEnvInfo {
Unit *unit;
const char *path;
} InvalidEnvInfo;
static void invalid_env(const char *p, void *userdata) {
InvalidEnvInfo *info = userdata;
log_unit_error(info->unit, "Ignoring invalid environment assignment '%s': %s", p, info->path);
}
int exec_context_load_environment(Unit *unit, const ExecContext *c, char ***l) {
char **i, **r = NULL;
assert(c);
assert(l);
STRV_FOREACH(i, c->environment_files) {
char *fn;
int k;
bool ignore = false;
char **p;
_cleanup_globfree_ glob_t pglob = {};
int count, n;
fn = *i;
if (fn[0] == '-') {
ignore = true;
fn ++;
}
if (!path_is_absolute(fn)) {
if (ignore)
continue;
strv_free(r);
return -EINVAL;
}
/* Filename supports globbing, take all matching files */
errno = 0;
if (glob(fn, 0, NULL, &pglob) != 0) {
if (ignore)
continue;
strv_free(r);
return errno ? -errno : -EINVAL;
}
count = pglob.gl_pathc;
if (count == 0) {
if (ignore)
continue;
strv_free(r);
return -EINVAL;
}
for (n = 0; n < count; n++) {
k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p);
if (k < 0) {
if (ignore)
continue;
strv_free(r);
return k;
}
/* Log invalid environment variables with filename */
if (p) {
InvalidEnvInfo info = {
.unit = unit,
.path = pglob.gl_pathv[n]
};
p = strv_env_clean_with_callback(p, invalid_env, &info);
}
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;
}
}
}
*l = r;
return 0;
}
static bool tty_may_match_dev_console(const char *tty) {
_cleanup_free_ char *active = NULL;
char *console;
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 */
return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
}
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;
assert(f);
STRV_FOREACH(g, l)
fprintf(f, " %s", *g);
}
void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
char **e;
unsigned i;
assert(c);
assert(f);
prefix = strempty(prefix);
fprintf(f,
"%sUMask: %04o\n"
"%sWorkingDirectory: %s\n"
"%sRootDirectory: %s\n"
"%sNonBlocking: %s\n"
"%sPrivateTmp: %s\n"
"%sPrivateNetwork: %s\n"
"%sPrivateDevices: %s\n"
"%sProtectHome: %s\n"
"%sProtectSystem: %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->private_network),
prefix, yes_no(c->private_devices),
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)
fprintf(f, "%sEnvironment: %s\n", prefix, *e);
STRV_FOREACH(e, c->environment_files)
fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e);
if (c->nice_set)
fprintf(f,
"%sNice: %i\n",
prefix, c->nice);
if (c->oom_score_adjust_set)
fprintf(f,
"%sOOMScoreAdjust: %i\n",
prefix, c->oom_score_adjust);
for (i = 0; i < RLIM_NLIMITS; i++)
if (c->rlimit[i])
fprintf(f, "%s%s: "RLIM_FMT"\n",
prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max);
if (c->ioprio_set) {
_cleanup_free_ char *class_str = NULL;
ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
fprintf(f,
"%sIOSchedulingClass: %s\n"
"%sIOPriority: %i\n",
prefix, strna(class_str),
prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
}
if (c->cpu_sched_set) {
_cleanup_free_ char *policy_str = NULL;
sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
fprintf(f,
"%sCPUSchedulingPolicy: %s\n"
"%sCPUSchedulingPriority: %i\n"
"%sCPUSchedulingResetOnFork: %s\n",
prefix, strna(policy_str),
prefix, c->cpu_sched_priority,
prefix, yes_no(c->cpu_sched_reset_on_fork));
}
if (c->cpuset) {
fprintf(f, "%sCPUAffinity:", prefix);
for (i = 0; i < c->cpuset_ncpus; i++)
if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
fprintf(f, " %u", i);
fputs("\n", f);
}
if (c->timer_slack_nsec != NSEC_INFINITY)
fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
fprintf(f,
"%sStandardInput: %s\n"
"%sStandardOutput: %s\n"
"%sStandardError: %s\n",
prefix, exec_input_to_string(c->std_input),
prefix, exec_output_to_string(c->std_output),
prefix, exec_output_to_string(c->std_error));
if (c->tty_path)
fprintf(f,
"%sTTYPath: %s\n"
"%sTTYReset: %s\n"
"%sTTYVHangup: %s\n"
"%sTTYVTDisallocate: %s\n",
prefix, c->tty_path,
prefix, yes_no(c->tty_reset),
prefix, yes_no(c->tty_vhangup),
prefix, yes_no(c->tty_vt_disallocate));
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) {
_cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
fprintf(f,
"%sSyslogFacility: %s\n"
"%sSyslogLevel: %s\n",
prefix, strna(fac_str),
prefix, strna(lvl_str));
}
if (c->capabilities) {
_cleanup_cap_free_charp_ char *t;
t = cap_to_text(c->capabilities, NULL);
if (t)
fprintf(f, "%sCapabilities: %s\n", prefix, t);
}
if (c->secure_bits)
fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
prefix,
(c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
(c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
(c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
(c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
(c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
(c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
if (c->capability_bounding_set_drop) {
unsigned long l;
fprintf(f, "%sCapabilityBoundingSet:", prefix);
for (l = 0; l <= cap_last_cap(); l++)
if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l)))
fprintf(f, " %s", strna(capability_to_name(l)));
fputs("\n", f);
}
if (c->user)
fprintf(f, "%sUser: %s\n", prefix, c->user);
if (c->group)
fprintf(f, "%sGroup: %s\n", prefix, c->group);
if (strv_length(c->supplementary_groups) > 0) {
fprintf(f, "%sSupplementaryGroups:", prefix);
strv_fprintf(f, c->supplementary_groups);
fputs("\n", f);
}
if (c->pam_name)
fprintf(f, "%sPAMName: %s\n", prefix, c->pam_name);
if (strv_length(c->read_write_dirs) > 0) {
fprintf(f, "%sReadWriteDirs:", prefix);
strv_fprintf(f, c->read_write_dirs);
fputs("\n", f);
}
if (strv_length(c->read_only_dirs) > 0) {
fprintf(f, "%sReadOnlyDirs:", prefix);
strv_fprintf(f, c->read_only_dirs);
fputs("\n", f);
}
if (strv_length(c->inaccessible_dirs) > 0) {
fprintf(f, "%sInaccessibleDirs:", prefix);
strv_fprintf(f, c->inaccessible_dirs);
fputs("\n", f);
}
if (c->utmp_id)
fprintf(f,
"%sUtmpIdentifier: %s\n",
prefix, c->utmp_id);
if (c->selinux_context)
fprintf(f,
"%sSELinuxContext: %s%s\n",
prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
if (c->personality != PERSONALITY_INVALID)
fprintf(f,
"%sPersonality: %s\n",
prefix, strna(personality_to_string(c->personality)));
if (c->syscall_filter) {
#ifdef HAVE_SECCOMP
Iterator j;
void *id;
bool first = true;
#endif
fprintf(f,
"%sSystemCallFilter: ",
prefix);
if (!c->syscall_whitelist)
fputc('~', f);
#ifdef HAVE_SECCOMP
SET_FOREACH(id, c->syscall_filter, j) {
_cleanup_free_ char *name = NULL;
if (first)
first = false;
else
fputc(' ', f);
name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
fputs(strna(name), f);
}
#endif
fputc('\n', f);
}
if (c->syscall_archs) {
#ifdef HAVE_SECCOMP
Iterator j;
void *id;
#endif
fprintf(f,
"%sSystemCallArchitectures:",
prefix);
#ifdef HAVE_SECCOMP
SET_FOREACH(id, c->syscall_archs, j)
fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
#endif
fputc('\n', f);
}
if (c->syscall_errno != 0)
fprintf(f,
"%sSystemCallErrorNumber: %s\n",
prefix, strna(errno_to_name(c->syscall_errno)));
if (c->apparmor_profile)
fprintf(f,
"%sAppArmorProfile: %s%s\n",
prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile);
}
bool exec_context_maintains_privileges(ExecContext *c) {
assert(c);
/* Returns true if the process forked off would run run under
* an unchanged UID or as root. */
if (!c->user)
return true;
if (streq(c->user, "root") || streq(c->user, "0"))
return true;
return false;
}
void exec_status_start(ExecStatus *s, pid_t pid) {
assert(s);
zero(*s);
s->pid = pid;
dual_timestamp_get(&s->start_timestamp);
}
void exec_status_exit(ExecStatus *s, ExecContext *context, pid_t pid, int code, int status) {
assert(s);
if (s->pid && s->pid != pid)
zero(*s);
s->pid = pid;
dual_timestamp_get(&s->exit_timestamp);
s->code = code;
s->status = status;
if (context) {
if (context->utmp_id)
utmp_put_dead_process(context->utmp_id, pid, code, status);
exec_context_tty_reset(context);
}
}
void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) {
char buf[FORMAT_TIMESTAMP_MAX];
assert(s);
assert(f);
if (s->pid <= 0)
return;
prefix = strempty(prefix);
fprintf(f,
"%sPID: "PID_FMT"\n",
prefix, s->pid);
if (s->start_timestamp.realtime > 0)
fprintf(f,
"%sStart Timestamp: %s\n",
prefix, format_timestamp(buf, sizeof(buf), s->start_timestamp.realtime));
if (s->exit_timestamp.realtime > 0)
fprintf(f,
"%sExit Timestamp: %s\n"
"%sExit Code: %s\n"
"%sExit Status: %i\n",
prefix, format_timestamp(buf, sizeof(buf), s->exit_timestamp.realtime),
prefix, sigchld_code_to_string(s->code),
prefix, s->status);
}
char *exec_command_line(char **argv) {
size_t k;
char *n, *p, **a;
bool first = true;
assert(argv);
k = 1;
STRV_FOREACH(a, argv)
k += strlen(*a)+3;
if (!(n = new(char, k)))
return NULL;
p = n;
STRV_FOREACH(a, argv) {
if (!first)
*(p++) = ' ';
else
first = false;
if (strpbrk(*a, WHITESPACE)) {
*(p++) = '\'';
p = stpcpy(p, *a);
*(p++) = '\'';
} else
p = stpcpy(p, *a);
}
*p = 0;
/* FIXME: this doesn't really handle arguments that have
* spaces and ticks in them */
return n;
}
void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) {
_cleanup_free_ char *cmd = NULL;
const char *prefix2;
assert(c);
assert(f);
prefix = strempty(prefix);
prefix2 = strjoina(prefix, "\t");
cmd = exec_command_line(c->argv);
fprintf(f,
"%sCommand Line: %s\n",
prefix, cmd ? cmd : strerror(ENOMEM));
exec_status_dump(&c->exec_status, f, prefix2);
}
void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) {
assert(f);
prefix = strempty(prefix);
LIST_FOREACH(command, c, c)
exec_command_dump(c, f, prefix);
}
void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
ExecCommand *end;
assert(l);
assert(e);
if (*l) {
/* It's kind of important, that we keep the order here */
LIST_FIND_TAIL(command, *l, end);
LIST_INSERT_AFTER(command, *l, end, e);
} else
*l = e;
}
int exec_command_set(ExecCommand *c, const char *path, ...) {
va_list ap;
char **l, *p;
assert(c);
assert(path);
va_start(ap, path);
l = strv_new_ap(path, ap);
va_end(ap);
if (!l)
return -ENOMEM;
p = strdup(path);
if (!p) {
strv_free(l);
return -ENOMEM;
}
free(c->path);
c->path = p;
strv_free(c->argv);
c->argv = l;
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)
return 0;
*rt = new0(ExecRuntime, 1);
if (!*rt)
return -ENOMEM;
(*rt)->n_ref = 1;
(*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
return 0;
}
int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
int r;
assert(rt);
assert(c);
assert(id);
if (*rt)
return 1;
if (!c->private_network && !c->private_tmp)
return 0;
r = exec_runtime_allocate(rt);
if (r < 0)
return r;
if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
return -errno;
}
if (c->private_tmp && !(*rt)->tmp_dir) {
r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
if (r < 0)
return r;
}
return 1;
}
ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
assert(r);
assert(r->n_ref > 0);
r->n_ref++;
return r;
}
ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
if (!r)
return NULL;
assert(r->n_ref > 0);
r->n_ref--;
if (r->n_ref > 0)
return NULL;
free(r->tmp_dir);
free(r->var_tmp_dir);
safe_close_pair(r->netns_storage_socket);
free(r);
return NULL;
}
int exec_runtime_serialize(Unit *u, ExecRuntime *rt, FILE *f, FDSet *fds) {
assert(u);
assert(f);
assert(fds);
if (!rt)
return 0;
if (rt->tmp_dir)
unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
if (rt->var_tmp_dir)
unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
if (rt->netns_storage_socket[0] >= 0) {
int copy;
copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
if (copy < 0)
return copy;
unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
}
if (rt->netns_storage_socket[1] >= 0) {
int copy;
copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
if (copy < 0)
return copy;
unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
}
return 0;
}
int exec_runtime_deserialize_item(Unit *u, ExecRuntime **rt, const char *key, const char *value, FDSet *fds) {
int r;
assert(rt);
assert(key);
assert(value);
if (streq(key, "tmp-dir")) {
char *copy;
r = exec_runtime_allocate(rt);
if (r < 0)
return log_oom();
copy = strdup(value);
if (!copy)
return log_oom();
free((*rt)->tmp_dir);
(*rt)->tmp_dir = copy;
} else if (streq(key, "var-tmp-dir")) {
char *copy;
r = exec_runtime_allocate(rt);
if (r < 0)
return log_oom();
copy = strdup(value);
if (!copy)
return log_oom();
free((*rt)->var_tmp_dir);
(*rt)->var_tmp_dir = copy;
} else if (streq(key, "netns-socket-0")) {
int fd;
r = exec_runtime_allocate(rt);
if (r < 0)
return log_oom();
if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
log_unit_debug(u, "Failed to parse netns socket value: %s", value);
else {
safe_close((*rt)->netns_storage_socket[0]);
(*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
}
} else if (streq(key, "netns-socket-1")) {
int fd;
r = exec_runtime_allocate(rt);
if (r < 0)
return log_oom();
if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
log_unit_debug(u, "Failed to parse netns socket value: %s", value);
else {
safe_close((*rt)->netns_storage_socket[1]);
(*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
}
} else
return 0;
return 1;
}
static void *remove_tmpdir_thread(void *p) {
_cleanup_free_ char *path = p;
(void) rm_rf(path, REMOVE_ROOT|REMOVE_PHYSICAL);
return NULL;
}
void exec_runtime_destroy(ExecRuntime *rt) {
int r;
if (!rt)
return;
/* If there are multiple users of this, let's leave the stuff around */
if (rt->n_ref > 1)
return;
if (rt->tmp_dir) {
log_debug("Spawning thread to nuke %s", rt->tmp_dir);
r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
if (r < 0) {
log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir);
free(rt->tmp_dir);
}
rt->tmp_dir = NULL;
}
if (rt->var_tmp_dir) {
log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
if (r < 0) {
log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir);
free(rt->var_tmp_dir);
}
rt->var_tmp_dir = NULL;
}
safe_close_pair(rt->netns_storage_socket);
}
static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
[EXEC_INPUT_NULL] = "null",
[EXEC_INPUT_TTY] = "tty",
[EXEC_INPUT_TTY_FORCE] = "tty-force",
[EXEC_INPUT_TTY_FAIL] = "tty-fail",
[EXEC_INPUT_SOCKET] = "socket"
};
DEFINE_STRING_TABLE_LOOKUP(exec_input, ExecInput);
static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
[EXEC_OUTPUT_INHERIT] = "inherit",
[EXEC_OUTPUT_NULL] = "null",
[EXEC_OUTPUT_TTY] = "tty",
[EXEC_OUTPUT_SYSLOG] = "syslog",
[EXEC_OUTPUT_SYSLOG_AND_CONSOLE] = "syslog+console",
[EXEC_OUTPUT_KMSG] = "kmsg",
[EXEC_OUTPUT_KMSG_AND_CONSOLE] = "kmsg+console",
[EXEC_OUTPUT_JOURNAL] = "journal",
[EXEC_OUTPUT_JOURNAL_AND_CONSOLE] = "journal+console",
[EXEC_OUTPUT_SOCKET] = "socket"
};
DEFINE_STRING_TABLE_LOOKUP(exec_output, ExecOutput);
static const char* const exec_utmp_mode_table[_EXEC_UTMP_MODE_MAX] = {
[EXEC_UTMP_INIT] = "init",
[EXEC_UTMP_LOGIN] = "login",
[EXEC_UTMP_USER] = "user",
};
DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode, ExecUtmpMode);