#include <errno.h>
#include <fcntl.h>
#include <poll.h>
-#include <sys/eventfd.h>
#include <sys/file.h>
-#include <sys/ioctl.h>
#include <sys/mman.h>
-#include <sys/mount.h>
#include <sys/personality.h>
#include <sys/prctl.h>
#include <sys/shm.h>
#include <linux/fs.h> /* Must be included after <sys/mount.h> */
-#if HAVE_PAM
-#include <security/pam_appl.h>
-#endif
-
-#if HAVE_SELINUX
-#include <selinux/selinux.h>
-#endif
-
-#if HAVE_APPARMOR
-#include <sys/apparmor.h>
-#endif
-
#include "sd-messages.h"
#include "af-list.h"
#include "alloc-util.h"
-#if HAVE_APPARMOR
-#include "apparmor-util.h"
-#endif
-#include "argv-util.h"
#include "async.h"
-#include "barrier.h"
-#include "bpf-lsm.h"
-#include "btrfs-util.h"
#include "cap-list.h"
#include "capability-util.h"
-#include "chattr-util.h"
#include "cgroup-setup.h"
-#include "chase.h"
-#include "chown-recursive.h"
#include "constants.h"
#include "cpu-set-util.h"
-#include "credential.h"
-#include "data-fd-util.h"
+#include "dev-setup.h"
#include "env-file.h"
#include "env-util.h"
#include "errno-list.h"
#include "escape.h"
+#include "exec-credential.h"
#include "execute.h"
+#include "execute-serialize.h"
#include "exit-status.h"
#include "fd-util.h"
+#include "fileio.h"
#include "format-util.h"
#include "glob-util.h"
#include "hexdecoct.h"
-#include "io-util.h"
#include "ioprio-util.h"
#include "lock-util.h"
#include "log.h"
#include "manager-dump.h"
#include "memory-util.h"
#include "missing_fs.h"
-#include "missing_ioprio.h"
#include "missing_prctl.h"
-#include "missing_syscall.h"
#include "mkdir-label.h"
#include "namespace.h"
#include "parse-util.h"
#include "path-util.h"
-#include "proc-cmdline.h"
#include "process-util.h"
-#include "psi-util.h"
#include "rlimit-util.h"
#include "rm-rf.h"
#include "seccomp-util.h"
#include "securebits-util.h"
#include "selinux-util.h"
-#include "signal-util.h"
-#include "smack-util.h"
-#include "socket-util.h"
+#include "serialize.h"
#include "sort-util.h"
#include "special.h"
#include "stat-util.h"
#include "user-util.h"
#include "utmp-wtmp.h"
-#define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
-#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
-
-#define SNDBUF_SIZE (8*1024*1024)
-
-static int shift_fds(int fds[], size_t n_fds) {
- if (n_fds <= 0)
- return 0;
-
- /* Modifies the fds array! (sorts it) */
-
- assert(fds);
-
- for (int start = 0;;) {
- int restart_from = -1;
-
- for (int i = start; i < (int) n_fds; i++) {
- int nfd;
-
- /* Already at right index? */
- if (fds[i] == i+3)
- continue;
-
- nfd = fcntl(fds[i], F_DUPFD, i + 3);
- if (nfd < 0)
- return -errno;
-
- safe_close(fds[i]);
- fds[i] = nfd;
-
- /* Hmm, the fd we wanted isn't free? Then
- * let's remember that and try again from here */
- if (nfd != i+3 && restart_from < 0)
- restart_from = i;
- }
-
- if (restart_from < 0)
- break;
-
- start = restart_from;
- }
-
- return 0;
-}
-
-static int flags_fds(
- const int fds[],
- size_t n_socket_fds,
- size_t n_fds,
- bool nonblock) {
-
- int r;
-
- if (n_fds <= 0)
- return 0;
-
- assert(fds);
-
- /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
- * O_NONBLOCK only applies to socket activation though. */
-
- for (size_t i = 0; i < n_fds; i++) {
-
- if (i < n_socket_fds) {
- r = fd_nonblock(fds[i], nonblock);
- if (r < 0)
- return r;
- }
-
- /* We unconditionally drop FD_CLOEXEC from the fds,
- * since after all we want to pass these fds to our
- * children */
-
- r = fd_cloexec(fds[i], false);
- if (r < 0)
- return r;
- }
-
- return 0;
-}
-
-static const char *exec_context_tty_path(const ExecContext *context) {
+const char *exec_context_tty_path(const ExecContext *context) {
assert(context);
if (context->stdio_as_fds)
return "/dev/console";
}
-static int exec_context_tty_size(const ExecContext *context, unsigned *ret_rows, unsigned *ret_cols) {
+int exec_context_tty_size(const ExecContext *context, unsigned *ret_rows, unsigned *ret_cols) {
unsigned rows, cols;
const char *tty;
return 0;
}
-static void exec_context_tty_reset(const ExecContext *context, const ExecParameters *p) {
- _cleanup_close_ int fd = -EBADF;
+void exec_context_tty_reset(const ExecContext *context, const ExecParameters *p) {
+ _cleanup_close_ int _fd = -EBADF, lock_fd = -EBADF;
const char *path = exec_context_tty_path(ASSERT_PTR(context));
+ int fd;
- /* Take a lock around the device for the duration of the setup that we do here.
- * systemd-vconsole-setup.service also takes the lock to avoid being interrupted.
- * We open a new fd that will be closed automatically, and operate on it for convenience.
- */
-
- if (p && p->stdin_fd >= 0) {
- fd = xopenat_lock(p->stdin_fd, NULL,
- O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY, 0, 0, LOCK_BSD, LOCK_EX);
+ if (p && p->stdin_fd >= 0)
+ fd = p->stdin_fd;
+ else if (path) {
+ fd = _fd = open_terminal(path, O_RDWR|O_NOCTTY|O_CLOEXEC|O_NONBLOCK);
if (fd < 0)
return;
- } else if (path) {
- fd = open_terminal(path, O_RDWR|O_NOCTTY|O_CLOEXEC|O_NONBLOCK);
- if (fd < 0)
- return;
-
- if (lock_generic(fd, LOCK_BSD, LOCK_EX) < 0)
- return;
} else
return; /* nothing to do */
+ /* Take a synchronization lock for the duration of the setup that we do here.
+ * systemd-vconsole-setup.service also takes the lock to avoid being interrupted.
+ * We open a new fd that will be closed automatically, and operate on it for convenience.
+ */
+ lock_fd = lock_dev_console();
+ if (lock_fd < 0)
+ return (void) log_debug_errno(lock_fd,
+ "Failed to lock /dev/console: %m");
+
if (context->tty_vhangup)
(void) terminal_vhangup_fd(fd);
EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
}
-static bool is_kmsg_output(ExecOutput o) {
- return IN_SET(o,
- EXEC_OUTPUT_KMSG,
- EXEC_OUTPUT_KMSG_AND_CONSOLE);
-}
-
-static bool exec_context_needs_term(const ExecContext *c) {
- assert(c);
-
- /* Return true if the execution context suggests we should set $TERM to something useful. */
-
- if (is_terminal_input(c->std_input))
- return true;
-
- if (is_terminal_output(c->std_output))
- return true;
-
- if (is_terminal_output(c->std_error))
- return true;
+bool exec_needs_network_namespace(const ExecContext *context) {
+ assert(context);
- return !!c->tty_path;
+ return context->private_network || context->network_namespace_path;
}
-static int open_null_as(int flags, int nfd) {
- int fd;
-
- assert(nfd >= 0);
-
- fd = open("/dev/null", flags|O_NOCTTY);
- if (fd < 0)
- return -errno;
-
- return move_fd(fd, nfd, false);
+static bool exec_needs_ephemeral(const ExecContext *context) {
+ return (context->root_image || context->root_directory) && context->root_ephemeral;
}
-static int connect_journal_socket(
- int fd,
- const char *log_namespace,
- uid_t uid,
- gid_t gid) {
-
- uid_t olduid = UID_INVALID;
- gid_t oldgid = GID_INVALID;
- const char *j;
- int r;
-
- j = log_namespace ?
- strjoina("/run/systemd/journal.", log_namespace, "/stdout") :
- "/run/systemd/journal/stdout";
-
- if (gid_is_valid(gid)) {
- oldgid = getgid();
-
- if (setegid(gid) < 0)
- return -errno;
- }
-
- if (uid_is_valid(uid)) {
- olduid = getuid();
-
- if (seteuid(uid) < 0) {
- r = -errno;
- goto restore_gid;
- }
- }
-
- r = connect_unix_path(fd, AT_FDCWD, j);
-
- /* If we fail to restore the uid or gid, things will likely fail later on. This should only happen if
- an LSM interferes. */
-
- if (uid_is_valid(uid))
- (void) seteuid(olduid);
-
- restore_gid:
- if (gid_is_valid(gid))
- (void) setegid(oldgid);
+bool exec_needs_ipc_namespace(const ExecContext *context) {
+ assert(context);
- return r;
+ return context->private_ipc || context->ipc_namespace_path;
}
-static int connect_logger_as(
- const Unit *unit,
+bool exec_needs_mount_namespace(
const ExecContext *context,
const ExecParameters *params,
- ExecOutput output,
- const char *ident,
- int nfd,
- uid_t uid,
- gid_t gid) {
-
- _cleanup_close_ int fd = -EBADF;
- int r;
+ const ExecRuntime *runtime) {
assert(context);
- assert(params);
- assert(output < _EXEC_OUTPUT_MAX);
- assert(ident);
- assert(nfd >= 0);
-
- fd = socket(AF_UNIX, SOCK_STREAM, 0);
- if (fd < 0)
- return -errno;
- r = connect_journal_socket(fd, context->log_namespace, uid, gid);
- if (r < 0)
- return r;
+ if (context->root_image)
+ return true;
- if (shutdown(fd, SHUT_RD) < 0)
- return -errno;
-
- (void) fd_inc_sndbuf(fd, SNDBUF_SIZE);
-
- if (dprintf(fd,
- "%s\n"
- "%s\n"
- "%i\n"
- "%i\n"
- "%i\n"
- "%i\n"
- "%i\n",
- context->syslog_identifier ?: ident,
- params->flags & EXEC_PASS_LOG_UNIT ? unit->id : "",
- context->syslog_priority,
- !!context->syslog_level_prefix,
- false,
- is_kmsg_output(output),
- is_terminal_output(output)) < 0)
- return -errno;
-
- return move_fd(TAKE_FD(fd), nfd, false);
-}
+ if (!strv_isempty(context->read_write_paths) ||
+ !strv_isempty(context->read_only_paths) ||
+ !strv_isempty(context->inaccessible_paths) ||
+ !strv_isempty(context->exec_paths) ||
+ !strv_isempty(context->no_exec_paths))
+ return true;
-static int open_terminal_as(const char *path, int flags, int nfd) {
- int fd;
+ if (context->n_bind_mounts > 0)
+ return true;
- assert(path);
- assert(nfd >= 0);
+ if (context->n_temporary_filesystems > 0)
+ return true;
- fd = open_terminal(path, flags | O_NOCTTY);
- if (fd < 0)
- return fd;
+ if (context->n_mount_images > 0)
+ return true;
- return move_fd(fd, nfd, false);
-}
+ if (context->n_extension_images > 0)
+ return true;
-static int acquire_path(const char *path, int flags, mode_t mode) {
- _cleanup_close_ int fd = -EBADF;
- int r;
+ if (!strv_isempty(context->extension_directories))
+ return true;
- assert(path);
+ if (!IN_SET(context->mount_propagation_flag, 0, MS_SHARED))
+ return true;
- if (IN_SET(flags & O_ACCMODE, O_WRONLY, O_RDWR))
- flags |= O_CREAT;
+ if (context->private_tmp && runtime && runtime->shared && (runtime->shared->tmp_dir || runtime->shared->var_tmp_dir))
+ return true;
- fd = open(path, flags|O_NOCTTY, mode);
- if (fd >= 0)
- return TAKE_FD(fd);
+ if (context->private_devices ||
+ context->private_mounts > 0 ||
+ (context->private_mounts < 0 && exec_needs_network_namespace(context)) ||
+ context->protect_system != PROTECT_SYSTEM_NO ||
+ context->protect_home != PROTECT_HOME_NO ||
+ context->protect_kernel_tunables ||
+ context->protect_kernel_modules ||
+ context->protect_kernel_logs ||
+ context->protect_control_groups ||
+ context->protect_proc != PROTECT_PROC_DEFAULT ||
+ context->proc_subset != PROC_SUBSET_ALL ||
+ exec_needs_ipc_namespace(context))
+ return true;
- if (errno != ENXIO) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
- return -errno;
+ if (context->root_directory) {
+ if (exec_context_get_effective_mount_apivfs(context))
+ return true;
- /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
+ for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
+ if (params && !params->prefix[t])
+ continue;
- fd = socket(AF_UNIX, SOCK_STREAM, 0);
- if (fd < 0)
- return -errno;
+ if (context->directories[t].n_items > 0)
+ return true;
+ }
+ }
- r = connect_unix_path(fd, AT_FDCWD, path);
- if (IN_SET(r, -ENOTSOCK, -EINVAL))
- /* Propagate initial error if we get ENOTSOCK or EINVAL, i.e. we have indication that this
- * wasn't an AF_UNIX socket after all */
- return -ENXIO;
- if (r < 0)
- return r;
+ if (context->dynamic_user &&
+ (context->directories[EXEC_DIRECTORY_STATE].n_items > 0 ||
+ context->directories[EXEC_DIRECTORY_CACHE].n_items > 0 ||
+ context->directories[EXEC_DIRECTORY_LOGS].n_items > 0))
+ return true;
- if ((flags & O_ACCMODE) == O_RDONLY)
- r = shutdown(fd, SHUT_WR);
- else if ((flags & O_ACCMODE) == O_WRONLY)
- r = shutdown(fd, SHUT_RD);
- else
- r = 0;
- if (r < 0)
- return -errno;
+ if (context->log_namespace)
+ return true;
- return TAKE_FD(fd);
+ return false;
}
-static int fixup_input(
- const ExecContext *context,
- int socket_fd,
- bool apply_tty_stdin) {
-
- ExecInput std_input;
-
+bool exec_directory_is_private(const ExecContext *context, ExecDirectoryType type) {
assert(context);
- std_input = context->std_input;
-
- if (is_terminal_input(std_input) && !apply_tty_stdin)
- return EXEC_INPUT_NULL;
-
- if (std_input == EXEC_INPUT_SOCKET && socket_fd < 0)
- return EXEC_INPUT_NULL;
-
- if (std_input == EXEC_INPUT_DATA && context->stdin_data_size == 0)
- return EXEC_INPUT_NULL;
-
- return std_input;
-}
+ if (!context->dynamic_user)
+ return false;
-static int fixup_output(ExecOutput output, int socket_fd) {
+ if (type == EXEC_DIRECTORY_CONFIGURATION)
+ return false;
- if (output == EXEC_OUTPUT_SOCKET && socket_fd < 0)
- return EXEC_OUTPUT_INHERIT;
+ if (type == EXEC_DIRECTORY_RUNTIME && context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO)
+ return false;
- return output;
+ return true;
}
-static int setup_input(
- const ExecContext *context,
+int exec_params_get_cgroup_path(
const ExecParameters *params,
- int socket_fd,
- const int named_iofds[static 3]) {
+ const CGroupContext *c,
+ char **ret) {
- ExecInput i;
- int r;
+ const char *subgroup = NULL;
+ char *p;
- assert(context);
assert(params);
- assert(named_iofds);
-
- if (params->stdin_fd >= 0) {
- if (dup2(params->stdin_fd, STDIN_FILENO) < 0)
- return -errno;
-
- /* Try to make this the controlling tty, if it is a tty, and reset it */
- if (isatty(STDIN_FILENO)) {
- unsigned rows = context->tty_rows, cols = context->tty_cols;
-
- (void) exec_context_tty_size(context, &rows, &cols);
- (void) ioctl(STDIN_FILENO, TIOCSCTTY, context->std_input == EXEC_INPUT_TTY_FORCE);
- (void) reset_terminal_fd(STDIN_FILENO, true);
- (void) terminal_set_size_fd(STDIN_FILENO, NULL, rows, cols);
- }
-
- return STDIN_FILENO;
- }
-
- i = fixup_input(context, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN);
-
- switch (i) {
-
- case EXEC_INPUT_NULL:
- return open_null_as(O_RDONLY, STDIN_FILENO);
-
- case EXEC_INPUT_TTY:
- case EXEC_INPUT_TTY_FORCE:
- case EXEC_INPUT_TTY_FAIL: {
- unsigned rows, cols;
- int fd;
-
- fd = acquire_terminal(exec_context_tty_path(context),
- i == EXEC_INPUT_TTY_FAIL ? ACQUIRE_TERMINAL_TRY :
- i == EXEC_INPUT_TTY_FORCE ? ACQUIRE_TERMINAL_FORCE :
- ACQUIRE_TERMINAL_WAIT,
- USEC_INFINITY);
- if (fd < 0)
- return fd;
+ assert(ret);
- r = exec_context_tty_size(context, &rows, &cols);
- if (r < 0)
- return r;
+ if (!params->cgroup_path)
+ return -EINVAL;
- r = terminal_set_size_fd(fd, exec_context_tty_path(context), rows, cols);
- if (r < 0)
- return r;
+ /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
+ * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
+ * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
+ * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
+ * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
+ * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
+ * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
+ * flag, which is only passed for the former statements, not for the latter. */
- return move_fd(fd, STDIN_FILENO, false);
+ if (FLAGS_SET(params->flags, EXEC_CGROUP_DELEGATE) && (FLAGS_SET(params->flags, EXEC_CONTROL_CGROUP) || c->delegate_subgroup)) {
+ if (FLAGS_SET(params->flags, EXEC_IS_CONTROL))
+ subgroup = ".control";
+ else
+ subgroup = c->delegate_subgroup;
}
- case EXEC_INPUT_SOCKET:
- assert(socket_fd >= 0);
-
- return RET_NERRNO(dup2(socket_fd, STDIN_FILENO));
-
- case EXEC_INPUT_NAMED_FD:
- assert(named_iofds[STDIN_FILENO] >= 0);
-
- (void) fd_nonblock(named_iofds[STDIN_FILENO], false);
- return RET_NERRNO(dup2(named_iofds[STDIN_FILENO], STDIN_FILENO));
+ if (subgroup)
+ p = path_join(params->cgroup_path, subgroup);
+ else
+ p = strdup(params->cgroup_path);
+ if (!p)
+ return -ENOMEM;
- case EXEC_INPUT_DATA: {
- int fd;
+ *ret = p;
+ return !!subgroup;
+}
- fd = acquire_data_fd(context->stdin_data, context->stdin_data_size, 0);
- if (fd < 0)
- return fd;
+bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c) {
+ assert(c);
- return move_fd(fd, STDIN_FILENO, false);
- }
+ return c->cpu_affinity_from_numa;
+}
- case EXEC_INPUT_FILE: {
- bool rw;
- int fd;
+static void log_command_line(Unit *unit, const char *msg, const char *executable, char **argv) {
+ assert(unit);
+ assert(msg);
+ assert(executable);
- assert(context->stdio_file[STDIN_FILENO]);
-
- rw = (context->std_output == EXEC_OUTPUT_FILE && streq_ptr(context->stdio_file[STDIN_FILENO], context->stdio_file[STDOUT_FILENO])) ||
- (context->std_error == EXEC_OUTPUT_FILE && streq_ptr(context->stdio_file[STDIN_FILENO], context->stdio_file[STDERR_FILENO]));
-
- fd = acquire_path(context->stdio_file[STDIN_FILENO], rw ? O_RDWR : O_RDONLY, 0666 & ~context->umask);
- if (fd < 0)
- return fd;
-
- return move_fd(fd, STDIN_FILENO, false);
- }
-
- default:
- assert_not_reached();
- }
-}
-
-static bool can_inherit_stderr_from_stdout(
- const ExecContext *context,
- ExecOutput o,
- ExecOutput e) {
-
- assert(context);
-
- /* Returns true, if given the specified STDERR and STDOUT output we can directly dup() the stdout fd to the
- * stderr fd */
-
- if (e == EXEC_OUTPUT_INHERIT)
- return true;
- if (e != o)
- return false;
-
- if (e == EXEC_OUTPUT_NAMED_FD)
- return streq_ptr(context->stdio_fdname[STDOUT_FILENO], context->stdio_fdname[STDERR_FILENO]);
-
- if (IN_SET(e, EXEC_OUTPUT_FILE, EXEC_OUTPUT_FILE_APPEND, EXEC_OUTPUT_FILE_TRUNCATE))
- return streq_ptr(context->stdio_file[STDOUT_FILENO], context->stdio_file[STDERR_FILENO]);
-
- return true;
-}
-
-static int setup_output(
- const Unit *unit,
- const ExecContext *context,
- const ExecParameters *params,
- int fileno,
- int socket_fd,
- const int named_iofds[static 3],
- const char *ident,
- uid_t uid,
- gid_t gid,
- dev_t *journal_stream_dev,
- ino_t *journal_stream_ino) {
-
- ExecOutput o;
- ExecInput i;
- int r;
-
- assert(unit);
- assert(context);
- assert(params);
- assert(ident);
- assert(journal_stream_dev);
- assert(journal_stream_ino);
-
- if (fileno == STDOUT_FILENO && params->stdout_fd >= 0) {
-
- if (dup2(params->stdout_fd, STDOUT_FILENO) < 0)
- return -errno;
-
- return STDOUT_FILENO;
- }
-
- if (fileno == STDERR_FILENO && params->stderr_fd >= 0) {
- if (dup2(params->stderr_fd, STDERR_FILENO) < 0)
- return -errno;
-
- return STDERR_FILENO;
- }
-
- i = fixup_input(context, socket_fd, params->flags & EXEC_APPLY_TTY_STDIN);
- o = fixup_output(context->std_output, socket_fd);
-
- if (fileno == STDERR_FILENO) {
- ExecOutput e;
- 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 fileno;
-
- /* Duplicate from stdout if possible */
- if (can_inherit_stderr_from_stdout(context, o, e))
- return RET_NERRNO(dup2(STDOUT_FILENO, fileno));
-
- 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(exec_context_tty_path(context), O_WRONLY, fileno);
-
- /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
- if (!IN_SET(i, EXEC_INPUT_NULL, EXEC_INPUT_DATA))
- return RET_NERRNO(dup2(STDIN_FILENO, fileno));
-
- /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
- if (getppid() != 1)
- return fileno;
-
- /* We need to open /dev/null here anew, to get the right access mode. */
- return open_null_as(O_WRONLY, fileno);
- }
-
- switch (o) {
-
- case EXEC_OUTPUT_NULL:
- return open_null_as(O_WRONLY, fileno);
-
- case EXEC_OUTPUT_TTY:
- if (is_terminal_input(i))
- return RET_NERRNO(dup2(STDIN_FILENO, fileno));
-
- /* We don't reset the terminal if this is just about output */
- return open_terminal_as(exec_context_tty_path(context), O_WRONLY, fileno);
-
- case EXEC_OUTPUT_KMSG:
- case EXEC_OUTPUT_KMSG_AND_CONSOLE:
- case EXEC_OUTPUT_JOURNAL:
- case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
- r = connect_logger_as(unit, context, params, o, ident, fileno, uid, gid);
- if (r < 0) {
- log_unit_warning_errno(unit, r, "Failed to connect %s to the journal socket, ignoring: %m",
- fileno == STDOUT_FILENO ? "stdout" : "stderr");
- r = open_null_as(O_WRONLY, fileno);
- } else {
- struct stat st;
-
- /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
- * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
- * services to detect whether they are connected to the journal or not.
- *
- * If both stdout and stderr are connected to a stream then let's make sure to store the data
- * about STDERR as that's usually the best way to do logging. */
-
- if (fstat(fileno, &st) >= 0 &&
- (*journal_stream_ino == 0 || fileno == STDERR_FILENO)) {
- *journal_stream_dev = st.st_dev;
- *journal_stream_ino = st.st_ino;
- }
- }
- return r;
-
- case EXEC_OUTPUT_SOCKET:
- assert(socket_fd >= 0);
-
- return RET_NERRNO(dup2(socket_fd, fileno));
-
- case EXEC_OUTPUT_NAMED_FD:
- assert(named_iofds[fileno] >= 0);
-
- (void) fd_nonblock(named_iofds[fileno], false);
- return RET_NERRNO(dup2(named_iofds[fileno], fileno));
-
- case EXEC_OUTPUT_FILE:
- case EXEC_OUTPUT_FILE_APPEND:
- case EXEC_OUTPUT_FILE_TRUNCATE: {
- bool rw;
- int fd, flags;
-
- assert(context->stdio_file[fileno]);
-
- rw = context->std_input == EXEC_INPUT_FILE &&
- streq_ptr(context->stdio_file[fileno], context->stdio_file[STDIN_FILENO]);
-
- if (rw)
- return RET_NERRNO(dup2(STDIN_FILENO, fileno));
-
- flags = O_WRONLY;
- if (o == EXEC_OUTPUT_FILE_APPEND)
- flags |= O_APPEND;
- else if (o == EXEC_OUTPUT_FILE_TRUNCATE)
- flags |= O_TRUNC;
-
- fd = acquire_path(context->stdio_file[fileno], flags, 0666 & ~context->umask);
- if (fd < 0)
- return fd;
-
- return move_fd(fd, fileno, 0);
- }
-
- default:
- assert_not_reached();
- }
-}
-
-static int chown_terminal(int fd, uid_t uid) {
- int r;
-
- assert(fd >= 0);
-
- /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
- if (isatty(fd) < 1) {
- if (IN_SET(errno, EINVAL, ENOTTY))
- return 0; /* not a tty */
-
- return -errno;
- }
-
- /* This might fail. What matters are the results. */
- r = fchmod_and_chown(fd, TTY_MODE, uid, GID_INVALID);
- if (r < 0)
- return r;
-
- return 1;
-}
-
-static int setup_confirm_stdio(
- const ExecContext *context,
- const char *vc,
- int *ret_saved_stdin,
- int *ret_saved_stdout) {
-
- _cleanup_close_ int fd = -EBADF, saved_stdin = -EBADF, saved_stdout = -EBADF;
- unsigned rows, cols;
- int r;
-
- assert(ret_saved_stdin);
- assert(ret_saved_stdout);
-
- saved_stdin = fcntl(STDIN_FILENO, F_DUPFD, 3);
- if (saved_stdin < 0)
- return -errno;
-
- saved_stdout = fcntl(STDOUT_FILENO, F_DUPFD, 3);
- if (saved_stdout < 0)
- return -errno;
-
- fd = acquire_terminal(vc, ACQUIRE_TERMINAL_WAIT, DEFAULT_CONFIRM_USEC);
- if (fd < 0)
- return fd;
-
- r = chown_terminal(fd, getuid());
- if (r < 0)
- return r;
-
- r = reset_terminal_fd(fd, true);
- if (r < 0)
- return r;
-
- r = exec_context_tty_size(context, &rows, &cols);
- if (r < 0)
- return r;
-
- r = terminal_set_size_fd(fd, vc, rows, cols);
- if (r < 0)
- return r;
-
- r = rearrange_stdio(fd, fd, STDERR_FILENO); /* Invalidates 'fd' also on failure */
- TAKE_FD(fd);
- if (r < 0)
- return r;
-
- *ret_saved_stdin = TAKE_FD(saved_stdin);
- *ret_saved_stdout = TAKE_FD(saved_stdout);
- return 0;
-}
-
-static void write_confirm_error_fd(int err, int fd, const Unit *u) {
- assert(err < 0);
-
- if (err == -ETIMEDOUT)
- dprintf(fd, "Confirmation question timed out for %s, assuming positive response.\n", u->id);
- else {
- errno = -err;
- dprintf(fd, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u->id);
- }
-}
-
-static void write_confirm_error(int err, const char *vc, const Unit *u) {
- _cleanup_close_ int fd = -EBADF;
-
- assert(vc);
-
- fd = open_terminal(vc, O_WRONLY|O_NOCTTY|O_CLOEXEC);
- if (fd < 0)
- return;
-
- write_confirm_error_fd(err, fd, u);
-}
-
-static int restore_confirm_stdio(int *saved_stdin, int *saved_stdout) {
- int r = 0;
-
- assert(saved_stdin);
- assert(saved_stdout);
-
- release_terminal();
-
- if (*saved_stdin >= 0)
- if (dup2(*saved_stdin, STDIN_FILENO) < 0)
- r = -errno;
-
- if (*saved_stdout >= 0)
- if (dup2(*saved_stdout, STDOUT_FILENO) < 0)
- r = -errno;
-
- *saved_stdin = safe_close(*saved_stdin);
- *saved_stdout = safe_close(*saved_stdout);
-
- return r;
-}
-
-enum {
- CONFIRM_PRETEND_FAILURE = -1,
- CONFIRM_PRETEND_SUCCESS = 0,
- CONFIRM_EXECUTE = 1,
-};
-
-static int ask_for_confirmation(const ExecContext *context, const char *vc, Unit *u, const char *cmdline) {
- int saved_stdout = -1, saved_stdin = -1, r;
- _cleanup_free_ char *e = NULL;
- char c;
-
- /* For any internal errors, assume a positive response. */
- r = setup_confirm_stdio(context, vc, &saved_stdin, &saved_stdout);
- if (r < 0) {
- write_confirm_error(r, vc, u);
- return CONFIRM_EXECUTE;
- }
-
- /* confirm_spawn might have been disabled while we were sleeping. */
- if (manager_is_confirm_spawn_disabled(u->manager)) {
- r = 1;
- goto restore_stdio;
- }
-
- e = ellipsize(cmdline, 60, 100);
- if (!e) {
- log_oom();
- r = CONFIRM_EXECUTE;
- goto restore_stdio;
- }
-
- for (;;) {
- r = ask_char(&c, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e);
- if (r < 0) {
- write_confirm_error_fd(r, STDOUT_FILENO, u);
- r = CONFIRM_EXECUTE;
- goto restore_stdio;
- }
-
- switch (c) {
- case 'c':
- printf("Resuming normal execution.\n");
- manager_disable_confirm_spawn();
- r = 1;
- break;
- case 'D':
- unit_dump(u, stdout, " ");
- continue; /* ask again */
- case 'f':
- printf("Failing execution.\n");
- r = CONFIRM_PRETEND_FAILURE;
- break;
- case 'h':
- printf(" c - continue, proceed without asking anymore\n"
- " D - dump, show the state of the unit\n"
- " f - fail, don't execute the command and pretend it failed\n"
- " h - help\n"
- " i - info, show a short summary of the unit\n"
- " j - jobs, show jobs that are in progress\n"
- " s - skip, don't execute the command and pretend it succeeded\n"
- " y - yes, execute the command\n");
- continue; /* ask again */
- case 'i':
- printf(" Description: %s\n"
- " Unit: %s\n"
- " Command: %s\n",
- u->id, u->description, cmdline);
- continue; /* ask again */
- case 'j':
- manager_dump_jobs(u->manager, stdout, /* patterns= */ NULL, " ");
- continue; /* ask again */
- case 'n':
- /* 'n' was removed in favor of 'f'. */
- printf("Didn't understand 'n', did you mean 'f'?\n");
- continue; /* ask again */
- case 's':
- printf("Skipping execution.\n");
- r = CONFIRM_PRETEND_SUCCESS;
- break;
- case 'y':
- r = CONFIRM_EXECUTE;
- break;
- default:
- assert_not_reached();
- }
- break;
- }
-
-restore_stdio:
- restore_confirm_stdio(&saved_stdin, &saved_stdout);
- return r;
-}
-
-static int get_fixed_user(const ExecContext *c, const char **user,
- uid_t *uid, gid_t *gid,
- const char **home, const char **shell) {
- int r;
- const char *name;
-
- assert(c);
-
- if (!c->user)
- return 0;
-
- /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
- * (i.e. are "/" or "/bin/nologin"). */
-
- name = c->user;
- r = get_user_creds(&name, uid, gid, home, shell, USER_CREDS_CLEAN);
- if (r < 0)
- return r;
-
- *user = name;
- return 0;
-}
-
-static int get_fixed_group(const ExecContext *c, const char **group, gid_t *gid) {
- int r;
- const char *name;
-
- assert(c);
-
- if (!c->group)
- return 0;
-
- name = c->group;
- r = get_group_creds(&name, gid, 0);
- if (r < 0)
- return r;
-
- *group = name;
- return 0;
-}
-
-static int get_supplementary_groups(const ExecContext *c, const char *user,
- const char *group, gid_t gid,
- gid_t **supplementary_gids, int *ngids) {
- int r, k = 0;
- int ngroups_max;
- bool keep_groups = false;
- gid_t *groups = NULL;
- _cleanup_free_ gid_t *l_gids = NULL;
-
- assert(c);
-
- /*
- * If user is given, then lookup GID and supplementary groups list.
- * We avoid NSS lookups for gid=0. Also we have to initialize groups
- * here and as early as possible so we keep the list of supplementary
- * groups of the caller.
- */
- if (user && gid_is_valid(gid) && gid != 0) {
- /* First step, initialize groups from /etc/groups */
- if (initgroups(user, gid) < 0)
- return -errno;
-
- keep_groups = true;
- }
-
- if (strv_isempty(c->supplementary_groups))
- return 0;
-
- /*
- * If SupplementaryGroups= was passed then NGROUPS_MAX has to
- * be positive, otherwise fail.
- */
- errno = 0;
- ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
- if (ngroups_max <= 0)
- return errno_or_else(EOPNOTSUPP);
-
- l_gids = new(gid_t, ngroups_max);
- if (!l_gids)
- return -ENOMEM;
-
- if (keep_groups) {
- /*
- * Lookup the list of groups that the user belongs to, we
- * avoid NSS lookups here too for gid=0.
- */
- k = ngroups_max;
- if (getgrouplist(user, gid, l_gids, &k) < 0)
- return -EINVAL;
- } else
- k = 0;
-
- STRV_FOREACH(i, c->supplementary_groups) {
- const char *g;
-
- if (k >= ngroups_max)
- return -E2BIG;
-
- g = *i;
- r = get_group_creds(&g, l_gids+k, 0);
- if (r < 0)
- return r;
-
- k++;
- }
-
- /*
- * Sets ngids to zero to drop all supplementary groups, happens
- * when we are under root and SupplementaryGroups= is empty.
- */
- if (k == 0) {
- *ngids = 0;
- return 0;
- }
-
- /* Otherwise get the final list of supplementary groups */
- groups = memdup(l_gids, sizeof(gid_t) * k);
- if (!groups)
- return -ENOMEM;
-
- *supplementary_gids = groups;
- *ngids = k;
-
- groups = NULL;
-
- return 0;
-}
-
-static int enforce_groups(gid_t gid, const gid_t *supplementary_gids, int ngids) {
- int r;
-
- /* Handle SupplementaryGroups= if it is not empty */
- if (ngids > 0) {
- r = maybe_setgroups(ngids, supplementary_gids);
- if (r < 0)
- return r;
- }
-
- if (gid_is_valid(gid)) {
- /* Then set our gids */
- if (setresgid(gid, gid, gid) < 0)
- return -errno;
- }
-
- return 0;
-}
-
-static int set_securebits(unsigned bits, unsigned mask) {
- unsigned applied;
- int current;
-
- current = prctl(PR_GET_SECUREBITS);
- if (current < 0)
- return -errno;
-
- /* Clear all securebits defined in mask and set bits */
- applied = ((unsigned) current & ~mask) | bits;
- if ((unsigned) current == applied)
- return 0;
-
- if (prctl(PR_SET_SECUREBITS, applied) < 0)
- return -errno;
-
- return 1;
-}
-
-static int enforce_user(
- const ExecContext *context,
- uid_t uid,
- uint64_t capability_ambient_set) {
- assert(context);
- int r;
-
- if (!uid_is_valid(uid))
- return 0;
-
- /* Sets (but doesn't look up) the UIS and makes sure we keep the capabilities while doing so. For
- * setting secure bits the capability CAP_SETPCAP is required, so we also need keep-caps in this
- * case. */
-
- if ((capability_ambient_set != 0 || context->secure_bits != 0) && uid != 0) {
-
- /* First step: If we need to keep capabilities but drop privileges we need to make sure we
- * keep our caps, while we drop privileges. Add KEEP_CAPS to the securebits */
- r = set_securebits(1U << SECURE_KEEP_CAPS, 0);
- if (r < 0)
- return r;
- }
-
- /* Second step: actually set the uids */
- if (setresuid(uid, uid, uid) < 0)
- return -errno;
-
- /* At this point we should have all necessary capabilities but are otherwise a normal user. However,
- * the caps might got corrupted due to the setresuid() so we need clean them up later. This is done
- * outside of this call. */
- return 0;
-}
-
-#if HAVE_PAM
-
-static int null_conv(
- int num_msg,
- const struct pam_message **msg,
- struct pam_response **resp,
- void *appdata_ptr) {
-
- /* We don't support conversations */
-
- return PAM_CONV_ERR;
-}
-
-#endif
-
-static int setup_pam(
- const char *name,
- const char *user,
- uid_t uid,
- gid_t gid,
- const char *tty,
- char ***env, /* updated on success */
- const int fds[], size_t n_fds) {
-
-#if HAVE_PAM
-
- static const struct pam_conv conv = {
- .conv = null_conv,
- .appdata_ptr = NULL
- };
-
- _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
- _cleanup_strv_free_ char **e = NULL;
- pam_handle_t *handle = NULL;
- sigset_t old_ss;
- int pam_code = PAM_SUCCESS, r;
- bool close_session = false;
- pid_t pam_pid = 0, parent_pid;
- int flags = 0;
-
- assert(name);
- assert(user);
- assert(env);
-
- /* We set up PAM in the parent process, then fork. The child
- * will then stay around until killed via PR_GET_PDEATHSIG or
- * systemd via the cgroup logic. It will then remove the PAM
- * session again. The parent process will exec() the actual
- * daemon. We do things this way to ensure that the main PID
- * of the daemon is the one we initially fork()ed. */
-
- r = barrier_create(&barrier);
- if (r < 0)
- goto fail;
-
- if (log_get_max_level() < LOG_DEBUG)
- flags |= PAM_SILENT;
-
- pam_code = pam_start(name, user, &conv, &handle);
- if (pam_code != PAM_SUCCESS) {
- handle = NULL;
- goto fail;
- }
-
- if (!tty) {
- _cleanup_free_ char *q = NULL;
-
- /* Hmm, so no TTY was explicitly passed, but an fd passed to us directly might be a TTY. Let's figure
- * out if that's the case, and read the TTY off it. */
-
- if (getttyname_malloc(STDIN_FILENO, &q) >= 0)
- tty = strjoina("/dev/", q);
- }
-
- if (tty) {
- pam_code = pam_set_item(handle, PAM_TTY, tty);
- if (pam_code != PAM_SUCCESS)
- goto fail;
- }
-
- STRV_FOREACH(nv, *env) {
- pam_code = pam_putenv(handle, *nv);
- if (pam_code != PAM_SUCCESS)
- goto fail;
- }
-
- pam_code = pam_acct_mgmt(handle, flags);
- if (pam_code != PAM_SUCCESS)
- goto fail;
-
- pam_code = pam_setcred(handle, PAM_ESTABLISH_CRED | flags);
- if (pam_code != PAM_SUCCESS)
- log_debug("pam_setcred() failed, ignoring: %s", pam_strerror(handle, pam_code));
-
- pam_code = pam_open_session(handle, flags);
- if (pam_code != PAM_SUCCESS)
- goto fail;
-
- close_session = true;
-
- e = pam_getenvlist(handle);
- if (!e) {
- pam_code = PAM_BUF_ERR;
- goto fail;
- }
-
- /* Block SIGTERM, so that we know that it won't get lost in the child */
-
- assert_se(sigprocmask_many(SIG_BLOCK, &old_ss, SIGTERM, -1) >= 0);
-
- parent_pid = getpid_cached();
-
- r = safe_fork("(sd-pam)", 0, &pam_pid);
- if (r < 0)
- goto fail;
- if (r == 0) {
- int sig, ret = EXIT_PAM;
-
- /* The child's job is to reset the PAM session on termination */
- barrier_set_role(&barrier, BARRIER_CHILD);
-
- /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only
- * those fds are open here that have been opened by PAM. */
- (void) close_many(fds, n_fds);
-
- /* Drop privileges - we don't need any to pam_close_session and this will make
- * PR_SET_PDEATHSIG work in most cases. If this fails, ignore the error - but expect sd-pam
- * threads to fail to exit normally */
-
- r = maybe_setgroups(0, NULL);
- if (r < 0)
- log_warning_errno(r, "Failed to setgroups() in sd-pam: %m");
- if (setresgid(gid, gid, gid) < 0)
- log_warning_errno(errno, "Failed to setresgid() in sd-pam: %m");
- if (setresuid(uid, uid, uid) < 0)
- log_warning_errno(errno, "Failed to setresuid() in sd-pam: %m");
-
- (void) ignore_signals(SIGPIPE);
-
- /* Wait until our parent died. This will only work if the above setresuid() succeeds,
- * otherwise the kernel will not allow unprivileged parents kill their privileged children
- * this way. We rely on the control groups kill logic to do the rest for us. */
- if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0)
- goto child_finish;
-
- /* Tell the parent that our setup is done. This is especially important regarding dropping
- * privileges. Otherwise, unit setup might race against our setresuid(2) call.
- *
- * If the parent aborted, we'll detect this below, hence ignore return failure here. */
- (void) barrier_place(&barrier);
-
- /* Check if our parent process might already have died? */
- if (getppid() == parent_pid) {
- sigset_t ss;
-
- assert_se(sigemptyset(&ss) >= 0);
- assert_se(sigaddset(&ss, SIGTERM) >= 0);
-
- for (;;) {
- if (sigwait(&ss, &sig) < 0) {
- if (errno == EINTR)
- continue;
-
- goto child_finish;
- }
-
- assert(sig == SIGTERM);
- break;
- }
- }
-
- pam_code = pam_setcred(handle, PAM_DELETE_CRED | flags);
- if (pam_code != PAM_SUCCESS)
- goto child_finish;
-
- /* If our parent died we'll end the session */
- if (getppid() != parent_pid) {
- pam_code = pam_close_session(handle, flags);
- if (pam_code != PAM_SUCCESS)
- goto child_finish;
- }
-
- ret = 0;
-
- child_finish:
- /* NB: pam_end() when called in child processes should set PAM_DATA_SILENT to let the module
- * know about this. See pam_end(3) */
- (void) pam_end(handle, pam_code | flags | PAM_DATA_SILENT);
- _exit(ret);
- }
-
- barrier_set_role(&barrier, BARRIER_PARENT);
-
- /* If the child was forked off successfully it will do all the cleanups, so forget about the handle
- * here. */
- handle = NULL;
-
- /* Unblock SIGTERM again in the parent */
- assert_se(sigprocmask(SIG_SETMASK, &old_ss, NULL) >= 0);
-
- /* We close the log explicitly here, since the PAM modules might have opened it, but we don't want
- * this fd around. */
- closelog();
-
- /* Synchronously wait for the child to initialize. We don't care for errors as we cannot
- * recover. However, warn loudly if it happens. */
- if (!barrier_place_and_sync(&barrier))
- log_error("PAM initialization failed");
-
- return strv_free_and_replace(*env, e);
-
-fail:
- if (pam_code != PAM_SUCCESS) {
- log_error("PAM failed: %s", pam_strerror(handle, pam_code));
- r = -EPERM; /* PAM errors do not map to errno */
- } else
- log_error_errno(r, "PAM failed: %m");
-
- if (handle) {
- if (close_session)
- pam_code = pam_close_session(handle, flags);
-
- (void) pam_end(handle, pam_code | flags);
- }
-
- closelog();
- return r;
-#else
- return 0;
-#endif
-}
-
-static void rename_process_from_path(const char *path) {
- _cleanup_free_ char *buf = NULL;
- const char *p;
-
- assert(path);
-
- /* This resulting string must fit in 10 chars (i.e. the length of "/sbin/init") to look pretty in
- * /bin/ps */
-
- if (path_extract_filename(path, &buf) < 0) {
- rename_process("(...)");
- return;
- }
-
- size_t l = strlen(buf);
- if (l > 8) {
- /* The end of the process name is usually more interesting, since the first bit might just be
- * "systemd-" */
- p = buf + l - 8;
- l = 8;
- } else
- p = buf;
-
- char process_name[11];
- process_name[0] = '(';
- memcpy(process_name+1, p, l);
- process_name[1+l] = ')';
- process_name[1+l+1] = 0;
-
- rename_process(process_name);
-}
-
-static bool context_has_address_families(const ExecContext *c) {
- assert(c);
-
- return c->address_families_allow_list ||
- !set_isempty(c->address_families);
-}
-
-static bool context_has_syscall_filters(const ExecContext *c) {
- assert(c);
-
- return c->syscall_allow_list ||
- !hashmap_isempty(c->syscall_filter);
-}
-
-static bool context_has_syscall_logs(const ExecContext *c) {
- assert(c);
-
- return c->syscall_log_allow_list ||
- !hashmap_isempty(c->syscall_log);
-}
-
-static bool context_has_no_new_privileges(const ExecContext *c) {
- assert(c);
-
- if (c->no_new_privileges)
- return true;
-
- if (have_effective_cap(CAP_SYS_ADMIN) > 0) /* if we are privileged, we don't need NNP */
- return false;
-
- /* We need NNP if we have any form of seccomp and are unprivileged */
- return c->lock_personality ||
- c->memory_deny_write_execute ||
- c->private_devices ||
- c->protect_clock ||
- c->protect_hostname ||
- c->protect_kernel_tunables ||
- c->protect_kernel_modules ||
- c->protect_kernel_logs ||
- context_has_address_families(c) ||
- exec_context_restrict_namespaces_set(c) ||
- c->restrict_realtime ||
- c->restrict_suid_sgid ||
- !set_isempty(c->syscall_archs) ||
- context_has_syscall_filters(c) ||
- context_has_syscall_logs(c);
-}
-
-#if HAVE_SECCOMP
-
-static bool skip_seccomp_unavailable(const Unit* u, const char* msg) {
-
- if (is_seccomp_available())
- return false;
-
- log_unit_debug(u, "SECCOMP features not detected in the kernel, skipping %s", msg);
- return true;
-}
-
-static int apply_syscall_filter(const Unit* u, const ExecContext *c, bool needs_ambient_hack) {
- uint32_t negative_action, default_action, action;
- int r;
-
- assert(u);
- assert(c);
-
- if (!context_has_syscall_filters(c))
- return 0;
-
- if (skip_seccomp_unavailable(u, "SystemCallFilter="))
- return 0;
-
- negative_action = c->syscall_errno == SECCOMP_ERROR_NUMBER_KILL ? scmp_act_kill_process() : SCMP_ACT_ERRNO(c->syscall_errno);
-
- if (c->syscall_allow_list) {
- default_action = negative_action;
- action = SCMP_ACT_ALLOW;
- } else {
- default_action = SCMP_ACT_ALLOW;
- action = negative_action;
- }
-
- if (needs_ambient_hack) {
- r = seccomp_filter_set_add(c->syscall_filter, c->syscall_allow_list, syscall_filter_sets + SYSCALL_FILTER_SET_SETUID);
- if (r < 0)
- return r;
- }
-
- return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_filter, action, false);
-}
-
-static int apply_syscall_log(const Unit* u, const ExecContext *c) {
-#ifdef SCMP_ACT_LOG
- uint32_t default_action, action;
-#endif
-
- assert(u);
- assert(c);
-
- if (!context_has_syscall_logs(c))
- return 0;
-
-#ifdef SCMP_ACT_LOG
- if (skip_seccomp_unavailable(u, "SystemCallLog="))
- return 0;
-
- if (c->syscall_log_allow_list) {
- /* Log nothing but the ones listed */
- default_action = SCMP_ACT_ALLOW;
- action = SCMP_ACT_LOG;
- } else {
- /* Log everything but the ones listed */
- default_action = SCMP_ACT_LOG;
- action = SCMP_ACT_ALLOW;
- }
-
- return seccomp_load_syscall_filter_set_raw(default_action, c->syscall_log, action, false);
-#else
- /* old libseccomp */
- log_unit_debug(u, "SECCOMP feature SCMP_ACT_LOG not available, skipping SystemCallLog=");
- return 0;
-#endif
-}
-
-static int apply_syscall_archs(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (set_isempty(c->syscall_archs))
- return 0;
-
- if (skip_seccomp_unavailable(u, "SystemCallArchitectures="))
- return 0;
-
- return seccomp_restrict_archs(c->syscall_archs);
-}
-
-static int apply_address_families(const Unit* u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!context_has_address_families(c))
- return 0;
-
- if (skip_seccomp_unavailable(u, "RestrictAddressFamilies="))
- return 0;
-
- return seccomp_restrict_address_families(c->address_families, c->address_families_allow_list);
-}
-
-static int apply_memory_deny_write_execute(const Unit* u, const ExecContext *c) {
- int r;
-
- assert(u);
- assert(c);
-
- if (!c->memory_deny_write_execute)
- return 0;
-
- /* use prctl() if kernel supports it (6.3) */
- r = prctl(PR_SET_MDWE, PR_MDWE_REFUSE_EXEC_GAIN, 0, 0, 0);
- if (r == 0) {
- log_unit_debug(u, "Enabled MemoryDenyWriteExecute= with PR_SET_MDWE");
- return 0;
- }
- if (r < 0 && errno != EINVAL)
- return log_unit_debug_errno(u, errno, "Failed to enable MemoryDenyWriteExecute= with PR_SET_MDWE: %m");
- /* else use seccomp */
- log_unit_debug(u, "Kernel doesn't support PR_SET_MDWE: falling back to seccomp");
-
- if (skip_seccomp_unavailable(u, "MemoryDenyWriteExecute="))
- return 0;
-
- return seccomp_memory_deny_write_execute();
-}
-
-static int apply_restrict_realtime(const Unit* u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!c->restrict_realtime)
- return 0;
-
- if (skip_seccomp_unavailable(u, "RestrictRealtime="))
- return 0;
-
- return seccomp_restrict_realtime();
-}
-
-static int apply_restrict_suid_sgid(const Unit* u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!c->restrict_suid_sgid)
- return 0;
-
- if (skip_seccomp_unavailable(u, "RestrictSUIDSGID="))
- return 0;
-
- return seccomp_restrict_suid_sgid();
-}
-
-static int apply_protect_sysctl(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
- * let's protect even those systems where this is left on in the kernel. */
-
- if (!c->protect_kernel_tunables)
- return 0;
-
- if (skip_seccomp_unavailable(u, "ProtectKernelTunables="))
- return 0;
-
- return seccomp_protect_sysctl();
-}
-
-static int apply_protect_kernel_modules(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- /* Turn off module syscalls on ProtectKernelModules=yes */
-
- if (!c->protect_kernel_modules)
- return 0;
-
- if (skip_seccomp_unavailable(u, "ProtectKernelModules="))
- return 0;
-
- return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_MODULE, SCMP_ACT_ERRNO(EPERM), false);
-}
-
-static int apply_protect_kernel_logs(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!c->protect_kernel_logs)
- return 0;
-
- if (skip_seccomp_unavailable(u, "ProtectKernelLogs="))
- return 0;
-
- return seccomp_protect_syslog();
-}
-
-static int apply_protect_clock(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!c->protect_clock)
- return 0;
-
- if (skip_seccomp_unavailable(u, "ProtectClock="))
- return 0;
-
- return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_CLOCK, SCMP_ACT_ERRNO(EPERM), false);
-}
-
-static int apply_private_devices(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
-
- if (!c->private_devices)
- return 0;
-
- if (skip_seccomp_unavailable(u, "PrivateDevices="))
- return 0;
-
- return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO, SCMP_ACT_ERRNO(EPERM), false);
-}
-
-static int apply_restrict_namespaces(const Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!exec_context_restrict_namespaces_set(c))
- return 0;
-
- if (skip_seccomp_unavailable(u, "RestrictNamespaces="))
- return 0;
-
- return seccomp_restrict_namespaces(c->restrict_namespaces);
-}
-
-static int apply_lock_personality(const Unit* u, const ExecContext *c) {
- unsigned long personality;
- int r;
-
- assert(u);
- assert(c);
-
- if (!c->lock_personality)
- return 0;
-
- if (skip_seccomp_unavailable(u, "LockPersonality="))
- return 0;
-
- personality = c->personality;
-
- /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
- if (personality == PERSONALITY_INVALID) {
-
- r = opinionated_personality(&personality);
- if (r < 0)
- return r;
- }
-
- return seccomp_lock_personality(personality);
-}
-
-#endif
-
-#if HAVE_LIBBPF
-static int apply_restrict_filesystems(Unit *u, const ExecContext *c) {
- assert(u);
- assert(c);
-
- if (!exec_context_restrict_filesystems_set(c))
- return 0;
-
- if (!u->manager->restrict_fs) {
- /* LSM BPF is unsupported or lsm_bpf_setup failed */
- log_unit_debug(u, "LSM BPF not supported, skipping RestrictFileSystems=");
- return 0;
- }
-
- return lsm_bpf_unit_restrict_filesystems(u, c->restrict_filesystems, c->restrict_filesystems_allow_list);
-}
-#endif
-
-static int apply_protect_hostname(const Unit *u, const ExecContext *c, int *ret_exit_status) {
- assert(u);
- assert(c);
-
- if (!c->protect_hostname)
- return 0;
-
- if (ns_type_supported(NAMESPACE_UTS)) {
- if (unshare(CLONE_NEWUTS) < 0) {
- if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) {
- *ret_exit_status = EXIT_NAMESPACE;
- return log_unit_error_errno(u, errno, "Failed to set up UTS namespacing: %m");
- }
-
- log_unit_warning(u, "ProtectHostname=yes is configured, but UTS namespace setup is prohibited (container manager?), ignoring namespace setup.");
- }
- } else
- log_unit_warning(u, "ProtectHostname=yes is configured, but the kernel does not support UTS namespaces, ignoring namespace setup.");
-
-#if HAVE_SECCOMP
- int r;
-
- if (skip_seccomp_unavailable(u, "ProtectHostname="))
- return 0;
-
- r = seccomp_protect_hostname();
- if (r < 0) {
- *ret_exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(u, r, "Failed to apply hostname restrictions: %m");
- }
-#endif
-
- return 0;
-}
-
-static void do_idle_pipe_dance(int idle_pipe[static 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. */
- (void) 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 const char *exec_directory_env_name_to_string(ExecDirectoryType t);
-
-static int build_environment(
- const Unit *u,
- const ExecContext *c,
- const ExecParameters *p,
- const CGroupContext *cgroup_context,
- size_t n_fds,
- char **fdnames,
- const char *home,
- const char *username,
- const char *shell,
- dev_t journal_stream_dev,
- ino_t journal_stream_ino,
- const char *memory_pressure_path,
- const char *creds_path,
- char ***ret) {
-
- _cleanup_strv_free_ char **our_env = NULL;
- size_t n_env = 0;
- char *x;
- int r;
-
- assert(u);
- assert(c);
- assert(p);
- assert(ret);
-
-#define N_ENV_VARS 19
- our_env = new0(char*, N_ENV_VARS + _EXEC_DIRECTORY_TYPE_MAX);
- if (!our_env)
- return -ENOMEM;
-
- if (n_fds > 0) {
- _cleanup_free_ char *joined = NULL;
-
- if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid_cached()) < 0)
- return -ENOMEM;
- our_env[n_env++] = x;
-
- if (asprintf(&x, "LISTEN_FDS=%zu", n_fds) < 0)
- return -ENOMEM;
- our_env[n_env++] = x;
-
- joined = strv_join(fdnames, ":");
- if (!joined)
- return -ENOMEM;
-
- x = strjoin("LISTEN_FDNAMES=", joined);
- if (!x)
- return -ENOMEM;
- our_env[n_env++] = x;
- }
-
- if ((p->flags & EXEC_SET_WATCHDOG) && p->watchdog_usec > 0) {
- if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid_cached()) < 0)
- return -ENOMEM;
- our_env[n_env++] = x;
-
- if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, p->watchdog_usec) < 0)
- return -ENOMEM;
- our_env[n_env++] = x;
- }
-
- /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use blocking
- * Varlink calls back to us for look up dynamic users in PID 1. Break the deadlock between D-Bus and
- * PID 1 by disabling use of PID1' NSS interface for looking up dynamic users. */
- if (p->flags & EXEC_NSS_DYNAMIC_BYPASS) {
- x = strdup("SYSTEMD_NSS_DYNAMIC_BYPASS=1");
- if (!x)
- return -ENOMEM;
- our_env[n_env++] = x;
- }
-
- if (home) {
- x = strjoin("HOME=", home);
- if (!x)
- return -ENOMEM;
-
- path_simplify(x + 5);
- our_env[n_env++] = x;
- }
-
- if (username) {
- x = strjoin("LOGNAME=", username);
- if (!x)
- return -ENOMEM;
- our_env[n_env++] = x;
-
- x = strjoin("USER=", username);
- if (!x)
- return -ENOMEM;
- our_env[n_env++] = x;
- }
-
- if (shell) {
- x = strjoin("SHELL=", shell);
- if (!x)
- return -ENOMEM;
-
- path_simplify(x + 6);
- our_env[n_env++] = x;
- }
-
- if (!sd_id128_is_null(u->invocation_id)) {
- if (asprintf(&x, "INVOCATION_ID=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id)) < 0)
- return -ENOMEM;
-
- our_env[n_env++] = x;
- }
-
- if (exec_context_needs_term(c)) {
- _cleanup_free_ char *cmdline = NULL;
- const char *tty_path, *term = NULL;
-
- tty_path = exec_context_tty_path(c);
-
- /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try
- * to inherit the $TERM set for PID 1. This is useful for containers so that the $TERM the
- * container manager passes to PID 1 ends up all the way in the console login shown. */
-
- if (path_equal_ptr(tty_path, "/dev/console") && getppid() == 1)
- term = getenv("TERM");
- else if (tty_path && in_charset(skip_dev_prefix(tty_path), ALPHANUMERICAL)) {
- _cleanup_free_ char *key = NULL;
-
- key = strjoin("systemd.tty.term.", skip_dev_prefix(tty_path));
- if (!key)
- return -ENOMEM;
-
- r = proc_cmdline_get_key(key, 0, &cmdline);
- if (r < 0)
- log_debug_errno(r, "Failed to read %s from kernel cmdline, ignoring: %m", key);
- else if (r > 0)
- term = cmdline;
- }
-
- if (!term)
- term = default_term_for_tty(tty_path);
-
- x = strjoin("TERM=", term);
- if (!x)
- return -ENOMEM;
- our_env[n_env++] = x;
- }
-
- if (journal_stream_dev != 0 && journal_stream_ino != 0) {
- if (asprintf(&x, "JOURNAL_STREAM=" DEV_FMT ":" INO_FMT, journal_stream_dev, journal_stream_ino) < 0)
- return -ENOMEM;
-
- our_env[n_env++] = x;
- }
-
- if (c->log_namespace) {
- x = strjoin("LOG_NAMESPACE=", c->log_namespace);
- if (!x)
- return -ENOMEM;
-
- our_env[n_env++] = x;
- }
-
- for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
- _cleanup_free_ char *joined = NULL;
- const char *n;
-
- if (!p->prefix[t])
- continue;
-
- if (c->directories[t].n_items == 0)
- continue;
-
- n = exec_directory_env_name_to_string(t);
- if (!n)
- continue;
-
- for (size_t i = 0; i < c->directories[t].n_items; i++) {
- _cleanup_free_ char *prefixed = NULL;
-
- prefixed = path_join(p->prefix[t], c->directories[t].items[i].path);
- if (!prefixed)
- return -ENOMEM;
-
- if (!strextend_with_separator(&joined, ":", prefixed))
- return -ENOMEM;
- }
-
- x = strjoin(n, "=", joined);
- if (!x)
- return -ENOMEM;
-
- our_env[n_env++] = x;
- }
-
- if (creds_path) {
- x = strjoin("CREDENTIALS_DIRECTORY=", creds_path);
- if (!x)
- return -ENOMEM;
-
- our_env[n_env++] = x;
- }
-
- if (asprintf(&x, "SYSTEMD_EXEC_PID=" PID_FMT, getpid_cached()) < 0)
- return -ENOMEM;
-
- our_env[n_env++] = x;
-
- if (memory_pressure_path) {
- x = strjoin("MEMORY_PRESSURE_WATCH=", memory_pressure_path);
- if (!x)
- return -ENOMEM;
-
- our_env[n_env++] = x;
-
- if (cgroup_context && !path_equal(memory_pressure_path, "/dev/null")) {
- _cleanup_free_ char *b = NULL, *e = NULL;
-
- if (asprintf(&b, "%s " USEC_FMT " " USEC_FMT,
- MEMORY_PRESSURE_DEFAULT_TYPE,
- cgroup_context->memory_pressure_threshold_usec == USEC_INFINITY ? MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC :
- CLAMP(cgroup_context->memory_pressure_threshold_usec, 1U, MEMORY_PRESSURE_DEFAULT_WINDOW_USEC),
- MEMORY_PRESSURE_DEFAULT_WINDOW_USEC) < 0)
- return -ENOMEM;
-
- if (base64mem(b, strlen(b) + 1, &e) < 0)
- return -ENOMEM;
-
- x = strjoin("MEMORY_PRESSURE_WRITE=", e);
- if (!x)
- return -ENOMEM;
-
- our_env[n_env++] = x;
- }
- }
-
- assert(n_env < N_ENV_VARS + _EXEC_DIRECTORY_TYPE_MAX);
-#undef N_ENV_VARS
-
- *ret = TAKE_PTR(our_env);
-
- return 0;
-}
-
-static int build_pass_environment(const ExecContext *c, char ***ret) {
- _cleanup_strv_free_ char **pass_env = NULL;
- size_t n_env = 0;
-
- STRV_FOREACH(i, c->pass_environment) {
- _cleanup_free_ char *x = NULL;
- char *v;
-
- v = getenv(*i);
- if (!v)
- continue;
- x = strjoin(*i, "=", v);
- if (!x)
- return -ENOMEM;
-
- if (!GREEDY_REALLOC(pass_env, n_env + 2))
- return -ENOMEM;
-
- pass_env[n_env++] = TAKE_PTR(x);
- pass_env[n_env] = NULL;
- }
-
- *ret = TAKE_PTR(pass_env);
-
- return 0;
-}
-
-bool exec_needs_network_namespace(const ExecContext *context) {
- assert(context);
-
- return context->private_network || context->network_namespace_path;
-}
-
-static bool exec_needs_ephemeral(const ExecContext *context) {
- return (context->root_image || context->root_directory) && context->root_ephemeral;
-}
-
-static bool exec_needs_ipc_namespace(const ExecContext *context) {
- assert(context);
-
- return context->private_ipc || context->ipc_namespace_path;
-}
-
-bool exec_needs_mount_namespace(
- const ExecContext *context,
- const ExecParameters *params,
- const ExecRuntime *runtime) {
-
- assert(context);
-
- if (context->root_image)
- return true;
-
- if (!strv_isempty(context->read_write_paths) ||
- !strv_isempty(context->read_only_paths) ||
- !strv_isempty(context->inaccessible_paths) ||
- !strv_isempty(context->exec_paths) ||
- !strv_isempty(context->no_exec_paths))
- return true;
-
- if (context->n_bind_mounts > 0)
- return true;
-
- if (context->n_temporary_filesystems > 0)
- return true;
-
- if (context->n_mount_images > 0)
- return true;
-
- if (context->n_extension_images > 0)
- return true;
-
- if (!strv_isempty(context->extension_directories))
- return true;
-
- if (!IN_SET(context->mount_propagation_flag, 0, MS_SHARED))
- return true;
-
- if (context->private_tmp && runtime && runtime->shared && (runtime->shared->tmp_dir || runtime->shared->var_tmp_dir))
- return true;
-
- if (context->private_devices ||
- context->private_mounts > 0 ||
- (context->private_mounts < 0 && exec_needs_network_namespace(context)) ||
- context->protect_system != PROTECT_SYSTEM_NO ||
- context->protect_home != PROTECT_HOME_NO ||
- context->protect_kernel_tunables ||
- context->protect_kernel_modules ||
- context->protect_kernel_logs ||
- context->protect_control_groups ||
- context->protect_proc != PROTECT_PROC_DEFAULT ||
- context->proc_subset != PROC_SUBSET_ALL ||
- exec_needs_ipc_namespace(context))
- return true;
-
- if (context->root_directory) {
- if (exec_context_get_effective_mount_apivfs(context))
- return true;
-
- for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
- if (params && !params->prefix[t])
- continue;
-
- if (context->directories[t].n_items > 0)
- return true;
- }
- }
-
- if (context->dynamic_user &&
- (context->directories[EXEC_DIRECTORY_STATE].n_items > 0 ||
- context->directories[EXEC_DIRECTORY_CACHE].n_items > 0 ||
- context->directories[EXEC_DIRECTORY_LOGS].n_items > 0))
- return true;
-
- if (context->log_namespace)
- return true;
-
- return false;
-}
-
-static int setup_private_users(uid_t ouid, gid_t ogid, uid_t uid, gid_t gid) {
- _cleanup_free_ char *uid_map = NULL, *gid_map = NULL;
- _cleanup_close_pair_ int errno_pipe[2] = PIPE_EBADF;
- _cleanup_close_ int unshare_ready_fd = -EBADF;
- _cleanup_(sigkill_waitp) pid_t pid = 0;
- uint64_t c = 1;
- ssize_t n;
- int r;
-
- /* Set up a user namespace and map the original UID/GID (IDs from before any user or group changes, i.e.
- * the IDs from the user or system manager(s)) to itself, the selected UID/GID to itself, and everything else to
- * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
- * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
- * which waits for the parent to create the new user namespace while staying in the original namespace. The
- * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
- * continues execution normally.
- * For unprivileged users (i.e. without capabilities), the root to root mapping is excluded. As such, it
- * does not need CAP_SETUID to write the single line mapping to itself. */
-
- /* Can only set up multiple mappings with CAP_SETUID. */
- if (have_effective_cap(CAP_SETUID) > 0 && uid != ouid && uid_is_valid(uid))
- r = asprintf(&uid_map,
- UID_FMT " " UID_FMT " 1\n" /* Map $OUID → $OUID */
- UID_FMT " " UID_FMT " 1\n", /* Map $UID → $UID */
- ouid, ouid, uid, uid);
- else
- r = asprintf(&uid_map,
- UID_FMT " " UID_FMT " 1\n", /* Map $OUID → $OUID */
- ouid, ouid);
-
- if (r < 0)
- return -ENOMEM;
-
- /* Can only set up multiple mappings with CAP_SETGID. */
- if (have_effective_cap(CAP_SETGID) > 0 && gid != ogid && gid_is_valid(gid))
- r = asprintf(&gid_map,
- GID_FMT " " GID_FMT " 1\n" /* Map $OGID → $OGID */
- GID_FMT " " GID_FMT " 1\n", /* Map $GID → $GID */
- ogid, ogid, gid, gid);
- else
- r = asprintf(&gid_map,
- GID_FMT " " GID_FMT " 1\n", /* Map $OGID -> $OGID */
- ogid, ogid);
-
- if (r < 0)
- return -ENOMEM;
-
- /* Create a communication channel so that the parent can tell the child when it finished creating the user
- * namespace. */
- unshare_ready_fd = eventfd(0, EFD_CLOEXEC);
- if (unshare_ready_fd < 0)
- return -errno;
-
- /* Create a communication channel so that the child can tell the parent a proper error code in case it
- * failed. */
- if (pipe2(errno_pipe, O_CLOEXEC) < 0)
- return -errno;
-
- r = safe_fork("(sd-userns)", FORK_RESET_SIGNALS|FORK_DEATHSIG, &pid);
- if (r < 0)
- return r;
- if (r == 0) {
- _cleanup_close_ int fd = -EBADF;
- const char *a;
- pid_t ppid;
-
- /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
- * here, after the parent opened its own user namespace. */
-
- ppid = getppid();
- errno_pipe[0] = safe_close(errno_pipe[0]);
-
- /* Wait until the parent unshared the user namespace */
- if (read(unshare_ready_fd, &c, sizeof(c)) < 0) {
- r = -errno;
- goto child_fail;
- }
-
- /* Disable the setgroups() system call in the child user namespace, for good. */
- a = procfs_file_alloca(ppid, "setgroups");
- fd = open(a, O_WRONLY|O_CLOEXEC);
- if (fd < 0) {
- if (errno != ENOENT) {
- r = -errno;
- goto child_fail;
- }
-
- /* If the file is missing the kernel is too old, let's continue anyway. */
- } else {
- if (write(fd, "deny\n", 5) < 0) {
- r = -errno;
- goto child_fail;
- }
-
- fd = safe_close(fd);
- }
-
- /* First write the GID map */
- a = procfs_file_alloca(ppid, "gid_map");
- fd = open(a, O_WRONLY|O_CLOEXEC);
- if (fd < 0) {
- r = -errno;
- goto child_fail;
- }
- if (write(fd, gid_map, strlen(gid_map)) < 0) {
- r = -errno;
- goto child_fail;
- }
- fd = safe_close(fd);
-
- /* The write the UID map */
- a = procfs_file_alloca(ppid, "uid_map");
- fd = open(a, O_WRONLY|O_CLOEXEC);
- if (fd < 0) {
- r = -errno;
- goto child_fail;
- }
- if (write(fd, uid_map, strlen(uid_map)) < 0) {
- r = -errno;
- goto child_fail;
- }
-
- _exit(EXIT_SUCCESS);
-
- child_fail:
- (void) write(errno_pipe[1], &r, sizeof(r));
- _exit(EXIT_FAILURE);
- }
-
- errno_pipe[1] = safe_close(errno_pipe[1]);
-
- if (unshare(CLONE_NEWUSER) < 0)
- return -errno;
-
- /* Let the child know that the namespace is ready now */
- if (write(unshare_ready_fd, &c, sizeof(c)) < 0)
- return -errno;
-
- /* Try to read an error code from the child */
- n = read(errno_pipe[0], &r, sizeof(r));
- if (n < 0)
- return -errno;
- if (n == sizeof(r)) { /* an error code was sent to us */
- if (r < 0)
- return r;
- return -EIO;
- }
- if (n != 0) /* on success we should have read 0 bytes */
- return -EIO;
-
- r = wait_for_terminate_and_check("(sd-userns)", TAKE_PID(pid), 0);
- if (r < 0)
- return r;
- if (r != EXIT_SUCCESS) /* If something strange happened with the child, let's consider this fatal, too */
- return -EIO;
-
- return 0;
-}
-
-static bool exec_directory_is_private(const ExecContext *context, ExecDirectoryType type) {
- assert(context);
-
- if (!context->dynamic_user)
- return false;
-
- if (type == EXEC_DIRECTORY_CONFIGURATION)
- return false;
-
- if (type == EXEC_DIRECTORY_RUNTIME && context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO)
- return false;
-
- return true;
-}
-
-static int create_many_symlinks(const char *root, const char *source, char **symlinks) {
- _cleanup_free_ char *src_abs = NULL;
- int r;
-
- assert(source);
-
- src_abs = path_join(root, source);
- if (!src_abs)
- return -ENOMEM;
-
- STRV_FOREACH(dst, symlinks) {
- _cleanup_free_ char *dst_abs = NULL;
-
- dst_abs = path_join(root, *dst);
- if (!dst_abs)
- return -ENOMEM;
-
- r = mkdir_parents_label(dst_abs, 0755);
- if (r < 0)
- return r;
-
- r = symlink_idempotent(src_abs, dst_abs, true);
- if (r < 0)
- return r;
- }
-
- return 0;
-}
-
-static int setup_exec_directory(
- Unit *u,
- const ExecContext *context,
- const ExecParameters *params,
- uid_t uid,
- gid_t gid,
- ExecDirectoryType type,
- bool needs_mount_namespace,
- int *exit_status) {
-
- static const int exit_status_table[_EXEC_DIRECTORY_TYPE_MAX] = {
- [EXEC_DIRECTORY_RUNTIME] = EXIT_RUNTIME_DIRECTORY,
- [EXEC_DIRECTORY_STATE] = EXIT_STATE_DIRECTORY,
- [EXEC_DIRECTORY_CACHE] = EXIT_CACHE_DIRECTORY,
- [EXEC_DIRECTORY_LOGS] = EXIT_LOGS_DIRECTORY,
- [EXEC_DIRECTORY_CONFIGURATION] = EXIT_CONFIGURATION_DIRECTORY,
- };
- int r;
-
- assert(context);
- assert(params);
- assert(type >= 0 && type < _EXEC_DIRECTORY_TYPE_MAX);
- assert(exit_status);
-
- if (!params->prefix[type])
- return 0;
-
- if (params->flags & EXEC_CHOWN_DIRECTORIES) {
- if (!uid_is_valid(uid))
- uid = 0;
- if (!gid_is_valid(gid))
- gid = 0;
- }
-
- for (size_t i = 0; i < context->directories[type].n_items; i++) {
- _cleanup_free_ char *p = NULL, *pp = NULL;
-
- p = path_join(params->prefix[type], context->directories[type].items[i].path);
- if (!p) {
- r = -ENOMEM;
- goto fail;
- }
-
- r = mkdir_parents_label(p, 0755);
- if (r < 0)
- goto fail;
-
- if (IN_SET(type, EXEC_DIRECTORY_STATE, EXEC_DIRECTORY_LOGS) && params->runtime_scope == RUNTIME_SCOPE_USER) {
-
- /* If we are in user mode, and a configuration directory exists but a state directory
- * doesn't exist, then we likely are upgrading from an older systemd version that
- * didn't know the more recent addition to the xdg-basedir spec: the $XDG_STATE_HOME
- * directory. In older systemd versions EXEC_DIRECTORY_STATE was aliased to
- * EXEC_DIRECTORY_CONFIGURATION, with the advent of $XDG_STATE_HOME is is now
- * separated. If a service has both dirs configured but only the configuration dir
- * exists and the state dir does not, we assume we are looking at an update
- * situation. Hence, create a compatibility symlink, so that all expectations are
- * met.
- *
- * (We also do something similar with the log directory, which still doesn't exist in
- * the xdg basedir spec. We'll make it a subdir of the state dir.) */
-
- /* this assumes the state dir is always created before the configuration dir */
- assert_cc(EXEC_DIRECTORY_STATE < EXEC_DIRECTORY_LOGS);
- assert_cc(EXEC_DIRECTORY_LOGS < EXEC_DIRECTORY_CONFIGURATION);
-
- r = laccess(p, F_OK);
- if (r == -ENOENT) {
- _cleanup_free_ char *q = NULL;
-
- /* OK, we know that the state dir does not exist. Let's see if the dir exists
- * under the configuration hierarchy. */
-
- if (type == EXEC_DIRECTORY_STATE)
- q = path_join(params->prefix[EXEC_DIRECTORY_CONFIGURATION], context->directories[type].items[i].path);
- else if (type == EXEC_DIRECTORY_LOGS)
- q = path_join(params->prefix[EXEC_DIRECTORY_CONFIGURATION], "log", context->directories[type].items[i].path);
- else
- assert_not_reached();
- if (!q) {
- r = -ENOMEM;
- goto fail;
- }
-
- r = laccess(q, F_OK);
- if (r >= 0) {
- /* It does exist! This hence looks like an update. Symlink the
- * configuration directory into the state directory. */
-
- r = symlink_idempotent(q, p, /* make_relative= */ true);
- if (r < 0)
- goto fail;
-
- log_unit_notice(u, "Unit state directory %s missing but matching configuration directory %s exists, assuming update from systemd 253 or older, creating compatibility symlink.", p, q);
- continue;
- } else if (r != -ENOENT)
- log_unit_warning_errno(u, r, "Unable to detect whether unit configuration directory '%s' exists, assuming not: %m", q);
-
- } else if (r < 0)
- log_unit_warning_errno(u, r, "Unable to detect whether unit state directory '%s' is missing, assuming it is: %m", p);
- }
-
- if (exec_directory_is_private(context, type)) {
- /* So, here's one extra complication when dealing with DynamicUser=1 units. In that
- * case we want to avoid leaving a directory around fully accessible that is owned by
- * a dynamic user whose UID is later on reused. To lock this down we use the same
- * trick used by container managers to prohibit host users to get access to files of
- * the same UID in containers: we place everything inside a directory that has an
- * access mode of 0700 and is owned root:root, so that it acts as security boundary
- * for unprivileged host code. We then use fs namespacing to make this directory
- * permeable for the service itself.
- *
- * Specifically: for a service which wants a special directory "foo/" we first create
- * a directory "private/" with access mode 0700 owned by root:root. Then we place
- * "foo" inside of that directory (i.e. "private/foo/"), and make "foo" a symlink to
- * "private/foo". This way, privileged host users can access "foo/" as usual, but
- * unprivileged host users can't look into it. Inside of the namespace of the unit
- * "private/" is replaced by a more liberally accessible tmpfs, into which the host's
- * "private/foo/" is mounted under the same name, thus disabling the access boundary
- * for the service and making sure it only gets access to the dirs it needs but no
- * others. Tricky? Yes, absolutely, but it works!
- *
- * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not
- * to be owned by the service itself.
- *
- * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used
- * for sharing files or sockets with other services. */
-
- pp = path_join(params->prefix[type], "private");
- if (!pp) {
- r = -ENOMEM;
- goto fail;
- }
-
- /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
- r = mkdir_safe_label(pp, 0700, 0, 0, MKDIR_WARN_MODE);
- if (r < 0)
- goto fail;
-
- if (!path_extend(&pp, context->directories[type].items[i].path)) {
- r = -ENOMEM;
- goto fail;
- }
-
- /* Create all directories between the configured directory and this private root, and mark them 0755 */
- r = mkdir_parents_label(pp, 0755);
- if (r < 0)
- goto fail;
-
- if (is_dir(p, false) > 0 &&
- (laccess(pp, F_OK) == -ENOENT)) {
-
- /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
- * it over. Most likely the service has been upgraded from one that didn't use
- * DynamicUser=1, to one that does. */
-
- log_unit_info(u, "Found pre-existing public %s= directory %s, migrating to %s.\n"
- "Apparently, service previously had DynamicUser= turned off, and has now turned it on.",
- exec_directory_type_to_string(type), p, pp);
-
- r = RET_NERRNO(rename(p, pp));
- if (r < 0)
- goto fail;
- } else {
- /* Otherwise, create the actual directory for the service */
-
- r = mkdir_label(pp, context->directories[type].mode);
- if (r < 0 && r != -EEXIST)
- goto fail;
- }
-
- if (!context->directories[type].items[i].only_create) {
- /* And link it up from the original place.
- * Notes
- * 1) If a mount namespace is going to be used, then this symlink remains on
- * the host, and a new one for the child namespace will be created later.
- * 2) It is not necessary to create this symlink when one of its parent
- * directories is specified and already created. E.g.
- * StateDirectory=foo foo/bar
- * In that case, the inode points to pp and p for "foo/bar" are the same:
- * pp = "/var/lib/private/foo/bar"
- * p = "/var/lib/foo/bar"
- * and, /var/lib/foo is a symlink to /var/lib/private/foo. So, not only
- * we do not need to create the symlink, but we cannot create the symlink.
- * See issue #24783. */
- r = symlink_idempotent(pp, p, true);
- if (r < 0)
- goto fail;
- }
-
- } else {
- _cleanup_free_ char *target = NULL;
-
- if (type != EXEC_DIRECTORY_CONFIGURATION &&
- readlink_and_make_absolute(p, &target) >= 0) {
- _cleanup_free_ char *q = NULL, *q_resolved = NULL, *target_resolved = NULL;
-
- /* This already exists and is a symlink? Interesting. Maybe it's one created
- * by DynamicUser=1 (see above)?
- *
- * We do this for all directory types except for ConfigurationDirectory=,
- * since they all support the private/ symlink logic at least in some
- * configurations, see above. */
-
- r = chase(target, NULL, 0, &target_resolved, NULL);
- if (r < 0)
- goto fail;
-
- q = path_join(params->prefix[type], "private", context->directories[type].items[i].path);
- if (!q) {
- r = -ENOMEM;
- goto fail;
- }
-
- /* /var/lib or friends may be symlinks. So, let's chase them also. */
- r = chase(q, NULL, CHASE_NONEXISTENT, &q_resolved, NULL);
- if (r < 0)
- goto fail;
-
- if (path_equal(q_resolved, target_resolved)) {
-
- /* Hmm, apparently DynamicUser= was once turned on for this service,
- * but is no longer. Let's move the directory back up. */
-
- log_unit_info(u, "Found pre-existing private %s= directory %s, migrating to %s.\n"
- "Apparently, service previously had DynamicUser= turned on, and has now turned it off.",
- exec_directory_type_to_string(type), q, p);
-
- r = RET_NERRNO(unlink(p));
- if (r < 0)
- goto fail;
-
- r = RET_NERRNO(rename(q, p));
- if (r < 0)
- goto fail;
- }
- }
-
- r = mkdir_label(p, context->directories[type].mode);
- if (r < 0) {
- if (r != -EEXIST)
- goto fail;
-
- if (type == EXEC_DIRECTORY_CONFIGURATION) {
- struct stat st;
-
- /* Don't change the owner/access mode of the configuration directory,
- * as in the common case it is not written to by a service, and shall
- * not be writable. */
-
- r = RET_NERRNO(stat(p, &st));
- if (r < 0)
- goto fail;
-
- /* Still complain if the access mode doesn't match */
- if (((st.st_mode ^ context->directories[type].mode) & 07777) != 0)
- log_unit_warning(u, "%s \'%s\' already exists but the mode is different. "
- "(File system: %o %sMode: %o)",
- exec_directory_type_to_string(type), context->directories[type].items[i].path,
- st.st_mode & 07777, exec_directory_type_to_string(type), context->directories[type].mode & 07777);
-
- continue;
- }
- }
- }
-
- /* Lock down the access mode (we use chmod_and_chown() to make this idempotent. We don't
- * specify UID/GID here, so that path_chown_recursive() can optimize things depending on the
- * current UID/GID ownership.) */
- r = chmod_and_chown(pp ?: p, context->directories[type].mode, UID_INVALID, GID_INVALID);
- if (r < 0)
- goto fail;
-
- /* Skip the rest (which deals with ownership) in user mode, since ownership changes are not
- * available to user code anyway */
- if (params->runtime_scope != RUNTIME_SCOPE_SYSTEM)
- continue;
-
- /* Then, change the ownership of the whole tree, if necessary. When dynamic users are used we
- * drop the suid/sgid bits, since we really don't want SUID/SGID files for dynamic UID/GID
- * assignments to exist. */
- r = path_chown_recursive(pp ?: p, uid, gid, context->dynamic_user ? 01777 : 07777, AT_SYMLINK_FOLLOW);
- if (r < 0)
- goto fail;
- }
-
- /* If we are not going to run in a namespace, set up the symlinks - otherwise
- * they are set up later, to allow configuring empty var/run/etc. */
- if (!needs_mount_namespace)
- for (size_t i = 0; i < context->directories[type].n_items; i++) {
- r = create_many_symlinks(params->prefix[type],
- context->directories[type].items[i].path,
- context->directories[type].items[i].symlinks);
- if (r < 0)
- goto fail;
- }
-
- return 0;
-
-fail:
- *exit_status = exit_status_table[type];
- return r;
-}
-
-#if ENABLE_SMACK
-static int setup_smack(
- const Manager *manager,
- const ExecContext *context,
- int executable_fd) {
- int r;
-
- assert(context);
- assert(executable_fd >= 0);
-
- if (context->smack_process_label) {
- r = mac_smack_apply_pid(0, context->smack_process_label);
- if (r < 0)
- return r;
- } else if (manager->default_smack_process_label) {
- _cleanup_free_ char *exec_label = NULL;
-
- r = mac_smack_read_fd(executable_fd, SMACK_ATTR_EXEC, &exec_label);
- if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r))
- return r;
-
- r = mac_smack_apply_pid(0, exec_label ?: manager->default_smack_process_label);
- if (r < 0)
- return r;
- }
-
- return 0;
-}
-#endif
-
-static int compile_bind_mounts(
- const ExecContext *context,
- const ExecParameters *params,
- BindMount **ret_bind_mounts,
- size_t *ret_n_bind_mounts,
- char ***ret_empty_directories) {
-
- _cleanup_strv_free_ char **empty_directories = NULL;
- BindMount *bind_mounts = NULL;
- size_t n, h = 0;
- int r;
-
- assert(context);
- assert(params);
- assert(ret_bind_mounts);
- assert(ret_n_bind_mounts);
- assert(ret_empty_directories);
-
- CLEANUP_ARRAY(bind_mounts, h, bind_mount_free_many);
-
- n = context->n_bind_mounts;
- for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
- if (!params->prefix[t])
- continue;
-
- for (size_t i = 0; i < context->directories[t].n_items; i++)
- n += !context->directories[t].items[i].only_create;
- }
-
- if (n <= 0) {
- *ret_bind_mounts = NULL;
- *ret_n_bind_mounts = 0;
- *ret_empty_directories = NULL;
- return 0;
- }
-
- bind_mounts = new(BindMount, n);
- if (!bind_mounts)
- return -ENOMEM;
-
- for (size_t i = 0; i < context->n_bind_mounts; i++) {
- BindMount *item = context->bind_mounts + i;
- _cleanup_free_ char *s = NULL, *d = NULL;
-
- s = strdup(item->source);
- if (!s)
- return -ENOMEM;
-
- d = strdup(item->destination);
- if (!d)
- return -ENOMEM;
-
- bind_mounts[h++] = (BindMount) {
- .source = TAKE_PTR(s),
- .destination = TAKE_PTR(d),
- .read_only = item->read_only,
- .recursive = item->recursive,
- .ignore_enoent = item->ignore_enoent,
- };
- }
-
- for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
- if (!params->prefix[t])
- continue;
-
- if (context->directories[t].n_items == 0)
- continue;
-
- if (exec_directory_is_private(context, t) &&
- !exec_context_with_rootfs(context)) {
- char *private_root;
-
- /* So this is for a dynamic user, and we need to make sure the process can access its own
- * directory. For that we overmount the usually inaccessible "private" subdirectory with a
- * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
-
- private_root = path_join(params->prefix[t], "private");
- if (!private_root)
- return -ENOMEM;
-
- r = strv_consume(&empty_directories, private_root);
- if (r < 0)
- return r;
- }
-
- for (size_t i = 0; i < context->directories[t].n_items; i++) {
- _cleanup_free_ char *s = NULL, *d = NULL;
-
- /* When one of the parent directories is in the list, we cannot create the symlink
- * for the child directory. See also the comments in setup_exec_directory(). */
- if (context->directories[t].items[i].only_create)
- continue;
-
- if (exec_directory_is_private(context, t))
- s = path_join(params->prefix[t], "private", context->directories[t].items[i].path);
- else
- s = path_join(params->prefix[t], context->directories[t].items[i].path);
- if (!s)
- return -ENOMEM;
-
- if (exec_directory_is_private(context, t) &&
- exec_context_with_rootfs(context))
- /* When RootDirectory= or RootImage= are set, then the symbolic link to the private
- * directory is not created on the root directory. So, let's bind-mount the directory
- * on the 'non-private' place. */
- d = path_join(params->prefix[t], context->directories[t].items[i].path);
- else
- d = strdup(s);
- if (!d)
- return -ENOMEM;
-
- bind_mounts[h++] = (BindMount) {
- .source = TAKE_PTR(s),
- .destination = TAKE_PTR(d),
- .read_only = false,
- .nosuid = context->dynamic_user, /* don't allow suid/sgid when DynamicUser= is on */
- .recursive = true,
- .ignore_enoent = false,
- };
- }
- }
-
- assert(h == n);
-
- *ret_bind_mounts = TAKE_PTR(bind_mounts);
- *ret_n_bind_mounts = n;
- *ret_empty_directories = TAKE_PTR(empty_directories);
-
- return (int) n;
-}
-
-/* ret_symlinks will contain a list of pairs src:dest that describes
- * the symlinks to create later on. For example, the symlinks needed
- * to safely give private directories to DynamicUser=1 users. */
-static int compile_symlinks(
- const ExecContext *context,
- const ExecParameters *params,
- bool setup_os_release_symlink,
- char ***ret_symlinks) {
-
- _cleanup_strv_free_ char **symlinks = NULL;
- int r;
-
- assert(context);
- assert(params);
- assert(ret_symlinks);
-
- for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
- for (size_t i = 0; i < context->directories[dt].n_items; i++) {
- _cleanup_free_ char *private_path = NULL, *path = NULL;
-
- STRV_FOREACH(symlink, context->directories[dt].items[i].symlinks) {
- _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL;
-
- src_abs = path_join(params->prefix[dt], context->directories[dt].items[i].path);
- dst_abs = path_join(params->prefix[dt], *symlink);
- if (!src_abs || !dst_abs)
- return -ENOMEM;
-
- r = strv_consume_pair(&symlinks, TAKE_PTR(src_abs), TAKE_PTR(dst_abs));
- if (r < 0)
- return r;
- }
-
- if (!exec_directory_is_private(context, dt) ||
- exec_context_with_rootfs(context) ||
- context->directories[dt].items[i].only_create)
- continue;
-
- private_path = path_join(params->prefix[dt], "private", context->directories[dt].items[i].path);
- if (!private_path)
- return -ENOMEM;
-
- path = path_join(params->prefix[dt], context->directories[dt].items[i].path);
- if (!path)
- return -ENOMEM;
-
- r = strv_consume_pair(&symlinks, TAKE_PTR(private_path), TAKE_PTR(path));
- if (r < 0)
- return r;
- }
- }
-
- /* We make the host's os-release available via a symlink, so that we can copy it atomically
- * and readers will never get a half-written version. Note that, while the paths specified here are
- * absolute, when they are processed in namespace.c they will be made relative automatically, i.e.:
- * 'os-release -> .os-release-stage/os-release' is what will be created. */
- if (setup_os_release_symlink) {
- r = strv_extend(&symlinks, "/run/host/.os-release-stage/os-release");
- if (r < 0)
- return r;
-
- r = strv_extend(&symlinks, "/run/host/os-release");
- if (r < 0)
- return r;
- }
-
- *ret_symlinks = TAKE_PTR(symlinks);
-
- return 0;
-}
-
-static bool insist_on_sandboxing(
- const ExecContext *context,
- const char *root_dir,
- const char *root_image,
- const BindMount *bind_mounts,
- size_t n_bind_mounts) {
-
- assert(context);
- assert(n_bind_mounts == 0 || bind_mounts);
-
- /* Checks whether we need to insist on fs namespacing. i.e. whether we have settings configured that
- * would alter the view on the file system beyond making things read-only or invisible, i.e. would
- * rearrange stuff in a way we cannot ignore gracefully. */
-
- if (context->n_temporary_filesystems > 0)
- return true;
-
- if (root_dir || root_image)
- return true;
-
- if (context->n_mount_images > 0)
- return true;
-
- if (context->dynamic_user)
- return true;
-
- if (context->n_extension_images > 0 || !strv_isempty(context->extension_directories))
- return true;
-
- /* If there are any bind mounts set that don't map back onto themselves, fs namespacing becomes
- * essential. */
- for (size_t i = 0; i < n_bind_mounts; i++)
- if (!path_equal(bind_mounts[i].source, bind_mounts[i].destination))
- return true;
-
- if (context->log_namespace)
- return true;
-
- return false;
-}
-
-static int setup_ephemeral(const ExecContext *context, ExecRuntime *runtime) {
- _cleanup_close_ int fd = -EBADF;
- int r;
-
- if (!runtime || !runtime->ephemeral_copy)
- return 0;
-
- r = posix_lock(runtime->ephemeral_storage_socket[0], LOCK_EX);
- if (r < 0)
- return log_debug_errno(r, "Failed to lock ephemeral storage socket: %m");
-
- CLEANUP_POSIX_UNLOCK(runtime->ephemeral_storage_socket[0]);
-
- fd = receive_one_fd(runtime->ephemeral_storage_socket[0], MSG_PEEK|MSG_DONTWAIT);
- if (fd >= 0)
- /* We got an fd! That means ephemeral has already been set up, so nothing to do here. */
- return 0;
-
- if (fd != -EAGAIN)
- return log_debug_errno(fd, "Failed to receive file descriptor queued on ephemeral storage socket: %m");
-
- log_debug("Making ephemeral snapshot of %s to %s",
- context->root_image ?: context->root_directory, runtime->ephemeral_copy);
-
- if (context->root_image)
- fd = copy_file(context->root_image, runtime->ephemeral_copy, O_EXCL, 0600,
- COPY_LOCK_BSD|COPY_REFLINK|COPY_CRTIME);
- else
- fd = btrfs_subvol_snapshot_at(AT_FDCWD, context->root_directory,
- AT_FDCWD, runtime->ephemeral_copy,
- BTRFS_SNAPSHOT_FALLBACK_COPY |
- BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
- BTRFS_SNAPSHOT_RECURSIVE |
- BTRFS_SNAPSHOT_LOCK_BSD);
- if (fd < 0)
- return log_debug_errno(fd, "Failed to snapshot %s to %s: %m",
- context->root_image ?: context->root_directory, runtime->ephemeral_copy);
-
- if (context->root_image) {
- /* A root image might be subject to lots of random writes so let's try to disable COW on it
- * which tends to not perform well in combination with lots of random writes.
- *
- * Note: btrfs actually isn't impressed by us setting the flag after making the reflink'ed
- * copy, but we at least want to make the intention clear.
- */
- r = chattr_fd(fd, FS_NOCOW_FL, FS_NOCOW_FL, NULL);
- if (r < 0)
- log_debug_errno(fd, "Failed to disable copy-on-write for %s, ignoring: %m", runtime->ephemeral_copy);
- }
-
- r = send_one_fd(runtime->ephemeral_storage_socket[1], fd, MSG_DONTWAIT);
- if (r < 0)
- return log_debug_errno(r, "Failed to queue file descriptor on ephemeral storage socket: %m");
-
- return 1;
-}
-
-static int verity_settings_prepare(
- VeritySettings *verity,
- const char *root_image,
- const void *root_hash,
- size_t root_hash_size,
- const char *root_hash_path,
- const void *root_hash_sig,
- size_t root_hash_sig_size,
- const char *root_hash_sig_path,
- const char *verity_data_path) {
-
- int r;
-
- assert(verity);
-
- if (root_hash) {
- void *d;
-
- d = memdup(root_hash, root_hash_size);
- if (!d)
- return -ENOMEM;
-
- free_and_replace(verity->root_hash, d);
- verity->root_hash_size = root_hash_size;
- verity->designator = PARTITION_ROOT;
- }
-
- if (root_hash_sig) {
- void *d;
-
- d = memdup(root_hash_sig, root_hash_sig_size);
- if (!d)
- return -ENOMEM;
-
- free_and_replace(verity->root_hash_sig, d);
- verity->root_hash_sig_size = root_hash_sig_size;
- verity->designator = PARTITION_ROOT;
- }
-
- if (verity_data_path) {
- r = free_and_strdup(&verity->data_path, verity_data_path);
- if (r < 0)
- return r;
- }
-
- r = verity_settings_load(
- verity,
- root_image,
- root_hash_path,
- root_hash_sig_path);
- if (r < 0)
- return log_debug_errno(r, "Failed to load root hash: %m");
-
- return 0;
-}
-
-static int apply_mount_namespace(
- const Unit *u,
- ExecCommandFlags command_flags,
- const ExecContext *context,
- const ExecParameters *params,
- ExecRuntime *runtime,
- const char *memory_pressure_path,
- const char *creds_path,
- int creds_fd,
- char **error_path) {
-
- _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
- _cleanup_strv_free_ char **empty_directories = NULL, **symlinks = NULL,
- **read_write_paths_cleanup = NULL;
- _cleanup_free_ char *incoming_dir = NULL, *propagate_dir = NULL,
- *extension_dir = NULL, *host_os_release_stage = NULL;
- const char *root_dir = NULL, *root_image = NULL, *tmp_dir = NULL, *var_tmp_dir = NULL;
- char **read_write_paths;
- NamespaceInfo ns_info;
- bool needs_sandboxing, setup_os_release_symlink;
- BindMount *bind_mounts = NULL;
- size_t n_bind_mounts = 0;
- int r;
-
- assert(context);
-
- CLEANUP_ARRAY(bind_mounts, n_bind_mounts, bind_mount_free_many);
-
- if (params->flags & EXEC_APPLY_CHROOT) {
- r = setup_ephemeral(context, runtime);
- if (r < 0)
- return r;
-
- if (context->root_image)
- root_image = (runtime ? runtime->ephemeral_copy : NULL) ?: context->root_image;
- else
- root_dir = (runtime ? runtime->ephemeral_copy : NULL) ?: context->root_directory;
- }
-
- r = compile_bind_mounts(context, params, &bind_mounts, &n_bind_mounts, &empty_directories);
- if (r < 0)
- return r;
-
- /* We need to make the pressure path writable even if /sys/fs/cgroups is made read-only, as the
- * service will need to write to it in order to start the notifications. */
- if (context->protect_control_groups && memory_pressure_path && !streq(memory_pressure_path, "/dev/null")) {
- read_write_paths_cleanup = strv_copy(context->read_write_paths);
- if (!read_write_paths_cleanup)
- return -ENOMEM;
-
- r = strv_extend(&read_write_paths_cleanup, memory_pressure_path);
- if (r < 0)
- return r;
-
- read_write_paths = read_write_paths_cleanup;
- } else
- read_write_paths = context->read_write_paths;
-
- needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command_flags & EXEC_COMMAND_FULLY_PRIVILEGED);
- if (needs_sandboxing) {
- /* 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.
- * This does not apply when we are using /run/systemd/empty as fallback. */
-
- if (context->private_tmp && runtime && runtime->shared) {
- if (streq_ptr(runtime->shared->tmp_dir, RUN_SYSTEMD_EMPTY))
- tmp_dir = runtime->shared->tmp_dir;
- else if (runtime->shared->tmp_dir)
- tmp_dir = strjoina(runtime->shared->tmp_dir, "/tmp");
-
- if (streq_ptr(runtime->shared->var_tmp_dir, RUN_SYSTEMD_EMPTY))
- var_tmp_dir = runtime->shared->var_tmp_dir;
- else if (runtime->shared->var_tmp_dir)
- var_tmp_dir = strjoina(runtime->shared->var_tmp_dir, "/tmp");
- }
-
- ns_info = (NamespaceInfo) {
- .ignore_protect_paths = false,
- .private_dev = context->private_devices,
- .protect_control_groups = context->protect_control_groups,
- .protect_kernel_tunables = context->protect_kernel_tunables,
- .protect_kernel_modules = context->protect_kernel_modules,
- .protect_kernel_logs = context->protect_kernel_logs,
- .protect_hostname = context->protect_hostname,
- .mount_apivfs = exec_context_get_effective_mount_apivfs(context),
- .protect_home = context->protect_home,
- .protect_system = context->protect_system,
- .protect_proc = context->protect_proc,
- .proc_subset = context->proc_subset,
- .private_network = exec_needs_network_namespace(context),
- .private_ipc = exec_needs_ipc_namespace(context),
- /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
- .mount_nosuid = context->no_new_privileges && !mac_selinux_use(),
- };
- } else if (!context->dynamic_user && root_dir)
- /*
- * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
- * sandbox info, otherwise enforce it, don't ignore protected paths and
- * fail if we are enable to apply the sandbox inside the mount namespace.
- */
- ns_info = (NamespaceInfo) {
- .ignore_protect_paths = true,
- };
- else
- ns_info = (NamespaceInfo) {};
-
- /* Symlinks (exec dirs, os-release) are set up after other mounts, before they are made read-only. */
- setup_os_release_symlink = ns_info.mount_apivfs && (root_dir || root_image);
- r = compile_symlinks(context, params, setup_os_release_symlink, &symlinks);
- if (r < 0)
- return r;
-
- if (context->mount_propagation_flag == MS_SHARED)
- log_unit_debug(u, "shared mount propagation hidden by other fs namespacing unit settings: ignoring");
-
- if (params->runtime_scope == RUNTIME_SCOPE_SYSTEM) {
- propagate_dir = path_join("/run/systemd/propagate/", u->id);
- if (!propagate_dir)
- return -ENOMEM;
-
- incoming_dir = strdup("/run/systemd/incoming");
- if (!incoming_dir)
- return -ENOMEM;
-
- extension_dir = strdup("/run/systemd/unit-extensions");
- if (!extension_dir)
- return -ENOMEM;
-
- /* If running under a different root filesystem, propagate the host's os-release. We make a
- * copy rather than just bind mounting it, so that it can be updated on soft-reboot. */
- if (setup_os_release_symlink) {
- host_os_release_stage = strdup("/run/systemd/propagate/.os-release-stage");
- if (!host_os_release_stage)
- return -ENOMEM;
- }
- } else {
- assert(params->runtime_scope == RUNTIME_SCOPE_USER);
-
- if (asprintf(&extension_dir, "/run/user/" UID_FMT "/systemd/unit-extensions", geteuid()) < 0)
- return -ENOMEM;
-
- if (setup_os_release_symlink) {
- if (asprintf(&host_os_release_stage,
- "/run/user/" UID_FMT "/systemd/propagate/.os-release-stage",
- geteuid()) < 0)
- return -ENOMEM;
- }
- }
-
- if (root_image) {
- r = verity_settings_prepare(
- &verity,
- root_image,
- context->root_hash, context->root_hash_size, context->root_hash_path,
- context->root_hash_sig, context->root_hash_sig_size, context->root_hash_sig_path,
- context->root_verity);
- if (r < 0)
- return r;
- }
-
- r = setup_namespace(
- root_dir,
- root_image,
- context->root_image_options,
- context->root_image_policy ?: &image_policy_service,
- &ns_info,
- read_write_paths,
- needs_sandboxing ? context->read_only_paths : NULL,
- needs_sandboxing ? context->inaccessible_paths : NULL,
- needs_sandboxing ? context->exec_paths : NULL,
- needs_sandboxing ? context->no_exec_paths : NULL,
- empty_directories,
- symlinks,
- bind_mounts,
- n_bind_mounts,
- context->temporary_filesystems,
- context->n_temporary_filesystems,
- context->mount_images,
- context->n_mount_images,
- context->mount_image_policy ?: &image_policy_service,
- tmp_dir,
- var_tmp_dir,
- creds_path,
- creds_fd,
- context->log_namespace,
- context->mount_propagation_flag,
- &verity,
- context->extension_images,
- context->n_extension_images,
- context->extension_image_policy ?: &image_policy_sysext,
- context->extension_directories,
- propagate_dir,
- incoming_dir,
- extension_dir,
- root_dir || root_image ? params->notify_socket : NULL,
- host_os_release_stage,
- error_path);
-
- /* If we couldn't set up the namespace this is probably due to a missing capability. setup_namespace() reports
- * that with a special, recognizable error ENOANO. In this case, silently proceed, but only if exclusively
- * sandboxing options were used, i.e. nothing such as RootDirectory= or BindMount= that would result in a
- * completely different execution environment. */
- if (r == -ENOANO) {
- if (insist_on_sandboxing(
- context,
- root_dir, root_image,
- bind_mounts,
- n_bind_mounts))
- return log_unit_debug_errno(u,
- SYNTHETIC_ERRNO(EOPNOTSUPP),
- "Failed to set up namespace, and refusing to continue since "
- "the selected namespacing options alter mount environment non-trivially.\n"
- "Bind mounts: %zu, temporary filesystems: %zu, root directory: %s, root image: %s, dynamic user: %s",
- n_bind_mounts,
- context->n_temporary_filesystems,
- yes_no(root_dir),
- yes_no(root_image),
- yes_no(context->dynamic_user));
-
- log_unit_debug(u, "Failed to set up namespace, assuming containerized execution and ignoring.");
- return 0;
- }
-
- return r;
-}
-
-static int apply_working_directory(
- const ExecContext *context,
- const ExecParameters *params,
- ExecRuntime *runtime,
- const char *home,
- int *exit_status) {
-
- const char *d, *wd;
-
- assert(context);
- assert(exit_status);
-
- if (context->working_directory_home) {
-
- if (!home) {
- *exit_status = EXIT_CHDIR;
- return -ENXIO;
- }
-
- wd = home;
-
- } else
- wd = empty_to_root(context->working_directory);
-
- if (params->flags & EXEC_APPLY_CHROOT)
- d = wd;
- else
- d = prefix_roota((runtime ? runtime->ephemeral_copy : NULL) ?: context->root_directory, wd);
-
- if (chdir(d) < 0 && !context->working_directory_missing_ok) {
- *exit_status = EXIT_CHDIR;
- return -errno;
- }
-
- return 0;
-}
-
-static int apply_root_directory(
- const ExecContext *context,
- const ExecParameters *params,
- ExecRuntime *runtime,
- const bool needs_mount_ns,
- int *exit_status) {
-
- assert(context);
- assert(exit_status);
-
- if (params->flags & EXEC_APPLY_CHROOT)
- if (!needs_mount_ns && context->root_directory)
- if (chroot((runtime ? runtime->ephemeral_copy : NULL) ?: context->root_directory) < 0) {
- *exit_status = EXIT_CHROOT;
- return -errno;
- }
-
- return 0;
-}
-
-static int setup_keyring(
- const Unit *u,
- const ExecContext *context,
- const ExecParameters *p,
- uid_t uid, gid_t gid) {
-
- key_serial_t keyring;
- int r = 0;
- uid_t saved_uid;
- gid_t saved_gid;
-
- assert(u);
- assert(context);
- assert(p);
-
- /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
- * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
- * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
- * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
- * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
- * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
-
- if (context->keyring_mode == EXEC_KEYRING_INHERIT)
- return 0;
-
- /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
- * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
- * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
- * & group is just as nasty as acquiring a reference to the user keyring. */
-
- saved_uid = getuid();
- saved_gid = getgid();
-
- if (gid_is_valid(gid) && gid != saved_gid) {
- if (setregid(gid, -1) < 0)
- return log_unit_error_errno(u, errno, "Failed to change GID for user keyring: %m");
- }
-
- if (uid_is_valid(uid) && uid != saved_uid) {
- if (setreuid(uid, -1) < 0) {
- r = log_unit_error_errno(u, errno, "Failed to change UID for user keyring: %m");
- goto out;
- }
- }
-
- keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
- if (keyring == -1) {
- if (errno == ENOSYS)
- log_unit_debug_errno(u, errno, "Kernel keyring not supported, ignoring.");
- else if (ERRNO_IS_PRIVILEGE(errno))
- log_unit_debug_errno(u, errno, "Kernel keyring access prohibited, ignoring.");
- else if (errno == EDQUOT)
- log_unit_debug_errno(u, errno, "Out of kernel keyrings to allocate, ignoring.");
- else
- r = log_unit_error_errno(u, errno, "Setting up kernel keyring failed: %m");
-
- goto out;
- }
-
- /* When requested link the user keyring into the session keyring. */
- if (context->keyring_mode == EXEC_KEYRING_SHARED) {
-
- if (keyctl(KEYCTL_LINK,
- KEY_SPEC_USER_KEYRING,
- KEY_SPEC_SESSION_KEYRING, 0, 0) < 0) {
- r = log_unit_error_errno(u, errno, "Failed to link user keyring into session keyring: %m");
- goto out;
- }
- }
-
- /* Restore uid/gid back */
- if (uid_is_valid(uid) && uid != saved_uid) {
- if (setreuid(saved_uid, -1) < 0) {
- r = log_unit_error_errno(u, errno, "Failed to change UID back for user keyring: %m");
- goto out;
- }
- }
-
- if (gid_is_valid(gid) && gid != saved_gid) {
- if (setregid(saved_gid, -1) < 0)
- return log_unit_error_errno(u, errno, "Failed to change GID back for user keyring: %m");
- }
-
- /* Populate they keyring with the invocation ID by default, as original saved_uid. */
- if (!sd_id128_is_null(u->invocation_id)) {
- key_serial_t key;
-
- key = add_key("user", "invocation_id", &u->invocation_id, sizeof(u->invocation_id), KEY_SPEC_SESSION_KEYRING);
- if (key == -1)
- log_unit_debug_errno(u, errno, "Failed to add invocation ID to keyring, ignoring: %m");
- else {
- if (keyctl(KEYCTL_SETPERM, key,
- KEY_POS_VIEW|KEY_POS_READ|KEY_POS_SEARCH|
- KEY_USR_VIEW|KEY_USR_READ|KEY_USR_SEARCH, 0, 0) < 0)
- r = log_unit_error_errno(u, errno, "Failed to restrict invocation ID permission: %m");
- }
- }
-
-out:
- /* Revert back uid & gid for the last time, and exit */
- /* no extra logging, as only the first already reported error matters */
- if (getuid() != saved_uid)
- (void) setreuid(saved_uid, -1);
-
- if (getgid() != saved_gid)
- (void) setregid(saved_gid, -1);
-
- return r;
-}
-
-static void append_socket_pair(int *array, size_t *n, const int pair[static 2]) {
- assert(array);
- assert(n);
- assert(pair);
-
- if (pair[0] >= 0)
- array[(*n)++] = pair[0];
- if (pair[1] >= 0)
- array[(*n)++] = pair[1];
-}
-
-static int close_remaining_fds(
- const ExecParameters *params,
- const ExecRuntime *runtime,
- int user_lookup_fd,
- int socket_fd,
- const int *fds, size_t n_fds) {
-
- size_t n_dont_close = 0;
- int dont_close[n_fds + 14];
-
- assert(params);
-
- if (params->stdin_fd >= 0)
- dont_close[n_dont_close++] = params->stdin_fd;
- if (params->stdout_fd >= 0)
- dont_close[n_dont_close++] = params->stdout_fd;
- if (params->stderr_fd >= 0)
- dont_close[n_dont_close++] = params->stderr_fd;
-
- if (socket_fd >= 0)
- dont_close[n_dont_close++] = socket_fd;
- if (n_fds > 0) {
- memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
- n_dont_close += n_fds;
- }
-
- if (runtime)
- append_socket_pair(dont_close, &n_dont_close, runtime->ephemeral_storage_socket);
-
- if (runtime && runtime->shared) {
- append_socket_pair(dont_close, &n_dont_close, runtime->shared->netns_storage_socket);
- append_socket_pair(dont_close, &n_dont_close, runtime->shared->ipcns_storage_socket);
- }
-
- if (runtime && runtime->dynamic_creds) {
- if (runtime->dynamic_creds->user)
- append_socket_pair(dont_close, &n_dont_close, runtime->dynamic_creds->user->storage_socket);
- if (runtime->dynamic_creds->group)
- append_socket_pair(dont_close, &n_dont_close, runtime->dynamic_creds->group->storage_socket);
- }
-
- if (user_lookup_fd >= 0)
- dont_close[n_dont_close++] = user_lookup_fd;
-
- return close_all_fds(dont_close, n_dont_close);
-}
-
-static int send_user_lookup(
- Unit *unit,
- int user_lookup_fd,
- uid_t uid,
- gid_t gid) {
-
- assert(unit);
-
- /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
- * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
- * specified. */
-
- if (user_lookup_fd < 0)
- return 0;
-
- if (!uid_is_valid(uid) && !gid_is_valid(gid))
- return 0;
-
- if (writev(user_lookup_fd,
- (struct iovec[]) {
- IOVEC_MAKE(&uid, sizeof(uid)),
- IOVEC_MAKE(&gid, sizeof(gid)),
- IOVEC_MAKE_STRING(unit->id) }, 3) < 0)
- return -errno;
-
- return 0;
-}
-
-static int acquire_home(const ExecContext *c, uid_t uid, const char** home, char **buf) {
- int r;
-
- assert(c);
- assert(home);
- assert(buf);
-
- /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
-
- if (*home)
- return 0;
-
- if (!c->working_directory_home)
- return 0;
-
- r = get_home_dir(buf);
- if (r < 0)
- return r;
-
- *home = *buf;
- return 1;
-}
-
-static int compile_suggested_paths(const ExecContext *c, const ExecParameters *p, char ***ret) {
- _cleanup_strv_free_ char ** list = NULL;
- int r;
-
- assert(c);
- assert(p);
- assert(ret);
-
- assert(c->dynamic_user);
-
- /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
- * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
- * directories. */
-
- for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
- if (t == EXEC_DIRECTORY_CONFIGURATION)
- continue;
-
- if (!p->prefix[t])
- continue;
-
- for (size_t i = 0; i < c->directories[t].n_items; i++) {
- char *e;
-
- if (exec_directory_is_private(c, t))
- e = path_join(p->prefix[t], "private", c->directories[t].items[i].path);
- else
- e = path_join(p->prefix[t], c->directories[t].items[i].path);
- if (!e)
- return -ENOMEM;
-
- r = strv_consume(&list, e);
- if (r < 0)
- return r;
- }
- }
-
- *ret = TAKE_PTR(list);
-
- return 0;
-}
-
-static int exec_parameters_get_cgroup_path(
- const ExecParameters *params,
- const CGroupContext *c,
- char **ret) {
-
- const char *subgroup = NULL;
- char *p;
-
- assert(params);
- assert(ret);
-
- if (!params->cgroup_path)
- return -EINVAL;
-
- /* If we are called for a unit where cgroup delegation is on, and the payload created its own populated
- * subcgroup (which we expect it to do, after all it asked for delegation), then we cannot place the control
- * processes started after the main unit's process in the unit's main cgroup because it is now an inner one,
- * and inner cgroups may not contain processes. Hence, if delegation is on, and this is a control process,
- * let's use ".control" as subcgroup instead. Note that we do so only for ExecStartPost=, ExecReload=,
- * ExecStop=, ExecStopPost=, i.e. for the commands where the main process is already forked. For ExecStartPre=
- * this is not necessary, the cgroup is still empty. We distinguish these cases with the EXEC_CONTROL_CGROUP
- * flag, which is only passed for the former statements, not for the latter. */
-
- if (FLAGS_SET(params->flags, EXEC_CGROUP_DELEGATE) && (FLAGS_SET(params->flags, EXEC_CONTROL_CGROUP) || c->delegate_subgroup)) {
- if (FLAGS_SET(params->flags, EXEC_IS_CONTROL))
- subgroup = ".control";
- else
- subgroup = c->delegate_subgroup;
- }
-
- if (subgroup)
- p = path_join(params->cgroup_path, subgroup);
- else
- p = strdup(params->cgroup_path);
- if (!p)
- return -ENOMEM;
-
- *ret = p;
- return !!subgroup;
-}
-
-static int exec_context_cpu_affinity_from_numa(const ExecContext *c, CPUSet *ret) {
- _cleanup_(cpu_set_reset) CPUSet s = {};
- int r;
-
- assert(c);
- assert(ret);
-
- if (!c->numa_policy.nodes.set) {
- log_debug("Can't derive CPU affinity mask from NUMA mask because NUMA mask is not set, ignoring");
- return 0;
- }
-
- r = numa_to_cpu_set(&c->numa_policy, &s);
- if (r < 0)
- return r;
-
- cpu_set_reset(ret);
-
- return cpu_set_add_all(ret, &s);
-}
-
-bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c) {
- assert(c);
-
- return c->cpu_affinity_from_numa;
-}
-
-static int add_shifted_fd(int *fds, size_t fds_size, size_t *n_fds, int fd, int *ret_fd) {
- int r;
-
- assert(fds);
- assert(n_fds);
- assert(*n_fds < fds_size);
- assert(ret_fd);
-
- if (fd < 0) {
- *ret_fd = -EBADF;
- return 0;
- }
-
- if (fd < 3 + (int) *n_fds) {
- /* Let's move the fd up, so that it's outside of the fd range we will use to store
- * the fds we pass to the process (or which are closed only during execve). */
-
- r = fcntl(fd, F_DUPFD_CLOEXEC, 3 + (int) *n_fds);
- if (r < 0)
- return -errno;
-
- close_and_replace(fd, r);
- }
-
- *ret_fd = fds[*n_fds] = fd;
- (*n_fds) ++;
- return 1;
-}
-
-static int connect_unix_harder(Unit *u, const OpenFile *of, int ofd) {
- union sockaddr_union addr = {
- .un.sun_family = AF_UNIX,
- };
- socklen_t sa_len;
- static const int socket_types[] = { SOCK_DGRAM, SOCK_STREAM, SOCK_SEQPACKET };
- int r;
-
- assert(u);
- assert(of);
- assert(ofd >= 0);
-
- r = sockaddr_un_set_path(&addr.un, FORMAT_PROC_FD_PATH(ofd));
- if (r < 0)
- return log_unit_error_errno(u, r, "Failed to set sockaddr for %s: %m", of->path);
-
- sa_len = r;
-
- for (size_t i = 0; i < ELEMENTSOF(socket_types); i++) {
- _cleanup_close_ int fd = -EBADF;
-
- fd = socket(AF_UNIX, socket_types[i] | SOCK_CLOEXEC, 0);
- if (fd < 0)
- return log_unit_error_errno(u, errno, "Failed to create socket for %s: %m", of->path);
-
- r = RET_NERRNO(connect(fd, &addr.sa, sa_len));
- if (r == -EPROTOTYPE)
- continue;
- if (r < 0)
- return log_unit_error_errno(u, r, "Failed to connect socket for %s: %m", of->path);
-
- return TAKE_FD(fd);
- }
-
- return log_unit_error_errno(u, SYNTHETIC_ERRNO(EPROTOTYPE), "Failed to connect socket for \"%s\".", of->path);
-}
-
-static int get_open_file_fd(Unit *u, const OpenFile *of) {
- struct stat st;
- _cleanup_close_ int fd = -EBADF, ofd = -EBADF;
-
- assert(u);
- assert(of);
-
- ofd = open(of->path, O_PATH | O_CLOEXEC);
- if (ofd < 0)
- return log_unit_error_errno(u, errno, "Could not open \"%s\": %m", of->path);
-
- if (fstat(ofd, &st) < 0)
- return log_unit_error_errno(u, errno, "Failed to stat %s: %m", of->path);
-
- if (S_ISSOCK(st.st_mode)) {
- fd = connect_unix_harder(u, of, ofd);
- if (fd < 0)
- return fd;
-
- if (FLAGS_SET(of->flags, OPENFILE_READ_ONLY) && shutdown(fd, SHUT_WR) < 0)
- return log_unit_error_errno(u, errno, "Failed to shutdown send for socket %s: %m",
- of->path);
-
- log_unit_debug(u, "socket %s opened (fd=%d)", of->path, fd);
- } else {
- int flags = FLAGS_SET(of->flags, OPENFILE_READ_ONLY) ? O_RDONLY : O_RDWR;
- if (FLAGS_SET(of->flags, OPENFILE_APPEND))
- flags |= O_APPEND;
- else if (FLAGS_SET(of->flags, OPENFILE_TRUNCATE))
- flags |= O_TRUNC;
-
- fd = fd_reopen(ofd, flags | O_CLOEXEC);
- if (fd < 0)
- return log_unit_error_errno(u, fd, "Failed to open file %s: %m", of->path);
-
- log_unit_debug(u, "file %s opened (fd=%d)", of->path, fd);
- }
-
- return TAKE_FD(fd);
-}
-
-static int collect_open_file_fds(
- Unit *u,
- OpenFile* open_files,
- int **fds,
- char ***fdnames,
- size_t *n_fds) {
- int r;
-
- assert(u);
- assert(fds);
- assert(fdnames);
- assert(n_fds);
-
- LIST_FOREACH(open_files, of, open_files) {
- _cleanup_close_ int fd = -EBADF;
-
- fd = get_open_file_fd(u, of);
- if (fd < 0) {
- if (FLAGS_SET(of->flags, OPENFILE_GRACEFUL)) {
- log_unit_debug_errno(u, fd, "Failed to get OpenFile= file descriptor for %s, ignoring: %m", of->path);
- continue;
- }
-
- return fd;
- }
-
- if (!GREEDY_REALLOC(*fds, *n_fds + 1))
- return -ENOMEM;
-
- r = strv_extend(fdnames, of->fdname);
- if (r < 0)
- return r;
-
- (*fds)[*n_fds] = TAKE_FD(fd);
-
- (*n_fds)++;
- }
-
- return 0;
-}
-
-static void log_command_line(Unit *unit, const char *msg, const char *executable, char **argv) {
- assert(unit);
- assert(msg);
- assert(executable);
-
- if (!DEBUG_LOGGING)
- return;
-
- _cleanup_free_ char *cmdline = quote_command_line(argv, SHELL_ESCAPE_EMPTY);
-
- log_unit_struct(unit, LOG_DEBUG,
- "EXECUTABLE=%s", executable,
- LOG_UNIT_MESSAGE(unit, "%s: %s", msg, strnull(cmdline)),
- LOG_UNIT_INVOCATION_ID(unit));
-}
-
-static bool exec_context_need_unprivileged_private_users(
- const ExecContext *context,
- const ExecParameters *params) {
-
- assert(context);
- assert(params);
-
- /* These options require PrivateUsers= when used in user units, as we need to be in a user namespace
- * to have permission to enable them when not running as root. If we have effective CAP_SYS_ADMIN
- * (system manager) then we have privileges and don't need this. */
- if (params->runtime_scope != RUNTIME_SCOPE_USER)
- return false;
-
- return context->private_users ||
- context->private_tmp ||
- context->private_devices ||
- context->private_network ||
- context->network_namespace_path ||
- context->private_ipc ||
- context->ipc_namespace_path ||
- context->private_mounts > 0 ||
- context->mount_apivfs ||
- context->n_bind_mounts > 0 ||
- context->n_temporary_filesystems > 0 ||
- context->root_directory ||
- !strv_isempty(context->extension_directories) ||
- context->protect_system != PROTECT_SYSTEM_NO ||
- context->protect_home != PROTECT_HOME_NO ||
- context->protect_kernel_tunables ||
- context->protect_kernel_modules ||
- context->protect_kernel_logs ||
- context->protect_control_groups ||
- context->protect_clock ||
- context->protect_hostname ||
- !strv_isempty(context->read_write_paths) ||
- !strv_isempty(context->read_only_paths) ||
- !strv_isempty(context->inaccessible_paths) ||
- !strv_isempty(context->exec_paths) ||
- !strv_isempty(context->no_exec_paths);
-}
-
-static int exec_child(
- Unit *unit,
- const ExecCommand *command,
- const ExecContext *context,
- const ExecParameters *params,
- ExecRuntime *runtime,
- const CGroupContext *cgroup_context,
- int socket_fd,
- const int named_iofds[static 3],
- int *params_fds,
- size_t n_socket_fds,
- size_t n_storage_fds,
- char **files_env,
- int user_lookup_fd,
- int *exit_status) {
-
- _cleanup_strv_free_ char **our_env = NULL, **pass_env = NULL, **joined_exec_search_path = NULL, **accum_env = NULL, **replaced_argv = NULL;
- int r, ngids = 0, exec_fd;
- _cleanup_free_ gid_t *supplementary_gids = NULL;
- const char *username = NULL, *groupname = NULL;
- _cleanup_free_ char *home_buffer = NULL, *memory_pressure_path = NULL, *creds_path = NULL;
- const char *home = NULL, *shell = NULL;
- char **final_argv = NULL;
- dev_t journal_stream_dev = 0;
- ino_t journal_stream_ino = 0;
- bool userns_set_up = false;
- bool needs_sandboxing, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
- needs_setuid, /* Do we need to do the actual setresuid()/setresgid() calls? */
- needs_mount_namespace, /* Do we need to set up a mount namespace for this kernel? */
- needs_ambient_hack; /* Do we need to apply the ambient capabilities hack? */
-#if HAVE_SELINUX
- _cleanup_free_ char *mac_selinux_context_net = NULL;
- bool use_selinux = false;
-#endif
-#if ENABLE_SMACK
- bool use_smack = false;
-#endif
-#if HAVE_APPARMOR
- bool use_apparmor = false;
-#endif
- uid_t saved_uid = getuid();
- gid_t saved_gid = getgid();
- uid_t uid = UID_INVALID;
- gid_t gid = GID_INVALID;
- size_t n_fds = n_socket_fds + n_storage_fds, /* fds to pass to the child */
- n_keep_fds; /* total number of fds not to close */
- int secure_bits;
- _cleanup_free_ gid_t *gids_after_pam = NULL;
- int ngids_after_pam = 0;
- _cleanup_free_ int *fds = NULL;
- _cleanup_strv_free_ char **fdnames = NULL;
- _cleanup_close_ int creds_fd = -EBADF;
-
- assert(unit);
- assert(command);
- assert(context);
- assert(params);
- assert(exit_status);
-
- /* Explicitly test for CVE-2021-4034 inspired invocations */
- assert(command->path);
- assert(!strv_isempty(command->argv));
-
- 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);
-
- if (context->ignore_sigpipe)
- (void) ignore_signals(SIGPIPE);
-
- r = reset_signal_mask();
- if (r < 0) {
- *exit_status = EXIT_SIGNAL_MASK;
- return log_unit_error_errno(unit, r, "Failed to set process signal mask: %m");
- }
-
- if (params->idle_pipe)
- do_idle_pipe_dance(params->idle_pipe);
-
- /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
- * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
- * any fds open we don't really want open during the transition. In order to make logging work, we switch the
- * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
-
- log_forget_fds();
- log_set_open_when_needed(true);
- log_settle_target();
-
- /* In case anything used libc syslog(), close this here, too */
- closelog();
-
- fds = newdup(int, params_fds, n_fds);
- if (!fds) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- fdnames = strv_copy((char**) params->fd_names);
- if (!fdnames) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- r = collect_open_file_fds(unit, params->open_files, &fds, &fdnames, &n_fds);
- if (r < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, r, "Failed to get OpenFile= file descriptors: %m");
- }
-
- int keep_fds[n_fds + 3];
- memcpy_safe(keep_fds, fds, n_fds * sizeof(int));
- n_keep_fds = n_fds;
-
- r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, params->exec_fd, &exec_fd);
- if (r < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m");
- }
-
-#if HAVE_LIBBPF
- if (unit->manager->restrict_fs) {
- int bpf_map_fd = lsm_bpf_map_restrict_fs_fd(unit);
- if (bpf_map_fd < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, bpf_map_fd, "Failed to get restrict filesystems BPF map fd: %m");
- }
-
- r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, bpf_map_fd, &bpf_map_fd);
- if (r < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m");
- }
- }
-#endif
-
- r = close_remaining_fds(params, runtime, user_lookup_fd, socket_fd, keep_fds, n_keep_fds);
- if (r < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, r, "Failed to close unwanted file descriptors: %m");
- }
-
- if (!context->same_pgrp &&
- setsid() < 0) {
- *exit_status = EXIT_SETSID;
- return log_unit_error_errno(unit, errno, "Failed to create new process session: %m");
- }
-
- exec_context_tty_reset(context, params);
-
- if (unit_shall_confirm_spawn(unit)) {
- _cleanup_free_ char *cmdline = NULL;
-
- cmdline = quote_command_line(command->argv, SHELL_ESCAPE_EMPTY);
- if (!cmdline) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- r = ask_for_confirmation(context, params->confirm_spawn, unit, cmdline);
- if (r != CONFIRM_EXECUTE) {
- if (r == CONFIRM_PRETEND_SUCCESS) {
- *exit_status = EXIT_SUCCESS;
- return 0;
- }
-
- *exit_status = EXIT_CONFIRM;
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ECANCELED),
- "Execution cancelled by the user");
- }
- }
-
- /* We are about to invoke NSS and PAM modules. Let's tell them what we are doing here, maybe they care. This is
- * used by nss-resolve to disable itself when we are about to start systemd-resolved, to avoid deadlocks. Note
- * that these env vars do not survive the execve(), which means they really only apply to the PAM and NSS
- * invocations themselves. Also note that while we'll only invoke NSS modules involved in user management they
- * might internally call into other NSS modules that are involved in hostname resolution, we never know. */
- if (setenv("SYSTEMD_ACTIVATION_UNIT", unit->id, true) != 0 ||
- setenv("SYSTEMD_ACTIVATION_SCOPE", runtime_scope_to_string(params->runtime_scope), true) != 0) {
- *exit_status = EXIT_MEMORY;
- return log_unit_error_errno(unit, errno, "Failed to update environment: %m");
- }
-
- if (context->dynamic_user && runtime && runtime->dynamic_creds) {
- _cleanup_strv_free_ char **suggested_paths = NULL;
-
- /* On top of that, make sure we bypass our own NSS module nss-systemd comprehensively for any NSS
- * checks, if DynamicUser=1 is used, as we shouldn't create a feedback loop with ourselves here. */
- if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, errno, "Failed to update environment: %m");
- }
-
- r = compile_suggested_paths(context, params, &suggested_paths);
- if (r < 0) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- r = dynamic_creds_realize(runtime->dynamic_creds, suggested_paths, &uid, &gid);
- if (r < 0) {
- *exit_status = EXIT_USER;
- if (r == -EILSEQ)
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP),
- "Failed to update dynamic user credentials: User or group with specified name already exists.");
- return log_unit_error_errno(unit, r, "Failed to update dynamic user credentials: %m");
- }
-
- if (!uid_is_valid(uid)) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ESRCH), "UID validation failed for \""UID_FMT"\"", uid);
- }
-
- if (!gid_is_valid(gid)) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(ESRCH), "GID validation failed for \""GID_FMT"\"", gid);
- }
-
- if (runtime->dynamic_creds->user)
- username = runtime->dynamic_creds->user->name;
-
- } else {
- r = get_fixed_user(context, &username, &uid, &gid, &home, &shell);
- if (r < 0) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, r, "Failed to determine user credentials: %m");
- }
-
- r = get_fixed_group(context, &groupname, &gid);
- if (r < 0) {
- *exit_status = EXIT_GROUP;
- return log_unit_error_errno(unit, r, "Failed to determine group credentials: %m");
- }
- }
-
- /* Initialize user supplementary groups and get SupplementaryGroups= ones */
- r = get_supplementary_groups(context, username, groupname, gid,
- &supplementary_gids, &ngids);
- if (r < 0) {
- *exit_status = EXIT_GROUP;
- return log_unit_error_errno(unit, r, "Failed to determine supplementary groups: %m");
- }
-
- r = send_user_lookup(unit, user_lookup_fd, uid, gid);
- if (r < 0) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, r, "Failed to send user credentials to PID1: %m");
- }
-
- user_lookup_fd = safe_close(user_lookup_fd);
-
- r = acquire_home(context, uid, &home, &home_buffer);
- if (r < 0) {
- *exit_status = EXIT_CHDIR;
- return log_unit_error_errno(unit, r, "Failed to determine $HOME for user: %m");
- }
-
- /* If a socket is connected to STDIN/STDOUT/STDERR, we must drop O_NONBLOCK */
- if (socket_fd >= 0)
- (void) fd_nonblock(socket_fd, false);
-
- /* Journald will try to look-up our cgroup in order to populate _SYSTEMD_CGROUP and _SYSTEMD_UNIT fields.
- * Hence we need to migrate to the target cgroup from init.scope before connecting to journald */
- if (params->cgroup_path) {
- _cleanup_free_ char *p = NULL;
-
- r = exec_parameters_get_cgroup_path(params, cgroup_context, &p);
- if (r < 0) {
- *exit_status = EXIT_CGROUP;
- return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m");
- }
-
- r = cg_attach_everywhere(params->cgroup_supported, p, 0, NULL, NULL);
- if (r == -EUCLEAN) {
- *exit_status = EXIT_CGROUP;
- return log_unit_error_errno(unit, r, "Failed to attach process to cgroup %s "
- "because the cgroup or one of its parents or "
- "siblings is in the threaded mode: %m", p);
- }
- if (r < 0) {
- *exit_status = EXIT_CGROUP;
- return log_unit_error_errno(unit, r, "Failed to attach to cgroup %s: %m", p);
- }
- }
-
- if (context->network_namespace_path && runtime && runtime->shared && runtime->shared->netns_storage_socket[0] >= 0) {
- r = open_shareable_ns_path(runtime->shared->netns_storage_socket, context->network_namespace_path, CLONE_NEWNET);
- if (r < 0) {
- *exit_status = EXIT_NETWORK;
- return log_unit_error_errno(unit, r, "Failed to open network namespace path %s: %m", context->network_namespace_path);
- }
- }
-
- if (context->ipc_namespace_path && runtime && runtime->shared && runtime->shared->ipcns_storage_socket[0] >= 0) {
- r = open_shareable_ns_path(runtime->shared->ipcns_storage_socket, context->ipc_namespace_path, CLONE_NEWIPC);
- if (r < 0) {
- *exit_status = EXIT_NAMESPACE;
- return log_unit_error_errno(unit, r, "Failed to open IPC namespace path %s: %m", context->ipc_namespace_path);
- }
- }
-
- r = setup_input(context, params, socket_fd, named_iofds);
- if (r < 0) {
- *exit_status = EXIT_STDIN;
- return log_unit_error_errno(unit, r, "Failed to set up standard input: %m");
- }
-
- r = setup_output(unit, context, params, STDOUT_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino);
- if (r < 0) {
- *exit_status = EXIT_STDOUT;
- return log_unit_error_errno(unit, r, "Failed to set up standard output: %m");
- }
-
- r = setup_output(unit, context, params, STDERR_FILENO, socket_fd, named_iofds, basename(command->path), uid, gid, &journal_stream_dev, &journal_stream_ino);
- if (r < 0) {
- *exit_status = EXIT_STDERR;
- return log_unit_error_errno(unit, r, "Failed to set up standard error output: %m");
- }
-
- if (context->oom_score_adjust_set) {
- /* 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. */
- r = set_oom_score_adjust(context->oom_score_adjust);
- if (ERRNO_IS_NEG_PRIVILEGE(r))
- log_unit_debug_errno(unit, r,
- "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
- else if (r < 0) {
- *exit_status = EXIT_OOM_ADJUST;
- return log_unit_error_errno(unit, r, "Failed to adjust OOM setting: %m");
- }
- }
-
- if (context->coredump_filter_set) {
- r = set_coredump_filter(context->coredump_filter);
- if (ERRNO_IS_NEG_PRIVILEGE(r))
- log_unit_debug_errno(unit, r, "Failed to adjust coredump_filter, ignoring: %m");
- else if (r < 0) {
- *exit_status = EXIT_LIMITS;
- return log_unit_error_errno(unit, r, "Failed to adjust coredump_filter: %m");
- }
- }
-
- if (context->nice_set) {
- r = setpriority_closest(context->nice);
- if (r < 0) {
- *exit_status = EXIT_NICE;
- return log_unit_error_errno(unit, r, "Failed to set up process scheduling priority (nice level): %m");
- }
- }
-
- 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),
- ¶m);
- if (r < 0) {
- *exit_status = EXIT_SETSCHEDULER;
- return log_unit_error_errno(unit, errno, "Failed to set up CPU scheduling: %m");
- }
- }
-
- if (context->cpu_affinity_from_numa || context->cpu_set.set) {
- _cleanup_(cpu_set_reset) CPUSet converted_cpu_set = {};
- const CPUSet *cpu_set;
-
- if (context->cpu_affinity_from_numa) {
- r = exec_context_cpu_affinity_from_numa(context, &converted_cpu_set);
- if (r < 0) {
- *exit_status = EXIT_CPUAFFINITY;
- return log_unit_error_errno(unit, r, "Failed to derive CPU affinity mask from NUMA mask: %m");
- }
-
- cpu_set = &converted_cpu_set;
- } else
- cpu_set = &context->cpu_set;
-
- if (sched_setaffinity(0, cpu_set->allocated, cpu_set->set) < 0) {
- *exit_status = EXIT_CPUAFFINITY;
- return log_unit_error_errno(unit, errno, "Failed to set up CPU affinity: %m");
- }
- }
-
- if (mpol_is_valid(numa_policy_get_type(&context->numa_policy))) {
- r = apply_numa_policy(&context->numa_policy);
- if (ERRNO_IS_NEG_NOT_SUPPORTED(r))
- log_unit_debug_errno(unit, r, "NUMA support not available, ignoring.");
- else if (r < 0) {
- *exit_status = EXIT_NUMA_POLICY;
- return log_unit_error_errno(unit, r, "Failed to set NUMA memory policy: %m");
- }
- }
-
- if (context->ioprio_set)
- if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) {
- *exit_status = EXIT_IOPRIO;
- return log_unit_error_errno(unit, errno, "Failed to set up IO scheduling priority: %m");
- }
-
- if (context->timer_slack_nsec != NSEC_INFINITY)
- if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) {
- *exit_status = EXIT_TIMERSLACK;
- return log_unit_error_errno(unit, errno, "Failed to set up timer slack: %m");
- }
-
- if (context->personality != PERSONALITY_INVALID) {
- r = safe_personality(context->personality);
- if (r < 0) {
- *exit_status = EXIT_PERSONALITY;
- return log_unit_error_errno(unit, r, "Failed to set up execution domain (personality): %m");
- }
- }
-
- if (context->utmp_id) {
- const char *line = context->tty_path ?
- (path_startswith(context->tty_path, "/dev/") ?: context->tty_path) :
- NULL;
- utmp_put_init_process(context->utmp_id, getpid_cached(), getsid(0),
- line,
- context->utmp_mode == EXEC_UTMP_INIT ? INIT_PROCESS :
- context->utmp_mode == EXEC_UTMP_LOGIN ? LOGIN_PROCESS :
- USER_PROCESS,
- username);
- }
-
- if (uid_is_valid(uid)) {
- r = chown_terminal(STDIN_FILENO, uid);
- if (r < 0) {
- *exit_status = EXIT_STDIN;
- return log_unit_error_errno(unit, r, "Failed to change ownership of terminal: %m");
- }
- }
-
- if (params->cgroup_path) {
- /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroup v1
- * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
- * safe. On cgroup v2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
- * touch a single hierarchy too. */
-
- if (params->flags & EXEC_CGROUP_DELEGATE) {
- _cleanup_free_ char *p = NULL;
-
- r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, uid, gid);
- if (r < 0) {
- *exit_status = EXIT_CGROUP;
- return log_unit_error_errno(unit, r, "Failed to adjust control group access: %m");
- }
-
- r = exec_parameters_get_cgroup_path(params, cgroup_context, &p);
- if (r < 0) {
- *exit_status = EXIT_CGROUP;
- return log_unit_error_errno(unit, r, "Failed to acquire cgroup path: %m");
- }
- if (r > 0) {
- r = cg_set_access_recursive(SYSTEMD_CGROUP_CONTROLLER, p, uid, gid);
- if (r < 0) {
- *exit_status = EXIT_CGROUP;
- return log_unit_error_errno(unit, r, "Failed to adjust control subgroup access: %m");
- }
- }
- }
-
- if (cgroup_context && cg_unified() > 0 && is_pressure_supported() > 0) {
- if (cgroup_context_want_memory_pressure(cgroup_context)) {
- r = cg_get_path("memory", params->cgroup_path, "memory.pressure", &memory_pressure_path);
- if (r < 0) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- r = chmod_and_chown(memory_pressure_path, 0644, uid, gid);
- if (r < 0) {
- log_unit_full_errno(unit, r == -ENOENT || ERRNO_IS_PRIVILEGE(r) ? LOG_DEBUG : LOG_WARNING, r,
- "Failed to adjust ownership of '%s', ignoring: %m", memory_pressure_path);
- memory_pressure_path = mfree(memory_pressure_path);
- }
- } else if (cgroup_context->memory_pressure_watch == CGROUP_PRESSURE_WATCH_OFF) {
- memory_pressure_path = strdup("/dev/null"); /* /dev/null is explicit indicator for turning of memory pressure watch */
- if (!memory_pressure_path) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
- }
- }
- }
-
- needs_mount_namespace = exec_needs_mount_namespace(context, params, runtime);
-
- for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
- r = setup_exec_directory(unit, context, params, uid, gid, dt, needs_mount_namespace, exit_status);
- if (r < 0)
- return log_unit_error_errno(unit, r, "Failed to set up special execution directory in %s: %m", params->prefix[dt]);
- }
-
- if (FLAGS_SET(params->flags, EXEC_WRITE_CREDENTIALS)) {
- r = setup_credentials(context, params, unit->id, uid, gid, &creds_path, &creds_fd);
- if (r < 0) {
- *exit_status = EXIT_CREDENTIALS;
- return log_unit_error_errno(unit, r, "Failed to set up credentials: %m");
- }
- }
-
- r = build_environment(
- unit,
- context,
- params,
- cgroup_context,
- n_fds,
- fdnames,
- home,
- username,
- shell,
- journal_stream_dev,
- journal_stream_ino,
- memory_pressure_path,
- creds_path,
- &our_env);
- if (r < 0) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- r = build_pass_environment(context, &pass_env);
- if (r < 0) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- /* The $PATH variable is set to the default path in params->environment. However, this is overridden
- * if user-specified fields have $PATH set. The intention is to also override $PATH if the unit does
- * not specify PATH but the unit has ExecSearchPath. */
- if (!strv_isempty(context->exec_search_path)) {
- _cleanup_free_ char *joined = NULL;
-
- joined = strv_join(context->exec_search_path, ":");
- if (!joined) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- r = strv_env_assign(&joined_exec_search_path, "PATH", joined);
- if (r < 0) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
- }
-
- accum_env = strv_env_merge(params->environment,
- our_env,
- joined_exec_search_path,
- pass_env,
- context->environment,
- files_env);
- if (!accum_env) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
- accum_env = strv_env_clean(accum_env);
-
- (void) umask(context->umask);
-
- r = setup_keyring(unit, context, params, uid, gid);
- if (r < 0) {
- *exit_status = EXIT_KEYRING;
- return log_unit_error_errno(unit, r, "Failed to set up kernel keyring: %m");
- }
-
- /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted
- * from it. */
- needs_sandboxing = (params->flags & EXEC_APPLY_SANDBOXING) && !(command->flags & EXEC_COMMAND_FULLY_PRIVILEGED);
-
- /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked
- * for it, and the kernel doesn't actually support ambient caps. */
- needs_ambient_hack = (params->flags & EXEC_APPLY_SANDBOXING) && (command->flags & EXEC_COMMAND_AMBIENT_MAGIC) && !ambient_capabilities_supported();
-
- /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly
- * excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not
- * desired. */
- if (needs_ambient_hack)
- needs_setuid = false;
- else
- needs_setuid = (params->flags & EXEC_APPLY_SANDBOXING) && !(command->flags & (EXEC_COMMAND_FULLY_PRIVILEGED|EXEC_COMMAND_NO_SETUID));
-
- uint64_t capability_ambient_set = context->capability_ambient_set;
-
- if (needs_sandboxing) {
- /* MAC enablement checks need to be done before a new mount ns is created, as they rely on
- * /sys being present. The actual MAC context application will happen later, as late as
- * possible, to avoid impacting our own code paths. */
-
-#if HAVE_SELINUX
- use_selinux = mac_selinux_use();
-#endif
-#if ENABLE_SMACK
- use_smack = mac_smack_use();
-#endif
-#if HAVE_APPARMOR
- use_apparmor = mac_apparmor_use();
-#endif
- }
-
- if (needs_sandboxing) {
- int which_failed;
-
- /* Let's set the resource limits before we call into PAM, so that pam_limits wins over what
- * is set here. (See below.) */
-
- r = setrlimit_closest_all((const struct rlimit* const *) context->rlimit, &which_failed);
- if (r < 0) {
- *exit_status = EXIT_LIMITS;
- return log_unit_error_errno(unit, r, "Failed to adjust resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
- }
- }
-
- if (needs_setuid && context->pam_name && username) {
- /* Let's call into PAM after we set up our own idea of resource limits to that pam_limits
- * wins here. (See above.) */
-
- /* All fds passed in the fds array will be closed in the pam child process. */
- r = setup_pam(context->pam_name, username, uid, gid, context->tty_path, &accum_env, fds, n_fds);
- if (r < 0) {
- *exit_status = EXIT_PAM;
- return log_unit_error_errno(unit, r, "Failed to set up PAM session: %m");
- }
-
- if (ambient_capabilities_supported()) {
- uint64_t ambient_after_pam;
-
- /* PAM modules might have set some ambient caps. Query them here and merge them into
- * the caps we want to set in the end, so that we don't end up unsetting them. */
- r = capability_get_ambient(&ambient_after_pam);
- if (r < 0) {
- *exit_status = EXIT_CAPABILITIES;
- return log_unit_error_errno(unit, r, "Failed to query ambient caps: %m");
- }
-
- capability_ambient_set |= ambient_after_pam;
- }
-
- ngids_after_pam = getgroups_alloc(&gids_after_pam);
- if (ngids_after_pam < 0) {
- *exit_status = EXIT_MEMORY;
- return log_unit_error_errno(unit, ngids_after_pam, "Failed to obtain groups after setting up PAM: %m");
- }
- }
-
- if (needs_sandboxing && exec_context_need_unprivileged_private_users(context, params)) {
- /* If we're unprivileged, set up the user namespace first to enable use of the other namespaces.
- * Users with CAP_SYS_ADMIN can set up user namespaces last because they will be able to
- * set up the all of the other namespaces (i.e. network, mount, UTS) without a user namespace. */
-
- r = setup_private_users(saved_uid, saved_gid, uid, gid);
- /* If it was requested explicitly and we can't set it up, fail early. Otherwise, continue and let
- * the actual requested operations fail (or silently continue). */
- if (r < 0 && context->private_users) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, r, "Failed to set up user namespacing for unprivileged user: %m");
- }
- if (r < 0)
- log_unit_info_errno(unit, r, "Failed to set up user namespacing for unprivileged user, ignoring: %m");
- else
- userns_set_up = true;
- }
-
- if (exec_needs_network_namespace(context) && runtime && runtime->shared && runtime->shared->netns_storage_socket[0] >= 0) {
-
- /* Try to enable network namespacing if network namespacing is available and we have
- * CAP_NET_ADMIN. We need CAP_NET_ADMIN to be able to configure the loopback device in the
- * new network namespace. And if we don't have that, then we could only create a network
- * namespace without the ability to set up "lo". Hence gracefully skip things then. */
- if (ns_type_supported(NAMESPACE_NET) && have_effective_cap(CAP_NET_ADMIN) > 0) {
- r = setup_shareable_ns(runtime->shared->netns_storage_socket, CLONE_NEWNET);
- if (ERRNO_IS_NEG_PRIVILEGE(r))
- log_unit_notice_errno(unit, r,
- "PrivateNetwork=yes is configured, but network namespace setup not permitted, proceeding without: %m");
- else if (r < 0) {
- *exit_status = EXIT_NETWORK;
- return log_unit_error_errno(unit, r, "Failed to set up network namespacing: %m");
- }
- } else if (context->network_namespace_path) {
- *exit_status = EXIT_NETWORK;
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP),
- "NetworkNamespacePath= is not supported, refusing.");
- } else
- log_unit_notice(unit, "PrivateNetwork=yes is configured, but the kernel does not support or we lack privileges for network namespace, proceeding without.");
- }
-
- if (exec_needs_ipc_namespace(context) && runtime && runtime->shared && runtime->shared->ipcns_storage_socket[0] >= 0) {
-
- if (ns_type_supported(NAMESPACE_IPC)) {
- r = setup_shareable_ns(runtime->shared->ipcns_storage_socket, CLONE_NEWIPC);
- if (r == -EPERM)
- log_unit_warning_errno(unit, r,
- "PrivateIPC=yes is configured, but IPC namespace setup failed, ignoring: %m");
- else if (r < 0) {
- *exit_status = EXIT_NAMESPACE;
- return log_unit_error_errno(unit, r, "Failed to set up IPC namespacing: %m");
- }
- } else if (context->ipc_namespace_path) {
- *exit_status = EXIT_NAMESPACE;
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EOPNOTSUPP),
- "IPCNamespacePath= is not supported, refusing.");
- } else
- log_unit_warning(unit, "PrivateIPC=yes is configured, but the kernel does not support IPC namespaces, ignoring.");
- }
-
- if (needs_mount_namespace) {
- _cleanup_free_ char *error_path = NULL;
-
- r = apply_mount_namespace(unit, command->flags, context, params, runtime, memory_pressure_path, creds_path, creds_fd, &error_path);
- if (r < 0) {
- *exit_status = EXIT_NAMESPACE;
- return log_unit_error_errno(unit, r, "Failed to set up mount namespacing%s%s: %m",
- error_path ? ": " : "", strempty(error_path));
- }
- }
-
- if (creds_fd >= 0) {
- assert(creds_path);
-
- /* When a mount namespace is not requested, then the target directory may not exist yet.
- * Here, we ignore the failure, as if it fails, the subsequent move_mount() will fail. */
- (void) mkdir_p_label(creds_path, 0755);
-
- if (move_mount(creds_fd, "", AT_FDCWD, creds_path, MOVE_MOUNT_F_EMPTY_PATH) < 0) {
- *exit_status = EXIT_CREDENTIALS;
- return log_unit_error_errno(unit, errno, "Failed to mount credentials directory on %s: %m", creds_path);
- }
- }
-
- if (needs_sandboxing) {
- r = apply_protect_hostname(unit, context, exit_status);
- if (r < 0)
- return r;
- }
-
- if (context->memory_ksm >= 0)
- if (prctl(PR_SET_MEMORY_MERGE, context->memory_ksm) < 0) {
- if (ERRNO_IS_NOT_SUPPORTED(errno))
- log_unit_debug_errno(unit, errno, "KSM support not available, ignoring.");
- else {
- *exit_status = EXIT_KSM;
- return log_unit_error_errno(unit, errno, "Failed to set KSM: %m");
- }
- }
-
- /* Drop groups as early as possible.
- * This needs to be done after PrivateDevices=y setup as device nodes should be owned by the host's root.
- * For non-root in a userns, devices will be owned by the user/group before the group change, and nobody. */
- if (needs_setuid) {
- _cleanup_free_ gid_t *gids_to_enforce = NULL;
- int ngids_to_enforce = 0;
-
- ngids_to_enforce = merge_gid_lists(supplementary_gids,
- ngids,
- gids_after_pam,
- ngids_after_pam,
- &gids_to_enforce);
- if (ngids_to_enforce < 0) {
- *exit_status = EXIT_MEMORY;
- return log_unit_error_errno(unit,
- ngids_to_enforce,
- "Failed to merge group lists. Group membership might be incorrect: %m");
- }
-
- r = enforce_groups(gid, gids_to_enforce, ngids_to_enforce);
- if (r < 0) {
- *exit_status = EXIT_GROUP;
- return log_unit_error_errno(unit, r, "Changing group credentials failed: %m");
- }
- }
-
- /* If the user namespace was not set up above, try to do it now.
- * It's preferred to set up the user namespace later (after all other namespaces) so as not to be
- * restricted by rules pertaining to combining user namespaces with other namespaces (e.g. in the
- * case of mount namespaces being less privileged when the mount point list is copied from a
- * different user namespace). */
-
- if (needs_sandboxing && context->private_users && !userns_set_up) {
- r = setup_private_users(saved_uid, saved_gid, uid, gid);
- if (r < 0) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, r, "Failed to set up user namespacing: %m");
- }
- }
-
- /* Now that the mount namespace has been set up and privileges adjusted, let's look for the thing we
- * shall execute. */
-
- _cleanup_free_ char *executable = NULL;
- _cleanup_close_ int executable_fd = -EBADF;
- r = find_executable_full(command->path, /* root= */ NULL, context->exec_search_path, false, &executable, &executable_fd);
- if (r < 0) {
- if (r != -ENOMEM && (command->flags & EXEC_COMMAND_IGNORE_FAILURE)) {
- log_unit_struct_errno(unit, LOG_INFO, r,
- "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
- LOG_UNIT_INVOCATION_ID(unit),
- LOG_UNIT_MESSAGE(unit, "Executable %s missing, skipping: %m",
- command->path),
- "EXECUTABLE=%s", command->path);
- *exit_status = EXIT_SUCCESS;
- return 0;
- }
-
- *exit_status = EXIT_EXEC;
- return log_unit_struct_errno(unit, LOG_INFO, r,
- "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
- LOG_UNIT_INVOCATION_ID(unit),
- LOG_UNIT_MESSAGE(unit, "Failed to locate executable %s: %m",
- command->path),
- "EXECUTABLE=%s", command->path);
- }
-
- r = add_shifted_fd(keep_fds, ELEMENTSOF(keep_fds), &n_keep_fds, executable_fd, &executable_fd);
- if (r < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, r, "Failed to shift fd and set FD_CLOEXEC: %m");
- }
-
-#if HAVE_SELINUX
- if (needs_sandboxing && use_selinux && params->selinux_context_net) {
- int fd = -EBADF;
-
- if (socket_fd >= 0)
- fd = socket_fd;
- else if (params->n_socket_fds == 1)
- /* If stdin is not connected to a socket but we are triggered by exactly one socket unit then we
- * use context from that fd to compute the label. */
- fd = params->fds[0];
-
- if (fd >= 0) {
- r = mac_selinux_get_child_mls_label(fd, executable, context->selinux_context, &mac_selinux_context_net);
- if (r < 0) {
- if (!context->selinux_context_ignore) {
- *exit_status = EXIT_SELINUX_CONTEXT;
- return log_unit_error_errno(unit, r, "Failed to determine SELinux context: %m");
- }
- log_unit_debug_errno(unit, r, "Failed to determine SELinux context, ignoring: %m");
- }
- }
- }
-#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 we don't need socket_fd and the netns and ipcns fds any
- * more. We do keep exec_fd however, if we have it, since we need to keep it open until the final
- * execve(). */
-
- r = close_all_fds(keep_fds, n_keep_fds);
- if (r >= 0)
- r = shift_fds(fds, n_fds);
- if (r >= 0)
- r = flags_fds(fds, n_socket_fds, n_fds, context->non_blocking);
- if (r < 0) {
- *exit_status = EXIT_FDS;
- return log_unit_error_errno(unit, r, "Failed to adjust passed file descriptors: %m");
- }
-
- /* At this point, the fds we want to pass to the program are all ready and set up, with O_CLOEXEC turned off
- * and at the right fd numbers. The are no other fds open, with one exception: the exec_fd if it is defined,
- * and it has O_CLOEXEC set, after all we want it to be closed by the execve(), so that our parent knows we
- * came this far. */
-
- secure_bits = context->secure_bits;
-
- if (needs_sandboxing) {
- uint64_t bset;
-
- /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested.
- * (Note this is placed after the general resource limit initialization, see above, in order
- * to take precedence.) */
- if (context->restrict_realtime && !context->rlimit[RLIMIT_RTPRIO]) {
- if (setrlimit(RLIMIT_RTPRIO, &RLIMIT_MAKE_CONST(0)) < 0) {
- *exit_status = EXIT_LIMITS;
- return log_unit_error_errno(unit, errno, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
- }
- }
-
-#if ENABLE_SMACK
- /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
- * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
- if (use_smack) {
- r = setup_smack(unit->manager, context, executable_fd);
- if (r < 0 && !context->smack_process_label_ignore) {
- *exit_status = EXIT_SMACK_PROCESS_LABEL;
- return log_unit_error_errno(unit, r, "Failed to set SMACK process label: %m");
- }
- }
-#endif
-
- bset = context->capability_bounding_set;
- /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
- * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
- * instead of us doing that */
- if (needs_ambient_hack)
- bset |= (UINT64_C(1) << CAP_SETPCAP) |
- (UINT64_C(1) << CAP_SETUID) |
- (UINT64_C(1) << CAP_SETGID);
-
- if (!cap_test_all(bset)) {
- r = capability_bounding_set_drop(bset, /* right_now= */ false);
- if (r < 0) {
- *exit_status = EXIT_CAPABILITIES;
- return log_unit_error_errno(unit, r, "Failed to drop capabilities: %m");
- }
- }
-
- /* Ambient capabilities are cleared during setresuid() (in enforce_user()) even with
- * keep-caps set.
- *
- * To be able to raise the ambient capabilities after setresuid() they have to be added to
- * the inherited set and keep caps has to be set (done in enforce_user()). After setresuid()
- * the ambient capabilities can be raised as they are present in the permitted and
- * inhertiable set. However it is possible that someone wants to set ambient capabilities
- * without changing the user, so we also set the ambient capabilities here.
- *
- * The requested ambient capabilities are raised in the inheritable set if the second
- * argument is true. */
- if (!needs_ambient_hack) {
- r = capability_ambient_set_apply(capability_ambient_set, /* also_inherit= */ true);
- if (r < 0) {
- *exit_status = EXIT_CAPABILITIES;
- return log_unit_error_errno(unit, r, "Failed to apply ambient capabilities (before UID change): %m");
- }
- }
- }
-
- /* chroot to root directory first, before we lose the ability to chroot */
- r = apply_root_directory(context, params, runtime, needs_mount_namespace, exit_status);
- if (r < 0)
- return log_unit_error_errno(unit, r, "Chrooting to the requested root directory failed: %m");
-
- if (needs_setuid) {
- if (uid_is_valid(uid)) {
- r = enforce_user(context, uid, capability_ambient_set);
- if (r < 0) {
- *exit_status = EXIT_USER;
- return log_unit_error_errno(unit, r, "Failed to change UID to " UID_FMT ": %m", uid);
- }
-
- if (!needs_ambient_hack && capability_ambient_set != 0) {
-
- /* Raise the ambient capabilities after user change. */
- r = capability_ambient_set_apply(capability_ambient_set, /* also_inherit= */ false);
- if (r < 0) {
- *exit_status = EXIT_CAPABILITIES;
- return log_unit_error_errno(unit, r, "Failed to apply ambient capabilities (after UID change): %m");
- }
- }
- }
- }
-
- /* Apply working directory here, because the working directory might be on NFS and only the user running
- * this service might have the correct privilege to change to the working directory */
- r = apply_working_directory(context, params, runtime, home, exit_status);
- if (r < 0)
- return log_unit_error_errno(unit, r, "Changing to the requested working directory failed: %m");
-
- if (needs_sandboxing) {
- /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
- * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
- * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
- * are restricted. */
-
-#if HAVE_SELINUX
- if (use_selinux) {
- char *exec_context = mac_selinux_context_net ?: context->selinux_context;
-
- if (exec_context) {
- r = setexeccon(exec_context);
- if (r < 0) {
- if (!context->selinux_context_ignore) {
- *exit_status = EXIT_SELINUX_CONTEXT;
- return log_unit_error_errno(unit, r, "Failed to change SELinux context to %s: %m", exec_context);
- }
- log_unit_debug_errno(unit, r, "Failed to change SELinux context to %s, ignoring: %m", exec_context);
- }
- }
- }
-#endif
-
-#if HAVE_APPARMOR
- if (use_apparmor && context->apparmor_profile) {
- r = aa_change_onexec(context->apparmor_profile);
- if (r < 0 && !context->apparmor_profile_ignore) {
- *exit_status = EXIT_APPARMOR_PROFILE;
- return log_unit_error_errno(unit, errno, "Failed to prepare AppArmor profile change to %s: %m", context->apparmor_profile);
- }
- }
-#endif
-
- /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential
- * EPERMs we'll try not to call PR_SET_SECUREBITS unless necessary. Setting securebits
- * requires CAP_SETPCAP. */
- if (prctl(PR_GET_SECUREBITS) != secure_bits) {
- /* CAP_SETPCAP is required to set securebits. This capability is raised into the
- * effective set here.
- *
- * The effective set is overwritten during execve() with the following values:
- *
- * - ambient set (for non-root processes)
- *
- * - (inheritable | bounding) set for root processes)
- *
- * Hence there is no security impact to raise it in the effective set before execve
- */
- r = capability_gain_cap_setpcap(/* return_caps= */ NULL);
- if (r < 0) {
- *exit_status = EXIT_CAPABILITIES;
- return log_unit_error_errno(unit, r, "Failed to gain CAP_SETPCAP for setting secure bits");
- }
- if (prctl(PR_SET_SECUREBITS, secure_bits) < 0) {
- *exit_status = EXIT_SECUREBITS;
- return log_unit_error_errno(unit, errno, "Failed to set process secure bits: %m");
- }
- }
-
- if (context_has_no_new_privileges(context))
- if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
- *exit_status = EXIT_NO_NEW_PRIVILEGES;
- return log_unit_error_errno(unit, errno, "Failed to disable new privileges: %m");
- }
-
-#if HAVE_SECCOMP
- r = apply_address_families(unit, context);
- if (r < 0) {
- *exit_status = EXIT_ADDRESS_FAMILIES;
- return log_unit_error_errno(unit, r, "Failed to restrict address families: %m");
- }
-
- r = apply_memory_deny_write_execute(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to disable writing to executable memory: %m");
- }
-
- r = apply_restrict_realtime(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply realtime restrictions: %m");
- }
-
- r = apply_restrict_suid_sgid(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply SUID/SGID restrictions: %m");
- }
-
- r = apply_restrict_namespaces(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply namespace restrictions: %m");
- }
-
- r = apply_protect_sysctl(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply sysctl restrictions: %m");
- }
-
- r = apply_protect_kernel_modules(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply module loading restrictions: %m");
- }
-
- r = apply_protect_kernel_logs(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply kernel log restrictions: %m");
- }
-
- r = apply_protect_clock(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply clock restrictions: %m");
- }
-
- r = apply_private_devices(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to set up private devices: %m");
- }
-
- r = apply_syscall_archs(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply syscall architecture restrictions: %m");
- }
-
- r = apply_lock_personality(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to lock personalities: %m");
- }
-
- r = apply_syscall_log(unit, context);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply system call log filters: %m");
- }
-
- /* This really should remain the last step before the execve(), to make sure our own code is unaffected
- * by the filter as little as possible. */
- r = apply_syscall_filter(unit, context, needs_ambient_hack);
- if (r < 0) {
- *exit_status = EXIT_SECCOMP;
- return log_unit_error_errno(unit, r, "Failed to apply system call filters: %m");
- }
-#endif
-
-#if HAVE_LIBBPF
- r = apply_restrict_filesystems(unit, context);
- if (r < 0) {
- *exit_status = EXIT_BPF;
- return log_unit_error_errno(unit, r, "Failed to restrict filesystems: %m");
- }
-#endif
-
- }
-
- if (!strv_isempty(context->unset_environment)) {
- char **ee = NULL;
-
- ee = strv_env_delete(accum_env, 1, context->unset_environment);
- if (!ee) {
- *exit_status = EXIT_MEMORY;
- return log_oom();
- }
-
- strv_free_and_replace(accum_env, ee);
- }
-
- if (!FLAGS_SET(command->flags, EXEC_COMMAND_NO_ENV_EXPAND)) {
- _cleanup_strv_free_ char **unset_variables = NULL, **bad_variables = NULL;
-
- r = replace_env_argv(command->argv, accum_env, &replaced_argv, &unset_variables, &bad_variables);
- if (r < 0) {
- *exit_status = EXIT_MEMORY;
- return log_unit_error_errno(unit, r, "Failed to replace environment variables: %m");
- }
- final_argv = replaced_argv;
-
- if (!strv_isempty(unset_variables)) {
- _cleanup_free_ char *ju = strv_join(unset_variables, ", ");
- log_unit_warning(unit, "Referenced but unset environment variable evaluates to an empty string: %s", strna(ju));
- }
-
- if (!strv_isempty(bad_variables)) {
- _cleanup_free_ char *jb = strv_join(bad_variables, ", ");
- log_unit_warning(unit, "Invalid environment variable name evaluates to an empty string: %s", strna(jb));;
- }
- } else
- final_argv = command->argv;
-
- log_command_line(unit, "Executing", executable, final_argv);
-
- if (exec_fd >= 0) {
- uint8_t hot = 1;
-
- /* We have finished with all our initializations. Let's now let the manager know that. From this point
- * on, if the manager sees POLLHUP on the exec_fd, then execve() was successful. */
-
- if (write(exec_fd, &hot, sizeof(hot)) < 0) {
- *exit_status = EXIT_EXEC;
- return log_unit_error_errno(unit, errno, "Failed to enable exec_fd: %m");
- }
- }
-
- r = fexecve_or_execve(executable_fd, executable, final_argv, accum_env);
-
- if (exec_fd >= 0) {
- uint8_t hot = 0;
-
- /* The execve() failed. This means the exec_fd is still open. Which means we need to tell the manager
- * that POLLHUP on it no longer means execve() succeeded. */
+ if (!DEBUG_LOGGING)
+ return;
- if (write(exec_fd, &hot, sizeof(hot)) < 0) {
- *exit_status = EXIT_EXEC;
- return log_unit_error_errno(unit, errno, "Failed to disable exec_fd: %m");
- }
- }
+ _cleanup_free_ char *cmdline = quote_command_line(argv, SHELL_ESCAPE_EMPTY);
- *exit_status = EXIT_EXEC;
- return log_unit_error_errno(unit, r, "Failed to execute %s: %m", executable);
+ log_unit_struct(unit, LOG_DEBUG,
+ "EXECUTABLE=%s", executable,
+ LOG_UNIT_MESSAGE(unit, "%s: %s", msg, strnull(cmdline)),
+ LOG_UNIT_INVOCATION_ID(unit));
}
static int exec_context_load_environment(const Unit *unit, const ExecContext *c, char ***l);
-static int exec_context_named_iofds(const ExecContext *c, const ExecParameters *p, int named_iofds[static 3]);
int exec_spawn(Unit *unit,
ExecCommand *command,
const ExecContext *context,
- const ExecParameters *params,
+ ExecParameters *params,
ExecRuntime *runtime,
const CGroupContext *cgroup_context,
pid_t *ret) {
- int socket_fd, r, named_iofds[3] = { -1, -1, -1 }, *fds = NULL;
- _cleanup_free_ char *subcgroup_path = NULL;
- _cleanup_strv_free_ char **files_env = NULL;
- size_t n_storage_fds = 0, n_socket_fds = 0;
+ char serialization_fd_number[DECIMAL_STR_MAX(int) + 1];
+ _cleanup_free_ char *subcgroup_path = NULL, *log_level = NULL, *executor_path = NULL;
+ _cleanup_fdset_free_ FDSet *fdset = NULL;
+ _cleanup_fclose_ FILE *f = NULL;
pid_t pid;
+ int r;
assert(unit);
+ assert(unit->manager);
+ assert(unit->manager->executor_fd >= 0);
assert(command);
assert(context);
assert(ret);
assert(params);
assert(params->fds || (params->n_socket_fds + params->n_storage_fds <= 0));
+ assert(!params->files_env); /* We fill this field, ensure it comes NULL-initialized to us */
LOG_CONTEXT_PUSH_UNIT(unit);
- if (context->std_input == EXEC_INPUT_SOCKET ||
- context->std_output == EXEC_OUTPUT_SOCKET ||
- context->std_error == EXEC_OUTPUT_SOCKET) {
-
- if (params->n_socket_fds > 1)
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EINVAL), "Got more than one socket.");
-
- if (params->n_socket_fds == 0)
- return log_unit_error_errno(unit, SYNTHETIC_ERRNO(EINVAL), "Got no socket.");
-
- socket_fd = params->fds[0];
- } else {
- socket_fd = -EBADF;
- fds = params->fds;
- n_socket_fds = params->n_socket_fds;
- n_storage_fds = params->n_storage_fds;
- }
-
- r = exec_context_named_iofds(context, params, named_iofds);
- if (r < 0)
- return log_unit_error_errno(unit, r, "Failed to load a named file descriptor: %m");
-
- r = exec_context_load_environment(unit, context, &files_env);
+ r = exec_context_load_environment(unit, context, ¶ms->files_env);
if (r < 0)
return log_unit_error_errno(unit, r, "Failed to load environment files: %m");
log_command_line(unit, "About to execute", command->path, command->argv);
if (params->cgroup_path) {
- r = exec_parameters_get_cgroup_path(params, cgroup_context, &subcgroup_path);
+ r = exec_params_get_cgroup_path(params, cgroup_context, &subcgroup_path);
if (r < 0)
return log_unit_error_errno(unit, r, "Failed to acquire subcgroup path: %m");
if (r > 0) {
}
}
- pid = fork();
- if (pid < 0)
- return log_unit_error_errno(unit, errno, "Failed to fork: %m");
-
- if (pid == 0) {
- int exit_status;
-
- r = exec_child(unit,
- command,
- context,
- params,
- runtime,
- cgroup_context,
- socket_fd,
- named_iofds,
- fds,
- n_socket_fds,
- n_storage_fds,
- files_env,
- unit->manager->user_lookup_fds[1],
- &exit_status);
+ /* In order to avoid copy-on-write traps and OOM-kills when pid1's memory.current is above the
+ * child's memory.max, serialize all the state needed to start the unit, and pass it to the
+ * systemd-executor binary. clone() with CLONE_VM + CLONE_VFORK will pause the parent until the exec
+ * and ensure all memory is shared. The child immediately execs the new binary so the delay should
+ * be minimal. Once glibc provides a clone3 wrapper we can switch to that, and clone directly in the
+ * target cgroup. */
- if (r < 0) {
- const char *status = ASSERT_PTR(
- exit_status_to_string(exit_status, EXIT_STATUS_LIBC | EXIT_STATUS_SYSTEMD));
-
- log_unit_struct_errno(unit, LOG_ERR, r,
- "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
- LOG_UNIT_INVOCATION_ID(unit),
- LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m",
- status, command->path),
- "EXECUTABLE=%s", command->path);
- } else
- assert(exit_status == EXIT_SUCCESS);
+ r = open_serialization_file("sd-executor-state", &f);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to open serialization stream: %m");
- _exit(exit_status);
- }
+ fdset = fdset_new();
+ if (!fdset)
+ return log_oom();
+
+ r = exec_serialize_invocation(f, fdset, context, command, params, runtime, cgroup_context);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to serialize parameters: %m");
+
+ if (fseeko(f, 0, SEEK_SET) == (off_t) -1)
+ return log_unit_error_errno(unit, errno, "Failed to reseek on serialization stream: %m");
+
+ r = fd_cloexec(fileno(f), false);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to set O_CLOEXEC on serialization fd: %m");
+
+ r = fdset_cloexec(fdset, false);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to set O_CLOEXEC on serialized fds: %m");
+
+ r = log_level_to_string_alloc(log_get_max_level(), &log_level);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to convert log level to string: %m");
+
+ r = fd_get_path(unit->manager->executor_fd, &executor_path);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to get executor path from fd: %m");
+
+ xsprintf(serialization_fd_number, "%i", fileno(f));
+
+ /* The executor binary is pinned, to avoid compatibility problems during upgrades. */
+ r = posix_spawn_wrapper(
+ FORMAT_PROC_FD_PATH(unit->manager->executor_fd),
+ STRV_MAKE(executor_path,
+ "--deserialize", serialization_fd_number,
+ "--log-level", log_level,
+ "--log-target", log_target_to_string(manager_get_executor_log_target(unit->manager))),
+ environ,
+ &pid);
+ if (r < 0)
+ return log_unit_error_errno(unit, r, "Failed to spawn executor: %m");
log_unit_debug(unit, "Forked %s as "PID_FMT, command->path, pid);
void exec_context_init(ExecContext *c) {
assert(c);
- c->umask = 0022;
- c->ioprio = IOPRIO_DEFAULT_CLASS_AND_PRIO;
- 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;
- for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++)
- c->directories[t].mode = 0755;
- c->timeout_clean_usec = USEC_INFINITY;
- c->capability_bounding_set = CAP_MASK_UNSET;
- assert_cc(NAMESPACE_FLAGS_INITIAL != NAMESPACE_FLAGS_ALL);
- c->restrict_namespaces = NAMESPACE_FLAGS_INITIAL;
- c->log_level_max = -1;
+ /* When initializing a bool member to 'true', make sure to serialize in execute-serialize.c using
+ * serialize_bool() instead of serialize_bool_elide(). */
+
+ *c = (ExecContext) {
+ .umask = 0022,
+ .ioprio = IOPRIO_DEFAULT_CLASS_AND_PRIO,
+ .cpu_sched_policy = SCHED_OTHER,
+ .syslog_priority = LOG_DAEMON|LOG_INFO,
+ .syslog_level_prefix = true,
+ .ignore_sigpipe = true,
+ .timer_slack_nsec = NSEC_INFINITY,
+ .personality = PERSONALITY_INVALID,
+ .timeout_clean_usec = USEC_INFINITY,
+ .capability_bounding_set = CAP_MASK_UNSET,
+ .restrict_namespaces = NAMESPACE_FLAGS_INITIAL,
+ .log_level_max = -1,
#if HAVE_SECCOMP
- c->syscall_errno = SECCOMP_ERROR_NUMBER_KILL;
+ .syscall_errno = SECCOMP_ERROR_NUMBER_KILL,
#endif
- c->tty_rows = UINT_MAX;
- c->tty_cols = UINT_MAX;
+ .tty_rows = UINT_MAX,
+ .tty_cols = UINT_MAX,
+ .private_mounts = -1,
+ .memory_ksm = -1,
+ .set_login_environment = -1,
+ };
+
+ FOREACH_ARRAY(d, c->directories, _EXEC_DIRECTORY_TYPE_MAX)
+ d->mode = 0755;
+
numa_policy_reset(&c->numa_policy);
- c->private_mounts = -1;
- c->memory_ksm = -1;
+
+ assert_cc(NAMESPACE_FLAGS_INITIAL != NAMESPACE_FLAGS_ALL);
}
void exec_context_done(ExecContext *c) {
if (!runtime_prefix)
return 0;
- for (size_t i = 0; i < c->directories[EXEC_DIRECTORY_RUNTIME].n_items; i++) {
+ FOREACH_ARRAY(i, c->directories[EXEC_DIRECTORY_RUNTIME].items, c->directories[EXEC_DIRECTORY_RUNTIME].n_items) {
_cleanup_free_ char *p = NULL;
if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME))
- p = path_join(runtime_prefix, "private", c->directories[EXEC_DIRECTORY_RUNTIME].items[i].path);
+ p = path_join(runtime_prefix, "private", i->path);
else
- p = path_join(runtime_prefix, c->directories[EXEC_DIRECTORY_RUNTIME].items[i].path);
+ p = path_join(runtime_prefix, i->path);
if (!p)
return -ENOMEM;
* service next. */
(void) rm_rf(p, REMOVE_ROOT);
- STRV_FOREACH(symlink, c->directories[EXEC_DIRECTORY_RUNTIME].items[i].symlinks) {
+ STRV_FOREACH(symlink, i->symlinks) {
_cleanup_free_ char *symlink_abs = NULL;
if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME))
return 0;
}
-static void exec_command_done(ExecCommand *c) {
+void exec_command_done(ExecCommand *c) {
assert(c);
c->path = mfree(c->path);
}
void exec_command_done_array(ExecCommand *c, size_t n) {
- for (size_t i = 0; i < n; i++)
- exec_command_done(c+i);
+ FOREACH_ARRAY(i, c, n)
+ exec_command_done(i);
}
ExecCommand* exec_command_free_list(ExecCommand *c) {
}
void exec_command_free_array(ExecCommand **c, size_t n) {
- for (size_t i = 0; i < n; i++)
- c[i] = exec_command_free_list(c[i]);
+ FOREACH_ARRAY(i, c, n)
+ *i = exec_command_free_list(*i);
}
void exec_command_reset_status_array(ExecCommand *c, size_t n) {
- for (size_t i = 0; i < n; i++)
- exec_status_reset(&c[i].exec_status);
+ FOREACH_ARRAY(i, c, n)
+ exec_status_reset(&i->exec_status);
}
void exec_command_reset_status_list_array(ExecCommand **c, size_t n) {
- for (size_t i = 0; i < n; i++)
- LIST_FOREACH(command, z, c[i])
+ FOREACH_ARRAY(i, c, n)
+ LIST_FOREACH(command, z, *i)
exec_status_reset(&z->exec_status);
}
}
}
-static int exec_context_named_iofds(
- const ExecContext *c,
- const ExecParameters *p,
- int named_iofds[static 3]) {
-
- size_t targets;
- const char* stdio_fdname[3];
- size_t n_fds;
-
- assert(c);
- assert(p);
- assert(named_iofds);
-
- targets = (c->std_input == EXEC_INPUT_NAMED_FD) +
- (c->std_output == EXEC_OUTPUT_NAMED_FD) +
- (c->std_error == EXEC_OUTPUT_NAMED_FD);
-
- for (size_t i = 0; i < 3; i++)
- stdio_fdname[i] = exec_context_fdname(c, i);
-
- n_fds = p->n_storage_fds + p->n_socket_fds;
-
- for (size_t i = 0; i < n_fds && targets > 0; i++)
- if (named_iofds[STDIN_FILENO] < 0 &&
- c->std_input == EXEC_INPUT_NAMED_FD &&
- stdio_fdname[STDIN_FILENO] &&
- streq(p->fd_names[i], stdio_fdname[STDIN_FILENO])) {
-
- named_iofds[STDIN_FILENO] = p->fds[i];
- targets--;
-
- } else if (named_iofds[STDOUT_FILENO] < 0 &&
- c->std_output == EXEC_OUTPUT_NAMED_FD &&
- stdio_fdname[STDOUT_FILENO] &&
- streq(p->fd_names[i], stdio_fdname[STDOUT_FILENO])) {
-
- named_iofds[STDOUT_FILENO] = p->fds[i];
- targets--;
-
- } else if (named_iofds[STDERR_FILENO] < 0 &&
- c->std_error == EXEC_OUTPUT_NAMED_FD &&
- stdio_fdname[STDERR_FILENO] &&
- streq(p->fd_names[i], stdio_fdname[STDERR_FILENO])) {
-
- named_iofds[STDERR_FILENO] = p->fds[i];
- targets--;
- }
-
- return targets == 0 ? 0 : -ENOENT;
-}
-
static int exec_context_load_environment(const Unit *unit, const ExecContext *c, char ***ret) {
_cleanup_strv_free_ char **v = NULL;
int r;
/* When we don't match anything, -ENOENT should be returned */
assert(pglob.gl_pathc > 0);
- for (size_t n = 0; n < pglob.gl_pathc; n++) {
+ FOREACH_ARRAY(path, pglob.gl_pathv, pglob.gl_pathc) {
_cleanup_strv_free_ char **p = NULL;
- r = load_env_file(NULL, pglob.gl_pathv[n], &p);
+ r = load_env_file(NULL, *path, &p);
if (r < 0) {
if (ignore)
continue;
if (p) {
InvalidEnvInfo info = {
.unit = unit,
- .path = pglob.gl_pathv[n]
+ .path = *path,
};
p = strv_env_clean_with_callback(p, invalid_env, &info);
}
}
+void exec_params_dump(const ExecParameters *p, FILE* f, const char *prefix) {
+ assert(p);
+ assert(f);
+
+ prefix = strempty(prefix);
+
+ fprintf(f,
+ "%sRuntimeScope: %s\n"
+ "%sExecFlags: %u\n"
+ "%sSELinuxContextNetwork: %s\n"
+ "%sCgroupSupportedMask: %u\n"
+ "%sCgroupPath: %s\n"
+ "%sCrededentialsDirectory: %s\n"
+ "%sEncryptedCredentialsDirectory: %s\n"
+ "%sConfirmSpawn: %s\n"
+ "%sShallConfirmSpawn: %s\n"
+ "%sWatchdogUSec: " USEC_FMT "\n"
+ "%sNotifySocket: %s\n"
+ "%sFallbackSmackProcessLabel: %s\n",
+ prefix, runtime_scope_to_string(p->runtime_scope),
+ prefix, p->flags,
+ prefix, yes_no(p->selinux_context_net),
+ prefix, p->cgroup_supported,
+ prefix, p->cgroup_path,
+ prefix, strempty(p->received_credentials_directory),
+ prefix, strempty(p->received_encrypted_credentials_directory),
+ prefix, strempty(p->confirm_spawn),
+ prefix, yes_no(p->shall_confirm_spawn),
+ prefix, p->watchdog_usec,
+ prefix, strempty(p->notify_socket),
+ prefix, strempty(p->fallback_smack_process_label));
+
+ strv_dump(f, prefix, "FdNames", p->fd_names);
+ strv_dump(f, prefix, "Environment", p->environment);
+ strv_dump(f, prefix, "Prefix", p->prefix);
+
+ LIST_FOREACH(open_files, file, p->open_files)
+ fprintf(f, "%sOpenFile: %s %s", prefix, file->path, open_file_flags_to_string(file->flags));
+
+ strv_dump(f, prefix, "FilesEnv", p->files_env);
+}
+
void exec_context_dump(const ExecContext *c, FILE* f, const char *prefix) {
int r;
prefix, protect_proc_to_string(c->protect_proc),
prefix, proc_subset_to_string(c->proc_subset));
+ if (c->set_login_environment >= 0)
+ fprintf(f, "%sSetLoginEnvironment: %s\n", prefix, yes_no(c->set_login_environment > 0));
+
if (c->root_image)
fprintf(f, "%sRootImage: %s\n", prefix, c->root_image);
fprintf(f, "%sTimeoutCleanSec: %s\n", prefix, FORMAT_TIMESPAN(c->timeout_clean_usec, USEC_PER_SEC));
+ if (c->memory_ksm >= 0)
+ fprintf(f, "%sMemoryKSM: %s\n", prefix, yes_no(c->memory_ksm > 0));
+
if (c->nice_set)
fprintf(f, "%sNice: %i\n", prefix, c->nice);
fputc('\n', f);
}
- for (size_t j = 0; j < c->n_log_extra_fields; j++) {
+ FOREACH_ARRAY(field, c->log_extra_fields, c->n_log_extra_fields) {
fprintf(f, "%sLogExtraFields: ", prefix);
- fwrite(c->log_extra_fields[j].iov_base,
- 1, c->log_extra_fields[j].iov_len,
- f);
+ fwrite(field->iov_base, 1, field->iov_len, f);
fputc('\n', f);
}
strv_dump(f, prefix, "NoExecPaths", c->no_exec_paths);
strv_dump(f, prefix, "ExecSearchPath", c->exec_search_path);
- for (size_t i = 0; i < c->n_bind_mounts; i++)
+ FOREACH_ARRAY(mount, c->bind_mounts, c->n_bind_mounts)
fprintf(f, "%s%s: %s%s:%s:%s\n", prefix,
- c->bind_mounts[i].read_only ? "BindReadOnlyPaths" : "BindPaths",
- c->bind_mounts[i].ignore_enoent ? "-": "",
- c->bind_mounts[i].source,
- c->bind_mounts[i].destination,
- c->bind_mounts[i].recursive ? "rbind" : "norbind");
-
- for (size_t i = 0; i < c->n_temporary_filesystems; i++) {
- const TemporaryFileSystem *t = c->temporary_filesystems + i;
+ mount->read_only ? "BindReadOnlyPaths" : "BindPaths",
+ mount->ignore_enoent ? "-": "",
+ mount->source,
+ mount->destination,
+ mount->recursive ? "rbind" : "norbind");
+ FOREACH_ARRAY(tmpfs, c->temporary_filesystems, c->n_temporary_filesystems)
fprintf(f, "%sTemporaryFileSystem: %s%s%s\n", prefix,
- t->path,
- isempty(t->options) ? "" : ":",
- strempty(t->options));
- }
+ tmpfs->path,
+ isempty(tmpfs->options) ? "" : ":",
+ strempty(tmpfs->options));
if (c->utmp_id)
fprintf(f,
fputc('\n', f);
}
- for (size_t i = 0; i < c->n_mount_images; i++) {
+ FOREACH_ARRAY(mount, c->mount_images, c->n_mount_images) {
fprintf(f, "%sMountImages: %s%s:%s", prefix,
- c->mount_images[i].ignore_enoent ? "-": "",
- c->mount_images[i].source,
- c->mount_images[i].destination);
- LIST_FOREACH(mount_options, o, c->mount_images[i].mount_options)
+ mount->ignore_enoent ? "-": "",
+ mount->source,
+ mount->destination);
+ LIST_FOREACH(mount_options, o, mount->mount_options)
fprintf(f, ":%s:%s",
partition_designator_to_string(o->partition_designator),
strempty(o->options));
fprintf(f, "\n");
}
- for (size_t i = 0; i < c->n_extension_images; i++) {
+ FOREACH_ARRAY(mount, c->extension_images, c->n_extension_images) {
fprintf(f, "%sExtensionImages: %s%s", prefix,
- c->extension_images[i].ignore_enoent ? "-": "",
- c->extension_images[i].source);
- LIST_FOREACH(mount_options, o, c->extension_images[i].mount_options)
+ mount->ignore_enoent ? "-": "",
+ mount->source);
+ LIST_FOREACH(mount_options, o, mount->mount_options)
fprintf(f, ":%s:%s",
partition_designator_to_string(o->partition_designator),
strempty(o->options));
void exec_context_free_log_extra_fields(ExecContext *c) {
assert(c);
- for (size_t l = 0; l < c->n_log_extra_fields; l++)
- free(c->log_extra_fields[l].iov_base);
+ FOREACH_ARRAY(field, c->log_extra_fields, c->n_log_extra_fields)
+ free(field->iov_base);
+
c->log_extra_fields = mfree(c->log_extra_fields);
c->n_log_extra_fields = 0;
}
if (!prefix[t])
continue;
- for (size_t i = 0; i < c->directories[t].n_items; i++) {
+ FOREACH_ARRAY(i, c->directories[t].items, c->directories[t].n_items) {
char *j;
- j = path_join(prefix[t], c->directories[t].items[i].path);
+ j = path_join(prefix[t], i->path);
if (!j)
return -ENOMEM;
/* Also remove private directories unconditionally. */
if (t != EXEC_DIRECTORY_CONFIGURATION) {
- j = path_join(prefix[t], "private", c->directories[t].items[i].path);
+ j = path_join(prefix[t], "private", i->path);
if (!j)
return -ENOMEM;
return r;
}
- STRV_FOREACH(symlink, c->directories[t].items[i].symlinks) {
+ STRV_FOREACH(symlink, i->symlinks) {
j = path_join(prefix[t], *symlink);
if (!j)
return -ENOMEM;
return 0;
}
+int exec_context_get_oom_score_adjust(const ExecContext *c) {
+ int n = 0, r;
+
+ assert(c);
+
+ if (c->oom_score_adjust_set)
+ return c->oom_score_adjust;
+
+ r = get_oom_score_adjust(&n);
+ if (r < 0)
+ log_debug_errno(r, "Failed to read /proc/self/oom_score_adj, ignoring: %m");
+
+ return n;
+}
+
+uint64_t exec_context_get_coredump_filter(const ExecContext *c) {
+ _cleanup_free_ char *t = NULL;
+ uint64_t n = COREDUMP_FILTER_MASK_DEFAULT;
+ int r;
+
+ assert(c);
+
+ if (c->coredump_filter_set)
+ return c->coredump_filter;
+
+ r = read_one_line_file("/proc/self/coredump_filter", &t);
+ if (r < 0)
+ log_debug_errno(r, "Failed to read /proc/self/coredump_filter, ignoring: %m");
+ else {
+ r = safe_atoux64(t, &n);
+ if (r < 0)
+ log_debug_errno(r, "Failed to parse \"%s\" from /proc/self/coredump_filter, ignoring: %m", t);
+ }
+
+ return n;
+}
+
+int exec_context_get_nice(const ExecContext *c) {
+ int n;
+
+ assert(c);
+
+ if (c->nice_set)
+ return c->nice;
+
+ errno = 0;
+ n = getpriority(PRIO_PROCESS, 0);
+ if (errno > 0) {
+ log_debug_errno(errno, "Failed to get process nice value, ignoring: %m");
+ n = 0;
+ }
+
+ return n;
+}
+
+int exec_context_get_cpu_sched_policy(const ExecContext *c) {
+ int n;
+
+ assert(c);
+
+ if (c->cpu_sched_set)
+ return c->cpu_sched_policy;
+
+ n = sched_getscheduler(0);
+ if (n < 0)
+ log_debug_errno(errno, "Failed to get scheduler policy, ignoring: %m");
+
+ return n < 0 ? SCHED_OTHER : n;
+}
+
+int exec_context_get_cpu_sched_priority(const ExecContext *c) {
+ struct sched_param p = {};
+ int r;
+
+ assert(c);
+
+ if (c->cpu_sched_set)
+ return c->cpu_sched_priority;
+
+ r = sched_getparam(0, &p);
+ if (r < 0)
+ log_debug_errno(errno, "Failed to get scheduler priority, ignoring: %m");
+
+ return r >= 0 ? p.sched_priority : 0;
+}
+
+uint64_t exec_context_get_timer_slack_nsec(const ExecContext *c) {
+ int r;
+
+ assert(c);
+
+ if (c->timer_slack_nsec != NSEC_INFINITY)
+ return c->timer_slack_nsec;
+
+ r = prctl(PR_GET_TIMERSLACK);
+ if (r < 0)
+ log_debug_errno(r, "Failed to get timer slack, ignoring: %m");
+
+ return (uint64_t) MAX(r, 0);
+}
+
+char** exec_context_get_syscall_filter(const ExecContext *c) {
+ _cleanup_strv_free_ char **l = NULL;
+
+ assert(c);
+
+#if HAVE_SECCOMP
+ void *id, *val;
+ HASHMAP_FOREACH_KEY(val, id, c->syscall_filter) {
+ _cleanup_free_ char *name = NULL;
+ const char *e = NULL;
+ char *s;
+ int num = PTR_TO_INT(val);
+
+ if (c->syscall_allow_list && num >= 0)
+ /* syscall with num >= 0 in allow-list is denied. */
+ continue;
+
+ name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
+ if (!name)
+ continue;
+
+ if (num >= 0) {
+ e = seccomp_errno_or_action_to_string(num);
+ if (e) {
+ s = strjoin(name, ":", e);
+ if (!s)
+ return NULL;
+ } else {
+ if (asprintf(&s, "%s:%d", name, num) < 0)
+ return NULL;
+ }
+ } else
+ s = TAKE_PTR(name);
+
+ if (strv_consume(&l, s) < 0)
+ return NULL;
+ }
+
+ strv_sort(l);
+#endif
+
+ return l ? TAKE_PTR(l) : strv_new(NULL);
+}
+
+char** exec_context_get_syscall_archs(const ExecContext *c) {
+ _cleanup_strv_free_ char **l = NULL;
+
+ assert(c);
+
+#if HAVE_SECCOMP
+ void *id;
+ SET_FOREACH(id, c->syscall_archs) {
+ const char *name;
+
+ name = seccomp_arch_to_string(PTR_TO_UINT32(id) - 1);
+ if (!name)
+ continue;
+
+ if (strv_extend(&l, name) < 0)
+ return NULL;
+ }
+
+ strv_sort(l);
+#endif
+
+ return l ? TAKE_PTR(l) : strv_new(NULL);
+}
+
+char** exec_context_get_syscall_log(const ExecContext *c) {
+ _cleanup_strv_free_ char **l = NULL;
+
+ assert(c);
+
+#if HAVE_SECCOMP
+ void *id, *val;
+ HASHMAP_FOREACH_KEY(val, id, c->syscall_log) {
+ char *name = NULL;
+
+ name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
+ if (!name)
+ continue;
+
+ if (strv_consume(&l, name) < 0)
+ return NULL;
+ }
+
+ strv_sort(l);
+#endif
+
+ return l ? TAKE_PTR(l) : strv_new(NULL);
+}
+
+char** exec_context_get_address_families(const ExecContext *c) {
+ _cleanup_strv_free_ char **l = NULL;
+ void *af;
+
+ assert(c);
+
+ SET_FOREACH(af, c->address_families) {
+ const char *name;
+
+ name = af_to_name(PTR_TO_INT(af));
+ if (!name)
+ continue;
+
+ if (strv_extend(&l, name) < 0)
+ return NULL;
+ }
+
+ strv_sort(l);
+
+ return l ? TAKE_PTR(l) : strv_new(NULL);
+}
+
+char** exec_context_get_restrict_filesystems(const ExecContext *c) {
+ _cleanup_strv_free_ char **l = NULL;
+
+ assert(c);
+
+#if HAVE_LIBBPF
+ l = set_get_strv(c->restrict_filesystems);
+ if (!l)
+ return NULL;
+
+ strv_sort(l);
+#endif
+
+ return l ? TAKE_PTR(l) : strv_new(NULL);
+}
+
void exec_status_start(ExecStatus *s, pid_t pid) {
assert(s);
return mfree(path);
}
-static ExecSharedRuntime* exec_shared_runtime_free(ExecSharedRuntime *rt) {
+void exec_shared_runtime_done(ExecSharedRuntime *rt) {
if (!rt)
- return NULL;
+ return;
if (rt->manager)
(void) hashmap_remove(rt->manager->exec_shared_runtime_by_id, rt->id);
rt->var_tmp_dir = mfree(rt->var_tmp_dir);
safe_close_pair(rt->netns_storage_socket);
safe_close_pair(rt->ipcns_storage_socket);
+}
+
+static ExecSharedRuntime* exec_shared_runtime_free(ExecSharedRuntime *rt) {
+ exec_shared_runtime_done(rt);
+
return mfree(rt);
}
int exec_shared_runtime_deserialize_compat(Unit *u, const char *key, const char *value, FDSet *fds) {
_cleanup_(exec_shared_runtime_freep) ExecSharedRuntime *rt_create = NULL;
- ExecSharedRuntime *rt;
+ ExecSharedRuntime *rt = NULL;
int r;
/* This is for the migration from old (v237 or earlier) deserialization text.
return 0;
}
- if (hashmap_ensure_allocated(&u->manager->exec_shared_runtime_by_id, &string_hash_ops) < 0)
- return log_oom();
+ if (u->manager) {
+ if (hashmap_ensure_allocated(&u->manager->exec_shared_runtime_by_id, &string_hash_ops) < 0)
+ return log_oom();
- rt = hashmap_get(u->manager->exec_shared_runtime_by_id, u->id);
+ rt = hashmap_get(u->manager->exec_shared_runtime_by_id, u->id);
+ }
if (!rt) {
if (exec_shared_runtime_allocate(&rt_create, u->id) < 0)
return log_oom();
return -ENOMEM;
} else if (streq(key, "netns-socket-0")) {
- int fd;
-
- if ((fd = parse_fd(value)) < 0 || !fdset_contains(fds, fd)) {
- log_unit_debug(u, "Failed to parse netns socket value: %s", value);
- return 0;
- }
safe_close(rt->netns_storage_socket[0]);
- rt->netns_storage_socket[0] = fdset_remove(fds, fd);
+ rt->netns_storage_socket[0] = deserialize_fd(fds, value);
+ if (rt->netns_storage_socket[0] < 0)
+ return 0;
} else if (streq(key, "netns-socket-1")) {
- int fd;
-
- if ((fd = parse_fd(value)) < 0 || !fdset_contains(fds, fd)) {
- log_unit_debug(u, "Failed to parse netns socket value: %s", value);
- return 0;
- }
safe_close(rt->netns_storage_socket[1]);
- rt->netns_storage_socket[1] = fdset_remove(fds, fd);
-
+ rt->netns_storage_socket[1] = deserialize_fd(fds, value);
+ if (rt->netns_storage_socket[1] < 0)
+ return 0;
} else
return 0;
/* If the object is newly created, then put it to the hashmap which manages ExecSharedRuntime objects. */
- if (rt_create) {
+ if (rt_create && u->manager) {
r = hashmap_put(u->manager->exec_shared_runtime_by_id, rt_create->id, rt_create);
if (r < 0) {
log_unit_debug_errno(u, r, "Failed to put runtime parameter to manager's storage: %m");
n = strcspn(v, " ");
buf = strndupa_safe(v, n);
- netns_fdpair[0] = parse_fd(buf);
+ netns_fdpair[0] = deserialize_fd(fds, buf);
if (netns_fdpair[0] < 0)
- return log_debug_errno(netns_fdpair[0], "Unable to parse exec-runtime specification netns-socket-0=%s: %m", buf);
- if (!fdset_contains(fds, netns_fdpair[0]))
- return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
- "exec-runtime specification netns-socket-0= refers to unknown fd %d: %m", netns_fdpair[0]);
- netns_fdpair[0] = fdset_remove(fds, netns_fdpair[0]);
+ return netns_fdpair[0];
if (v[n] != ' ')
goto finalize;
p = v + n + 1;
n = strcspn(v, " ");
buf = strndupa_safe(v, n);
- netns_fdpair[1] = parse_fd(buf);
+ netns_fdpair[1] = deserialize_fd(fds, buf);
if (netns_fdpair[1] < 0)
- return log_debug_errno(netns_fdpair[1], "Unable to parse exec-runtime specification netns-socket-1=%s: %m", buf);
- if (!fdset_contains(fds, netns_fdpair[1]))
- return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
- "exec-runtime specification netns-socket-1= refers to unknown fd %d: %m", netns_fdpair[1]);
- netns_fdpair[1] = fdset_remove(fds, netns_fdpair[1]);
+ return netns_fdpair[1];
if (v[n] != ' ')
goto finalize;
p = v + n + 1;
n = strcspn(v, " ");
buf = strndupa_safe(v, n);
- ipcns_fdpair[0] = parse_fd(buf);
+ ipcns_fdpair[0] = deserialize_fd(fds, buf);
if (ipcns_fdpair[0] < 0)
- return log_debug_errno(ipcns_fdpair[0], "Unable to parse exec-runtime specification ipcns-socket-0=%s: %m", buf);
- if (!fdset_contains(fds, ipcns_fdpair[0]))
- return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
- "exec-runtime specification ipcns-socket-0= refers to unknown fd %d: %m", ipcns_fdpair[0]);
- ipcns_fdpair[0] = fdset_remove(fds, ipcns_fdpair[0]);
+ return ipcns_fdpair[0];
if (v[n] != ' ')
goto finalize;
p = v + n + 1;
n = strcspn(v, " ");
buf = strndupa_safe(v, n);
- ipcns_fdpair[1] = parse_fd(buf);
+ ipcns_fdpair[1] = deserialize_fd(fds, buf);
if (ipcns_fdpair[1] < 0)
- return log_debug_errno(ipcns_fdpair[1], "Unable to parse exec-runtime specification ipcns-socket-1=%s: %m", buf);
- if (!fdset_contains(fds, ipcns_fdpair[1]))
- return log_debug_errno(SYNTHETIC_ERRNO(EBADF),
- "exec-runtime specification ipcns-socket-1= refers to unknown fd %d: %m", ipcns_fdpair[1]);
- ipcns_fdpair[1] = fdset_remove(fds, ipcns_fdpair[1]);
+ return ipcns_fdpair[1];
}
finalize:
return exec_runtime_free(rt);
}
+void exec_runtime_clear(ExecRuntime *rt) {
+ if (!rt)
+ return;
+
+ safe_close_pair(rt->ephemeral_storage_socket);
+ rt->ephemeral_copy = mfree(rt->ephemeral_copy);
+}
+
void exec_params_clear(ExecParameters *p) {
if (!p)
return;
p->environment = strv_free(p->environment);
p->fd_names = strv_free(p->fd_names);
+ p->files_env = strv_free(p->files_env);
p->fds = mfree(p->fds);
p->exec_fd = safe_close(p->exec_fd);
+ p->user_lookup_fd = -EBADF;
+ p->bpf_outer_map_fd = -EBADF;
+ p->unit_id = mfree(p->unit_id);
+ p->invocation_id = SD_ID128_NULL;
+ p->invocation_id_string[0] = '\0';
+ p->confirm_spawn = mfree(p->confirm_spawn);
+}
+
+void exec_params_serialized_done(ExecParameters *p) {
+ if (!p)
+ return;
+
+ close_many_unset(p->fds, p->n_socket_fds + p->n_storage_fds);
+
+ p->cgroup_path = mfree(p->cgroup_path);
+
+ if (p->prefix) {
+ for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++)
+ free(p->prefix[t]);
+
+ p->prefix = mfree(p->prefix);
+ }
+
+ p->received_credentials_directory = mfree(p->received_credentials_directory);
+ p->received_encrypted_credentials_directory = mfree(p->received_encrypted_credentials_directory);
+
+ if (p->idle_pipe) {
+ close_many_and_free(p->idle_pipe, 4);
+ p->idle_pipe = NULL;
+ }
+
+ p->stdin_fd = safe_close(p->stdin_fd);
+ p->stdout_fd = safe_close(p->stdout_fd);
+ p->stderr_fd = safe_close(p->stderr_fd);
+
+ p->notify_socket = mfree(p->notify_socket);
+
+ open_file_free_many(&p->open_files);
+
+ p->fallback_smack_process_label = mfree(p->fallback_smack_process_label);
+
+ exec_params_clear(p);
}
void exec_directory_done(ExecDirectory *d) {
if (!d)
return;
- for (size_t i = 0; i < d->n_items; i++) {
- free(d->items[i].path);
- strv_free(d->items[i].symlinks);
+ FOREACH_ARRAY(i, d->items, d->n_items) {
+ free(i->path);
+ strv_free(i->symlinks);
}
d->items = mfree(d->items);
assert(d);
assert(path);
- for (size_t i = 0; i < d->n_items; i++)
- if (path_equal(d->items[i].path, path))
- return &d->items[i];
+ FOREACH_ARRAY(i, d->items, d->n_items)
+ if (path_equal(i->path, path))
+ return i;
return NULL;
}
DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_symlink, ExecDirectoryType);
+static const char* const exec_directory_type_mode_table[_EXEC_DIRECTORY_TYPE_MAX] = {
+ [EXEC_DIRECTORY_RUNTIME] = "RuntimeDirectoryMode",
+ [EXEC_DIRECTORY_STATE] = "StateDirectoryMode",
+ [EXEC_DIRECTORY_CACHE] = "CacheDirectoryMode",
+ [EXEC_DIRECTORY_LOGS] = "LogsDirectoryMode",
+ [EXEC_DIRECTORY_CONFIGURATION] = "ConfigurationDirectoryMode",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(exec_directory_type_mode, ExecDirectoryType);
+
/* And this table maps ExecDirectoryType too, but to a generic term identifying the type of resource. This
* one is supposed to be generic enough to be used for unit types that don't use ExecContext and per-unit
* directories, specifically .timer units with their timestamp touch file. */
DEFINE_STRING_TABLE_LOOKUP(exec_resource_type, ExecDirectoryType);
-/* And this table also maps ExecDirectoryType, to the environment variable we pass the selected directory to
- * the service payload in. */
-static const char* const exec_directory_env_name_table[_EXEC_DIRECTORY_TYPE_MAX] = {
- [EXEC_DIRECTORY_RUNTIME] = "RUNTIME_DIRECTORY",
- [EXEC_DIRECTORY_STATE] = "STATE_DIRECTORY",
- [EXEC_DIRECTORY_CACHE] = "CACHE_DIRECTORY",
- [EXEC_DIRECTORY_LOGS] = "LOGS_DIRECTORY",
- [EXEC_DIRECTORY_CONFIGURATION] = "CONFIGURATION_DIRECTORY",
-};
-
-DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(exec_directory_env_name, ExecDirectoryType);
-
static const char* const exec_keyring_mode_table[_EXEC_KEYRING_MODE_MAX] = {
[EXEC_KEYRING_INHERIT] = "inherit",
[EXEC_KEYRING_PRIVATE] = "private",