#include <sys/poll.h>
#include <linux/seccomp-bpf.h>
#include <glob.h>
+#include <libgen.h>
+#undef basename
#ifdef HAVE_PAM
#include <security/pam_appl.h>
#include "sd-messages.h"
#include "ioprio.h"
#include "securebits.h"
-#include "cgroup.h"
#include "namespace.h"
#include "tcpwrap.h"
#include "exit-status.h"
#include "missing.h"
#include "utmp-wtmp.h"
#include "def.h"
-#include "loopback-setup.h"
#include "path-util.h"
#include "syscall-list.h"
#include "env-util.h"
#include "fileio.h"
+#include "unit.h"
+#include "async.h"
#define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
+#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
/* This assumes there is a 'tty' group */
#define TTY_MODE 0620
+#define SNDBUF_SIZE (8*1024*1024)
+
static int shift_fds(int fds[], unsigned n_fds) {
int start, restart_from;
return 0;
}
-static const char *tty_path(const ExecContext *context) {
+_pure_ static const char *tty_path(const ExecContext *context) {
assert(context);
if (context->tty_path)
return "/dev/console";
}
-void exec_context_tty_reset(const ExecContext *context) {
+static void exec_context_tty_reset(const ExecContext *context) {
assert(context);
if (context->tty_vhangup)
vt_disallocate(context->tty_path);
}
+static bool is_terminal_output(ExecOutput o) {
+ return
+ o == EXEC_OUTPUT_TTY ||
+ o == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
+ o == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
+ o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE;
+}
+
static int open_null_as(int flags, int nfd) {
int fd, r;
assert(nfd >= 0);
- if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0)
+ fd = open("/dev/null", flags|O_NOCTTY);
+ if (fd < 0)
return -errno;
if (fd != nfd) {
static int connect_logger_as(const ExecContext *context, ExecOutput output, const char *ident, const char *unit_id, int nfd) {
int fd, r;
- union sockaddr_union sa;
+ union sockaddr_union sa = {
+ .un.sun_family = AF_UNIX,
+ .un.sun_path = "/run/systemd/journal/stdout",
+ };
assert(context);
assert(output < _EXEC_OUTPUT_MAX);
if (fd < 0)
return -errno;
- zero(sa);
- sa.un.sun_family = AF_UNIX;
- strncpy(sa.un.sun_path, "/run/systemd/journal/stdout", sizeof(sa.un.sun_path));
-
r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
if (r < 0) {
close_nointr_nofail(fd);
return -errno;
}
+ fd_inc_sndbuf(fd, SNDBUF_SIZE);
+
dprintf(fd,
"%s\n"
"%s\n"
!!context->syslog_level_prefix,
output == EXEC_OUTPUT_SYSLOG || output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE,
output == EXEC_OUTPUT_KMSG || output == EXEC_OUTPUT_KMSG_AND_CONSOLE,
- output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || output == EXEC_OUTPUT_KMSG_AND_CONSOLE || output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE);
+ is_terminal_output(output));
if (fd != nfd) {
r = dup2(fd, nfd) < 0 ? -errno : nfd;
case EXEC_INPUT_TTY_FAIL: {
int fd, r;
- if ((fd = acquire_terminal(
- tty_path(context),
- i == EXEC_INPUT_TTY_FAIL,
- i == EXEC_INPUT_TTY_FORCE,
- false,
- (usec_t) -1)) < 0)
+ fd = acquire_terminal(tty_path(context),
+ i == EXEC_INPUT_TTY_FAIL,
+ i == EXEC_INPUT_TTY_FORCE,
+ false,
+ (usec_t) -1);
+ if (fd < 0)
return fd;
if (fd != STDIN_FILENO) {
}
}
-static int setup_output(const ExecContext *context, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
+static int setup_output(const ExecContext *context, int fileno, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
ExecOutput o;
ExecInput i;
int r;
i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
o = fixup_output(context->std_output, socket_fd);
- /* This expects the input is already set up */
+ if (fileno == STDERR_FILENO) {
+ ExecOutput e;
+ e = fixup_output(context->std_error, socket_fd);
- switch (o) {
+ /* This expects the input and output are already set up */
- case EXEC_OUTPUT_INHERIT:
+ /* Don't change the stderr file descriptor if we inherit all
+ * the way and are not on a tty */
+ if (e == EXEC_OUTPUT_INHERIT &&
+ o == EXEC_OUTPUT_INHERIT &&
+ i == EXEC_INPUT_NULL &&
+ !is_terminal_input(context->std_input) &&
+ getppid () != 1)
+ return fileno;
+ /* Duplicate from stdout if possible */
+ if (e == o || e == EXEC_OUTPUT_INHERIT)
+ return dup2(STDOUT_FILENO, fileno) < 0 ? -errno : fileno;
+
+ o = e;
+
+ } else if (o == EXEC_OUTPUT_INHERIT) {
/* If input got downgraded, inherit the original value */
if (i == EXEC_INPUT_NULL && is_terminal_input(context->std_input))
- return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
+ return open_terminal_as(tty_path(context), O_WRONLY, fileno);
/* If the input is connected to anything that's not a /dev/null, inherit that... */
if (i != EXEC_INPUT_NULL)
- return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
+ return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
/* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
if (getppid() != 1)
- return STDOUT_FILENO;
-
- /* We need to open /dev/null here anew, to get the
- * right access mode. So we fall through */
-
- case EXEC_OUTPUT_NULL:
- return open_null_as(O_WRONLY, STDOUT_FILENO);
-
- case EXEC_OUTPUT_TTY:
- if (is_terminal_input(i))
- return dup2(STDIN_FILENO, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
-
- /* We don't reset the terminal if this is just about output */
- return open_terminal_as(tty_path(context), O_WRONLY, STDOUT_FILENO);
-
- case EXEC_OUTPUT_SYSLOG:
- case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
- case EXEC_OUTPUT_KMSG:
- case EXEC_OUTPUT_KMSG_AND_CONSOLE:
- case EXEC_OUTPUT_JOURNAL:
- case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
- r = connect_logger_as(context, o, ident, unit_id, STDOUT_FILENO);
- if (r < 0) {
- log_error("Failed to connect stdout of %s to the journal socket: %s", unit_id, strerror(-r));
- r = open_null_as(O_WRONLY, STDOUT_FILENO);
- }
- return r;
-
- case EXEC_OUTPUT_SOCKET:
- assert(socket_fd >= 0);
- return dup2(socket_fd, STDOUT_FILENO) < 0 ? -errno : STDOUT_FILENO;
+ return fileno;
- default:
- assert_not_reached("Unknown output type");
+ /* We need to open /dev/null here anew, to get the right access mode. */
+ return open_null_as(O_WRONLY, fileno);
}
-}
-
-static int setup_error(const ExecContext *context, int socket_fd, const char *ident, const char *unit_id, bool apply_tty_stdin) {
- ExecOutput o, e;
- ExecInput i;
- int r;
-
- assert(context);
- assert(ident);
-
- i = fixup_input(context->std_input, socket_fd, apply_tty_stdin);
- o = fixup_output(context->std_output, socket_fd);
- e = fixup_output(context->std_error, socket_fd);
-
- /* This expects the input and output are already set up */
-
- /* Don't change the stderr file descriptor if we inherit all
- * the way and are not on a tty */
- if (e == EXEC_OUTPUT_INHERIT &&
- o == EXEC_OUTPUT_INHERIT &&
- i == EXEC_INPUT_NULL &&
- !is_terminal_input(context->std_input) &&
- getppid () != 1)
- return STDERR_FILENO;
- /* Duplicate from stdout if possible */
- if (e == o || e == EXEC_OUTPUT_INHERIT)
- return dup2(STDOUT_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
-
- switch (e) {
+ switch (o) {
case EXEC_OUTPUT_NULL:
- return open_null_as(O_WRONLY, STDERR_FILENO);
+ return open_null_as(O_WRONLY, fileno);
case EXEC_OUTPUT_TTY:
if (is_terminal_input(i))
- return dup2(STDIN_FILENO, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
+ return dup2(STDIN_FILENO, fileno) < 0 ? -errno : fileno;
/* We don't reset the terminal if this is just about output */
- return open_terminal_as(tty_path(context), O_WRONLY, STDERR_FILENO);
+ return open_terminal_as(tty_path(context), O_WRONLY, fileno);
case EXEC_OUTPUT_SYSLOG:
case EXEC_OUTPUT_SYSLOG_AND_CONSOLE:
case EXEC_OUTPUT_KMSG_AND_CONSOLE:
case EXEC_OUTPUT_JOURNAL:
case EXEC_OUTPUT_JOURNAL_AND_CONSOLE:
- r = connect_logger_as(context, e, ident, unit_id, STDERR_FILENO);
+ r = connect_logger_as(context, o, ident, unit_id, fileno);
if (r < 0) {
- log_error("Failed to connect stderr of %s to the journal socket: %s", unit_id, strerror(-r));
- r = open_null_as(O_WRONLY, STDERR_FILENO);
+ log_struct_unit(LOG_CRIT, unit_id,
+ "MESSAGE=Failed to connect std%s of %s to the journal socket: %s",
+ fileno == STDOUT_FILENO ? "out" : "err",
+ unit_id, strerror(-r),
+ "ERRNO=%d", -r,
+ NULL);
+ r = open_null_as(O_WRONLY, fileno);
}
return r;
case EXEC_OUTPUT_SOCKET:
assert(socket_fd >= 0);
- return dup2(socket_fd, STDERR_FILENO) < 0 ? -errno : STDERR_FILENO;
+ return dup2(socket_fd, fileno) < 0 ? -errno : fileno;
default:
assert_not_reached("Unknown error type");
return r;
}
-static int write_confirm_message(const char *format, ...) {
+_printf_(1, 2) static int write_confirm_message(const char *format, ...) {
int fd;
va_list ap;
/* First step: If we need to keep capabilities but
* drop privileges we need to make sure we keep our
- * caps, whiel we drop privileges. */
+ * caps, while we drop privileges. */
if (uid != 0) {
- int sb = context->secure_bits|SECURE_KEEP_CAPS;
+ int sb = context->secure_bits | 1<<SECURE_KEEP_CAPS;
if (prctl(PR_GET_SECUREBITS) != sb)
if (prctl(PR_SET_SECUREBITS, sb) < 0)
char **e = NULL;
bool close_session = false;
pid_t pam_pid = 0, parent_pid;
+ int flags = 0;
assert(name);
assert(user);
* daemon. We do things this way to ensure that the main PID
* of the daemon is the one we initially fork()ed. */
- if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) {
+ if (log_get_max_level() < LOG_PRI(LOG_DEBUG))
+ flags |= PAM_SILENT;
+
+ pam_code = pam_start(name, user, &conv, &handle);
+ if (pam_code != PAM_SUCCESS) {
handle = NULL;
goto fail;
}
- if (tty)
- if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS)
+ if (tty) {
+ pam_code = pam_set_item(handle, PAM_TTY, tty);
+ if (pam_code != PAM_SUCCESS)
goto fail;
+ }
- if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS)
+ pam_code = pam_acct_mgmt(handle, flags);
+ if (pam_code != PAM_SUCCESS)
goto fail;
- if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS)
+ pam_code = pam_open_session(handle, flags);
+ if (pam_code != PAM_SUCCESS)
goto fail;
close_session = true;
- if ((!(e = pam_getenvlist(handle)))) {
+ e = pam_getenvlist(handle);
+ if (!e) {
pam_code = PAM_BUF_ERR;
goto fail;
}
parent_pid = getpid();
- if ((pam_pid = fork()) < 0)
+ pam_pid = fork();
+ if (pam_pid < 0)
goto fail;
if (pam_pid == 0) {
}
/* If our parent died we'll end the session */
- if (getppid() != parent_pid)
- if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS)
+ if (getppid() != parent_pid) {
+ pam_code = pam_close_session(handle, flags);
+ if (pam_code != PAM_SUCCESS)
goto child_finish;
+ }
r = 0;
child_finish:
- pam_end(handle, pam_code | PAM_DATA_SILENT);
+ pam_end(handle, pam_code | flags);
_exit(r);
}
return 0;
fail:
- if (pam_code != PAM_SUCCESS)
+ if (pam_code != PAM_SUCCESS) {
+ log_error("PAM failed: %s", pam_strerror(handle, pam_code));
err = -EPERM; /* PAM errors do not map to errno */
- else
+ } else {
+ log_error("PAM failed: %m");
err = -errno;
+ }
if (handle) {
if (close_session)
- pam_code = pam_close_session(handle, PAM_DATA_SILENT);
+ pam_code = pam_close_session(handle, flags);
- pam_end(handle, pam_code | PAM_DATA_SILENT);
+ pam_end(handle, pam_code | flags);
}
strv_free(e);
/* This resulting string must fit in 10 chars (i.e. the length
* of "/sbin/init") to look pretty in /bin/ps */
- p = path_get_file_name(path);
+ p = basename(path);
if (isempty(p)) {
rename_process("(...)");
return;
int i;
unsigned n;
struct sock_filter *f;
- struct sock_fprog prog;
+ struct sock_fprog prog = {};
assert(syscall_filter);
memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer));
/* Third: install the filter */
- zero(prog);
prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n;
prog.filter = f;
if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0)
return 0;
}
+static void do_idle_pipe_dance(int idle_pipe[4]) {
+ assert(idle_pipe);
+
+ if (idle_pipe[1] >= 0)
+ close_nointr_nofail(idle_pipe[1]);
+ if (idle_pipe[2] >= 0)
+ close_nointr_nofail(idle_pipe[2]);
+
+ if (idle_pipe[0] >= 0) {
+ int r;
+
+ r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
+
+ if (idle_pipe[3] >= 0 && r == 0 /* timeout */) {
+ /* Signal systemd that we are bored and want to continue. */
+ write(idle_pipe[3], "x", 1);
+
+ /* Wait for systemd to react to the signal above. */
+ fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
+ }
+
+ close_nointr_nofail(idle_pipe[0]);
+
+ }
+
+ if (idle_pipe[3] >= 0)
+ close_nointr_nofail(idle_pipe[3]);
+}
+
+static int build_environment(
+ ExecContext *c,
+ unsigned n_fds,
+ const char *home,
+ const char *username,
+ const char *shell,
+ char ***ret) {
+
+ _cleanup_strv_free_ char **our_env = NULL;
+ unsigned n_env = 0;
+ char *x;
+
+ assert(c);
+ assert(ret);
+
+ our_env = new(char*, 8);
+ if (!our_env)
+ return -ENOMEM;
+
+ if (n_fds > 0) {
+ if (asprintf(&x, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+
+ if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+ }
+
+ if (home) {
+ x = strappend("HOME=", home);
+ if (!x)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+ }
+
+ if (username) {
+ x = strappend("LOGNAME=", username);
+ if (!x)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+
+ x = strappend("USER=", username);
+ if (!x)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+ }
+
+ if (shell) {
+ x = strappend("SHELL=", shell);
+ if (!x)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+ }
+
+ if (is_terminal_input(c->std_input) ||
+ c->std_output == EXEC_OUTPUT_TTY ||
+ c->std_error == EXEC_OUTPUT_TTY ||
+ c->tty_path) {
+
+ x = strdup(default_term_for_tty(tty_path(c)));
+ if (!x)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+ }
+
+ our_env[n_env++] = NULL;
+ assert(n_env <= 8);
+
+ *ret = our_env;
+ our_env = NULL;
+
+ return 0;
+}
+
int exec_spawn(ExecCommand *command,
char **argv,
- const ExecContext *context,
+ ExecContext *context,
int fds[], unsigned n_fds,
char **environment,
bool apply_permissions,
bool apply_chroot,
bool apply_tty_stdin,
bool confirm_spawn,
- CGroupBonding *cgroup_bondings,
- CGroupAttribute *cgroup_attributes,
- const char *cgroup_suffix,
+ CGroupControllerMask cgroup_supported,
+ const char *cgroup_path,
const char *unit_id,
- int idle_pipe[2],
+ int idle_pipe[4],
+ ExecRuntime *runtime,
pid_t *ret) {
+ _cleanup_strv_free_ char **files_env = NULL;
+ int socket_fd;
+ char *line;
pid_t pid;
int r;
- char *line;
- int socket_fd;
- char _cleanup_strv_free_ **files_env = NULL;
assert(command);
assert(context);
return log_oom();
log_struct_unit(LOG_DEBUG,
- unit_id,
- "MESSAGE=About to execute %s", line,
- NULL);
+ unit_id,
+ "EXECUTABLE=%s", command->path,
+ "MESSAGE=About to execute: %s", line,
+ NULL);
free(line);
- r = cgroup_bonding_realize_list(cgroup_bondings);
- if (r < 0)
- return r;
-
- cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings);
-
pid = fork();
if (pid < 0)
return -errno;
if (pid == 0) {
- int i, err;
- sigset_t ss;
- const char *username = NULL, *home = NULL;
+ _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL;
+ const char *username = NULL, *home = NULL, *shell = NULL;
+ unsigned n_dont_close = 0;
+ int dont_close[n_fds + 3];
uid_t uid = (uid_t) -1;
gid_t gid = (gid_t) -1;
- char _cleanup_strv_free_ **our_env = NULL, **pam_env = NULL,
- **final_env = NULL, **final_argv = NULL;
- unsigned n_env = 0;
- bool set_access = false;
+ sigset_t ss;
+ int i, err;
/* child */
goto fail_child;
}
- if (idle_pipe) {
- if (idle_pipe[1] >= 0)
- close_nointr_nofail(idle_pipe[1]);
- if (idle_pipe[0] >= 0) {
- fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC);
- close_nointr_nofail(idle_pipe[0]);
- }
- }
+ if (idle_pipe)
+ do_idle_pipe_dance(idle_pipe);
/* Close sockets very early to make sure we don't
* block init reexecution because it cannot bind its
* sockets */
log_forget_fds();
- err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds,
- socket_fd >= 0 ? 1 : n_fds);
+
+ if (socket_fd >= 0)
+ dont_close[n_dont_close++] = socket_fd;
+ if (n_fds > 0) {
+ memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds);
+ n_dont_close += n_fds;
+ }
+ if (runtime) {
+ if (runtime->netns_storage_socket[0] >= 0)
+ dont_close[n_dont_close++] = runtime->netns_storage_socket[0];
+ if (runtime->netns_storage_socket[1] >= 0)
+ dont_close[n_dont_close++] = runtime->netns_storage_socket[1];
+ }
+
+ err = close_all_fds(dont_close, n_dont_close);
if (err < 0) {
r = EXIT_FDS;
goto fail_child;
goto fail_child;
}
- err = setup_output(context, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
+ err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
if (err < 0) {
r = EXIT_STDOUT;
goto fail_child;
}
- err = setup_error(context, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin);
+ err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), unit_id, apply_tty_stdin);
if (err < 0) {
r = EXIT_STDERR;
goto fail_child;
}
- if (cgroup_bondings) {
- err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix);
+ if (cgroup_path) {
+ err = cg_attach_everywhere(cgroup_supported, cgroup_path, 0);
if (err < 0) {
r = EXIT_CGROUP;
goto fail_child;
snprintf(t, sizeof(t), "%i", context->oom_score_adjust);
char_array_0(t);
- if (write_one_line_file("/proc/self/oom_score_adj", t) < 0) {
+ if (write_string_file("/proc/self/oom_score_adj", t) < 0) {
err = -errno;
r = EXIT_OOM_ADJUST;
goto fail_child;
}
if (context->cpu_sched_set) {
- struct sched_param param;
-
- zero(param);
- param.sched_priority = context->cpu_sched_priority;
+ struct sched_param param = {
+ .sched_priority = context->cpu_sched_priority,
+ };
- if (sched_setscheduler(0, context->cpu_sched_policy |
- (context->cpu_sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0), ¶m) < 0) {
+ r = sched_setscheduler(0,
+ context->cpu_sched_policy |
+ (context->cpu_sched_reset_on_fork ?
+ SCHED_RESET_ON_FORK : 0),
+ ¶m);
+ if (r < 0) {
err = -errno;
r = EXIT_SETSCHEDULER;
goto fail_child;
if (context->user) {
username = context->user;
- err = get_user_creds(&username, &uid, &gid, &home, NULL);
+ err = get_user_creds(&username, &uid, &gid, &home, &shell);
if (err < 0) {
r = EXIT_USER;
goto fail_child;
goto fail_child;
}
}
+ }
- if (cgroup_bondings && context->control_group_modify) {
- err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid);
- if (err >= 0)
- err = cgroup_bonding_set_task_access_list(cgroup_bondings, 0644, uid, gid, context->control_group_persistent);
- if (err < 0) {
- r = EXIT_CGROUP;
- goto fail_child;
- }
-
- set_access = true;
+#ifdef HAVE_PAM
+ if (cgroup_path && context->user && context->pam_name) {
+ err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0644, uid, gid);
+ if (err < 0) {
+ r = EXIT_CGROUP;
+ goto fail_child;
}
- }
- if (cgroup_bondings && !set_access && context->control_group_persistent >= 0) {
- err = cgroup_bonding_set_task_access_list(cgroup_bondings, (mode_t) -1, (uid_t) -1, (uid_t) -1, context->control_group_persistent);
+
+ err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, 0755, uid, gid);
if (err < 0) {
r = EXIT_CGROUP;
goto fail_child;
}
}
+#endif
if (apply_permissions) {
err = enforce_groups(context, username, gid);
}
}
#endif
- if (context->private_network) {
- if (unshare(CLONE_NEWNET) < 0) {
- err = -errno;
+ if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) {
+ err = setup_netns(runtime->netns_storage_socket);
+ if (err < 0) {
r = EXIT_NETWORK;
goto fail_child;
}
-
- loopback_setup();
}
- if (strv_length(context->read_write_dirs) > 0 ||
- strv_length(context->read_only_dirs) > 0 ||
- strv_length(context->inaccessible_dirs) > 0 ||
+ if (!strv_isempty(context->read_write_dirs) ||
+ !strv_isempty(context->read_only_dirs) ||
+ !strv_isempty(context->inaccessible_dirs) ||
context->mount_flags != 0 ||
- context->private_tmp) {
- err = setup_namespace(context->read_write_dirs,
- context->read_only_dirs,
- context->inaccessible_dirs,
- context->private_tmp,
- context->mount_flags);
+ (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir))) {
+
+ char *tmp = NULL, *var = NULL;
+
+ /* The runtime struct only contains the parent
+ * of the private /tmp, which is
+ * non-accessible to world users. Inside of it
+ * there's a /tmp that is sticky, and that's
+ * the one we want to use here. */
+
+ if (context->private_tmp && runtime) {
+ if (runtime->tmp_dir)
+ tmp = strappenda(runtime->tmp_dir, "/tmp");
+ if (runtime->var_tmp_dir)
+ var = strappenda(runtime->var_tmp_dir, "/tmp");
+ }
+
+ err = setup_namespace(
+ context->read_write_dirs,
+ context->read_only_dirs,
+ context->inaccessible_dirs,
+ tmp,
+ var,
+ context->mount_flags);
+
if (err < 0) {
r = EXIT_NAMESPACE;
goto fail_child;
goto fail_child;
}
} else {
- char _cleanup_free_ *d = NULL;
+ _cleanup_free_ char *d = NULL;
if (asprintf(&d, "%s/%s",
context->root_directory ? context->root_directory : "",
}
}
- if (!(our_env = new0(char*, 7))) {
- err = -ENOMEM;
+ err = build_environment(context, n_fds, home, username, shell, &our_env);
+ if (r < 0) {
r = EXIT_MEMORY;
goto fail_child;
}
- if (n_fds > 0)
- if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 ||
- asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
- err = -ENOMEM;
- r = EXIT_MEMORY;
- goto fail_child;
- }
-
- if (home)
- if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
- err = -ENOMEM;
- r = EXIT_MEMORY;
- goto fail_child;
- }
-
- if (username)
- if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
- asprintf(our_env + n_env++, "USER=%s", username) < 0) {
- err = -ENOMEM;
- r = EXIT_MEMORY;
- goto fail_child;
- }
-
- if (is_terminal_input(context->std_input) ||
- context->std_output == EXEC_OUTPUT_TTY ||
- context->std_error == EXEC_OUTPUT_TTY)
- if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) {
- err = -ENOMEM;
- r = EXIT_MEMORY;
- goto fail_child;
- }
-
- assert(n_env <= 7);
-
- if (!(final_env = strv_env_merge(
- 5,
- environment,
- our_env,
- context->environment,
- files_env,
- pam_env,
- NULL))) {
+ final_env = strv_env_merge(5,
+ environment,
+ our_env,
+ context->environment,
+ files_env,
+ pam_env,
+ NULL);
+ if (!final_env) {
err = -ENOMEM;
r = EXIT_MEMORY;
goto fail_child;
}
- if (!(final_argv = replace_env_argv(argv, final_env))) {
+ final_argv = replace_env_argv(argv, final_env);
+ if (!final_argv) {
err = -ENOMEM;
r = EXIT_MEMORY;
goto fail_child;
final_env = strv_env_clean(final_env);
+ if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) {
+ line = exec_command_line(final_argv);
+ if (line) {
+ log_open();
+ log_struct_unit(LOG_DEBUG,
+ unit_id,
+ "EXECUTABLE=%s", command->path,
+ "MESSAGE=Executing: %s", line,
+ NULL);
+ log_close();
+ free(line);
+ line = NULL;
+ }
+ }
execve(command->path, final_argv, final_env);
err = -errno;
r = EXIT_EXEC;
}
log_struct_unit(LOG_DEBUG,
- unit_id,
- "MESSAGE=Forked %s as %lu",
- command->path, (unsigned long) pid,
- NULL);
+ unit_id,
+ "MESSAGE=Forked %s as %lu",
+ command->path, (unsigned long) pid,
+ NULL);
/* We add the new process to the cgroup both in the child (so
* that we can be sure that no user code is ever executed
* outside of the cgroup) and in the parent (so that we can be
* sure that when we kill the cgroup the process will be
* killed too). */
- if (cgroup_bondings)
- cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix);
+ if (cgroup_path)
+ cg_attach(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, pid);
exec_status_start(&command->exec_status, pid);
c->cpu_sched_policy = SCHED_OTHER;
c->syslog_priority = LOG_DAEMON|LOG_INFO;
c->syslog_level_prefix = true;
- c->control_group_persistent = -1;
c->ignore_sigpipe = true;
c->timer_slack_nsec = (nsec_t) -1;
}
ExecCommand *i;
while ((i = c)) {
- LIST_REMOVE(ExecCommand, command, c, i);
+ LIST_REMOVE(command, c, i);
exec_command_done(i);
free(i);
}
int k;
bool ignore = false;
char **p;
- glob_t pglob;
+ _cleanup_globfree_ glob_t pglob = {};
int count, n;
fn = *i;
}
if (!path_is_absolute(fn)) {
-
if (ignore)
continue;
}
/* Filename supports globbing, take all matching files */
- zero(pglob);
errno = 0;
if (glob(fn, 0, NULL, &pglob) != 0) {
- globfree(&pglob);
if (ignore)
continue;
}
count = pglob.gl_pathc;
if (count == 0) {
- globfree(&pglob);
if (ignore)
continue;
return -EINVAL;
}
for (n = 0; n < count; n++) {
- k = load_env_file(pglob.gl_pathv[n], &p);
+ k = load_env_file(pglob.gl_pathv[n], NULL, &p);
if (k < 0) {
if (ignore)
continue;
strv_free(r);
- globfree(&pglob);
return k;
- }
+ }
+ /* Log invalid environment variables with filename */
+ if (p)
+ p = strv_env_clean_log(p, pglob.gl_pathv[n]);
if (r == NULL)
r = p;
m = strv_env_merge(2, r, p);
strv_free(r);
strv_free(p);
-
- if (!m) {
- globfree(&pglob);
+ if (!m)
return -ENOMEM;
- }
r = m;
}
}
- globfree(&pglob);
}
*l = r;
return 0;
}
+static bool tty_may_match_dev_console(const char *tty) {
+ char *active = NULL, *console;
+ bool b;
+
+ if (startswith(tty, "/dev/"))
+ tty += 5;
+
+ /* trivial identity? */
+ if (streq(tty, "console"))
+ return true;
+
+ console = resolve_dev_console(&active);
+ /* if we could not resolve, assume it may */
+ if (!console)
+ return true;
+
+ /* "tty0" means the active VC, so it may be the same sometimes */
+ b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty));
+ free(active);
+
+ return b;
+}
+
+bool exec_context_may_touch_console(ExecContext *ec) {
+ return (ec->tty_reset || ec->tty_vhangup || ec->tty_vt_disallocate ||
+ is_terminal_input(ec->std_input) ||
+ is_terminal_output(ec->std_output) ||
+ is_terminal_output(ec->std_error)) &&
+ tty_may_match_dev_console(tty_path(ec));
+}
+
static void strv_fprintf(FILE *f, char **l) {
char **g;
}
void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
- char ** e;
+ char **e;
unsigned i;
assert(c);
assert(f);
- if (!prefix)
- prefix = "";
+ prefix = strempty(prefix);
fprintf(f,
"%sUMask: %04o\n"
"%sRootDirectory: %s\n"
"%sNonBlocking: %s\n"
"%sPrivateTmp: %s\n"
- "%sControlGroupModify: %s\n"
- "%sControlGroupPersistent: %s\n"
"%sPrivateNetwork: %s\n"
"%sIgnoreSIGPIPE: %s\n",
prefix, c->umask,
prefix, c->root_directory ? c->root_directory : "/",
prefix, yes_no(c->non_blocking),
prefix, yes_no(c->private_tmp),
- prefix, yes_no(c->control_group_modify),
- prefix, yes_no(c->control_group_persistent),
prefix, yes_no(c->private_network),
prefix, yes_no(c->ignore_sigpipe));
if (c->secure_bits)
fprintf(f, "%sSecure Bits:%s%s%s%s%s%s\n",
prefix,
- (c->secure_bits & SECURE_KEEP_CAPS) ? " keep-caps" : "",
- (c->secure_bits & SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
- (c->secure_bits & SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
- (c->secure_bits & SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
- (c->secure_bits & SECURE_NOROOT) ? " noroot" : "",
- (c->secure_bits & SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
+ (c->secure_bits & 1<<SECURE_KEEP_CAPS) ? " keep-caps" : "",
+ (c->secure_bits & 1<<SECURE_KEEP_CAPS_LOCKED) ? " keep-caps-locked" : "",
+ (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP) ? " no-setuid-fixup" : "",
+ (c->secure_bits & 1<<SECURE_NO_SETUID_FIXUP_LOCKED) ? " no-setuid-fixup-locked" : "",
+ (c->secure_bits & 1<<SECURE_NOROOT) ? " noroot" : "",
+ (c->secure_bits & 1<<SECURE_NOROOT_LOCKED) ? "noroot-locked" : "");
if (c->capability_bounding_set_drop) {
unsigned long l;
if (*l) {
/* It's kind of important, that we keep the order here */
- LIST_FIND_TAIL(ExecCommand, command, *l, end);
- LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
+ LIST_FIND_TAIL(command, *l, end);
+ LIST_INSERT_AFTER(command, *l, end, e);
} else
*l = e;
}
if (!l)
return -ENOMEM;
- if (!(p = strdup(path))) {
+ p = strdup(path);
+ if (!p) {
strv_free(l);
return -ENOMEM;
}
return 0;
}
+static int exec_runtime_allocate(ExecRuntime **rt) {
+
+ if (*rt)
+ return 0;
+
+ *rt = new0(ExecRuntime, 1);
+ if (!rt)
+ return -ENOMEM;
+
+ (*rt)->n_ref = 1;
+ (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1;
+
+ return 0;
+}
+
+int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) {
+ int r;
+
+ assert(rt);
+ assert(c);
+ assert(id);
+
+ if (*rt)
+ return 1;
+
+ if (!c->private_network && !c->private_tmp)
+ return 0;
+
+ r = exec_runtime_allocate(rt);
+ if (r < 0)
+ return r;
+
+ if (c->private_network && (*rt)->netns_storage_socket[0] < 0) {
+ if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0)
+ return -errno;
+ }
+
+ if (c->private_tmp && !(*rt)->tmp_dir) {
+ r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir);
+ if (r < 0)
+ return r;
+ }
+
+ return 1;
+}
+
+ExecRuntime *exec_runtime_ref(ExecRuntime *r) {
+ assert(r);
+ assert(r->n_ref > 0);
+
+ r->n_ref++;
+ return r;
+}
+
+ExecRuntime *exec_runtime_unref(ExecRuntime *r) {
+
+ if (!r)
+ return NULL;
+
+ assert(r->n_ref > 0);
+
+ r->n_ref--;
+ if (r->n_ref <= 0) {
+ free(r->tmp_dir);
+ free(r->var_tmp_dir);
+ close_pipe(r->netns_storage_socket);
+ free(r);
+ }
+
+ return NULL;
+}
+
+int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) {
+ assert(u);
+ assert(f);
+ assert(fds);
+
+ if (!rt)
+ return 0;
+
+ if (rt->tmp_dir)
+ unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir);
+
+ if (rt->var_tmp_dir)
+ unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir);
+
+ if (rt->netns_storage_socket[0] >= 0) {
+ int copy;
+
+ copy = fdset_put_dup(fds, rt->netns_storage_socket[0]);
+ if (copy < 0)
+ return copy;
+
+ unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy);
+ }
+
+ if (rt->netns_storage_socket[1] >= 0) {
+ int copy;
+
+ copy = fdset_put_dup(fds, rt->netns_storage_socket[1]);
+ if (copy < 0)
+ return copy;
+
+ unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy);
+ }
+
+ return 0;
+}
+
+int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) {
+ int r;
+
+ assert(rt);
+ assert(key);
+ assert(value);
+
+ if (streq(key, "tmp-dir")) {
+ char *copy;
+
+ r = exec_runtime_allocate(rt);
+ if (r < 0)
+ return r;
+
+ copy = strdup(value);
+ if (!copy)
+ return log_oom();
+
+ free((*rt)->tmp_dir);
+ (*rt)->tmp_dir = copy;
+
+ } else if (streq(key, "var-tmp-dir")) {
+ char *copy;
+
+ r = exec_runtime_allocate(rt);
+ if (r < 0)
+ return r;
+
+ copy = strdup(value);
+ if (!copy)
+ return log_oom();
+
+ free((*rt)->var_tmp_dir);
+ (*rt)->var_tmp_dir = copy;
+
+ } else if (streq(key, "netns-socket-0")) {
+ int fd;
+
+ r = exec_runtime_allocate(rt);
+ if (r < 0)
+ return r;
+
+ if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
+ log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
+ else {
+ if ((*rt)->netns_storage_socket[0] >= 0)
+ close_nointr_nofail((*rt)->netns_storage_socket[0]);
+
+ (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd);
+ }
+ } else if (streq(key, "netns-socket-1")) {
+ int fd;
+
+ r = exec_runtime_allocate(rt);
+ if (r < 0)
+ return r;
+
+ if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd))
+ log_debug_unit(u->id, "Failed to parse netns socket value %s", value);
+ else {
+ if ((*rt)->netns_storage_socket[1] >= 0)
+ close_nointr_nofail((*rt)->netns_storage_socket[1]);
+
+ (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd);
+ }
+ } else
+ return 0;
+
+ return 1;
+}
+
+static void *remove_tmpdir_thread(void *p) {
+ _cleanup_free_ char *path = p;
+
+ rm_rf_dangerous(path, false, true, false);
+ return NULL;
+}
+
+void exec_runtime_destroy(ExecRuntime *rt) {
+ if (!rt)
+ return;
+
+ /* If there are multiple users of this, let's leave the stuff around */
+ if (rt->n_ref > 1)
+ return;
+
+ if (rt->tmp_dir) {
+ log_debug("Spawning thread to nuke %s", rt->tmp_dir);
+ asynchronous_job(remove_tmpdir_thread, rt->tmp_dir);
+ rt->tmp_dir = NULL;
+ }
+
+ if (rt->var_tmp_dir) {
+ log_debug("Spawning thread to nuke %s", rt->var_tmp_dir);
+ asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir);
+ rt->var_tmp_dir = NULL;
+ }
+
+ close_pipe(rt->netns_storage_socket);
+}
+
static const char* const exec_input_table[_EXEC_INPUT_MAX] = {
[EXEC_INPUT_NULL] = "null",
[EXEC_INPUT_TTY] = "tty",
~ianmdlvl / elogind.git / blobdiff