#include <grp.h>
#include <poll.h>
#include <signal.h>
-#include <string.h>
#include <sys/capability.h>
#include <sys/eventfd.h>
+#include <sys/ioctl.h>
#include <sys/mman.h>
#include <sys/personality.h>
#include <sys/prctl.h>
#include "cap-list.h"
#include "capability-util.h"
#include "chown-recursive.h"
+#include "cgroup-setup.h"
#include "cpu-set-util.h"
#include "def.h"
#include "env-file.h"
#include "macro.h"
#include "manager.h"
#include "memory-util.h"
-#include "missing.h"
+#include "missing_fs.h"
#include "mkdir.h"
#include "namespace.h"
#include "parse-util.h"
const ExecContext *context,
const ExecParameters *params,
int socket_fd,
- int named_iofds[3]) {
+ const int named_iofds[static 3]) {
ExecInput i;
assert(context);
assert(params);
+ assert(named_iofds);
if (params->stdin_fd >= 0) {
if (dup2(params->stdin_fd, STDIN_FILENO) < 0)
const ExecParameters *params,
int fileno,
int socket_fd,
- int named_iofds[3],
+ const int named_iofds[static 3],
const char *ident,
uid_t uid,
gid_t gid,
*/
errno = 0;
ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
- if (ngroups_max <= 0) {
- if (errno > 0)
- return -errno;
- else
- return -EOPNOTSUPP; /* For all other values */
- }
+ if (ngroups_max <= 0)
+ return errno_or_else(EOPNOTSUPP);
l_gids = new(gid_t, ngroups_max);
if (!l_gids)
n = write(idle_pipe[3], "x", 1);
if (n > 0)
/* Wait for systemd to react to the signal above. */
- fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
+ (void) fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC);
}
idle_pipe[0] = safe_close(idle_pipe[0]);
}
if (home) {
- x = strappend("HOME=", home);
+ x = strjoin("HOME=", home);
if (!x)
return -ENOMEM;
}
if (username) {
- x = strappend("LOGNAME=", username);
+ x = strjoin("LOGNAME=", username);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
- x = strappend("USER=", username);
+ x = strjoin("USER=", username);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
}
if (shell) {
- x = strappend("SHELL=", shell);
+ x = strjoin("SHELL=", shell);
if (!x)
return -ENOMEM;
if (!term)
term = default_term_for_tty(tty_path);
- x = strappend("TERM=", term);
+ x = strjoin("TERM=", term);
if (!x)
return -ENOMEM;
our_env[n_env++] = x;
return 0;
}
+static bool exec_directory_is_private(const ExecContext *context, ExecDirectoryType type) {
+ 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 setup_exec_directory(
const ExecContext *context,
const ExecParameters *params,
if (r < 0)
goto fail;
- if (context->dynamic_user &&
- (!IN_SET(type, EXEC_DIRECTORY_RUNTIME, EXEC_DIRECTORY_CONFIGURATION) ||
- (type == EXEC_DIRECTORY_RUNTIME && context->runtime_directory_preserve_mode != EXEC_PRESERVE_NO))) {
+ if (exec_directory_is_private(context, type)) {
_cleanup_free_ char *private_root = NULL;
- /* 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.
+ /* 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 namespaceof the container "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!
+ * 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. */
+ * 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. */
private_root = path_join(params->prefix[type], "private");
if (!private_root) {
* it over. Most likely the service has been upgraded from one that didn't use
* DynamicUser=1, to one that does. */
+ log_info("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);
+
if (rename(p, pp) < 0) {
r = -errno;
goto fail;
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;
+
+ /* 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. */
+
+ q = path_join(params->prefix[type], "private", *rt);
+ if (!q) {
+ r = -ENOMEM;
+ goto fail;
+ }
+
+ if (path_equal(q, target)) {
+
+ /* Hmm, apparently DynamicUser= was once turned on for this service,
+ * but is no longer. Let's move the directory back up. */
+
+ log_info("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);
+
+ if (unlink(p) < 0) {
+ r = -errno;
+ goto fail;
+ }
+
+ if (rename(q, p) < 0) {
+ r = -errno;
+ goto fail;
+ }
+ }
+ }
+
r = mkdir_label(p, context->directories[type].mode);
if (r < 0) {
if (r != -EEXIST)
if (strv_isempty(context->directories[t].paths))
continue;
- if (context->dynamic_user &&
- !IN_SET(t, EXEC_DIRECTORY_RUNTIME, EXEC_DIRECTORY_CONFIGURATION) &&
+ if (exec_directory_is_private(context, t) &&
!(context->root_directory || context->root_image)) {
char *private_root;
STRV_FOREACH(suffix, context->directories[t].paths) {
char *s, *d;
- if (context->dynamic_user &&
- !IN_SET(t, EXEC_DIRECTORY_RUNTIME, EXEC_DIRECTORY_CONFIGURATION))
+ if (exec_directory_is_private(context, t))
s = path_join(params->prefix[t], "private", *suffix);
else
s = path_join(params->prefix[t], *suffix);
goto finish;
}
- if (context->dynamic_user &&
- !IN_SET(t, EXEC_DIRECTORY_RUNTIME, EXEC_DIRECTORY_CONFIGURATION) &&
+ if (exec_directory_is_private(context, t) &&
(context->root_directory || context->root_image))
/* 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
const ExecContext *context,
const ExecParameters *params,
const char *home,
- const bool needs_mount_ns,
int *exit_status) {
const char *d, *wd;
else
wd = "/";
- if (params->flags & EXEC_APPLY_CHROOT) {
- if (!needs_mount_ns && context->root_directory)
- if (chroot(context->root_directory) < 0) {
- *exit_status = EXIT_CHROOT;
- return -errno;
- }
-
+ if (params->flags & EXEC_APPLY_CHROOT)
d = wd;
- } else
+ else
d = prefix_roota(context->root_directory, wd);
if (chdir(d) < 0 && !context->working_directory_missing_ok) {
return 0;
}
+static int apply_root_directory(
+ const ExecContext *context,
+ const ExecParameters *params,
+ 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(context->root_directory) < 0) {
+ *exit_status = EXIT_CHROOT;
+ return -errno;
+ }
+ }
+
+ return 0;
+}
+
static int setup_keyring(
const Unit *u,
const ExecContext *context,
static void append_socket_pair(int *array, size_t *n, const int pair[static 2]) {
assert(array);
assert(n);
-
- if (!pair)
- return;
+ assert(pair);
if (pair[0] >= 0)
array[(*n)++] = pair[0];
STRV_FOREACH(i, c->directories[t].paths) {
char *e;
- if (t == EXEC_DIRECTORY_RUNTIME)
- e = path_join(p->prefix[t], *i);
- else
+ if (exec_directory_is_private(c, t))
e = path_join(p->prefix[t], "private", *i);
+ else
+ e = path_join(p->prefix[t], *i);
if (!e)
return -ENOMEM;
ExecRuntime *runtime,
DynamicCreds *dcreds,
int socket_fd,
- int named_iofds[3],
+ const int named_iofds[static 3],
int *fds,
size_t n_socket_fds,
size_t n_storage_fds,
if (mpol_is_valid(numa_policy_get_type(&context->numa_policy))) {
r = apply_numa_policy(&context->numa_policy);
if (r == -EOPNOTSUPP)
- log_unit_debug_errno(unit, SYNTHETIC_ERRNO(r), "NUMA support not available, ignoring.");
+ 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");
}
}
+ /* chroot to root directory first, before we lose the ability to chroot */
+ r = apply_root_directory(context, params, 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);
/* 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, home, needs_mount_namespace, exit_status);
+ r = apply_working_directory(context, params, home, exit_status);
if (r < 0)
return log_unit_error_errno(unit, r, "Changing to the requested working directory failed: %m");
}
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[3]);
+static int exec_context_named_iofds(const ExecContext *c, const ExecParameters *p, int named_iofds[static 3]);
int exec_spawn(Unit *unit,
ExecCommand *command,
unit->manager->user_lookup_fds[1],
&exit_status);
- if (r < 0)
+ if (r < 0) {
+ const char *status =
+ exit_status_to_string(exit_status,
+ EXIT_STATUS_LIBC | EXIT_STATUS_SYSTEMD);
+
log_struct_errno(LOG_ERR, r,
"MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR,
LOG_UNIT_ID(unit),
LOG_UNIT_INVOCATION_ID(unit),
LOG_UNIT_MESSAGE(unit, "Failed at step %s spawning %s: %m",
- exit_status_to_string(exit_status, EXIT_STATUS_SYSTEMD),
- command->path),
+ status, command->path),
"EXECUTABLE=%s", command->path);
+ }
_exit(exit_status);
}
c->personality = PERSONALITY_INVALID;
for (i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++)
c->directories[i].mode = 0755;
+ c->timeout_clean_usec = USEC_INFINITY;
c->capability_bounding_set = CAP_ALL;
assert_cc(NAMESPACE_FLAGS_INITIAL != NAMESPACE_FLAGS_ALL);
c->restrict_namespaces = NAMESPACE_FLAGS_INITIAL;
exec_context_free_log_extra_fields(c);
- c->log_rate_limit_interval_usec = 0;
- c->log_rate_limit_burst = 0;
+ c->log_ratelimit_interval_usec = 0;
+ c->log_ratelimit_burst = 0;
c->stdin_data = mfree(c->stdin_data);
c->stdin_data_size = 0;
STRV_FOREACH(i, c->directories[EXEC_DIRECTORY_RUNTIME].paths) {
_cleanup_free_ char *p;
- p = path_join(runtime_prefix, *i);
+ if (exec_directory_is_private(c, EXEC_DIRECTORY_RUNTIME))
+ p = path_join(runtime_prefix, "private", *i);
+ else
+ p = path_join(runtime_prefix, *i);
if (!p)
return -ENOMEM;
}
}
-static int exec_context_named_iofds(const ExecContext *c, const ExecParameters *p, int named_iofds[static 3]) {
+static int exec_context_named_iofds(
+ const ExecContext *c,
+ const ExecParameters *p,
+ int named_iofds[static 3]) {
+
size_t i, 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) +
}
void exec_context_dump(const ExecContext *c, FILE* f, const char *prefix) {
+ char **e, **d, buf_clean[FORMAT_TIMESPAN_MAX];
ExecDirectoryType dt;
- char **e, **d;
unsigned i;
int r;
fprintf(f, "%s%s: %s\n", prefix, exec_directory_type_to_string(dt), *d);
}
+ fprintf(f,
+ "%sTimeoutCleanSec: %s\n",
+ prefix, format_timespan(buf_clean, sizeof(buf_clean), c->timeout_clean_usec, USEC_PER_SEC));
+
if (c->nice_set)
fprintf(f,
"%sNice: %i\n",
fprintf(f, "%sLogLevelMax: %s\n", prefix, strna(t));
}
- if (c->log_rate_limit_interval_usec > 0) {
+ if (c->log_ratelimit_interval_usec > 0) {
char buf_timespan[FORMAT_TIMESPAN_MAX];
fprintf(f,
"%sLogRateLimitIntervalSec: %s\n",
- prefix, format_timespan(buf_timespan, sizeof(buf_timespan), c->log_rate_limit_interval_usec, USEC_PER_SEC));
+ prefix, format_timespan(buf_timespan, sizeof(buf_timespan), c->log_ratelimit_interval_usec, USEC_PER_SEC));
}
- if (c->log_rate_limit_burst > 0)
- fprintf(f, "%sLogRateLimitBurst: %u\n", prefix, c->log_rate_limit_burst);
+ if (c->log_ratelimit_burst > 0)
+ fprintf(f, "%sLogRateLimitBurst: %u\n", prefix, c->log_ratelimit_burst);
if (c->n_log_extra_fields > 0) {
size_t j;
}
}
+int exec_context_get_clean_directories(
+ ExecContext *c,
+ char **prefix,
+ ExecCleanMask mask,
+ char ***ret) {
+
+ _cleanup_strv_free_ char **l = NULL;
+ ExecDirectoryType t;
+ int r;
+
+ assert(c);
+ assert(prefix);
+ assert(ret);
+
+ for (t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++) {
+ char **i;
+
+ if (!FLAGS_SET(mask, 1U << t))
+ continue;
+
+ if (!prefix[t])
+ continue;
+
+ STRV_FOREACH(i, c->directories[t].paths) {
+ char *j;
+
+ j = path_join(prefix[t], *i);
+ if (!j)
+ return -ENOMEM;
+
+ r = strv_consume(&l, j);
+ if (r < 0)
+ return r;
+
+ /* Also remove private directories unconditionally. */
+ if (t != EXEC_DIRECTORY_CONFIGURATION) {
+ j = path_join(prefix[t], "private", *i);
+ if (!j)
+ return -ENOMEM;
+
+ r = strv_consume(&l, j);
+ if (r < 0)
+ return r;
+ }
+ }
+ }
+
+ *ret = TAKE_PTR(l);
+ return 0;
+}
+
+int exec_context_get_clean_mask(ExecContext *c, ExecCleanMask *ret) {
+ ExecCleanMask mask = 0;
+
+ assert(c);
+ assert(ret);
+
+ for (ExecDirectoryType t = 0; t < _EXEC_DIRECTORY_TYPE_MAX; t++)
+ if (!strv_isempty(c->directories[t].paths))
+ mask |= 1U << t;
+
+ *ret = mask;
+ return 0;
+}
+
void exec_status_start(ExecStatus *s, pid_t pid) {
assert(s);
cmd = exec_command_line(c->argv);
fprintf(f,
"%sCommand Line: %s\n",
- prefix, cmd ? cmd : strerror(ENOMEM));
+ prefix, cmd ? cmd : strerror_safe(ENOMEM));
exec_status_dump(&c->exec_status, f, prefix2);
}
DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode, ExecPreserveMode, EXEC_PRESERVE_YES);
+/* This table maps ExecDirectoryType to the setting it is configured with in the unit */
static const char* const exec_directory_type_table[_EXEC_DIRECTORY_TYPE_MAX] = {
[EXEC_DIRECTORY_RUNTIME] = "RuntimeDirectory",
[EXEC_DIRECTORY_STATE] = "StateDirectory",
DEFINE_STRING_TABLE_LOOKUP(exec_directory_type, 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. */
+static const char* const exec_resource_type_table[_EXEC_DIRECTORY_TYPE_MAX] = {
+ [EXEC_DIRECTORY_RUNTIME] = "runtime",
+ [EXEC_DIRECTORY_STATE] = "state",
+ [EXEC_DIRECTORY_CACHE] = "cache",
+ [EXEC_DIRECTORY_LOGS] = "logs",
+ [EXEC_DIRECTORY_CONFIGURATION] = "configuration",
+};
+
+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",
projects / thirdparty / systemd.git / blobdiff