/*-*- Mode: C; c-basic-offset: 8 -*-*/
-#pragma GCC diagnostic ignored "-Wattributes"
+/***
+ This file is part of systemd.
+
+ Copyright 2010 Lennart Poettering
+
+ systemd is free software; you can redistribute it and/or modify it
+ under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ systemd is distributed in the hope that it will be useful, but
+ WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
#include <assert.h>
#include <dirent.h>
#include <sys/un.h>
#include <sys/prctl.h>
#include <linux/sched.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <grp.h>
+#include <pwd.h>
#include "execute.h"
#include "strv.h"
#include "log.h"
#include "ioprio.h"
#include "securebits.h"
-
-static int close_fds(int except[], unsigned n_except) {
- DIR *d;
- struct dirent *de;
- int r = 0;
-
- /* Modifies the fds array! (sorts it) */
-
- if (!(d = opendir("/proc/self/fd")))
- return -errno;
-
- while ((de = readdir(d))) {
- int fd;
-
- if (de->d_name[0] == '.')
- continue;
-
- if ((r = safe_atoi(de->d_name, &fd)) < 0)
- goto finish;
-
- if (fd < 3)
- continue;
-
- if (fd == dirfd(d))
- continue;
-
- if (except) {
- bool found;
- unsigned i;
-
- found = false;
- for (i = 0; i < n_except; i++)
- if (except[i] == fd) {
- found = true;
- break;
- }
-
- if (found)
- continue;
- }
-
- if ((r = close_nointr(fd)) < 0)
- goto finish;
- }
-
-finish:
- closedir(d);
- return r;
-}
+#include "cgroup.h"
static int shift_fds(int fds[], unsigned n_fds) {
int start, restart_from;
if (n_fds <= 0)
return 0;
+ /* Modifies the fds array! (sorts it) */
+
assert(fds);
start = 0;
return 0;
}
-static int flags_fds(int fds[], unsigned n_fds) {
+static int flags_fds(int fds[], unsigned n_fds, bool nonblock) {
unsigned i;
+ int r;
if (n_fds <= 0)
return 0;
assert(fds);
- /* Drops O_NONBLOCK and FD_CLOEXEC from the file flags */
+ /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags */
for (i = 0; i < n_fds; i++) {
- int flags;
-
- if ((flags = fcntl(fds[i], F_GETFL, 0)) < 0)
- return -errno;
- /* Since we are at it, let's make sure that nobody
- * forgot setting O_NONBLOCK for all our fds */
-
- if (fcntl(fds[i], F_SETFL, flags &~O_NONBLOCK) < 0)
- return -errno;
+ if ((r = fd_nonblock(fds[i], nonblock)) < 0)
+ return r;
- if ((flags = fcntl(fds[i], F_GETFD, 0)) < 0)
- return -errno;
+ /* We unconditionally drop FD_CLOEXEC from the fds,
+ * since after all we want to pass these fds to our
+ * children */
- /* Also make sure nobody forgot O_CLOEXEC for all our
- * fds */
- if (fcntl(fds[i], F_SETFD, flags &~FD_CLOEXEC) < 0)
- return -errno;
+ if ((r = fd_cloexec(fds[i], false)) < 0)
+ return r;
}
return 0;
}
}
-int setup_input(const ExecContext *context) {
+static int setup_input(const ExecContext *context) {
int r;
assert(context);
}
}
-int exec_spawn(const ExecCommand *command, const ExecContext *context, int *fds, unsigned n_fds, pid_t *ret) {
+static int get_group_creds(const char *groupname, gid_t *gid) {
+ struct group *g;
+ unsigned long lu;
+
+ assert(groupname);
+ assert(gid);
+
+ /* We enforce some special rules for gid=0: in order to avoid
+ * NSS lookups for root we hardcode its data. */
+
+ if (streq(groupname, "root") || streq(groupname, "0")) {
+ *gid = 0;
+ return 0;
+ }
+
+ if (safe_atolu(groupname, &lu) >= 0) {
+ errno = 0;
+ g = getgrgid((gid_t) lu);
+ } else {
+ errno = 0;
+ g = getgrnam(groupname);
+ }
+
+ if (!g)
+ return errno != 0 ? -errno : -ESRCH;
+
+ *gid = g->gr_gid;
+ return 0;
+}
+
+static int get_user_creds(const char **username, uid_t *uid, gid_t *gid, const char **home) {
+ struct passwd *p;
+ unsigned long lu;
+
+ assert(username);
+ assert(*username);
+ assert(uid);
+ assert(gid);
+ assert(home);
+
+ /* We enforce some special rules for uid=0: in order to avoid
+ * NSS lookups for root we hardcode its data. */
+
+ if (streq(*username, "root") || streq(*username, "0")) {
+ *username = "root";
+ *uid = 0;
+ *gid = 0;
+ *home = "/root";
+ return 0;
+ }
+
+ if (safe_atolu(*username, &lu) >= 0) {
+ errno = 0;
+ p = getpwuid((uid_t) lu);
+
+ /* If there are multiple users with the same id, make
+ * sure to leave $USER to the configured value instead
+ * of the first occurence in the database. However if
+ * the uid was configured by a numeric uid, then let's
+ * pick the real username from /etc/passwd. */
+ if (*username && p)
+ *username = p->pw_name;
+ } else {
+ errno = 0;
+ p = getpwnam(*username);
+ }
+
+ if (!p)
+ return errno != 0 ? -errno : -ESRCH;
+
+ *uid = p->pw_uid;
+ *gid = p->pw_gid;
+ *home = p->pw_dir;
+ return 0;
+}
+
+static int enforce_groups(const ExecContext *context, const char *username, gid_t gid) {
+ bool keep_groups = false;
+ int r;
+
+ assert(context);
+
+ /* Lookup and ser GID and supplementary group list. Here too
+ * we avoid NSS lookups for gid=0. */
+
+ if (context->group || username) {
+
+ if (context->group)
+ if ((r = get_group_creds(context->group, &gid)) < 0)
+ return r;
+
+ /* First step, initialize groups from /etc/groups */
+ if (username && gid != 0) {
+ if (initgroups(username, gid) < 0)
+ return -errno;
+
+ keep_groups = true;
+ }
+
+ /* Second step, set our gids */
+ if (setresgid(gid, gid, gid) < 0)
+ return -errno;
+ }
+
+ if (context->supplementary_groups) {
+ int ngroups_max, k;
+ gid_t *gids;
+ char **i;
+
+ /* Final step, initialize any manually set supplementary groups */
+ ngroups_max = (int) sysconf(_SC_NGROUPS_MAX);
+
+ if (!(gids = new(gid_t, ngroups_max)))
+ return -ENOMEM;
+
+ if (keep_groups) {
+ if ((k = getgroups(ngroups_max, gids)) < 0) {
+ free(gids);
+ return -errno;
+ }
+ } else
+ k = 0;
+
+ STRV_FOREACH(i, context->supplementary_groups) {
+
+ if (k >= ngroups_max) {
+ free(gids);
+ return -E2BIG;
+ }
+
+ if ((r = get_group_creds(*i, gids+k)) < 0) {
+ free(gids);
+ return r;
+ }
+
+ k++;
+ }
+
+ if (setgroups(k, gids) < 0) {
+ free(gids);
+ return -errno;
+ }
+
+ free(gids);
+ }
+
+ return 0;
+}
+
+static int enforce_user(const ExecContext *context, uid_t uid) {
+ int r;
+ assert(context);
+
+ /* Sets (but doesn't lookup) the uid and make sure we keep the
+ * capabilities while doing so. */
+
+ if (context->capabilities) {
+ cap_t d;
+ static const cap_value_t bits[] = {
+ CAP_SETUID, /* Necessary so that we can run setresuid() below */
+ CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */
+ };
+
+ /* First step: If we need to keep capabilities but
+ * drop privileges we need to make sure we keep our
+ * caps, whiel we drop priviliges. */
+ if (uid != 0) {
+ int sb = context->secure_bits|SECURE_KEEP_CAPS;
+
+ if (prctl(PR_GET_SECUREBITS) != sb)
+ if (prctl(PR_SET_SECUREBITS, sb) < 0)
+ return -errno;
+ }
+
+ /* Second step: set the capabilites. This will reduce
+ * the capabilities to the minimum we need. */
+
+ if (!(d = cap_dup(context->capabilities)))
+ return -errno;
+
+ if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 ||
+ cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) {
+ r = -errno;
+ cap_free(d);
+ return r;
+ }
+
+ if (cap_set_proc(d) < 0) {
+ r = -errno;
+ cap_free(d);
+ return r;
+ }
+
+ cap_free(d);
+ }
+
+ /* Third 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;
+}
+
+int exec_spawn(const ExecCommand *command,
+ const ExecContext *context,
+ int *fds, unsigned n_fds,
+ bool apply_permissions,
+ bool apply_chroot,
+ CGroupBonding *cgroup_bondings,
+ pid_t *ret) {
+
pid_t pid;
+ int r;
assert(command);
assert(context);
assert(ret);
assert(fds || n_fds <= 0);
- log_debug("about to execute %s", command->path);
+ log_debug("About to execute %s", command->path);
+
+ if (cgroup_bondings)
+ if ((r = cgroup_bonding_realize_list(cgroup_bondings)))
+ return r;
if ((pid = fork()) < 0)
return -errno;
if (pid == 0) {
- char **e, **f = NULL;
- int i, r;
+ int i;
sigset_t ss;
+ const char *username = NULL, *home = NULL;
+ uid_t uid = (uid_t) -1;
+ gid_t gid = (gid_t) -1;
+ char **our_env = NULL, **final_env = NULL;
+ unsigned n_env = 0;
/* child */
goto fail;
}
- if (setpgid(0, 0) < 0) {
- r = EXIT_PGID;
- goto fail;
+ if (context->new_session) {
+ if (setsid() < 0) {
+ r = EXIT_SETSID;
+ goto fail;
+ }
+ } else {
+ if (setpgid(0, 0) < 0) {
+ r = EXIT_PGID;
+ goto fail;
+ }
}
umask(context->umask);
goto fail;
}
+ if (cgroup_bondings)
+ if ((r = cgroup_bonding_install_list(cgroup_bondings, 0)) < 0) {
+ r = EXIT_CGROUP;
+ goto fail;
+ }
+
if (context->oom_adjust_set) {
char t[16];
}
}
- if (context->root_directory)
- if (chroot(context->root_directory) < 0) {
- r = EXIT_CHROOT;
- goto fail;
- }
-
- if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
- r = EXIT_CHDIR;
- goto fail;
- }
-
if (context->nice_set)
if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) {
r = EXIT_NICE;
goto fail;
}
- if (close_fds(fds, n_fds) < 0 ||
- shift_fds(fds, n_fds) < 0 ||
- flags_fds(fds, n_fds) < 0) {
- r = EXIT_FDS;
- goto fail;
+ if (context->user) {
+ username = context->user;
+ if (get_user_creds(&username, &uid, &gid, &home) < 0) {
+ r = EXIT_USER;
+ goto fail;
+ }
}
- for (i = 0; i < RLIMIT_NLIMITS; i++) {
- if (!context->rlimit[i])
- continue;
+ if (apply_permissions)
+ if (enforce_groups(context, username, uid) < 0) {
+ r = EXIT_GROUP;
+ goto fail;
+ }
+
+ if (apply_chroot) {
+ if (context->root_directory)
+ if (chroot(context->root_directory) < 0) {
+ r = EXIT_CHROOT;
+ goto fail;
+ }
- if (setrlimit(i, context->rlimit[i]) < 0) {
- r = EXIT_LIMITS;
+ if (chdir(context->working_directory ? context->working_directory : "/") < 0) {
+ r = EXIT_CHDIR;
goto fail;
}
- }
+ } else {
- if (context->secure_bits) {
- if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
- r = EXIT_SECUREBITS;
+ char *d;
+
+ if (asprintf(&d, "%s/%s",
+ context->root_directory ? context->root_directory : "",
+ context->working_directory ? context->working_directory : "") < 0) {
+ r = EXIT_MEMORY;
+ goto fail;
+ }
+
+ if (chdir(d) < 0) {
+ free(d);
+ r = EXIT_CHDIR;
goto fail;
}
+
+ free(d);
+ }
+
+ if (close_all_fds(fds, n_fds) < 0 ||
+ shift_fds(fds, n_fds) < 0 ||
+ flags_fds(fds, n_fds, context->non_blocking) < 0) {
+ r = EXIT_FDS;
+ goto fail;
}
- if (n_fds > 0) {
- char a[64], b[64];
- char *listen_env[3] = {
- a,
- b,
- NULL
- };
+ if (apply_permissions) {
- snprintf(a, sizeof(a), "LISTEN_PID=%llu", (unsigned long long) getpid());
- snprintf(b, sizeof(b), "LISTEN_FDS=%u", n_fds);
+ for (i = 0; i < RLIMIT_NLIMITS; i++) {
+ if (!context->rlimit[i])
+ continue;
- a[sizeof(a)-1] = 0;
- b[sizeof(b)-1] = 0;
+ if (setrlimit(i, context->rlimit[i]) < 0) {
+ r = EXIT_LIMITS;
+ goto fail;
+ }
+ }
- if (context->environment) {
- if (!(f = strv_merge(listen_env, context->environment))) {
- r = EXIT_MEMORY;
+ if (context->user)
+ if (enforce_user(context, uid) < 0) {
+ r = EXIT_USER;
goto fail;
}
- e = f;
- } else
- e = listen_env;
- } else
- e = context->environment;
+ /* PR_GET_SECUREBITS is not priviliged, while
+ * PR_SET_SECUREBITS is. So to suppress
+ * potential EPERMs we'll try not to call
+ * PR_SET_SECUREBITS unless necessary. */
+ if (prctl(PR_GET_SECUREBITS) != context->secure_bits)
+ if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) {
+ r = EXIT_SECUREBITS;
+ goto fail;
+ }
- execve(command->path, command->argv, e);
+ if (context->capabilities)
+ if (cap_set_proc(context->capabilities) < 0) {
+ r = EXIT_CAPABILITIES;
+ goto fail;
+ }
+ }
+
+ if (!(our_env = new0(char*, 6))) {
+ r = EXIT_MEMORY;
+ goto fail;
+ }
+
+ if (n_fds > 0)
+ if (asprintf(our_env + n_env++, "LISTEN_PID=%llu", (unsigned long long) getpid()) < 0 ||
+ asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) {
+ r = EXIT_MEMORY;
+ goto fail;
+ }
+
+ if (home)
+ if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) {
+ r = EXIT_MEMORY;
+ goto fail;
+ }
+
+ if (username)
+ if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 ||
+ asprintf(our_env + n_env++, "USER=%s", username) < 0) {
+ r = EXIT_MEMORY;
+ goto fail;
+ }
+
+ if (!(final_env = strv_env_merge(environ, our_env, context->environment, NULL))) {
+ r = EXIT_MEMORY;
+ goto fail;
+ }
+
+ execve(command->path, command->argv, final_env);
r = EXIT_EXEC;
fail:
- strv_free(f);
+ strv_free(our_env);
+ strv_free(final_env);
+
_exit(r);
}
- log_debug("executed %s as %llu", command->path, (unsigned long long) pid);
+ log_debug("Forked %s as %llu", command->path, (unsigned long long) pid);
*ret = pid;
return 0;
c->cpu_sched_set = false;
CPU_ZERO(&c->cpu_affinity);
c->cpu_affinity_set = false;
+ c->timer_slack_ns = 0;
+ c->timer_slack_ns_set = false;
+
+ c->cpu_sched_reset_on_fork = false;
+ c->non_blocking = false;
+ c->new_session = false;
c->input = 0;
c->output = 0;
fprintf(f,
"%sUMask: %04o\n"
"%sWorkingDirectory: %s\n"
- "%sRootDirectory: %s\n",
+ "%sRootDirectory: %s\n"
+ "%sNonBlocking: %s\n"
+ "%sNewSession: %s\n",
prefix, c->umask,
prefix, c->working_directory ? c->working_directory : "/",
- prefix, c->root_directory ? c->root_directory : "/");
+ prefix, c->root_directory ? c->root_directory : "/",
+ prefix, yes_no(c->non_blocking),
+ prefix, yes_no(c->new_session));
if (c->environment)
for (e = c->environment; *e; e++)
exec_command_dump(c, f, prefix);
}
+void exec_command_append_list(ExecCommand **l, ExecCommand *e) {
+ ExecCommand *end;
+
+ assert(l);
+ assert(e);
+
+ if (*l) {
+ /* It's kinda important that we keep the order here */
+ LIST_FIND_TAIL(ExecCommand, command, *l, end);
+ LIST_INSERT_AFTER(ExecCommand, command, *l, end, e);
+ } else
+ *l = e;
+}
+
static const char* const exec_output_table[_EXEC_OUTPUT_MAX] = {
[EXEC_OUTPUT_CONSOLE] = "console",
[EXEC_OUTPUT_NULL] = "null",