#include <linux/fs.h>
#include <linux/oom.h>
#include <sys/poll.h>
-#include <linux/seccomp-bpf.h>
#include <glob.h>
+#include <sys/personality.h>
#include <libgen.h>
#undef basename
#include <security/pam_appl.h>
#endif
+#ifdef HAVE_SELINUX
+#include <selinux/selinux.h>
+#endif
+
+#ifdef HAVE_SECCOMP
+#include <seccomp.h>
+#endif
+
#include "execute.h"
#include "strv.h"
#include "macro.h"
#include "utmp-wtmp.h"
#include "def.h"
#include "path-util.h"
-#include "syscall-list.h"
#include "env-util.h"
#include "fileio.h"
#include "unit.h"
#include "async.h"
+#include "selinux-util.h"
+#include "errno-list.h"
+
+#ifdef HAVE_SECCOMP
+#include "seccomp-util.h"
+#endif
#define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
}
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;
+ _cleanup_cap_free_ cap_t d = NULL;
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 */
/* Second step: set the capabilities. This will reduce
* the capabilities to the minimum we need. */
- if (!(d = cap_dup(context->capabilities)))
+ d = cap_dup(context->capabilities);
+ if (!d)
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_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0)
+ return -errno;
- cap_free(d);
+ if (cap_set_proc(d) < 0)
+ return -errno;
}
/* Third step: actually set the uids */
rename_process(process_name);
}
-static int apply_seccomp(uint32_t *syscall_filter) {
- static const struct sock_filter header[] = {
- VALIDATE_ARCHITECTURE,
- EXAMINE_SYSCALL
- };
- static const struct sock_filter footer[] = {
- _KILL_PROCESS
- };
-
- int i;
- unsigned n;
- struct sock_filter *f;
- struct sock_fprog prog = {};
+#ifdef HAVE_SECCOMP
- assert(syscall_filter);
+static int apply_seccomp(ExecContext *c) {
+ uint32_t negative_action, action;
+ scmp_filter_ctx *seccomp;
+ Iterator i;
+ void *id;
+ int r;
- /* First: count the syscalls to check for */
- for (i = 0, n = 0; i < syscall_max(); i++)
- if (syscall_filter[i >> 4] & (1 << (i & 31)))
- n++;
+ assert(c);
- /* Second: build the filter program from a header the syscall
- * matches and the footer */
- f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer)));
- memcpy(f, header, sizeof(header));
+ negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno);
- for (i = 0, n = 0; i < syscall_max(); i++)
- if (syscall_filter[i >> 4] & (1 << (i & 31))) {
- struct sock_filter item[] = {
- BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1),
- BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW)
- };
+ seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW);
+ if (!seccomp)
+ return -ENOMEM;
- assert_cc(ELEMENTSOF(item) == 2);
+ if (c->syscall_archs) {
- f[ELEMENTSOF(header) + 2*n] = item[0];
- f[ELEMENTSOF(header) + 2*n+1] = item[1];
+ SET_FOREACH(id, c->syscall_archs, i) {
+ r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
+ if (r == -EEXIST)
+ continue;
+ if (r < 0) {
+ seccomp_release(seccomp);
+ return r;
+ }
+ }
+ } else {
- n++;
+ r = seccomp_add_secondary_archs(seccomp);
+ if (r < 0) {
+ seccomp_release(seccomp);
+ return r;
}
+ }
- memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer));
+ action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action;
+ SET_FOREACH(id, c->syscall_filter, i) {
+ r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0);
+ if (r < 0) {
+ seccomp_release(seccomp);
+ return r;
+ }
+ }
- /* Third: install the filter */
- prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n;
- prog.filter = f;
- if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0)
- return -errno;
+ r = seccomp_load(seccomp);
+ seccomp_release(seccomp);
- return 0;
+ return r;
}
+#endif
static void do_idle_pipe_dance(int idle_pipe[4]) {
assert(idle_pipe);
static int build_environment(
ExecContext *c,
unsigned n_fds,
+ usec_t watchdog_usec,
const char *home,
const char *username,
const char *shell,
assert(c);
assert(ret);
- our_env = new(char*, 8);
+ our_env = new0(char*, 10);
if (!our_env)
return -ENOMEM;
if (n_fds > 0) {
- if (asprintf(&x, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0)
+ if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0)
return -ENOMEM;
our_env[n_env++] = x;
our_env[n_env++] = x;
}
+ if (watchdog_usec > 0) {
+ if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+
+ if (asprintf(&x, "WATCHDOG_USEC=%llu", (unsigned long long) watchdog_usec) < 0)
+ return -ENOMEM;
+ our_env[n_env++] = x;
+ }
+
if (home) {
x = strappend("HOME=", home);
if (!x)
}
our_env[n_env++] = NULL;
- assert(n_env <= 8);
+ assert(n_env <= 10);
*ret = our_env;
our_env = NULL;
CGroupControllerMask cgroup_supported,
const char *cgroup_path,
const char *unit_id,
+ usec_t watchdog_usec,
int idle_pipe[4],
ExecRuntime *runtime,
pid_t *ret) {
goto fail_child;
}
+ if (context->personality != 0xffffffffUL)
+ if (personality(context->personality) < 0) {
+ err = -errno;
+ r = EXIT_PERSONALITY;
+ goto fail_child;
+ }
+
if (context->utmp_id)
utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path);
!strv_isempty(context->read_only_dirs) ||
!strv_isempty(context->inaccessible_dirs) ||
context->mount_flags != 0 ||
- (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir))) {
+ (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) ||
+ context->private_devices) {
char *tmp = NULL, *var = NULL;
context->inaccessible_dirs,
tmp,
var,
+ context->private_devices,
context->mount_flags);
if (err < 0) {
goto fail_child;
}
- if (context->syscall_filter) {
- err = apply_seccomp(context->syscall_filter);
+#ifdef HAVE_SECCOMP
+ if (context->syscall_filter || context->syscall_archs) {
+ err = apply_seccomp(context);
if (err < 0) {
r = EXIT_SECCOMP;
goto fail_child;
}
}
+#endif
+
+#ifdef HAVE_SELINUX
+ if (context->selinux_context && use_selinux()) {
+ err = setexeccon(context->selinux_context);
+ if (err < 0 && !context->selinux_context_ignore) {
+ r = EXIT_SELINUX_CONTEXT;
+ goto fail_child;
+ }
+ }
+#endif
}
- err = build_environment(context, n_fds, home, username, shell, &our_env);
+ err = build_environment(context, n_fds, watchdog_usec, home, username, shell, &our_env);
if (r < 0) {
r = EXIT_MEMORY;
goto fail_child;
log_struct_unit(LOG_DEBUG,
unit_id,
- "MESSAGE=Forked %s as %lu",
- command->path, (unsigned long) pid,
+ "MESSAGE=Forked %s as "PID_FMT,
+ command->path, pid,
NULL);
/* We add the new process to the cgroup both in the child (so
c->syslog_level_prefix = true;
c->ignore_sigpipe = true;
c->timer_slack_nsec = (nsec_t) -1;
+ c->personality = 0xffffffffUL;
}
void exec_context_done(ExecContext *c) {
free(c->utmp_id);
c->utmp_id = NULL;
- free(c->syscall_filter);
+ free(c->selinux_context);
+ c->selinux_context = NULL;
+
+#ifdef HAVE_SECCOMP
+ set_free(c->syscall_filter);
c->syscall_filter = NULL;
+
+ set_free(c->syscall_archs);
+ c->syscall_archs = NULL;
+#endif
}
void exec_command_done(ExecCommand *c) {
"%sNonBlocking: %s\n"
"%sPrivateTmp: %s\n"
"%sPrivateNetwork: %s\n"
+ "%sPrivateDevices: %s\n"
"%sIgnoreSIGPIPE: %s\n",
prefix, c->umask,
prefix, c->working_directory ? c->working_directory : "/",
prefix, yes_no(c->non_blocking),
prefix, yes_no(c->private_tmp),
prefix, yes_no(c->private_network),
+ prefix, yes_no(c->private_devices),
prefix, yes_no(c->ignore_sigpipe));
STRV_FOREACH(e, c->environment)
fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max);
if (c->ioprio_set) {
- char *class_str;
- int r;
+ _cleanup_free_ char *class_str = NULL;
- r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
- if (r < 0)
- class_str = NULL;
+ ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str);
fprintf(f,
"%sIOSchedulingClass: %s\n"
"%sIOPriority: %i\n",
prefix, strna(class_str),
prefix, (int) IOPRIO_PRIO_DATA(c->ioprio));
- free(class_str);
}
if (c->cpu_sched_set) {
- char *policy_str;
- int r;
+ _cleanup_free_ char *policy_str = NULL;
- r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
- if (r < 0)
- policy_str = NULL;
+ sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str);
fprintf(f,
"%sCPUSchedulingPolicy: %s\n"
"%sCPUSchedulingPriority: %i\n"
prefix, strna(policy_str),
prefix, c->cpu_sched_priority,
prefix, yes_no(c->cpu_sched_reset_on_fork));
- free(policy_str);
}
if (c->cpuset) {
fprintf(f, "%sCPUAffinity:", prefix);
for (i = 0; i < c->cpuset_ncpus; i++)
if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset))
- fprintf(f, " %i", i);
+ fprintf(f, " %u", i);
fputs("\n", f);
}
if (c->timer_slack_nsec != (nsec_t) -1)
- fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, (unsigned long)c->timer_slack_nsec);
+ fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec);
fprintf(f,
"%sStandardInput: %s\n"
prefix, yes_no(c->tty_vhangup),
prefix, yes_no(c->tty_vt_disallocate));
- if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL ||
- c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
- c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL ||
- c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
- char *fac_str, *lvl_str;
- int r;
+ if (c->std_output == EXEC_OUTPUT_SYSLOG ||
+ c->std_output == EXEC_OUTPUT_KMSG ||
+ c->std_output == EXEC_OUTPUT_JOURNAL ||
+ c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
+ c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
+ c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE ||
+ c->std_error == EXEC_OUTPUT_SYSLOG ||
+ c->std_error == EXEC_OUTPUT_KMSG ||
+ c->std_error == EXEC_OUTPUT_JOURNAL ||
+ c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE ||
+ c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE ||
+ c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) {
- r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
- if (r < 0)
- fac_str = NULL;
+ _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL;
- r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
- if (r < 0)
- lvl_str = NULL;
+ log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str);
+ log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str);
fprintf(f,
"%sSyslogFacility: %s\n"
"%sSyslogLevel: %s\n",
prefix, strna(fac_str),
prefix, strna(lvl_str));
- free(lvl_str);
- free(fac_str);
}
if (c->capabilities) {
- char *t;
- if ((t = cap_to_text(c->capabilities, NULL))) {
- fprintf(f, "%sCapabilities: %s\n",
- prefix, t);
- cap_free(t);
- }
+ _cleanup_cap_free_charp_ char *t;
+
+ t = cap_to_text(c->capabilities, NULL);
+ if (t)
+ fprintf(f, "%sCapabilities: %s\n", prefix, t);
}
if (c->secure_bits)
for (l = 0; l <= cap_last_cap(); l++)
if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) {
- char *t;
+ _cleanup_cap_free_charp_ char *t;
- if ((t = cap_to_name(l))) {
+ t = cap_to_name(l);
+ if (t)
fprintf(f, " %s", t);
- cap_free(t);
- }
}
fputs("\n", f);
fprintf(f,
"%sUtmpIdentifier: %s\n",
prefix, c->utmp_id);
+
+ if (c->selinux_context)
+ fprintf(f,
+ "%sSELinuxContext: %s%s\n",
+ prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context);
+
+ if (c->personality != 0xffffffffUL)
+ fprintf(f,
+ "%sPersonality: %s\n",
+ prefix, strna(personality_to_string(c->personality)));
+
+ if (c->syscall_filter) {
+#ifdef HAVE_SECCOMP
+ Iterator j;
+ void *id;
+ bool first = true;
+#endif
+
+ fprintf(f,
+ "%sSystemCallFilter: ",
+ prefix);
+
+ if (!c->syscall_whitelist)
+ fputc('~', f);
+
+#ifdef HAVE_SECCOMP
+ SET_FOREACH(id, c->syscall_filter, j) {
+ _cleanup_free_ char *name = NULL;
+
+ if (first)
+ first = false;
+ else
+ fputc(' ', f);
+
+ name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1);
+ fputs(strna(name), f);
+ }
+#endif
+
+ fputc('\n', f);
+ }
+
+ if (c->syscall_archs) {
+#ifdef HAVE_SECCOMP
+ Iterator j;
+ void *id;
+#endif
+
+ fprintf(f,
+ "%sSystemCallArchitectures:",
+ prefix);
+
+#ifdef HAVE_SECCOMP
+ SET_FOREACH(id, c->syscall_archs, j)
+ fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1)));
+#endif
+ fputc('\n', f);
+ }
+
+ if (c->syscall_errno != 0)
+ fprintf(f,
+ "%sSystemCallErrorNumber: %s\n",
+ prefix, strna(errno_to_name(c->syscall_errno)));
}
void exec_status_start(ExecStatus *s, pid_t pid) {
return;
fprintf(f,
- "%sPID: %lu\n",
- prefix, (unsigned long) s->pid);
+ "%sPID: "PID_FMT"\n",
+ prefix, s->pid);
if (s->start_timestamp.realtime > 0)
fprintf(f,
return 0;
*rt = new0(ExecRuntime, 1);
- if (!rt)
+ if (!*rt)
return -ENOMEM;
(*rt)->n_ref = 1;
~ianmdlvl / elogind.git / blobdiff