+/* SPDX-License-Identifier: LGPL-2.1+ */
/***
- 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 Lesser General Public License as published by
- the Free Software Foundation; either version 2.1 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public License
- along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
#include <ctype.h>
#include <signal.h>
#include <stdbool.h>
#include <stdio.h>
+#include <stdio_ext.h>
#include <stdlib.h>
#include <string.h>
+#include <sys/mman.h>
+//#include <sys/mount.h>
#include <sys/personality.h>
#include <sys/prctl.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <syslog.h>
#include <unistd.h>
-#ifdef HAVE_VALGRIND_VALGRIND_H
+#if HAVE_VALGRIND_VALGRIND_H
#include <valgrind/valgrind.h>
#endif
#include "macro.h"
#include "missing.h"
#include "process-util.h"
+#include "raw-clone.h"
#include "signal-util.h"
//#include "stat-util.h"
#include "string-table.h"
#include "string-util.h"
+//#include "terminal-util.h"
#include "user-util.h"
#include "util.h"
return (unsigned char) state;
}
-int get_process_comm(pid_t pid, char **name) {
+int get_process_comm(pid_t pid, char **ret) {
+ _cleanup_free_ char *escaped = NULL, *comm = NULL;
const char *p;
int r;
- assert(name);
+ assert(ret);
assert(pid >= 0);
+ escaped = new(char, TASK_COMM_LEN);
+ if (!escaped)
+ return -ENOMEM;
+
p = procfs_file_alloca(pid, "comm");
- r = read_one_line_file(p, name);
+ r = read_one_line_file(p, &comm);
if (r == -ENOENT)
return -ESRCH;
+ if (r < 0)
+ return r;
- return r;
+ /* Escape unprintable characters, just in case, but don't grow the string beyond the underlying size */
+ cellescape(escaped, TASK_COMM_LEN, comm);
+
+ *ret = TAKE_PTR(escaped);
+ return 0;
}
int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line) {
_cleanup_fclose_ FILE *f = NULL;
bool space = false;
- char *r = NULL, *k;
+ char *k, *ans = NULL;
const char *p;
int c;
* command line that resolves to the empty string will return the "comm" name of the process instead.
*
* Returns -ESRCH if the process doesn't exist, and -ENOENT if the process has no command line (and
- * comm_fallback is false). */
+ * comm_fallback is false). Returns 0 and sets *line otherwise. */
p = procfs_file_alloca(pid, "cmdline");
return -errno;
}
+ (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
+
if (max_length == 1) {
/* If there's only room for one byte, return the empty string */
- r = new0(char, 1);
- if (!r)
+ ans = new0(char, 1);
+ if (!ans)
return -ENOMEM;
- *line = r;
+ *line = ans;
return 0;
} else if (max_length == 0) {
while ((c = getc(f)) != EOF) {
- if (!GREEDY_REALLOC(r, allocated, len+3)) {
- free(r);
+ if (!GREEDY_REALLOC(ans, allocated, len+3)) {
+ free(ans);
return -ENOMEM;
}
if (isprint(c)) {
if (space) {
- r[len++] = ' ';
+ ans[len++] = ' ';
space = false;
}
- r[len++] = c;
+ ans[len++] = c;
} else if (len > 0)
space = true;
}
if (len > 0)
- r[len] = 0;
+ ans[len] = '\0';
else
- r = mfree(r);
+ ans = mfree(ans);
} else {
bool dotdotdot = false;
size_t left;
- r = new(char, max_length);
- if (!r)
+ ans = new(char, max_length);
+ if (!ans)
return -ENOMEM;
- k = r;
+ k = ans;
left = max_length;
while ((c = getc(f)) != EOF) {
*(k++) = (char) c;
left--;
- } else if (k > r)
+ } else if (k > ans)
space = true;
}
if (dotdotdot) {
if (max_length <= 4) {
- k = r;
+ k = ans;
left = max_length;
} else {
- k = r + max_length - 4;
+ k = ans + max_length - 4;
left = 4;
/* Eat up final spaces */
- while (k > r && isspace(k[-1])) {
+ while (k > ans && isspace(k[-1])) {
k--;
left++;
}
}
/* Kernel threads have no argv[] */
- if (isempty(r)) {
+ if (isempty(ans)) {
_cleanup_free_ char *t = NULL;
int h;
- free(r);
+ free(ans);
if (!comm_fallback)
return -ENOENT;
return h;
if (max_length == 0)
- r = strjoin("[", t, "]", NULL);
+ ans = strjoin("[", t, "]");
else {
size_t l;
l = strlen(t);
if (l + 3 <= max_length)
- r = strjoin("[", t, "]", NULL);
+ ans = strjoin("[", t, "]");
else if (max_length <= 6) {
- r = new(char, max_length);
- if (!r)
+ ans = new(char, max_length);
+ if (!ans)
return -ENOMEM;
- memcpy(r, "[...]", max_length-1);
- r[max_length-1] = 0;
+ memcpy(ans, "[...]", max_length-1);
+ ans[max_length-1] = 0;
} else {
- char *e;
-
t[max_length - 6] = 0;
/* Chop off final spaces */
- e = strchr(t, 0);
- while (e > t && isspace(e[-1]))
- e--;
- *e = 0;
+ delete_trailing_chars(t, WHITESPACE);
- r = strjoin("[", t, "...]", NULL);
+ ans = strjoin("[", t, "...]");
}
}
- if (!r)
+ if (!ans)
return -ENOMEM;
}
- *line = r;
+ *line = ans;
return 0;
}
-#if 0 /// UNNEEDED by elogind
-void rename_process(const char name[8]) {
- assert(name);
+int rename_process(const char name[]) {
+ static size_t mm_size = 0;
+ static char *mm = NULL;
+ bool truncated = false;
+ size_t l;
+
+ /* This is a like a poor man's setproctitle(). It changes the comm field, argv[0], and also the glibc's
+ * internally used name of the process. For the first one a limit of 16 chars applies; to the second one in
+ * many cases one of 10 (i.e. length of "/sbin/init") — however if we have CAP_SYS_RESOURCES it is unbounded;
+ * to the third one 7 (i.e. the length of "systemd". If you pass a longer string it will likely be
+ * truncated.
+ *
+ * Returns 0 if a name was set but truncated, > 0 if it was set but not truncated. */
- /* This is a like a poor man's setproctitle(). It changes the
- * comm field, argv[0], and also the glibc's internally used
- * name of the process. For the first one a limit of 16 chars
- * applies, to the second one usually one of 10 (i.e. length
- * of "/sbin/init"), to the third one one of 7 (i.e. length of
- * "systemd"). If you pass a longer string it will be
- * truncated */
+ if (isempty(name))
+ return -EINVAL; /* let's not confuse users unnecessarily with an empty name */
- (void) prctl(PR_SET_NAME, name);
+ if (!is_main_thread())
+ return -EPERM; /* Let's not allow setting the process name from other threads than the main one, as we
+ * cache things without locking, and we make assumptions that PR_SET_NAME sets the
+ * process name that isn't correct on any other threads */
- if (program_invocation_name)
- strncpy(program_invocation_name, name, strlen(program_invocation_name));
+ l = strlen(name);
+
+ /* First step, change the comm field. The main thread's comm is identical to the process comm. This means we
+ * can use PR_SET_NAME, which sets the thread name for the calling thread. */
+ if (prctl(PR_SET_NAME, name) < 0)
+ log_debug_errno(errno, "PR_SET_NAME failed: %m");
+ if (l >= TASK_COMM_LEN) /* Linux process names can be 15 chars at max */
+ truncated = true;
+
+ /* Second step, change glibc's ID of the process name. */
+ if (program_invocation_name) {
+ size_t k;
+
+ k = strlen(program_invocation_name);
+ strncpy(program_invocation_name, name, k);
+ if (l > k)
+ truncated = true;
+ }
+
+ /* Third step, completely replace the argv[] array the kernel maintains for us. This requires privileges, but
+ * has the advantage that the argv[] array is exactly what we want it to be, and not filled up with zeros at
+ * the end. This is the best option for changing /proc/self/cmdline. */
+
+ /* Let's not bother with this if we don't have euid == 0. Strictly speaking we should check for the
+ * CAP_SYS_RESOURCE capability which is independent of the euid. In our own code the capability generally is
+ * present only for euid == 0, hence let's use this as quick bypass check, to avoid calling mmap() if
+ * PR_SET_MM_ARG_{START,END} fails with EPERM later on anyway. After all geteuid() is dead cheap to call, but
+ * mmap() is not. */
+ if (geteuid() != 0)
+ log_debug("Skipping PR_SET_MM, as we don't have privileges.");
+ else if (mm_size < l+1) {
+ size_t nn_size;
+ char *nn;
+
+ nn_size = PAGE_ALIGN(l+1);
+ nn = mmap(NULL, nn_size, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+ if (nn == MAP_FAILED) {
+ log_debug_errno(errno, "mmap() failed: %m");
+ goto use_saved_argv;
+ }
+
+ strncpy(nn, name, nn_size);
+
+ /* Now, let's tell the kernel about this new memory */
+ if (prctl(PR_SET_MM, PR_SET_MM_ARG_START, (unsigned long) nn, 0, 0) < 0) {
+ log_debug_errno(errno, "PR_SET_MM_ARG_START failed, proceeding without: %m");
+ (void) munmap(nn, nn_size);
+ goto use_saved_argv;
+ }
+
+ /* And update the end pointer to the new end, too. If this fails, we don't really know what to do, it's
+ * pretty unlikely that we can rollback, hence we'll just accept the failure, and continue. */
+ if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) nn + l + 1, 0, 0) < 0)
+ log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
+
+ if (mm)
+ (void) munmap(mm, mm_size);
+
+ mm = nn;
+ mm_size = nn_size;
+ } else {
+ strncpy(mm, name, mm_size);
+
+ /* Update the end pointer, continuing regardless of any failure. */
+ if (prctl(PR_SET_MM, PR_SET_MM_ARG_END, (unsigned long) mm + l + 1, 0, 0) < 0)
+ log_debug_errno(errno, "PR_SET_MM_ARG_END failed, proceeding without: %m");
+ }
+
+use_saved_argv:
+ /* Fourth step: in all cases we'll also update the original argv[], so that our own code gets it right too if
+ * it still looks here */
if (saved_argc > 0) {
int i;
- if (saved_argv[0])
- strncpy(saved_argv[0], name, strlen(saved_argv[0]));
+ if (saved_argv[0]) {
+ size_t k;
+
+ k = strlen(saved_argv[0]);
+ strncpy(saved_argv[0], name, k);
+ if (l > k)
+ truncated = true;
+ }
for (i = 1; i < saved_argc; i++) {
if (!saved_argv[i])
memzero(saved_argv[i], strlen(saved_argv[i]));
}
}
+
+ return !truncated;
}
-#endif // 0
int is_kernel_thread(pid_t pid) {
+ _cleanup_free_ char *line = NULL;
+ unsigned long long flags;
+ size_t l, i;
const char *p;
- size_t count;
- char c;
- bool eof;
- FILE *f;
+ char *q;
+ int r;
- if (pid == 0 || pid == 1) /* pid 1, and we ourselves certainly aren't a kernel thread */
+ if (IN_SET(pid, 0, 1) || pid == getpid_cached()) /* pid 1, and we ourselves certainly aren't a kernel thread */
return 0;
+ if (!pid_is_valid(pid))
+ return -EINVAL;
- assert(pid > 1);
+ p = procfs_file_alloca(pid, "stat");
+ r = read_one_line_file(p, &line);
+ if (r == -ENOENT)
+ return -ESRCH;
+ if (r < 0)
+ return r;
- p = procfs_file_alloca(pid, "cmdline");
- f = fopen(p, "re");
- if (!f) {
- if (errno == ENOENT)
- return -ESRCH;
- return -errno;
+ /* Skip past the comm field */
+ q = strrchr(line, ')');
+ if (!q)
+ return -EINVAL;
+ q++;
+
+ /* Skip 6 fields to reach the flags field */
+ for (i = 0; i < 6; i++) {
+ l = strspn(q, WHITESPACE);
+ if (l < 1)
+ return -EINVAL;
+ q += l;
+
+ l = strcspn(q, WHITESPACE);
+ if (l < 1)
+ return -EINVAL;
+ q += l;
}
- count = fread(&c, 1, 1, f);
- eof = feof(f);
- fclose(f);
+ /* Skip preceeding whitespace */
+ l = strspn(q, WHITESPACE);
+ if (l < 1)
+ return -EINVAL;
+ q += l;
- /* Kernel threads have an empty cmdline */
+ /* Truncate the rest */
+ l = strcspn(q, WHITESPACE);
+ if (l < 1)
+ return -EINVAL;
+ q[l] = 0;
- if (count <= 0)
- return eof ? 1 : -errno;
+ r = safe_atollu(q, &flags);
+ if (r < 0)
+ return r;
- return 0;
+ return !!(flags & PF_KTHREAD);
}
#if 0 /// UNNEEDED by elogind
assert(field);
assert(uid);
+ if (pid < 0)
+ return -EINVAL;
+
p = procfs_file_alloca(pid, "status");
f = fopen(p, "re");
if (!f) {
return -errno;
}
+ (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
+
FOREACH_LINE(line, f, return -errno) {
char *l;
}
int get_process_uid(pid_t pid, uid_t *uid) {
+
+ if (pid == 0 || pid == getpid_cached()) {
+ *uid = getuid();
+ return 0;
+ }
+
return get_process_id(pid, "Uid:", uid);
}
int get_process_gid(pid_t pid, gid_t *gid) {
+
+ if (pid == 0 || pid == getpid_cached()) {
+ *gid = getgid();
+ return 0;
+ }
+
assert_cc(sizeof(uid_t) == sizeof(gid_t));
return get_process_id(pid, "Gid:", gid);
}
return -errno;
}
+ (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
+
while ((c = fgetc(f)) != EOF) {
if (!GREEDY_REALLOC(outcome, allocated, sz + 5))
return -ENOMEM;
} else
outcome[sz] = '\0';
- *env = outcome;
- outcome = NULL;
+ *env = TAKE_PTR(outcome);
return 0;
}
assert(pid >= 0);
assert(_ppid);
- if (pid == 0) {
+ if (pid == 0 || pid == getpid_cached()) {
*_ppid = getppid();
return 0;
}
* A warning is emitted if the process terminates abnormally,
* and also if it returns non-zero unless check_exit_code is true.
*/
-int wait_for_terminate_and_warn(const char *name, pid_t pid, bool check_exit_code) {
- int r;
+int wait_for_terminate_and_check(const char *name, pid_t pid, WaitFlags flags) {
+ _cleanup_free_ char *buffer = NULL;
siginfo_t status;
+ int r, prio;
- assert(name);
assert(pid > 1);
+ if (!name) {
+ r = get_process_comm(pid, &buffer);
+ if (r < 0)
+ log_debug_errno(r, "Failed to acquire process name of " PID_FMT ", ignoring: %m", pid);
+ else
+ name = buffer;
+ }
+
+ prio = flags & WAIT_LOG_ABNORMAL ? LOG_ERR : LOG_DEBUG;
+
r = wait_for_terminate(pid, &status);
if (r < 0)
- return log_warning_errno(r, "Failed to wait for %s: %m", name);
+ return log_full_errno(prio, r, "Failed to wait for %s: %m", strna(name));
if (status.si_code == CLD_EXITED) {
- if (status.si_status != 0)
- log_full(check_exit_code ? LOG_WARNING : LOG_DEBUG,
- "%s failed with error code %i.", name, status.si_status);
+ if (status.si_status != EXIT_SUCCESS)
+ log_full(flags & WAIT_LOG_NON_ZERO_EXIT_STATUS ? LOG_ERR : LOG_DEBUG,
+ "%s failed with exit status %i.", strna(name), status.si_status);
else
log_debug("%s succeeded.", name);
return status.si_status;
- } else if (status.si_code == CLD_KILLED ||
- status.si_code == CLD_DUMPED) {
- log_warning("%s terminated by signal %s.", name, signal_to_string(status.si_status));
+ } else if (IN_SET(status.si_code, CLD_KILLED, CLD_DUMPED)) {
+
+ log_full(prio, "%s terminated by signal %s.", strna(name), signal_to_string(status.si_status));
return -EPROTO;
}
- log_warning("%s failed due to unknown reason.", name);
+ log_full(prio, "%s failed due to unknown reason.", strna(name));
+ return -EPROTO;
+}
+
+/*
+ * Return values:
+ *
+ * < 0 : wait_for_terminate_with_timeout() failed to get the state of the process, the process timed out, the process
+ * was terminated by a signal, or failed for an unknown reason.
+ *
+ * >=0 : The process terminated normally with no failures.
+ *
+ * Success is indicated by a return value of zero, a timeout is indicated by ETIMEDOUT, and all other child failure
+ * states are indicated by error is indicated by a non-zero value.
+ *
+ * This call assumes SIGCHLD has been blocked already, in particular before the child to wait for has been forked off
+ * to remain entirely race-free.
+ */
+int wait_for_terminate_with_timeout(pid_t pid, usec_t timeout) {
+ sigset_t mask;
+ int r;
+ usec_t until;
+
+ assert_se(sigemptyset(&mask) == 0);
+ assert_se(sigaddset(&mask, SIGCHLD) == 0);
+
+ /* Drop into a sigtimewait-based timeout. Waiting for the
+ * pid to exit. */
+ until = now(CLOCK_MONOTONIC) + timeout;
+ for (;;) {
+ usec_t n;
+ siginfo_t status = {};
+ struct timespec ts;
+
+ n = now(CLOCK_MONOTONIC);
+ if (n >= until)
+ break;
+
+ r = sigtimedwait(&mask, NULL, timespec_store(&ts, until - n)) < 0 ? -errno : 0;
+ /* Assuming we woke due to the child exiting. */
+ if (waitid(P_PID, pid, &status, WEXITED|WNOHANG) == 0) {
+ if (status.si_pid == pid) {
+ /* This is the correct child.*/
+ if (status.si_code == CLD_EXITED)
+ return (status.si_status == 0) ? 0 : -EPROTO;
+ else
+ return -EPROTO;
+ }
+ }
+ /* Not the child, check for errors and proceed appropriately */
+ if (r < 0) {
+ switch (r) {
+ case -EAGAIN:
+ /* Timed out, child is likely hung. */
+ return -ETIMEDOUT;
+ case -EINTR:
+ /* Received a different signal and should retry */
+ continue;
+ default:
+ /* Return any unexpected errors */
+ return r;
+ }
+ }
+ }
+
return -EPROTO;
}
}
void sigkill_waitp(pid_t *pid) {
+ PROTECT_ERRNO;
+
if (!pid)
return;
if (*pid <= 1)
sigkill_wait(*pid);
}
+#endif // 0
+
+void sigterm_wait(pid_t pid) {
+ assert(pid > 1);
+
+ if (kill_and_sigcont(pid, SIGTERM) > 0)
+ (void) wait_for_terminate(pid, NULL);
+}
int kill_and_sigcont(pid_t pid, int sig) {
int r;
/* If this worked, also send SIGCONT, unless we already just sent a SIGCONT, or SIGKILL was sent which isn't
* affected by a process being suspended anyway. */
- if (r >= 0 && !IN_SET(SIGCONT, SIGKILL))
+ if (r >= 0 && !IN_SET(sig, SIGCONT, SIGKILL))
(void) kill(pid, SIGCONT);
return r;
}
-#endif // 0
-int getenv_for_pid(pid_t pid, const char *field, char **_value) {
+int getenv_for_pid(pid_t pid, const char *field, char **ret) {
_cleanup_fclose_ FILE *f = NULL;
char *value = NULL;
- int r;
bool done = false;
- size_t l;
const char *path;
+ size_t l;
assert(pid >= 0);
assert(field);
- assert(_value);
+ assert(ret);
+
+ if (pid == 0 || pid == getpid_cached()) {
+ const char *e;
+
+ e = getenv(field);
+ if (!e) {
+ *ret = NULL;
+ return 0;
+ }
+
+ value = strdup(e);
+ if (!value)
+ return -ENOMEM;
+
+ *ret = value;
+ return 1;
+ }
path = procfs_file_alloca(pid, "environ");
if (!f) {
if (errno == ENOENT)
return -ESRCH;
+
return -errno;
}
+ (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
+
l = strlen(field);
- r = 0;
do {
char line[LINE_MAX];
- unsigned i;
+ size_t i;
for (i = 0; i < sizeof(line)-1; i++) {
int c;
}
line[i] = 0;
- if (memcmp(line, field, l) == 0 && line[l] == '=') {
+ if (strneq(line, field, l) && line[l] == '=') {
value = strdup(line + l + 1);
if (!value)
return -ENOMEM;
- r = 1;
- break;
+ *ret = value;
+ return 1;
}
} while (!done);
- *_value = value;
- return r;
+ *ret = NULL;
+ return 0;
}
bool pid_is_unwaited(pid_t pid) {
if (pid <= 1) /* If we or PID 1 would be dead and have been waited for, this code would not be running */
return true;
+ if (pid == getpid_cached())
+ return true;
+
if (kill(pid, 0) >= 0)
return true;
if (pid <= 1) /* If we or PID 1 would be a zombie, this code would not be running */
return true;
+ if (pid == getpid_cached())
+ return true;
+
r = get_process_state(pid);
- if (r == -ESRCH || r == 'Z')
+ if (IN_SET(r, -ESRCH, 'Z'))
return false;
return true;
const char *root;
if (pid < 0)
- return 0;
+ return false;
+
+ if (pid == 0 || pid == getpid_cached())
+ return true;
root = procfs_file_alloca(pid, "root");
- return files_same(root, "/proc/1/root");
+ return files_same(root, "/proc/1/root", 0);
}
#endif // 0
static thread_local int cached = 0;
if (_unlikely_(cached == 0))
- cached = getpid() == gettid() ? 1 : -1;
+ cached = getpid_cached() == gettid() ? 1 : -1;
return cached > 0;
}
#if 0 /// UNNEEDED by elogind
-noreturn void freeze(void) {
+_noreturn_ void freeze(void) {
log_close();
sync();
+ /* Let's not freeze right away, but keep reaping zombies. */
+ for (;;) {
+ int r;
+ siginfo_t si = {};
+
+ r = waitid(P_ALL, 0, &si, WEXITED);
+ if (r < 0 && errno != EINTR)
+ break;
+ }
+
+ /* waitid() failed with an unexpected error, things are really borked. Freeze now! */
for (;;)
pause();
}
return architecture_to_string(architecture);
}
+int safe_personality(unsigned long p) {
+ int ret;
+
+ /* So here's the deal, personality() is weirdly defined by glibc. In some cases it returns a failure via errno,
+ * and in others as negative return value containing an errno-like value. Let's work around this: this is a
+ * wrapper that uses errno if it is set, and uses the return value otherwise. And then it sets both errno and
+ * the return value indicating the same issue, so that we are definitely on the safe side.
+ *
+ * See https://github.com/systemd/systemd/issues/6737 */
+
+ errno = 0;
+ ret = personality(p);
+ if (ret < 0) {
+ if (errno != 0)
+ return -errno;
+
+ errno = -ret;
+ }
+
+ return ret;
+}
+
+int opinionated_personality(unsigned long *ret) {
+ int current;
+
+ /* Returns the current personality, or PERSONALITY_INVALID if we can't determine it. This function is a bit
+ * opinionated though, and ignores all the finer-grained bits and exotic personalities, only distinguishing the
+ * two most relevant personalities: PER_LINUX and PER_LINUX32. */
+
+ current = safe_personality(PERSONALITY_INVALID);
+ if (current < 0)
+ return current;
+
+ if (((unsigned long) current & 0xffff) == PER_LINUX32)
+ *ret = PER_LINUX32;
+ else
+ *ret = PER_LINUX;
+
+ return 0;
+}
+
void valgrind_summary_hack(void) {
-#ifdef HAVE_VALGRIND_VALGRIND_H
- if (getpid() == 1 && RUNNING_ON_VALGRIND) {
+#if HAVE_VALGRIND_VALGRIND_H
+ if (getpid_cached() == 1 && RUNNING_ON_VALGRIND) {
pid_t pid;
- pid = raw_clone(SIGCHLD, NULL);
+ pid = raw_clone(SIGCHLD);
if (pid < 0)
log_emergency_errno(errno, "Failed to fork off valgrind helper: %m");
else if (pid == 0)
return 0;
}
+int ioprio_parse_priority(const char *s, int *ret) {
+ int i, r;
+
+ assert(s);
+ assert(ret);
+
+ r = safe_atoi(s, &i);
+ if (r < 0)
+ return r;
+
+ if (!ioprio_priority_is_valid(i))
+ return -EINVAL;
+
+ *ret = i;
+ return 0;
+}
+#endif // 0
+
+/* The cached PID, possible values:
+ *
+ * == UNSET [0] → cache not initialized yet
+ * == BUSY [-1] → some thread is initializing it at the moment
+ * any other → the cached PID
+ */
+
+#define CACHED_PID_UNSET ((pid_t) 0)
+#define CACHED_PID_BUSY ((pid_t) -1)
+
+static pid_t cached_pid = CACHED_PID_UNSET;
+
+void reset_cached_pid(void) {
+ /* Invoked in the child after a fork(), i.e. at the first moment the PID changed */
+ cached_pid = CACHED_PID_UNSET;
+}
+
+/* We use glibc __register_atfork() + __dso_handle directly here, as they are not included in the glibc
+ * headers. __register_atfork() is mostly equivalent to pthread_atfork(), but doesn't require us to link against
+ * libpthread, as it is part of glibc anyway. */
+#ifdef __GLIBC__
+extern int __register_atfork(void (*prepare) (void), void (*parent) (void), void (*child) (void), void * __dso_handle);
+extern void* __dso_handle __attribute__ ((__weak__));
+#endif // ifdef __GLIBC__
+
+pid_t getpid_cached(void) {
+ static bool installed = false;
+ pid_t current_value;
+
+ /* getpid_cached() is much like getpid(), but caches the value in local memory, to avoid having to invoke a
+ * system call each time. This restores glibc behaviour from before 2.24, when getpid() was unconditionally
+ * cached. Starting with 2.24 getpid() started to become prohibitively expensive when used for detecting when
+ * objects were used across fork()s. With this caching the old behaviour is somewhat restored.
+ *
+ * https://bugzilla.redhat.com/show_bug.cgi?id=1443976
+ * https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c579f48edba88380635ab98cb612030e3ed8691e
+ */
+
+ current_value = __sync_val_compare_and_swap(&cached_pid, CACHED_PID_UNSET, CACHED_PID_BUSY);
+
+ switch (current_value) {
+
+ case CACHED_PID_UNSET: { /* Not initialized yet, then do so now */
+ pid_t new_pid;
+
+ new_pid = raw_getpid();
+
+ if (!installed) {
+ /* __register_atfork() either returns 0 or -ENOMEM, in its glibc implementation. Since it's
+ * only half-documented (glibc doesn't document it but LSB does — though only superficially)
+ * we'll check for errors only in the most generic fashion possible. */
+
+ if (__register_atfork(NULL, NULL, reset_cached_pid, __dso_handle) != 0) {
+ /* OOM? Let's try again later */
+ cached_pid = CACHED_PID_UNSET;
+ return new_pid;
+ }
+
+ installed = true;
+ }
+
+ cached_pid = new_pid;
+ return new_pid;
+ }
+
+ case CACHED_PID_BUSY: /* Somebody else is currently initializing */
+ return raw_getpid();
+
+ default: /* Properly initialized */
+ return current_value;
+ }
+}
+
+int must_be_root(void) {
+
+ if (geteuid() == 0)
+ return 0;
+
+ log_error("Need to be root.");
+ return -EPERM;
+}
+
+int safe_fork_full(
+ const char *name,
+ const int except_fds[],
+ size_t n_except_fds,
+ ForkFlags flags,
+ pid_t *ret_pid) {
+
+ pid_t original_pid, pid;
+ sigset_t saved_ss, ss;
+ bool block_signals = false;
+ int prio, r;
+
+ /* A wrapper around fork(), that does a couple of important initializations in addition to mere forking. Always
+ * returns the child's PID in *ret_pid. Returns == 0 in the child, and > 0 in the parent. */
+
+ prio = flags & FORK_LOG ? LOG_ERR : LOG_DEBUG;
+
+ original_pid = getpid_cached();
+
+ if (flags & (FORK_RESET_SIGNALS|FORK_DEATHSIG)) {
+
+ /* We temporarily block all signals, so that the new child has them blocked initially. This way, we can
+ * be sure that SIGTERMs are not lost we might send to the child. */
+
+ if (sigfillset(&ss) < 0)
+ return log_full_errno(prio, errno, "Failed to reset signal set: %m");
+
+ block_signals = true;
+
+ } else if (flags & FORK_WAIT) {
+
+ /* Let's block SIGCHLD at least, so that we can safely watch for the child process */
+
+ if (sigemptyset(&ss) < 0)
+ return log_full_errno(prio, errno, "Failed to clear signal set: %m");
+
+ if (sigaddset(&ss, SIGCHLD) < 0)
+ return log_full_errno(prio, errno, "Failed to add SIGCHLD to signal set: %m");
+
+ block_signals = true;
+ }
+
+ if (block_signals)
+ if (sigprocmask(SIG_SETMASK, &ss, &saved_ss) < 0)
+ return log_full_errno(prio, errno, "Failed to set signal mask: %m");
+
+ if (flags & FORK_NEW_MOUNTNS)
+ pid = raw_clone(SIGCHLD|CLONE_NEWNS);
+ else
+ pid = fork();
+ if (pid < 0) {
+ r = -errno;
+
+ if (block_signals) /* undo what we did above */
+ (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
+
+ return log_full_errno(prio, r, "Failed to fork: %m");
+ }
+ if (pid > 0) {
+ /* We are in the parent process */
+
+ log_debug("Successfully forked off '%s' as PID " PID_FMT ".", strna(name), pid);
+
+ if (flags & FORK_WAIT) {
+ r = wait_for_terminate_and_check(name, pid, (flags & FORK_LOG ? WAIT_LOG : 0));
+ if (r < 0)
+ return r;
+ if (r != EXIT_SUCCESS) /* exit status > 0 should be treated as failure, too */
+ return -EPROTO;
+ }
+
+ if (block_signals) /* undo what we did above */
+ (void) sigprocmask(SIG_SETMASK, &saved_ss, NULL);
+
+ if (ret_pid)
+ *ret_pid = pid;
+
+ return 1;
+ }
+
+ /* We are in the child process */
+
+ if (flags & FORK_REOPEN_LOG) {
+ /* Close the logs if requested, before we log anything. And make sure we reopen it if needed. */
+ log_close();
+ log_set_open_when_needed(true);
+ }
+
+ if (name) {
+ r = rename_process(name);
+ if (r < 0)
+ log_full_errno(flags & FORK_LOG ? LOG_WARNING : LOG_DEBUG,
+ r, "Failed to rename process, ignoring: %m");
+ }
+
+ if (flags & FORK_DEATHSIG)
+ if (prctl(PR_SET_PDEATHSIG, SIGTERM) < 0) {
+ log_full_errno(prio, errno, "Failed to set death signal: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ if (flags & FORK_RESET_SIGNALS) {
+ r = reset_all_signal_handlers();
+ if (r < 0) {
+ log_full_errno(prio, r, "Failed to reset signal handlers: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ /* This implicitly undoes the signal mask stuff we did before the fork()ing above */
+ r = reset_signal_mask();
+ if (r < 0) {
+ log_full_errno(prio, r, "Failed to reset signal mask: %m");
+ _exit(EXIT_FAILURE);
+ }
+ } else if (block_signals) { /* undo what we did above */
+ if (sigprocmask(SIG_SETMASK, &saved_ss, NULL) < 0) {
+ log_full_errno(prio, errno, "Failed to restore signal mask: %m");
+ _exit(EXIT_FAILURE);
+ }
+ }
+
+ if (flags & FORK_DEATHSIG) {
+ pid_t ppid;
+ /* Let's see if the parent PID is still the one we started from? If not, then the parent
+ * already died by the time we set PR_SET_PDEATHSIG, hence let's emulate the effect */
+
+ ppid = getppid();
+ if (ppid == 0)
+ /* Parent is in a differn't PID namespace. */;
+ else if (ppid != original_pid) {
+ log_debug("Parent died early, raising SIGTERM.");
+ (void) raise(SIGTERM);
+ _exit(EXIT_FAILURE);
+ }
+ }
+
+ if (FLAGS_SET(flags, FORK_NEW_MOUNTNS | FORK_MOUNTNS_SLAVE)) {
+
+ /* Optionally, make sure we never propagate mounts to the host. */
+
+ if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
+ log_full_errno(prio, errno, "Failed to remount root directory as MS_SLAVE: %m");
+ _exit(EXIT_FAILURE);
+ }
+ }
+
+ if (flags & FORK_CLOSE_ALL_FDS) {
+ /* Close the logs here in case it got reopened above, as close_all_fds() would close them for us */
+ log_close();
+
+ r = close_all_fds(except_fds, n_except_fds);
+ if (r < 0) {
+ log_full_errno(prio, r, "Failed to close all file descriptors: %m");
+ _exit(EXIT_FAILURE);
+ }
+ }
+
+ /* When we were asked to reopen the logs, do so again now */
+ if (flags & FORK_REOPEN_LOG) {
+ log_open();
+ log_set_open_when_needed(false);
+ }
+
+ if (flags & FORK_NULL_STDIO) {
+ r = make_null_stdio();
+ if (r < 0) {
+ log_full_errno(prio, r, "Failed to connect stdin/stdout to /dev/null: %m");
+ _exit(EXIT_FAILURE);
+ }
+ }
+
+ if (ret_pid)
+ *ret_pid = getpid_cached();
+
+ return 0;
+}
+
+int fork_agent(const char *name, const int except[], size_t n_except, pid_t *ret_pid, const char *path, ...) {
+ bool stdout_is_tty, stderr_is_tty;
+ size_t n, i;
+ va_list ap;
+ char **l;
+ int r;
+
+ assert(path);
+
+ /* Spawns a temporary TTY agent, making sure it goes away when we go away */
+
+ r = safe_fork_full(name, except, n_except, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS, ret_pid);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ return 0;
+
+ /* In the child: */
+
+ stdout_is_tty = isatty(STDOUT_FILENO);
+ stderr_is_tty = isatty(STDERR_FILENO);
+
+ if (!stdout_is_tty || !stderr_is_tty) {
+ int fd;
+
+ /* Detach from stdout/stderr. and reopen
+ * /dev/tty for them. This is important to
+ * ensure that when systemctl is started via
+ * popen() or a similar call that expects to
+ * read EOF we actually do generate EOF and
+ * not delay this indefinitely by because we
+ * keep an unused copy of stdin around. */
+ fd = open("/dev/tty", O_WRONLY);
+ if (fd < 0) {
+ log_error_errno(errno, "Failed to open /dev/tty: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ if (!stdout_is_tty && dup2(fd, STDOUT_FILENO) < 0) {
+ log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ if (!stderr_is_tty && dup2(fd, STDERR_FILENO) < 0) {
+ log_error_errno(errno, "Failed to dup2 /dev/tty: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ safe_close_above_stdio(fd);
+ }
+
+ /* Count arguments */
+ va_start(ap, path);
+ for (n = 0; va_arg(ap, char*); n++)
+ ;
+ va_end(ap);
+
+ /* Allocate strv */
+ l = newa(char*, n + 1);
+
+ /* Fill in arguments */
+ va_start(ap, path);
+ for (i = 0; i <= n; i++)
+ l[i] = va_arg(ap, char*);
+ va_end(ap);
+
+ execv(path, l);
+ _exit(EXIT_FAILURE);
+}
+
+int set_oom_score_adjust(int value) {
+ char t[DECIMAL_STR_MAX(int)];
+
+ sprintf(t, "%i", value);
+
+ return write_string_file("/proc/self/oom_score_adj", t,
+ WRITE_STRING_FILE_VERIFY_ON_FAILURE|WRITE_STRING_FILE_DISABLE_BUFFER);
+}
+
+#if 0 /// UNNEEDED by elogind
static const char *const ioprio_class_table[] = {
[IOPRIO_CLASS_NONE] = "none",
[IOPRIO_CLASS_RT] = "realtime",
[IOPRIO_CLASS_IDLE] = "idle"
};
-DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, INT_MAX);
+DEFINE_STRING_TABLE_LOOKUP_WITH_FALLBACK(ioprio_class, int, IOPRIO_N_CLASSES);
static const char *const sigchld_code_table[] = {
[CLD_EXITED] = "exited",