along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
-#include <assert.h>
#include <string.h>
#include <unistd.h>
#include <errno.h>
#include <stdlib.h>
#include <signal.h>
+#include <libintl.h>
+#include <locale.h>
#include <stdio.h>
#include <syslog.h>
#include <sched.h>
#include <linux/tiocl.h>
#include <termios.h>
#include <stdarg.h>
-#include <sys/inotify.h>
-#include <sys/poll.h>
+#include <poll.h>
#include <ctype.h>
#include <sys/prctl.h>
#include <sys/utsname.h>
#include <pwd.h>
#include <netinet/ip.h>
#include <linux/kd.h>
-#include <dlfcn.h>
#include <sys/wait.h>
#include <sys/time.h>
#include <glob.h>
#include <langinfo.h>
#include <locale.h>
#include <sys/personality.h>
+#include <sys/xattr.h>
+#include <sys/statvfs.h>
+#include <sys/file.h>
+#include <linux/fs.h>
+
+/* When we include libgen.h because we need dirname() we immediately
+ * undefine basename() since libgen.h defines it as a macro to the XDG
+ * version which is really broken. */
#include <libgen.h>
#undef basename
#include <sys/auxv.h>
#endif
+#include "config.h"
#include "macro.h"
#include "util.h"
#include "ioprio.h"
#include "missing.h"
#include "log.h"
#include "strv.h"
-#include "label.h"
#include "mkdir.h"
#include "path-util.h"
#include "exit-status.h"
#include "gunicode.h"
#include "virt.h"
#include "def.h"
+#include "sparse-endian.h"
int saved_argc = 0;
char **saved_argv = NULL;
int safe_atod(const char *s, double *ret_d) {
char *x = NULL;
double d = 0;
+ locale_t loc;
assert(s);
assert(ret_d);
- RUN_WITH_LOCALE(LC_NUMERIC_MASK, "C") {
- errno = 0;
- d = strtod(s, &x);
- }
+ loc = newlocale(LC_NUMERIC_MASK, "C", (locale_t) 0);
+ if (loc == (locale_t) 0)
+ return -errno;
- if (!x || x == s || *x || errno)
+ errno = 0;
+ d = strtod_l(s, &x, loc);
+
+ if (!x || x == s || *x || errno) {
+ freelocale(loc);
return errno ? -errno : -EINVAL;
+ }
+ freelocale(loc);
*ret_d = (double) d;
return 0;
}
}
char *file_in_same_dir(const char *path, const char *filename) {
- char *e, *r;
+ char *e, *ret;
size_t k;
assert(path);
if (path_is_absolute(filename))
return strdup(filename);
- if (!(e = strrchr(path, '/')))
+ e = strrchr(path, '/');
+ if (!e)
return strdup(filename);
k = strlen(filename);
- if (!(r = new(char, e-path+1+k+1)))
+ ret = new(char, (e + 1 - path) + k + 1);
+ if (!ret)
return NULL;
- memcpy(r, path, e-path+1);
- memcpy(r+(e-path)+1, filename, k+1);
-
- return r;
+ memcpy(mempcpy(ret, path, e + 1 - path), filename, k + 1);
+ return ret;
}
int rmdir_parents(const char *path, const char *stop) {
memcpy(r, prefix, pl);
for (f = s, t = r + pl; f < s + length; f++) {
+ size_t remaining = s + length - f;
+ assert(remaining > 0);
- if (*f != '\\') {
+ if (*f != '\\') { /* a literal literal */
*(t++) = *f;
continue;
}
+ if (--remaining == 0) { /* copy trailing backslash verbatim */
+ *(t++) = *f;
+ break;
+ }
+
f++;
switch (*f) {
case 'x': {
/* hexadecimal encoding */
- int a, b;
+ int a = -1, b = -1;
- a = unhexchar(f[1]);
- b = unhexchar(f[2]);
+ if (remaining >= 2) {
+ a = unhexchar(f[1]);
+ b = unhexchar(f[2]);
+ }
if (a < 0 || b < 0 || (a == 0 && b == 0)) {
/* Invalid escape code, let's take it literal then */
case '6':
case '7': {
/* octal encoding */
- int a, b, c;
+ int a = -1, b = -1, c = -1;
- a = unoctchar(f[0]);
- b = unoctchar(f[1]);
- c = unoctchar(f[2]);
+ if (remaining >= 3) {
+ a = unoctchar(f[0]);
+ b = unoctchar(f[1]);
+ c = unoctchar(f[2]);
+ }
if (a < 0 || b < 0 || c < 0 || (a == 0 && b == 0 && c == 0)) {
/* Invalid escape code, let's take it literal then */
break;
}
- case 0:
- /* premature end of string. */
- *(t++) = '\\';
- goto finish;
-
default:
/* Invalid escape code, let's take it literal then */
*(t++) = '\\';
}
}
-finish:
*t = 0;
return r;
}
return t;
}
-_pure_ static bool ignore_file_allow_backup(const char *filename) {
+_pure_ static bool hidden_file_allow_backup(const char *filename) {
assert(filename);
return
endswith(filename, ".dpkg-old") ||
endswith(filename, ".dpkg-new") ||
endswith(filename, ".dpkg-tmp") ||
+ endswith(filename, ".dpkg-dist") ||
+ endswith(filename, ".dpkg-bak") ||
+ endswith(filename, ".dpkg-backup") ||
+ endswith(filename, ".dpkg-remove") ||
endswith(filename, ".swp");
}
-bool ignore_file(const char *filename) {
+bool hidden_file(const char *filename) {
assert(filename);
if (endswith(filename, "~"))
return true;
- return ignore_file_allow_backup(filename);
+ return hidden_file_allow_backup(filename);
}
int fd_nonblock(int fd, bool nonblock) {
while ((de = readdir(d))) {
int fd = -1;
- if (ignore_file(de->d_name))
+ if (hidden_file(de->d_name))
continue;
if (safe_atoi(de->d_name, &fd) < 0)
bool fstype_is_network(const char *fstype) {
static const char table[] =
+ "afs\0"
"cifs\0"
"smbfs\0"
"sshfs\0"
assert(notify >= 0);
for (;;) {
- uint8_t buffer[INOTIFY_EVENT_MAX] _alignas_(struct inotify_event);
+ union inotify_event_buffer buffer;
struct inotify_event *e;
ssize_t l;
}
}
- l = read(notify, buffer, sizeof(buffer));
+ l = read(notify, &buffer, sizeof(buffer));
if (l < 0) {
if (errno == EINTR || errno == EAGAIN)
continue;
return n;
}
+int loop_read_exact(int fd, void *buf, size_t nbytes, bool do_poll) {
+ ssize_t n;
+
+ n = loop_read(fd, buf, nbytes, do_poll);
+ if (n < 0)
+ return n;
+ if ((size_t) n != nbytes)
+ return -EIO;
+ return 0;
+}
+
int loop_write(int fd, const void *buf, size_t nbytes, bool do_poll) {
const uint8_t *p = buf;
if (!de)
return 1;
- if (!ignore_file(de->d_name))
+ if (!hidden_file(de->d_name))
return 0;
}
}
int dev_urandom(void *p, size_t n) {
static int have_syscall = -1;
- int r, fd;
- ssize_t k;
+
+ _cleanup_close_ int fd = -1;
+ int r;
/* Gathers some randomness from the kernel. This call will
* never block, and will always return some data from the
return -errno;
} else
/* too short read? */
- return -EIO;
+ return -ENODATA;
}
fd = open("/dev/urandom", O_RDONLY|O_CLOEXEC|O_NOCTTY);
if (fd < 0)
return errno == ENOENT ? -ENOSYS : -errno;
- k = loop_read(fd, p, n, true);
- safe_close(fd);
-
- if (k < 0)
- return (int) k;
- if ((size_t) k != n)
- return -EIO;
-
- return 0;
+ return loop_read_exact(fd, p, n, true);
}
void initialize_srand(void) {
return lookup_uid(getuid());
}
-int getttyname_malloc(int fd, char **r) {
- char path[PATH_MAX], *c;
- int k;
+int getttyname_malloc(int fd, char **ret) {
+ size_t l = 100;
+ int r;
- assert(r);
+ assert(fd >= 0);
+ assert(ret);
- k = ttyname_r(fd, path, sizeof(path));
- if (k > 0)
- return -k;
+ for (;;) {
+ char path[l];
- char_array_0(path);
+ r = ttyname_r(fd, path, sizeof(path));
+ if (r == 0) {
+ const char *p;
+ char *c;
- c = strdup(startswith(path, "/dev/") ? path + 5 : path);
- if (!c)
- return -ENOMEM;
+ p = startswith(path, "/dev/");
+ c = strdup(p ?: path);
+ if (!c)
+ return -ENOMEM;
+
+ *ret = c;
+ return 0;
+ }
+
+ if (r != ERANGE)
+ return -r;
+
+ l *= 2;
+ }
- *r = c;
return 0;
}
c = 80;
cached_columns = c;
- return c;
+ return cached_columns;
}
int fd_lines(int fd) {
unsigned lines(void) {
const char *e;
- unsigned l;
+ int l;
if (_likely_(cached_lines > 0))
return cached_lines;
l = 0;
e = getenv("LINES");
if (e)
- (void) safe_atou(e, &l);
+ (void) safe_atoi(e, &l);
if (l <= 0)
l = fd_lines(STDOUT_FILENO);
const char *default_term_for_tty(const char *tty) {
assert(tty);
- return tty_is_vc_resolve(tty) ? "TERM=linux" : "TERM=vt102";
+ return tty_is_vc_resolve(tty) ? "TERM=linux" : "TERM=vt220";
}
bool dirent_is_file(const struct dirent *de) {
assert(de);
- if (ignore_file(de->d_name))
+ if (hidden_file(de->d_name))
return false;
if (de->d_type != DT_REG &&
de->d_type != DT_UNKNOWN)
return false;
- if (ignore_file_allow_backup(de->d_name))
+ if (hidden_file_allow_backup(de->d_name))
return false;
return endswith(de->d_name, suffix);
}
-void execute_directory(const char *directory, DIR *d, usec_t timeout, char *argv[]) {
- pid_t executor_pid;
- int r;
-
- assert(directory);
+static int do_execute(char **directories, usec_t timeout, char *argv[]) {
+ _cleanup_hashmap_free_free_ Hashmap *pids = NULL;
+ _cleanup_set_free_free_ Set *seen = NULL;
+ char **directory;
- /* Executes all binaries in a directory in parallel and waits
- * for them to finish. Optionally a timeout is applied. */
+ /* We fork this all off from a child process so that we can
+ * somewhat cleanly make use of SIGALRM to set a time limit */
- executor_pid = fork();
- if (executor_pid < 0) {
- log_error_errno(errno, "Failed to fork: %m");
- return;
+ reset_all_signal_handlers();
+ reset_signal_mask();
- } else if (executor_pid == 0) {
- _cleanup_hashmap_free_free_ Hashmap *pids = NULL;
- _cleanup_closedir_ DIR *_d = NULL;
- struct dirent *de;
+ assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0);
- /* We fork this all off from a child process so that
- * we can somewhat cleanly make use of SIGALRM to set
- * a time limit */
+ pids = hashmap_new(NULL);
+ if (!pids)
+ return log_oom();
- reset_all_signal_handlers();
- reset_signal_mask();
+ seen = set_new(&string_hash_ops);
+ if (!seen)
+ return log_oom();
- assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0);
+ STRV_FOREACH(directory, directories) {
+ _cleanup_closedir_ DIR *d;
+ struct dirent *de;
+ d = opendir(*directory);
if (!d) {
- d = _d = opendir(directory);
- if (!d) {
- if (errno == ENOENT)
- _exit(EXIT_SUCCESS);
-
- log_error_errno(errno, "Failed to enumerate directory %s: %m", directory);
- _exit(EXIT_FAILURE);
- }
- }
+ if (errno == ENOENT)
+ continue;
- pids = hashmap_new(NULL);
- if (!pids) {
- log_oom();
- _exit(EXIT_FAILURE);
+ return log_error_errno(errno, "Failed to open directory %s: %m", *directory);
}
FOREACH_DIRENT(de, d, break) {
_cleanup_free_ char *path = NULL;
pid_t pid;
+ int r;
if (!dirent_is_file(de))
continue;
- path = strjoin(directory, "/", de->d_name, NULL);
- if (!path) {
- log_oom();
- _exit(EXIT_FAILURE);
+ if (set_contains(seen, de->d_name)) {
+ log_debug("%1$s/%2$s skipped (%2$s was already seen).", *directory, de->d_name);
+ continue;
}
+ r = set_put_strdup(seen, de->d_name);
+ if (r < 0)
+ return log_oom();
+
+ path = strjoin(*directory, "/", de->d_name, NULL);
+ if (!path)
+ return log_oom();
+
+ if (null_or_empty_path(path)) {
+ log_debug("%s is empty (a mask).", path);
+ continue;
+ } else
+ log_debug("%s will be executed.", path);
+
pid = fork();
if (pid < 0) {
log_error_errno(errno, "Failed to fork: %m");
argv[0] = path;
execv(path, argv);
- log_error_errno(errno, "Failed to execute %s: %m", path);
- _exit(EXIT_FAILURE);
+ return log_error_errno(errno, "Failed to execute %s: %m", path);
}
log_debug("Spawned %s as " PID_FMT ".", path, pid);
r = hashmap_put(pids, UINT_TO_PTR(pid), path);
- if (r < 0) {
- log_oom();
- _exit(EXIT_FAILURE);
- }
-
+ if (r < 0)
+ return log_oom();
path = NULL;
}
+ }
- /* Abort execution of this process after the
- * timout. We simply rely on SIGALRM as default action
- * terminating the process, and turn on alarm(). */
+ /* Abort execution of this process after the timout. We simply
+ * rely on SIGALRM as default action terminating the process,
+ * and turn on alarm(). */
- if (timeout != USEC_INFINITY)
- alarm((timeout + USEC_PER_SEC - 1) / USEC_PER_SEC);
+ if (timeout != USEC_INFINITY)
+ alarm((timeout + USEC_PER_SEC - 1) / USEC_PER_SEC);
- while (!hashmap_isempty(pids)) {
- _cleanup_free_ char *path = NULL;
- pid_t pid;
+ while (!hashmap_isempty(pids)) {
+ _cleanup_free_ char *path = NULL;
+ pid_t pid;
- pid = PTR_TO_UINT(hashmap_first_key(pids));
- assert(pid > 0);
+ pid = PTR_TO_UINT(hashmap_first_key(pids));
+ assert(pid > 0);
- path = hashmap_remove(pids, UINT_TO_PTR(pid));
- assert(path);
+ path = hashmap_remove(pids, UINT_TO_PTR(pid));
+ assert(path);
- wait_for_terminate_and_warn(path, pid, true);
- }
+ wait_for_terminate_and_warn(path, pid, true);
+ }
- _exit(EXIT_SUCCESS);
+ return 0;
+}
+
+void execute_directories(const char* const* directories, usec_t timeout, char *argv[]) {
+ pid_t executor_pid;
+ int r;
+ char *name;
+ char **dirs = (char**) directories;
+
+ assert(!strv_isempty(dirs));
+
+ name = basename(dirs[0]);
+ assert(!isempty(name));
+
+ /* Executes all binaries in the directories in parallel and waits
+ * for them to finish. Optionally a timeout is applied. If a file
+ * with the same name exists in more than one directory, the
+ * earliest one wins. */
+
+ executor_pid = fork();
+ if (executor_pid < 0) {
+ log_error_errno(errno, "Failed to fork: %m");
+ return;
+
+ } else if (executor_pid == 0) {
+ r = do_execute(dirs, timeout, argv);
+ _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
}
- wait_for_terminate_and_warn(directory, executor_pid, true);
+ wait_for_terminate_and_warn(name, executor_pid, true);
}
int kill_and_sigcont(pid_t pid, int sig) {
if (isempty(s))
return false;
+ /* Doesn't accept empty hostnames, hostnames with trailing or
+ * leading dots, and hostnames with multiple dots in a
+ * sequence. Also ensures that the length stays below
+ * HOST_NAME_MAX. */
+
for (p = s, dot = true; *p; p++) {
if (*p == '.') {
if (dot)
{ "K", 1024ULL },
};
+ if (t == (off_t) -1)
+ return NULL;
+
for (i = 0; i < ELEMENTSOF(table); i++) {
if (t >= table[i].factor) {
return r;
}
-bool is_valid_documentation_url(const char *url) {
- assert(url);
+bool http_etag_is_valid(const char *etag) {
+ if (isempty(etag))
+ return false;
- if (startswith(url, "http://") && url[7])
- return true;
+ if (!endswith(etag, "\""))
+ return false;
- if (startswith(url, "https://") && url[8])
- return true;
+ if (!startswith(etag, "\"") && !startswith(etag, "W/\""))
+ return false;
- if (startswith(url, "file:") && url[5])
- return true;
+ return true;
+}
- if (startswith(url, "info:") && url[5])
- return true;
+bool http_url_is_valid(const char *url) {
+ const char *p;
+
+ if (isempty(url))
+ return false;
- if (startswith(url, "man:") && url[4])
+ p = startswith(url, "http://");
+ if (!p)
+ p = startswith(url, "https://");
+ if (!p)
+ return false;
+
+ if (isempty(p))
+ return false;
+
+ return ascii_is_valid(p);
+}
+
+bool documentation_url_is_valid(const char *url) {
+ const char *p;
+
+ if (isempty(url))
+ return false;
+
+ if (http_url_is_valid(url))
return true;
- return false;
+ p = startswith(url, "file:/");
+ if (!p)
+ p = startswith(url, "info:");
+ if (!p)
+ p = startswith(url, "man:");
+
+ if (isempty(p))
+ return false;
+
+ return ascii_is_valid(p);
}
bool in_initrd(void) {
return NULL;
}
+void init_gettext(void) {
+ setlocale(LC_ALL, "");
+ textdomain(GETTEXT_PACKAGE);
+}
+
bool is_locale_utf8(void) {
const char *set;
static int cached_answer = -1;
d = opendir("/sys/class/power_supply");
if (!d)
- return -errno;
+ return errno == ENOENT ? true : -errno;
for (;;) {
struct dirent *de;
if (!de)
break;
- if (ignore_file(de->d_name))
+ if (hidden_file(de->d_name))
continue;
device = openat(dirfd(d), de->d_name, O_DIRECTORY|O_RDONLY|O_CLOEXEC|O_NOCTTY);
assert(machine);
assert(pid);
- p = strappenda("/run/systemd/machines/", machine);
+ p = strjoina("/run/systemd/machines/", machine);
r = parse_env_file(p, NEWLINE, "LEADER", &s, "CLASS", &class, NULL);
if (r == -ENOENT)
return -EHOSTDOWN;
if (isempty(s)) {
free(s);
- return -ENOTSUP;
+ return -EOPNOTSUPP;
}
*ret = s;
#endif
/* Fall back to unguessable name + unlinking */
- p = strappenda(path, "/systemd-tmp-XXXXXX");
+ p = strjoina(path, "/systemd-tmp-XXXXXX");
fd = mkostemp_safe(p, flags);
if (fd < 0)
return (uint64_t) mem * (uint64_t) page_size();
}
-char* mount_test_option(const char *haystack, const char *needle) {
-
- struct mntent me = {
- .mnt_opts = (char*) haystack
- };
-
- assert(needle);
-
- /* Like glibc's hasmntopt(), but works on a string, not a
- * struct mntent */
-
- if (!haystack)
- return NULL;
-
- return hasmntopt(&me, needle);
-}
-
void hexdump(FILE *f, const void *p, size_t s) {
const uint8_t *b = p;
unsigned n = 0;
return r ? r : n;
}
+static int get_mount_flags(const char *path, unsigned long *flags) {
+ struct statvfs buf;
+
+ if (statvfs(path, &buf) < 0)
+ return -errno;
+ *flags = buf.f_flag;
+ return 0;
+}
+
int bind_remount_recursive(const char *prefix, bool ro) {
_cleanup_set_free_free_ Set *done = NULL;
_cleanup_free_ char *cleaned = NULL;
_cleanup_set_free_free_ Set *todo = NULL;
bool top_autofs = false;
char *x;
+ unsigned long orig_flags;
todo = set_new(&string_hash_ops);
if (!todo)
if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
return -errno;
- if (mount(NULL, prefix, NULL, MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
+ orig_flags = 0;
+ (void) get_mount_flags(cleaned, &orig_flags);
+ orig_flags &= ~MS_RDONLY;
+
+ if (mount(NULL, prefix, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0)
return -errno;
x = strdup(cleaned);
if (r < 0)
return r;
- if (mount(NULL, x, NULL, MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0) {
+ /* Try to reuse the original flag set, but
+ * don't care for errors, in case of
+ * obstructed mounts */
+ orig_flags = 0;
+ (void) get_mount_flags(x, &orig_flags);
+ orig_flags &= ~MS_RDONLY;
+
+ if (mount(NULL, x, NULL, orig_flags|MS_BIND|MS_REMOUNT|(ro ? MS_RDONLY : 0), NULL) < 0) {
/* Deal with mount points that are
* obstructed by a later mount */
* /foo/bar/waldo
*
* Into this:
- * /foo/bar/.waldoXXXXXX
+ * /foo/bar/.#waldoXXXXXX
*/
fn = basename(p);
if (!filename_is_valid(fn))
return -EINVAL;
- t = new(char, strlen(p) + 1 + 6 + 1);
+ t = new(char, strlen(p) + 2 + 6 + 1);
if (!t)
return -ENOMEM;
- strcpy(stpcpy(stpcpy(mempcpy(t, p, fn - p), "."), fn), "XXXXXX");
+ strcpy(stpcpy(stpcpy(mempcpy(t, p, fn - p), ".#"), fn), "XXXXXX");
*ret = path_kill_slashes(t);
return 0;
* /foo/bar/waldo
*
* Into this:
- * /foo/bar/.waldobaa2a261115984a9
+ * /foo/bar/.#waldobaa2a261115984a9
*/
fn = basename(p);
if (!filename_is_valid(fn))
return -EINVAL;
- t = new(char, strlen(p) + 1 + 16 + 1);
+ t = new(char, strlen(p) + 2 + 16 + 1);
if (!t)
return -ENOMEM;
- x = stpcpy(stpcpy(mempcpy(t, p, fn - p), "."), fn);
+ x = stpcpy(stpcpy(mempcpy(t, p, fn - p), ".#"), fn);
u = random_u64();
for (i = 0; i < 16; i++) {
/* Turns this:
* /foo/bar/waldo
* Into this:
- * /foo/bar/waldo/.3c2b6219aa75d7d0
+ * /foo/bar/waldo/.#3c2b6219aa75d7d0
*/
- t = new(char, strlen(p) + 2 + 16 + 1);
+ t = new(char, strlen(p) + 3 + 16 + 1);
if (!t)
return -ENOMEM;
- x = stpcpy(stpcpy(t, p), "/.");
+ x = stpcpy(stpcpy(t, p), "/.#");
u = random_u64();
for (i = 0; i < 16; i++) {
* awfully racy, and thus we just won't do them. */
if (root)
- path = strappenda(root, "/etc/.pwd.lock");
+ path = strjoina(root, "/etc/.pwd.lock");
else
path = "/etc/.pwd.lock";
return 1;
}
+
+int ptsname_malloc(int fd, char **ret) {
+ size_t l = 100;
+
+ assert(fd >= 0);
+ assert(ret);
+
+ for (;;) {
+ char *c;
+
+ c = new(char, l);
+ if (!c)
+ return -ENOMEM;
+
+ if (ptsname_r(fd, c, l) == 0) {
+ *ret = c;
+ return 0;
+ }
+ if (errno != ERANGE) {
+ free(c);
+ return -errno;
+ }
+
+ free(c);
+ l *= 2;
+ }
+}
+
+int openpt_in_namespace(pid_t pid, int flags) {
+ _cleanup_close_ int pidnsfd = -1, mntnsfd = -1, rootfd = -1;
+ _cleanup_close_pair_ int pair[2] = { -1, -1 };
+ union {
+ struct cmsghdr cmsghdr;
+ uint8_t buf[CMSG_SPACE(sizeof(int))];
+ } control = {};
+ struct msghdr mh = {
+ .msg_control = &control,
+ .msg_controllen = sizeof(control),
+ };
+ struct cmsghdr *cmsg;
+ siginfo_t si;
+ pid_t child;
+ int r;
+
+ assert(pid > 0);
+
+ r = namespace_open(pid, &pidnsfd, &mntnsfd, NULL, &rootfd);
+ if (r < 0)
+ return r;
+
+ if (socketpair(AF_UNIX, SOCK_DGRAM, 0, pair) < 0)
+ return -errno;
+
+ child = fork();
+ if (child < 0)
+ return -errno;
+
+ if (child == 0) {
+ int master;
+
+ pair[0] = safe_close(pair[0]);
+
+ r = namespace_enter(pidnsfd, mntnsfd, -1, rootfd);
+ if (r < 0)
+ _exit(EXIT_FAILURE);
+
+ master = posix_openpt(flags);
+ if (master < 0)
+ _exit(EXIT_FAILURE);
+
+ cmsg = CMSG_FIRSTHDR(&mh);
+ cmsg->cmsg_level = SOL_SOCKET;
+ cmsg->cmsg_type = SCM_RIGHTS;
+ cmsg->cmsg_len = CMSG_LEN(sizeof(int));
+ memcpy(CMSG_DATA(cmsg), &master, sizeof(int));
+
+ mh.msg_controllen = cmsg->cmsg_len;
+
+ if (sendmsg(pair[1], &mh, MSG_NOSIGNAL) < 0)
+ _exit(EXIT_FAILURE);
+
+ _exit(EXIT_SUCCESS);
+ }
+
+ pair[1] = safe_close(pair[1]);
+
+ r = wait_for_terminate(child, &si);
+ if (r < 0)
+ return r;
+ if (si.si_code != CLD_EXITED || si.si_status != EXIT_SUCCESS)
+ return -EIO;
+
+ if (recvmsg(pair[0], &mh, MSG_NOSIGNAL|MSG_CMSG_CLOEXEC) < 0)
+ return -errno;
+
+ for (cmsg = CMSG_FIRSTHDR(&mh); cmsg; cmsg = CMSG_NXTHDR(&mh, cmsg))
+ if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
+ int *fds;
+ unsigned n_fds;
+
+ fds = (int*) CMSG_DATA(cmsg);
+ n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
+
+ if (n_fds != 1) {
+ close_many(fds, n_fds);
+ return -EIO;
+ }
+
+ return fds[0];
+ }
+
+ return -EIO;
+}
+
+ssize_t fgetxattrat_fake(int dirfd, const char *filename, const char *attribute, void *value, size_t size, int flags) {
+ _cleanup_close_ int fd = -1;
+ ssize_t l;
+
+ /* The kernel doesn't have a fgetxattrat() command, hence let's emulate one */
+
+ fd = openat(dirfd, filename, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOATIME|(flags & AT_SYMLINK_NOFOLLOW ? O_NOFOLLOW : 0));
+ if (fd < 0)
+ return -errno;
+
+ l = fgetxattr(fd, attribute, value, size);
+ if (l < 0)
+ return -errno;
+
+ return l;
+}
+
+static int parse_crtime(le64_t le, usec_t *usec) {
+ uint64_t u;
+
+ assert(usec);
+
+ u = le64toh(le);
+ if (u == 0 || u == (uint64_t) -1)
+ return -EIO;
+
+ *usec = (usec_t) u;
+ return 0;
+}
+
+int fd_getcrtime(int fd, usec_t *usec) {
+ le64_t le;
+ ssize_t n;
+
+ assert(fd >= 0);
+ assert(usec);
+
+ /* Until Linux gets a real concept of birthtime/creation time,
+ * let's fake one with xattrs */
+
+ n = fgetxattr(fd, "user.crtime_usec", &le, sizeof(le));
+ if (n < 0)
+ return -errno;
+ if (n != sizeof(le))
+ return -EIO;
+
+ return parse_crtime(le, usec);
+}
+
+int fd_getcrtime_at(int dirfd, const char *name, usec_t *usec, int flags) {
+ le64_t le;
+ ssize_t n;
+
+ n = fgetxattrat_fake(dirfd, name, "user.crtime_usec", &le, sizeof(le), flags);
+ if (n < 0)
+ return -errno;
+ if (n != sizeof(le))
+ return -EIO;
+
+ return parse_crtime(le, usec);
+}
+
+int path_getcrtime(const char *p, usec_t *usec) {
+ le64_t le;
+ ssize_t n;
+
+ assert(p);
+ assert(usec);
+
+ n = getxattr(p, "user.crtime_usec", &le, sizeof(le));
+ if (n < 0)
+ return -errno;
+ if (n != sizeof(le))
+ return -EIO;
+
+ return parse_crtime(le, usec);
+}
+
+int fd_setcrtime(int fd, usec_t usec) {
+ le64_t le;
+
+ assert(fd >= 0);
+
+ if (usec <= 0)
+ usec = now(CLOCK_REALTIME);
+
+ le = htole64((uint64_t) usec);
+ if (fsetxattr(fd, "user.crtime_usec", &le, sizeof(le), 0) < 0)
+ return -errno;
+
+ return 0;
+}
+
+int same_fd(int a, int b) {
+ struct stat sta, stb;
+ pid_t pid;
+ int r, fa, fb;
+
+ assert(a >= 0);
+ assert(b >= 0);
+
+ /* Compares two file descriptors. Note that semantics are
+ * quite different depending on whether we have kcmp() or we
+ * don't. If we have kcmp() this will only return true for
+ * dup()ed file descriptors, but not otherwise. If we don't
+ * have kcmp() this will also return true for two fds of the same
+ * file, created by separate open() calls. Since we use this
+ * call mostly for filtering out duplicates in the fd store
+ * this difference hopefully doesn't matter too much. */
+
+ if (a == b)
+ return true;
+
+ /* Try to use kcmp() if we have it. */
+ pid = getpid();
+ r = kcmp(pid, pid, KCMP_FILE, a, b);
+ if (r == 0)
+ return true;
+ if (r > 0)
+ return false;
+ if (errno != ENOSYS)
+ return -errno;
+
+ /* We don't have kcmp(), use fstat() instead. */
+ if (fstat(a, &sta) < 0)
+ return -errno;
+
+ if (fstat(b, &stb) < 0)
+ return -errno;
+
+ if ((sta.st_mode & S_IFMT) != (stb.st_mode & S_IFMT))
+ return false;
+
+ /* We consider all device fds different, since two device fds
+ * might refer to quite different device contexts even though
+ * they share the same inode and backing dev_t. */
+
+ if (S_ISCHR(sta.st_mode) || S_ISBLK(sta.st_mode))
+ return false;
+
+ if (sta.st_dev != stb.st_dev || sta.st_ino != stb.st_ino)
+ return false;
+
+ /* The fds refer to the same inode on disk, let's also check
+ * if they have the same fd flags. This is useful to
+ * distuingish the read and write side of a pipe created with
+ * pipe(). */
+ fa = fcntl(a, F_GETFL);
+ if (fa < 0)
+ return -errno;
+
+ fb = fcntl(b, F_GETFL);
+ if (fb < 0)
+ return -errno;
+
+ return fa == fb;
+}
+
+int chattr_fd(int fd, bool b, unsigned mask) {
+ unsigned old_attr, new_attr;
+
+ assert(fd >= 0);
+
+ if (mask == 0)
+ return 0;
+
+ if (ioctl(fd, FS_IOC_GETFLAGS, &old_attr) < 0)
+ return -errno;
+
+ if (b)
+ new_attr = old_attr | mask;
+ else
+ new_attr = old_attr & ~mask;
+
+ if (new_attr == old_attr)
+ return 0;
+
+ if (ioctl(fd, FS_IOC_SETFLAGS, &new_attr) < 0)
+ return -errno;
+
+ return 0;
+}
+
+int chattr_path(const char *p, bool b, unsigned mask) {
+ _cleanup_close_ int fd = -1;
+
+ assert(p);
+
+ if (mask == 0)
+ return 0;
+
+ fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
+ if (fd < 0)
+ return -errno;
+
+ return chattr_fd(fd, b, mask);
+}
+
+int read_attr_fd(int fd, unsigned *ret) {
+ assert(fd >= 0);
+
+ if (ioctl(fd, FS_IOC_GETFLAGS, ret) < 0)
+ return -errno;
+
+ return 0;
+}
+
+int read_attr_path(const char *p, unsigned *ret) {
+ _cleanup_close_ int fd = -1;
+
+ assert(p);
+ assert(ret);
+
+ fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
+ if (fd < 0)
+ return -errno;
+
+ return read_attr_fd(fd, ret);
+}
+
+int make_lock_file(const char *p, int operation, LockFile *ret) {
+ _cleanup_close_ int fd = -1;
+ _cleanup_free_ char *t = NULL;
+ int r;
+
+ /*
+ * We use UNPOSIX locks if they are available. They have nice
+ * semantics, and are mostly compatible with NFS. However,
+ * they are only available on new kernels. When we detect we
+ * are running on an older kernel, then we fall back to good
+ * old BSD locks. They also have nice semantics, but are
+ * slightly problematic on NFS, where they are upgraded to
+ * POSIX locks, even though locally they are orthogonal to
+ * POSIX locks.
+ */
+
+ t = strdup(p);
+ if (!t)
+ return -ENOMEM;
+
+ for (;;) {
+ struct flock fl = {
+ .l_type = (operation & ~LOCK_NB) == LOCK_EX ? F_WRLCK : F_RDLCK,
+ .l_whence = SEEK_SET,
+ };
+ struct stat st;
+
+ fd = open(p, O_CREAT|O_RDWR|O_NOFOLLOW|O_CLOEXEC|O_NOCTTY, 0600);
+ if (fd < 0)
+ return -errno;
+
+ r = fcntl(fd, (operation & LOCK_NB) ? F_OFD_SETLK : F_OFD_SETLKW, &fl);
+ if (r < 0) {
+
+ /* If the kernel is too old, use good old BSD locks */
+ if (errno == EINVAL)
+ r = flock(fd, operation);
+
+ if (r < 0)
+ return errno == EAGAIN ? -EBUSY : -errno;
+ }
+
+ /* If we acquired the lock, let's check if the file
+ * still exists in the file system. If not, then the
+ * previous exclusive owner removed it and then closed
+ * it. In such a case our acquired lock is worthless,
+ * hence try again. */
+
+ r = fstat(fd, &st);
+ if (r < 0)
+ return -errno;
+ if (st.st_nlink > 0)
+ break;
+
+ fd = safe_close(fd);
+ }
+
+ ret->path = t;
+ ret->fd = fd;
+ ret->operation = operation;
+
+ fd = -1;
+ t = NULL;
+
+ return r;
+}
+
+int make_lock_file_for(const char *p, int operation, LockFile *ret) {
+ const char *fn;
+ char *t;
+
+ assert(p);
+ assert(ret);
+
+ fn = basename(p);
+ if (!filename_is_valid(fn))
+ return -EINVAL;
+
+ t = newa(char, strlen(p) + 2 + 4 + 1);
+ stpcpy(stpcpy(stpcpy(mempcpy(t, p, fn - p), ".#"), fn), ".lck");
+
+ return make_lock_file(t, operation, ret);
+}
+
+void release_lock_file(LockFile *f) {
+ int r;
+
+ if (!f)
+ return;
+
+ if (f->path) {
+
+ /* If we are the exclusive owner we can safely delete
+ * the lock file itself. If we are not the exclusive
+ * owner, we can try becoming it. */
+
+ if (f->fd >= 0 &&
+ (f->operation & ~LOCK_NB) == LOCK_SH) {
+ static const struct flock fl = {
+ .l_type = F_WRLCK,
+ .l_whence = SEEK_SET,
+ };
+
+ r = fcntl(f->fd, F_OFD_SETLK, &fl);
+ if (r < 0 && errno == EINVAL)
+ r = flock(f->fd, LOCK_EX|LOCK_NB);
+
+ if (r >= 0)
+ f->operation = LOCK_EX|LOCK_NB;
+ }
+
+ if ((f->operation & ~LOCK_NB) == LOCK_EX)
+ unlink_noerrno(f->path);
+
+ free(f->path);
+ f->path = NULL;
+ }
+
+ f->fd = safe_close(f->fd);
+ f->operation = 0;
+}
+
+static size_t nul_length(const uint8_t *p, size_t sz) {
+ size_t n = 0;
+
+ while (sz > 0) {
+ if (*p != 0)
+ break;
+
+ n++;
+ p++;
+ sz--;
+ }
+
+ return n;
+}
+
+ssize_t sparse_write(int fd, const void *p, size_t sz, size_t run_length) {
+ const uint8_t *q, *w, *e;
+ ssize_t l;
+
+ q = w = p;
+ e = q + sz;
+ while (q < e) {
+ size_t n;
+
+ n = nul_length(q, e - q);
+
+ /* If there are more than the specified run length of
+ * NUL bytes, or if this is the beginning or the end
+ * of the buffer, then seek instead of write */
+ if ((n > run_length) ||
+ (n > 0 && q == p) ||
+ (n > 0 && q + n >= e)) {
+ if (q > w) {
+ l = write(fd, w, q - w);
+ if (l < 0)
+ return -errno;
+ if (l != q -w)
+ return -EIO;
+ }
+
+ if (lseek(fd, n, SEEK_CUR) == (off_t) -1)
+ return -errno;
+
+ q += n;
+ w = q;
+ } else if (n > 0)
+ q += n;
+ else
+ q ++;
+ }
+
+ if (q > w) {
+ l = write(fd, w, q - w);
+ if (l < 0)
+ return -errno;
+ if (l != q - w)
+ return -EIO;
+ }
+
+ return q - (const uint8_t*) p;
+}
+
+void sigkill_wait(pid_t *pid) {
+ if (!pid)
+ return;
+ if (*pid <= 1)
+ return;
+
+ if (kill(*pid, SIGKILL) > 0)
+ (void) wait_for_terminate(*pid, NULL);
+}
+
+int syslog_parse_priority(const char **p, int *priority, bool with_facility) {
+ int a = 0, b = 0, c = 0;
+ int k;
+
+ assert(p);
+ assert(*p);
+ assert(priority);
+
+ if ((*p)[0] != '<')
+ return 0;
+
+ if (!strchr(*p, '>'))
+ return 0;
+
+ if ((*p)[2] == '>') {
+ c = undecchar((*p)[1]);
+ k = 3;
+ } else if ((*p)[3] == '>') {
+ b = undecchar((*p)[1]);
+ c = undecchar((*p)[2]);
+ k = 4;
+ } else if ((*p)[4] == '>') {
+ a = undecchar((*p)[1]);
+ b = undecchar((*p)[2]);
+ c = undecchar((*p)[3]);
+ k = 5;
+ } else
+ return 0;
+
+ if (a < 0 || b < 0 || c < 0 ||
+ (!with_facility && (a || b || c > 7)))
+ return 0;
+
+ if (with_facility)
+ *priority = a*100 + b*10 + c;
+ else
+ *priority = (*priority & LOG_FACMASK) | c;
+
+ *p += k;
+ return 1;
+}
+
+ssize_t string_table_lookup(const char * const *table, size_t len, const char *key) {
+ size_t i;
+
+ if (!key)
+ return -1;
+
+ for (i = 0; i < len; ++i)
+ if (streq_ptr(table[i], key))
+ return (ssize_t)i;
+
+ return -1;
+}
+
+void cmsg_close_all(struct msghdr *mh) {
+ struct cmsghdr *cmsg;
+
+ assert(mh);
+
+ for (cmsg = CMSG_FIRSTHDR(mh); cmsg; cmsg = CMSG_NXTHDR(mh, cmsg))
+ if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS)
+ close_many((int*) CMSG_DATA(cmsg), (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int));
+}
+
+int rename_noreplace(int olddirfd, const char *oldpath, int newdirfd, const char *newpath) {
+ struct stat buf;
+ int ret;
+
+ ret = renameat2(olddirfd, oldpath, newdirfd, newpath, RENAME_NOREPLACE);
+ if (ret >= 0)
+ return 0;
+
+ /* Even though renameat2() exists since Linux 3.15, btrfs added
+ * support for it later. If it is not implemented, fallback to another
+ * method. */
+ if (errno != EINVAL)
+ return -errno;
+
+ /* The link()/unlink() fallback does not work on directories. But
+ * renameat() without RENAME_NOREPLACE gives the same semantics on
+ * directories, except when newpath is an *empty* directory. This is
+ * good enough. */
+ ret = fstatat(olddirfd, oldpath, &buf, AT_SYMLINK_NOFOLLOW);
+ if (ret >= 0 && S_ISDIR(buf.st_mode)) {
+ ret = renameat(olddirfd, oldpath, newdirfd, newpath);
+ return ret >= 0 ? 0 : -errno;
+ }
+
+ /* If it is not a directory, use the link()/unlink() fallback. */
+ ret = linkat(olddirfd, oldpath, newdirfd, newpath, 0);
+ if (ret < 0)
+ return -errno;
+
+ ret = unlinkat(olddirfd, oldpath, 0);
+ if (ret < 0) {
+ /* backup errno before the following unlinkat() alters it */
+ ret = errno;
+ (void) unlinkat(newdirfd, newpath, 0);
+ errno = ret;
+ return -errno;
+ }
+
+ return 0;
+}