tree-wide: drop support for kernels without pidfd_open() and pidfd_send_signal()

Our baseline is v5.4 now.
This commit is contained in:
Mike Yuan
2025-01-08 13:14:23 +01:00
parent a342c71d36
commit 6e14c46bac
6 changed files with 35 additions and 83 deletions

View File

@@ -299,12 +299,7 @@ static inline long missing_get_mempolicy(int *mode, unsigned long *nodemask,
#if !HAVE_PIDFD_SEND_SIGNAL
static inline int missing_pidfd_send_signal(int fd, int sig, siginfo_t *info, unsigned flags) {
# ifdef __NR_pidfd_send_signal
return syscall(__NR_pidfd_send_signal, fd, sig, info, flags);
# else
errno = ENOSYS;
return -1;
# endif
}
# define pidfd_send_signal missing_pidfd_send_signal
@@ -312,12 +307,7 @@ static inline int missing_pidfd_send_signal(int fd, int sig, siginfo_t *info, un
#if !HAVE_PIDFD_OPEN
static inline int missing_pidfd_open(pid_t pid, unsigned flags) {
# ifdef __NR_pidfd_open
return syscall(__NR_pidfd_open, pid, flags);
# else
errno = ENOSYS;
return -1;
# endif
}
# define pidfd_open missing_pidfd_open

View File

@@ -24,12 +24,8 @@ static int pidfd_check_pidfs(void) {
return have_pidfs;
_cleanup_close_ int fd = pidfd_open(getpid_cached(), 0);
if (fd < 0) {
if (ERRNO_IS_NOT_SUPPORTED(errno))
return (have_pidfs = false);
if (fd < 0)
return -errno;
}
return (have_pidfs = fd_is_fs_type(fd, PID_FS_MAGIC));
}

View File

@@ -84,8 +84,8 @@ int pidref_set_pid(PidRef *pidref, pid_t pid) {
fd = pidfd_open(pid, 0);
if (fd < 0) {
/* Graceful fallback in case the kernel doesn't support pidfds or is out of fds */
if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno) && !ERRNO_IS_RESOURCE(errno))
/* Graceful fallback in case the kernel is out of fds */
if (!ERRNO_IS_RESOURCE(errno))
return log_debug_errno(errno, "Failed to open pidfd for pid " PID_FMT ": %m", pid);
fd = -EBADF;

View File

@@ -25,6 +25,7 @@
#include "missing_magic.h"
#include "missing_syscall.h"
#include "missing_threads.h"
#include "missing_wait.h"
#include "origin-id.h"
#include "path-util.h"
#include "prioq.h"
@@ -1074,6 +1075,8 @@ static void source_disconnect(sd_event_source *s) {
}
static sd_event_source* source_free(sd_event_source *s) {
int r;
assert(s);
source_disconnect(s);
@@ -1087,31 +1090,23 @@ static sd_event_source* source_free(sd_event_source *s) {
if (s->child.process_owned) {
if (!s->child.exited) {
bool sent = false;
if (s->child.pidfd >= 0) {
if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) {
if (errno == ESRCH) /* Already dead */
sent = true;
else if (!ERRNO_IS_NOT_SUPPORTED(errno))
log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m",
s->child.pid);
} else
sent = true;
}
if (!sent)
if (kill(s->child.pid, SIGKILL) < 0)
if (errno != ESRCH) /* Already dead */
log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m",
s->child.pid);
if (s->child.pidfd >= 0)
r = RET_NERRNO(pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0));
else
r = RET_NERRNO(kill(s->child.pid, SIGKILL));
if (r < 0 && r != -ESRCH)
log_debug_errno(r, "Failed to kill process " PID_FMT ", ignoring: %m",
s->child.pid);
}
if (!s->child.waited) {
siginfo_t si = {};
/* Reap the child if we can */
(void) waitid(P_PID, s->child.pid, &si, WEXITED);
if (s->child.pidfd >= 0)
(void) waitid(P_PIDFD, s->child.pidfd, &si, WEXITED);
else
(void) waitid(P_PID, s->child.pid, &si, WEXITED);
}
}
@@ -1578,11 +1573,6 @@ static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *us
return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata));
}
static bool shall_use_pidfd(void) {
/* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */
return secure_getenv_bool("SYSTEMD_PIDFD") != 0;
}
_public_ int sd_event_add_child(
sd_event *e,
sd_event_source **ret,
@@ -1630,34 +1620,29 @@ _public_ int sd_event_add_child(
if (!s)
return -ENOMEM;
/* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we
* pin the PID, and make regular waitid() handling race-free. */
s->child.pidfd = pidfd_open(pid, 0);
if (s->child.pidfd < 0)
return -errno;
s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */
s->wakeup = WAKEUP_EVENT_SOURCE;
s->child.options = options;
s->child.callback = callback;
s->userdata = userdata;
s->enabled = SD_EVENT_ONESHOT;
/* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we
* pin the PID, and make regular waitid() handling race-free. */
if (shall_use_pidfd()) {
s->child.pidfd = pidfd_open(pid, 0);
if (s->child.pidfd < 0) {
/* Propagate errors unless the syscall is not supported or blocked */
if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno))
return -errno;
} else
s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */
} else
s->child.pidfd = -EBADF;
if (EVENT_SOURCE_WATCH_PIDFD(s)) {
/* We have a pidfd and we only want to watch for exit */
/* We only want to watch for exit */
r = source_child_pidfd_register(s, s->enabled);
if (r < 0)
return r;
} else {
/* We have no pidfd or we shall wait for some other event than WEXITED */
/* We shall wait for some other event than WEXITED */
r = event_make_signal_data(e, SIGCHLD, NULL);
if (r < 0)
return r;
@@ -3239,12 +3224,10 @@ _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, cons
if (si)
copy = *si;
if (pidfd_send_signal(s->child.pidfd, sig, si ? &copy : NULL, 0) < 0) {
/* Let's propagate the error only if the system call is not implemented or prohibited */
if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno))
return -errno;
} else
return 0;
if (pidfd_send_signal(s->child.pidfd, sig, si ? &copy : NULL, 0) < 0)
return -errno;
return 0;
}
/* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse

View File

@@ -198,7 +198,7 @@ static int post_handler(sd_event_source *s, void *userdata) {
return 2;
}
static void test_basic_one(bool with_pidfd) {
TEST(basic) {
sd_event *e = NULL;
sd_event_source *w = NULL, *x = NULL, *y = NULL, *z = NULL, *q = NULL, *t = NULL;
static const char ch = 'x';
@@ -207,10 +207,6 @@ static void test_basic_one(bool with_pidfd) {
uint64_t event_now;
int64_t priority;
log_info("/* %s(pidfd=%s) */", __func__, yes_no(with_pidfd));
assert_se(setenv("SYSTEMD_PIDFD", yes_no(with_pidfd), 1) >= 0);
assert_se(pipe(a) >= 0);
assert_se(pipe(b) >= 0);
assert_se(pipe(d) >= 0);
@@ -301,13 +297,6 @@ static void test_basic_one(bool with_pidfd) {
safe_close_pair(b);
safe_close_pair(d);
safe_close_pair(k);
assert_se(unsetenv("SYSTEMD_PIDFD") >= 0);
}
TEST(basic) {
test_basic_one(true); /* test with pidfd */
test_basic_one(false); /* test without pidfd */
}
TEST(sd_event_now) {
@@ -583,13 +572,7 @@ TEST(pidfd) {
assert_se(pid > 1);
pidfd = pidfd_open(pid, 0);
if (pidfd < 0) {
/* No pidfd_open() supported or blocked? */
assert_se(ERRNO_IS_NOT_SUPPORTED(errno) || ERRNO_IS_PRIVILEGE(errno));
(void) wait_for_terminate(pid, NULL);
return;
}
ASSERT_OK(pidfd = pidfd_open(pid, 0));
pid2 = fork();
if (pid2 == 0)

View File

@@ -876,7 +876,7 @@ static int create_session_message(
if (!avoid_pidfd) {
pidfd = pidfd_open(getpid_cached(), 0);
if (pidfd < 0 && !ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno))
if (pidfd < 0)
return -errno;
}