|
 |
9fc0f6 |
From f04488d2842544de263e245ee1a21e789ec818a8 Mon Sep 17 00:00:00 2001
|
|
|
9fc0f6 |
From: Yuxuan Shui <yshuiv7@gmail.com>
|
|
|
9fc0f6 |
Date: Sat, 15 Feb 2014 02:38:50 +0800
|
|
|
9fc0f6 |
Subject: [PATCH] core: fix detection of dead processes
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
Commit 5ba6985b moves the UNIT_VTABLE(u)->sigchld_event before systemd
|
|
|
9fc0f6 |
actually reaps the zombie. Which leads to service_load_pid_file accepting
|
|
|
9fc0f6 |
zombie as a valid pid.
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
This fixes timeouts like:
|
|
|
9fc0f6 |
[ 2746.602243] systemd[1]: chronyd.service stop-sigterm timed out. Killing.
|
|
|
9fc0f6 |
[ 2836.852545] systemd[1]: chronyd.service still around after SIGKILL. Ignoring.
|
|
|
9fc0f6 |
[ 2927.102187] systemd[1]: chronyd.service stop-final-sigterm timed out. Killing.
|
|
|
9fc0f6 |
[ 3017.352560] systemd[1]: chronyd.service still around after final SIGKILL. Entering failed mode.
|
|
|
9fc0f6 |
---
|
|
|
9fc0f6 |
src/core/service.c | 8 ++++++++
|
|
|
9fc0f6 |
src/shared/util.c | 25 +++++++++++++++++++++++++
|
|
|
9fc0f6 |
src/shared/util.h | 1 +
|
|
|
9fc0f6 |
3 files changed, 34 insertions(+)
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
diff --git a/src/core/service.c b/src/core/service.c
|
|
|
9fc0f6 |
index dff5286..3eacf3b 100644
|
|
|
9fc0f6 |
--- a/src/core/service.c
|
|
|
9fc0f6 |
+++ b/src/core/service.c
|
|
|
9fc0f6 |
@@ -1429,6 +1429,14 @@ static int service_load_pid_file(Service *s, bool may_warn) {
|
|
|
9fc0f6 |
return -ESRCH;
|
|
|
9fc0f6 |
}
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
+ if (get_process_state(pid) == 'Z') {
|
|
|
9fc0f6 |
+ if (may_warn)
|
|
|
9fc0f6 |
+ log_info_unit(UNIT(s)->id,
|
|
|
9fc0f6 |
+ "PID "PID_FMT" read from file %s is a zombie.",
|
|
|
9fc0f6 |
+ pid, s->pid_file);
|
|
|
9fc0f6 |
+ return -ESRCH;
|
|
|
9fc0f6 |
+ }
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
if (s->main_pid_known) {
|
|
|
9fc0f6 |
if (pid == s->main_pid)
|
|
|
9fc0f6 |
return 0;
|
|
|
9fc0f6 |
diff --git a/src/shared/util.c b/src/shared/util.c
|
|
|
9fc0f6 |
index 2086847..090a204 100644
|
|
|
9fc0f6 |
--- a/src/shared/util.c
|
|
|
9fc0f6 |
+++ b/src/shared/util.c
|
|
|
9fc0f6 |
@@ -558,6 +558,31 @@ char *truncate_nl(char *s) {
|
|
|
9fc0f6 |
return s;
|
|
|
9fc0f6 |
}
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
+int get_process_state(pid_t pid) {
|
|
|
9fc0f6 |
+ const char *p;
|
|
|
9fc0f6 |
+ char state;
|
|
|
9fc0f6 |
+ int r;
|
|
|
9fc0f6 |
+ _cleanup_free_ char *line = NULL;
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
+ assert(pid >= 0);
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
+ p = procfs_file_alloca(pid, "stat");
|
|
|
9fc0f6 |
+ r = read_one_line_file(p, &line);
|
|
|
9fc0f6 |
+ if (r < 0)
|
|
|
9fc0f6 |
+ return r;
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
+ p = strrchr(line, ')');
|
|
|
9fc0f6 |
+ if (!p)
|
|
|
9fc0f6 |
+ return -EIO;
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
+ p++;
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
+ if (sscanf(p, " %c", &state) != 1)
|
|
|
9fc0f6 |
+ return -EIO;
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
+ return (unsigned char) state;
|
|
|
9fc0f6 |
+}
|
|
|
9fc0f6 |
+
|
|
|
9fc0f6 |
int get_process_comm(pid_t pid, char **name) {
|
|
|
9fc0f6 |
const char *p;
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
diff --git a/src/shared/util.h b/src/shared/util.h
|
|
|
9fc0f6 |
index 222abe0..1d60687 100644
|
|
|
9fc0f6 |
--- a/src/shared/util.h
|
|
|
9fc0f6 |
+++ b/src/shared/util.h
|
|
|
9fc0f6 |
@@ -215,6 +215,7 @@ char *file_in_same_dir(const char *path, const char *filename);
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
int rmdir_parents(const char *path, const char *stop);
|
|
|
9fc0f6 |
|
|
|
9fc0f6 |
+char get_process_state(pid_t pid);
|
|
|
9fc0f6 |
int get_process_comm(pid_t pid, char **name);
|
|
|
9fc0f6 |
int get_process_cmdline(pid_t pid, size_t max_length, bool comm_fallback, char **line);
|
|
|
9fc0f6 |
int get_process_exe(pid_t pid, char **name);
|