问题现象
大量僵尸进程
root 32278 0.0 0.0 0 0 ? Z 05:39 0:00 [runuser] <defunct>
root 32280 0.0 0.0 0 0 ? Z Aug22 0:00 [runuser] <defunct>
root 32304 0.0 0.0 0 0 ? Z Aug21 0:00 [runuser] <defunct>
root 32362 0.0 0.0 0 0 ? Z Aug22 0:00 [runuser] <defunct>
root 32448 0.0 0.0 0 0 ? Z Aug22 0:00 [runuser] <defunct>
root 32475 0.0 0.0 0 0 ? Z Aug20 0:00 [runuser] <defunct>
root 32498 0.0 0.0 0 0 ? Z Aug22 0:00 [runuser] <defunct>
root 32509 0.0 0.0 0 0 ? Z Aug18 0:00 [runuser] <defunct>
root 32551 0.0 0.0 0 0 ? Z 09:54 0:00 [runuser] <defunct>
root 32556 0.0 0.0 0 0 ? Z Aug18 0:00 [runuser] <defunct>
root 32557 0.0 0.0 0 0 ? Z Aug18 0:00 [runuser] <defunct>
root 32611 0.0 0.0 0 0 ? Z Aug22 0:00 [runuser] <defunct>
root 32612 0.0 0.0 0 0 ? Z Aug22 0:00 [runuser] <defunct>
root 32647 0.0 0.0 0 0 ? Z Aug19 0:00 [runuser] <defunct>
root 32699 0.0 0.0 0 0 ? Z Aug18 0:00 [runuser] <defunct>
root 32700 0.0 0.0 0 0 ? Z Aug18 0:00 [runuser] <defunct>
root 32716 0.0 0.0 0 0 ? Z Aug19 0:00 [runuser] <defunct>
root 32753 0.0 0.0 0 0 ? Z Aug18 0:00 [runuser] <defunct>
systemd crash
#pstack 1
#0 0x00007fba436a4a2d in pause () from /lib64/libpthread.so.0
#1 0x00007fba44c66ecd in freeze ()
#2 0x00007fba44c14eef in crash.2986 ()
#3 <signal handler called>
#4 0x00007fba433095f7 in raise () from /lib64/libc.so.6
#5 0x00007fba4330ace8 in abort () from /lib64/libc.so.6
#6 0x00007fba44c85c72 in log_assert_failed ()
#7 0x00007fba44c4635d in unit_watch_pid ()
#8 0x00007fba44be3767 in service_spawn.9373 ()
#9 0x00007fba44c5ecab in service_start.9527 ()
#10 0x00007fba44bdc1ba in manager_dispatch_run_queue.3886 ()
#11 0x00007fba44c37db2 in source_dispatch.52334.3984 ()
#12 0x00007fba44c3adba in sd_event_dispatch ()
#13 0x00007fba44c0c9ef in manager_loop ()
#14 0x00007fba44bcb86c in main ()
(gdb)
(gdb) bt
#0 0x00007fba436a4a2d in pause () from /lib64/libpthread.so.0
#1 0x00007fba44c66ecd in freeze () at src/shared/util.c:3833
#2 0x00007fba44c14eef in crash.2986 (sig=<optimized out>) at src/core/main.c:223
#3 <signal handler called>
#4 0x00007fba433095f7 in raise () from /lib64/libc.so.6
#5 0x00007fba4330ace8 in abort () from /lib64/libc.so.6
#6 0x00007fba44c85c72 in log_assert_failed (text=<optimized out>, file=<optimized out>, line=<optimized out>, func=<optimized out>) at src/shared/log.c:753
#7 0x00007fba44c4635d in unit_watch_pid (u=0x7fba44e44250, pid=0) at src/core/unit.c:2017
#8 0x00007fba44be3767 in service_spawn.9373 (s=0x7fba44e44250, c=0x7fba68f07ad0, timeout=90000000, pass_fds=<optimized out>, apply_permissions=<optimized out>, apply_chroot=<optimized out>,
apply_tty_stdin=true, is_control=true, _pid=0x7fba44e4486c) at src/core/service.c:1170
#9 0x00007fba44c5ecab in service_enter_start_pre (s=0x7fba44e44250) at src/core/service.c:1596
#10 service_start.9527 (u=0x7fba44e44250) at src/core/service.c:1841
#11 0x00007fba44bdc1ba in manager_dispatch_run_queue.3886 (source=<optimized out>, userdata=0x7fba44d132e0) at src/core/unit.c:1481
#12 0x00007fba44c37db2 in source_dispatch (s=s@entry=0x7fba44d11ea0) at src/libsystemd/sd-event/sd-event.c:2155
#13 0x00007fba44c3adba in sd_event_dispatch (e=0x7fba44d11c30) at src/libsystemd/sd-event/sd-event.c:2472
#14 0x00007fba44c0c9ef in sd_event_run (timeout=18446744073709551615, e=0x7fba44d11c30) at src/libsystemd/sd-event/sd-event.c:2501
#15 manager_loop (m=0x7fba44d132e0) at src/core/manager.c:2070
#16 0x00007fba44bcb86c in main (argc=5, argv=0x7ffc45ee3a38) at src/core/main.c:1755
(gdb)
问题代码定位
(gdb) list unit_watch_pid
2008
2009 unit_add_to_dbus_queue(u);
2010 unit_add_to_gc_queue(u);
2011 }
2012
2013 int unit_watch_pid(Unit *u, pid_t pid) {
2014 int q, r;
2015
2016 assert(u);
2017 assert(pid >= 1); # 这里触发
(gdb)
(gdb) list src/core/service.c:1170
return r;
static int service_load_pid_file(Service *s, bool may_warn) {
1165 s->exec_runtime,
1166 &pid);
1167 if (r < 0)
1168 goto fail;
1169
1170 r = unit_watch_pid(UNIT(s), pid); # 这里调用
1171 if (r < 0)
1172 /* FIXME: we need to do something here */
1173 goto fail;
1174
(gdb) list src/core/service.c:1596
static void service_enter_start_pre(Service *s) {
1591 * be left from previous runs. */
1592 service_kill_control_processes(s);
1593
1594 s->control_command_id = SERVICE_EXEC_START_PRE;
1595
1596 r = service_spawn(s,
1597 s->control_command,
1598 s->timeout_start_usec,
1599 false,
1600 !s->permissions_start_only
复现方法
- 创建一个vm中测试
- 想办法耗尽内存
# mount -t tmpfs -o size=20G tmpfs /mnt
# dd if=/dev/zero of=/mnt/file bs=1M
- while :; do systemctl restart iptables.service; sleep 5; done
分析
问题发生在当系统内存耗尽，恰好在此时，有个systemd的service重启，此时，systemd需要分配一个pid，但是因为内存耗尽，无法分配pid。然后触发了systemd一个bug：
BUG存在于: 创建pid函数exec_spawn(), 当创建pid失败的时候，理应返回错误值，但是这里返回的是0。 这个0是r决定的。 这个r是r = exec_context_load_environment() 函数的返回值，然而这个函数永远都是返回成功，导致，pid < 0 pid创建失败后，依然返回0
创建pid函数exec_spawn():
int exec_spawn
pid = fork();
if (pid < 0)
return log_unit_error_errno(params->unit_id, r, "Failed to fork: %m");
service重启，systemd会为service创建一个pid，在pid<0 的情况下，执行到unit_watch_pid() 函数，unit_watch_pid()函数判断assert(pid >= 1);
service_spawn()函数-> 调用exec_spawn()
重启service 会调用service_spawn()函数
static int service_spawn(...
r = exec_spawn(c,
&s->exec_context,
&exec_params,
s->exec_runtime,
&pid);
if (r < 0)
goto fail;
r = unit_watch_pid(UNIT(s), pid);
if (r < 0)
/* FIXME: we need to do something here */
goto fail;
*_pid = pid;
int unit_watch_pid(Unit *u, pid_t pid) {
int q, r;
assert(u);
assert(pid >= 1);
REF
patch:
https://github.com/lnykryn/systemd-rhel/pull/119/commits/e18a5fa348aae0800807c3121a1ccf750eca206f
https://github.com/systemd/systemd/pull/5886
solution: