SIGTERM unable to kill process

Question

I have a single-threaded process which is not dying on kill -TERM. The process signal mask does not show that SIGTERM is blocked. I'm executing 'kill' as root. I'm able to kill the process using SIGKILL but this is part of a larger system and I'd like SIGTERM to work.

$ cat /proc/5105/status
Name:   task_root.nginx
State:  S (sleeping)
Tgid:   5105
Pid:    5105
PPid:   1
TracerPid:  0
Uid:    1000    1000    1000    1000
Gid:    1000    1000    1000    1000
FDSize: 256
Groups: 4 24 27 30 46 109 124 1000 
VmPeak:  3304480 kB
VmSize:  3304472 kB
VmLck:         0 kB
VmPin:         0 kB
VmHWM:    249060 kB
VmRSS:       320 kB
VmData:  3228468 kB
VmStk:      1012 kB
VmExe:      3020 kB
VmLib:     30232 kB
VmPTE:      1076 kB
VmSwap:   248288 kB
Threads:    1
SigQ:   0/63014
SigPnd: 0000000000000000
ShdPnd: 0000000000000000
SigBlk: 0000000000000004
SigIgn: 0000000000000000
SigCgt: 2000000181001cef
CapInh: 0000000000000000
CapPrm: 0000000000200000
CapEff: 0000000000200000
CapBnd: ffffffffffffffff
Cpus_allowed:   ff
Cpus_allowed_list:  0-7
Mems_allowed:   00000000,00000001
Mems_allowed_list:  0
voluntary_ctxt_switches:    16
nonvoluntary_ctxt_switches: 1

Note the Sig* properties. SigCgt, SigIgn, and SigBlk indicate that SIGTERM is neither caught, ignored, or blocked (bit #15 is unset - counting least significant bit as #1). Since the default disposition for SIGTERM is to terminate the process, I'd expect it to get killed. But that does not happen -

$ sudo kill -TERM 5105

$ cat /proc/5105/status | grep Name
Name:   task_root.nginx

The process is blocked in the wait() system call, waiting for termination of a sub-process:

$ sudo cat /proc/5105/stack
[<ffffffff8106bfc4>] do_wait+0x1e4/0x260
[<ffffffff8106d230>] sys_wait4+0xa0/0xf0
[<ffffffff81668d02>] system_call_fastpath+0x16/0x1b
[<ffffffffffffffff>] 0xffffffffffffffff

The process is a single threaded process that was created using clone() with flags SIGCHLD | CLONE_NEWPID | CLONE_NEWNS. The PID 5105 is the top level PID as viewed from the parent (default) PID namespace.

Killing the process with SIGKILL works showing that PID or process tracking is not a problem.

Answer 1

Are you sure SIGTERM is signal 15 on your system? You didn't specify OS or architecture in the question, but it's very likely that you can use "kill -l" to list signals (that's a lowercase letter L, not numeral one). What happens if you strace the process, then kill -TERM it? For example:

strace -Tttf -o strace.out -p 5105 &
kill -TERM 5105

Are the signal masks changing? strace should also show if this is happening.