1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
|
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<meta http-equiv="Content-Language" content="en" />
<title>execline: exit codes</title>
<meta name="Description" content="execline: exit codes" />
<meta name="Keywords" content="execline exit codes" />
<!-- <link rel="stylesheet" type="text/css" href="http://skarnet.org/default.css" /> -->
</head>
<body>
<p>
<a href="index.html">execline</a><br />
<a href="http://skarnet.org/software/">Software</a><br />
<a href="http://skarnet.org/">skarnet.org</a>
</p>
<h1> How to propagate exit codes up a process dynasty </h1>
<p>
Say we have a parent process <em>P</em>, child of a grandparent process
<em>G</em>, spawning a child process <em>C</em> and waiting for it.
Either <em>C</em> dies normally with an exit code from 0 to 255, or it is
killed by a signal.
How can we make sure that <em>P</em> reports to <em>G</em> what happened
to <em>C</em>, with as much precision as possible ?
</p>
<p>
The problem is, there's more information in a wstat (the
structure filled in by
<a href="http://pubs.opengroup.org/onlinepubs/9699919799/functions/waitpid.html">waitpid()</a>)
than a process can report by
simply exiting. <em>P</em> could exit with the same exit code as <em>C</em>,
but then what should it do if C has been killed by a signal ?
</p>
<p>
An idea is to have <em>P</em> kill itself with the same signal that killed
<em>C</em>.
But that's actually not right, because <em>P</em> itself could be killed by a
signal from another source, and G needs that information. "<em>P</em> has been
killed by a signal" and "<em>C</em> has been killed by a signal" are two
different informations, so they should not be reported in the same way.
</p>
<p>
So, any way you look at it, there is always more information than we
can report.
</p>
<p>
Shells have their own
<a href="http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18_08_02">convention</a>
for reporting crashes, but since any exit code greater than 127 is reported
as is, the information given by the shell is unreliable: "child exited 129"
and "child was killed by SIGHUP" are indistinguishable. When shells get
nested, all bets are off - the information conveyed by exit codes becomes
devoid of meaning pretty fast. We need something better.
</p>
<h2> execline's solution </h2>
<p>
execline commands such as <a href="forx.html">forx</a>, that can report
a child's exit code, proceed that way when they're in the position of
<em>P</em>:
</p>
<ul>
<li> If <em>C</em> was killed by a signal: <em>P</em> exits 128 plus the
signal number. </li>
<li> If <em>C</em> exited 128 or more: <em>P</em> exits 128. </li>
<li> Else, <em>P</em> exits with the same code as <em>C</em>. </li>
</ul>
<p>
Rationale:
</p>
<ul>
<li> 128+ exit codes are extremely rare and should report really
problematic conditions; commands usually exit 127 or less. If <em>C</em>
exits 128+, it's more important to convey the information
"something really bad happened, but the <em>C</em> process itself was not
killed by a signal" than the exact nature of the event. </li>
<li> Commands following that convention can be nested. If <em>P</em> exits
129+, <em>G</em> knows that <em>C</em> was
killed by a signal. If <em>G</em> also needs to report that to its parent,
it will exit 128: <em>G</em>'s parent will not know the signal number, but
it will know that <em>P</em> reported 128 or more, so either <em>C</em> or
a scion of <em>C</em> had problems. </li>
<li> Exact information is reported in the common case. </li>
</ul>
</body>
</html>
|
