On 3 Feb 1999 07:56:25 -0000, D. J. Bernstein wrote:
>This is a bug in your operating system.
Yes. This is where a small change in qmail could cause it to be more
robust, even on a less-than-perfect OS.
>On bug-free systems, the only way for qmail-rspawn to generate that
>message is for execve() to return an error that fails error_temp():
>normally ENOTDIR, ENAMETOOLONG, ENOENT, ELOOP, EACCES, ENOEXEC, E2BIG,
>or EFAULT. None of these can be caused by temporary failures; they are
>permanent (and quite serious) configuration errors.
Permanent(OS) and Permanent(mail) are 2 different things. I don't see
why e.g. ENOENT should cause the message to be bounced. It's not a
permanent problem with the message, and over the queue-life of the
message it's not a permanent config problem either. If the sysadmins
doesn't fix it, it becomes a (mail) permanent error when the message
times out in the queue.
>What happened to you, presumably, is that crt0.o tried to load a shared
>library, failed because it was out of memory, and incorrectly decided to
>exit with some arbitrary code, never mind the fact that exit codes have
>meanings. It should instead have terminated the process with SIGKILL.
More or less. Again, the error is permanence from the point of view of
the application, but shouldn't cause the message to bounce immediately.
-Sincerely, Fred
(Frederik Lindberg, Infectious Diseases, WashU, St. Louis, MO, USA)
