[PATCH] Close a race between exit1() and SIGSTOP

[John Baldwin]

On Friday, September 07, 2012 12:35:52 pm Garrett Cooper wrote:
> Does this case continue to function normally?
> 
> $ sh -c 'while :; do sleep 1; done'
> ^Z
> [1]+  Stopped                 sh -c 'while :; do sleep 1; done'
> $ kill %1
> $ fg
> bash: fg: job has terminated
> [1]+  Terminated: 15          sh -c 'while :; do sleep 1; done'
> 
> $ sh -c 'while :; do sleep 1; done'
> ^Z
> [1]+  Stopped                 sh -c 'while :; do sleep 1; done'
> $ kill %1
> $ kill %1
> bash: kill: (3522) - No such process
> [1]+  Terminated: 15          sh -c 'while :; do sleep 1; done'
> 
> In particular, a single kill results in the signal being sent after
> the process wakes up, and a double kill results in the immediate death
> of the process (in this case a shell job) (in part because of the
> signals /bin/sh masks).

This still works in bash:

> bash
$ sh -c 'while :; do sleep 1; done'
^Z
[1]+  Stopped                 sh -c 'while :; do sleep 1; done'
$ kill %1
$ fg
bash: fg: job has terminated
[1]+  Terminated: 15          sh -c 'while :; do sleep 1; done'
$ sh -c 'while :; do sleep 1; done'
^Z
[1]+  Stopped                 sh -c 'while :; do sleep 1; done'
$ kill %1
$ kill %1
bash: kill: (61009) - No such process
[1]+  Terminated: 15          sh -c 'while :; do sleep 1; done'

I get different behavior in tcsh:

> sh -c 'while :; do sleep 1; done'
^Z
Suspended
> kill %1
> 
[1]    Terminated                    sh -c while :; do sleep 1; done

Note that this patch merey inhibits SIGSTOP from having an effect while
a process is exiting, it does not affect SIGTERM at all, so I don't think
it could have any effect on this case anyway.

-- 
John Baldwin