Porting on FreeBSD 53

[Seán C. Farley]

On Fri, 13 May 2005, Hervé Kergourlay wrote:

> Seán C. Farley a écrit :
>
>> On Thu, 12 May 2005, Seán C. Farley wrote:
>> 
>>> On Thu, 12 May 2005, Hervé Kergourlay wrote:
>>> 
>>> <snip>
>>> 
>>>> 4) wait() API
>>>> 
>>>> 2 problems, the first is a ECHILD error on a wait call after a fork
>>>> fork The code is generic for most of unix system. Is there any
>>>> specific problems to manage the fork and wait APIs ?  the second
>>>> problem with calls is a blocking wait() call in the same condition
>>>> but this time the son process is finished but the wait call in the
>>>> father stays blocked, again it's a generic Unix code
>>>> 
>>>> If there is no evidence, ask me for more informations
>>> 
>>> 
>>> The second problem sounds like what I am encountering
>>> (http://www.freebsd.org/cgi/query-pr.cgi?pr=77818) with zsh for my
>>> shell.  You did suspend (sigsuspend()) SIGCHLD before the fork?  By
>>> "fork fork", do you mean you fork twice?
>> 
>> 
>> Ah ha!  I see the problem that has been causing me this problem and
>> probably you too.  Signal suspensions (only these?) are not being
>> copied with a double fork().  Here is an example program[1] to
>> illustrate.  They do get copied on FreeBSD-4.10 and Linux.  I just do
>> not know if they are supposed to be copied.
>> 
>> Seán
>> 
>> P.S.  I included David since he has been trying to help me with this
>> bug.
>> 
>>   1. http://www.farley.org/freebsd/tmp/grandparent.c
>
>
> I test your sample
>
> it's working on Linux and FreeBSD 4.0 but failing on FreeBDS 5.2 et
> 5.3.  the process stays blocked on the suspend call
>
> I rewrite another sample with the same model, joined here, as we wrote
> it in our main program. It's working also on Linux but failing on all
> FreeBSD included FreeBSD 4.0, 5.2 et 5.3
>
> trace when working
>
> avant FORK : pid 71995.
> avant FORK : pid 71996.
> son pid 71997.
> Child calling execve(/bin/ls).
> sigw    sigw.c  sigw2   sigw2.c
> Grand parent pid 71996.
> Grand parent waiting for a signal.
> Parent pid 71997.
> Parent waiting for a signal.
> Parent get status 100000.
> Grand parent get status 100000.
>
> trace when failing
>
> avant FORK : pid 71995.
> avant FORK : pid 71996.
> son pid 71997.
> Child calling execve(/bin/ls).
> sigw    sigw.c  sigw2   sigw2.c
> Grand parent pid 71996.
> Grand parent waiting for a signal.
> Parent pid 71997.
> Parent waiting for a signal.
> wait returned No child processes.
> wait returned No child processes.
>
> the wait call return with an errno ECHILD ??

The children are exiting before the parents (due to sleep()'s) get to
their wait()'s.

> did you have any idea if the problem will be solve by the FreeBSD team
> or not ?

I updated my bug report and tried to notify David Xu but the e-mail
server rejected the e-mail (too many "Received" headers).  I am also
Cc'ing Greg Lehey since he ran into a possibly similar bug[1] in
February.

> or if there is a workarouund ?

Reapplying the sigsuspend() before each fork() may work.

Seán
   1. http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/77537
-- 
sean-freebsd at farley.org