runsvdir problem + patch

runsvdir problem + patch

[Eric Lammerts]

Hi,
I found a problem with runsvdir: If you create two subdirectories in 
/service within the same second, runsvdir doesn't always see the 2nd 
one.

For example:
- create /service/foo
- runsvdir sees mtime of /service has changed, runs runsv for foo
- create /service/bar

If these three things happen within the same second, the creation of 
/service/bar doesn't change the mtime of /service, and therefore 
runsvdir won't rescan it.

This patch fixes the problem by forcing a check on the next iteration 
after a service was started.

Eric

--- runit-1.0.3/src/runsvdir.c.orig	2004-07-22 19:36:35.000000000 -0400
+++ runit-1.0.3/src/runsvdir.c	2004-07-22 19:37:19.000000000 -0400
@@ -117,6 +117,7 @@
         sv[i].isgone =0;
         svnum++;
         runsv(i, d->d_name);
+      check =1;
       }
     }
     if (errno) {

Re: runsvdir problem + patch

[Gerrit Pape]

On Thu, Jul 22, 2004 at 10:15:48PM -0400, Eric Lammerts wrote:
> I found a problem with runsvdir: If you create two subdirectories in 
> /service within the same second, runsvdir doesn't always see the 2nd 
> one.
> 
> For example:
> - create /service/foo
> - runsvdir sees mtime of /service has changed, runs runsv for foo
> - create /service/bar
> 
> If these three things happen within the same second, the creation of 
> /service/bar doesn't change the mtime of /service, and therefore 
> runsvdir won't rescan it.

Yes, I know about this for some time, but actually was hit by the
problem just once, on a system where the service directory was located
on a ramdisk, and the subdirectories were created in background after
stage 1 finished.

Did you actually run into this problem, or found it by reading the
source?

> This patch fixes the problem by forcing a check on the next iteration 
> after a service was started.

The patch looks fine to me at the first glance, thanks for posting it.

Regards, Gerrit.

> --- runit-1.0.3/src/runsvdir.c.orig	2004-07-22 19:36:35.000000000 -0400
> +++ runit-1.0.3/src/runsvdir.c	2004-07-22 19:37:19.000000000 -0400
> @@ -117,6 +117,7 @@
>         sv[i].isgone =0;
>         svnum++;
>         runsv(i, d->d_name);
> +      check =1;
>       }
>     }
>     if (errno) {

Re: runsvdir problem + patch

[Eric Lammerts]

Gerrit Pape wrote:
>On Thu, Jul 22, 2004 at 10:15:48PM -0400, Eric Lammerts wrote:
>> I found a problem with runsvdir: If you create two subdirectories 
>> in /service within the same second, runsvdir doesn't always see the 
>> 2nd one.

> Yes, I know about this for some time, but actually was hit by the 
> problem just once, on a system where the service directory was 
> located on a ramdisk, and the subdirectories were created in 
> background after stage 1 finished.
>
> Did you actually run into this problem, or found it by reading the
> source?

I ran into it while testing out stuff. Usually the service directories 
are created during stage 1, so it's not actually a problem during 
normal operation.

Eric