Re: monitoring FIFOs permissions

[Martin Pala]

Jan-Henrik Haukeland wrote:
> On 26. des. 2005, at 00.16, Martin Pala wrote:
>
> > Example syntax:
> >
> >   check fifo qmgr-fifo with path /var/spool/postfix/public/qmgr
> >     if failed permission 622 then alert
> >     if failed uid postfix then alert
> >     if failed gid postdrop then alert
> >
> > If developers will agree, i can add it to the cvs.
>
>
> Hmm, my initial thought on this is that instead of creating a new  
> specialized file check-service and "duplicate" code, another option  is 
> to, during parse-time, issue an error when if-checks that is not  
> supported by the concrete file is used. Since there are many  different 
> files on Unix we may end up with lots of new specialized  statements if 
> we start. We already have three specialized statements  for file; 
> directory, device and file which should suffice IMHO. The  
> end-functionality should be the same and we keep the name-space in  the 
> control file small.

I though about it this way:

Monit watch the monitored object type (in the case of file it checks 
whether it is regular file, etc.) In the case that the type changed, 
monit can handle the event. This is important since you need to check 
that for example the fifo was not replaced by regular file, the file 
with directory, etc. Because also some tests are specific for given 
object type (for example checksum make sense only for regular file) => 
you need to know the object type to support just relevant tests. In the 
case that we will reuse the 'check file' statement for fifo, we will 
need to add some option which will specify the expected object type and 
extend the parser so that it will check the syntax of 'check file' 
statement based on object type context. I think the standalone 'check 
fifo' is more simple in this case (the context is clearly given by 
service type and just related options/tests are naturaly supported). And 
last but not least ;) ... we already use different check statements for 
device, directory and file (FIFO is just another standalone type).

We already covered almost all basic types (stat system call - types):

check file:
 S_ISREG(m)  is it a regular file?

check directory:
 S_ISDIR(m)  directory?

check device:
 S_ISCHR(m)  character device?
 S_ISBLK(m)  block device?

check fifo (maybe?):
 S_ISFIFO(m) FIFO (named pipe)?

not implemented currently (and probably not needed):
 S_ISLNK(m)  symbolic link? (Not in POSIX.1-1996.)
 S_ISSOCK(m) socket? (Not in POSIX.1-1996.)


For these reasons, i think the new 'check fifo' is better then sharing 
the 'check file' statement.

What do you think?


Thanks :)
Martin