Thanks for your help on this Stefano. I will add the null terminator
when writing strings, but I have also restructured my code so it will
only happen rarely, so should not be a problem anyway.
On 14/10/2019 21:21, Stefano Miccoli via Owfs-developers wrote:
There is for sure something broken in the owserver, and it can be very
dangerous from a strict security point of view (buffer overflows and
so on). However I think that most owserver instances are run in
private networks so this should not be a major concern.
For now I would advise you to add a null terminator when you are
writing “string” data to owserver (something like the alias name) and
let us know if the problem resurfaces again.
The commit in which Paul added the possibility of non null terminated
strings on the owserver payload is
https://github.com/owfs/owfs/commit/379f4e926
(but let me stress again that it is only my guess that your problem is
linked to the null terminator, maybe it is something completely
different.)
S.
On 14 Oct 2019, at 00:46, Mick Sulley <m...@sulley.info
<mailto:m...@sulley.info>> wrote:
Hi Stefano,
I ran test.py and it completed without error. I then ran it again
and it failed at B9, I ran it several times more and it failed at B9
each time.
I then made the change you suggested and ran it again, it still
failed at B9. I then rebooted and ran it again and it completed
without error. I then ran it many more times and it still completed
without error. So it would appear that your fix together with a
reboot fixes the problem :)
Is this something which should be fixed in owserver or should it be a
discipline on the user to add b'\x00' to any string when writing?
Mick
On 13/10/2019 21:25, Stefano Miccoli via Owfs-developers wrote:
I had no time to stress test my installation, but let me add some
thoughts here.
The fact that failures are sporadic, and that a reboot makes the
system more resilient, makes me guess that there is a memory leak in
owserver.
I suspect that this could be linked to the way in which strings are
treated in owserver. I clearly remember that when I implemented the
write feature I asked on this list if data sent in a write request
should be null-terminated. Paul replied that this was not necessary,
since the length of the payload is known to owserver.
Therefore the payload for an alias is sent by pyownet as something like
b”/26.123412341234/alias\x00T0”
(where b”\x00” is a null byte in python). This means that on the
receiving side the C code has to add a null terminator to the
received alias. So I would suggest to test writes likes this
owp.write(addr, newa + b'\x00')
just to see explicitly adding a null terminator on the sending side
mitigates the problem. (Please note that this is just a wild guess,
maybe the problem is in another portion of the code.)
Stefano
_______________________________________________
Owfs-developers mailing list
Owfs-developers@lists.sourceforge.net
<mailto:Owfs-developers@lists.sourceforge.net>
https://lists.sourceforge.net/lists/listinfo/owfs-developers
_______________________________________________
Owfs-developers mailing list
Owfs-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/owfs-developers
_______________________________________________
Owfs-developers mailing list
Owfs-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/owfs-developers
