Hi Paul, Hugh,

On 26/10/17 19:13, Paul Wouters wrote:
> On Thu, 26 Oct 2017, D. Hugh Redelmeier wrote:
> 
>> I'm just glancing at unbound_ctx_config() because Paul made a minor
>> change to it to silence coverity.  A failed stat is now logged (good)
>> but behaviour is not otherwise changed.  I've changed it to not use
>> the pathame if stat fails on it.
> 
> Thanks, I should have done that.
> 
>> In several cases (unaffectd by Paul's change), when an error is
>> detected, the code seems to log that error and then continue as if the
>> error had not happened.  I would expect this pattern to be a bug, at
>> least in most cases.
> 
> We try to read whatever files specified via glob and wildcards as we
> can, and ignore the rest. That seems better then complete failure on
> startup. So if you remove a trust anchor file from your DNS setup,
> things still work as best they can without the file.
> 
>> libunbound(3) documents nothing about errno and yet unbound_ctx_config()
>> displays strerror(errno) in a couple of cases after a call to libunbound
>> fails.
>>
>> Is errno meaningful after a failed call to libunbound?  If so, the
>> libunbound manpage should be updated.
> 
> I've CC:ed Wouter :)

Yes it is.  Specifically for the error-to-read-file case.  Not other
cases (eg. socket errors happen too far away in the code).

A manpage update may be meaningful here, yes.

Best regards, Wouter

> 
>> If not, this is an error in our code.
>>
>> At a minimum, I think errno should be initialized to 0 before these
>> calls.  I've done that.
> 
> Yes, thanks!
> 
> Paul


Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
Swan-dev mailing list
[email protected]
https://lists.libreswan.org/mailman/listinfo/swan-dev

Reply via email to