Re: lib/46111: yplib will hang forever if no server can be found

[christos%zoulas.com@localhost (Christos Zoulas)]

On Feb 29,  4:35pm, Wolfgang.Stukenbrock%nagler-company.com@localhost (Wolfgang 
Stukenbrock) wrote:
-- Subject: Re: lib/46111: yplib will hang forever if no server can be found

| Hi,
| 
| I'm not shure if a function is a good idea, because there is already a 
| global variable _yplib_nerrs that is set to 5 as default.
| As far as I've figured out _yplib_nerrs is used in some files in 
| src/lib/libc/yp and is not documented in any headerfiles and/or the 
| manual. It seems to be a global libc-internal variable.
| A disadvantage of a function is the missing ability to query the current 
| state.
| So I would prefer a variable, but I can implement a function too, if you 
| prefer this.

make it return the current state then, and ignore the setting if the setting
is negative.

| Currently after _yplib_nerrs number of retries the message "YP server 
| for ... still trying" is printed, but only if the domain is already 
| found in the _ypbindlist. No printing is done for the first contact with 
| ypbind - or I've overlooked something.
| This looks like someone has changed the original Sun behaviour in the 
| past that is printing this message as far as I remember even for the 
| first connection attemp.
| remark: an entry is added to the _ypbindlist after successfull 
| contacting ypbind or if a binding file is present and valid.
| 
| If there is a binding file present for the domain, and the file locking 
| fails, YPERR_YPBIND is returned without contacting ypbind.
| When contacting ypbind, YPERR_YPBIND is returned too if clnttcp_create() 
| fails.
| Theese both should be very rare situations, but will already currently 
| not wait until a server is available.
| 
| In e.g. yp_first and yp_next the variable _yplib_nerrs is used to print 
| a message too if the binding has succeeded but the clnt_call() has 
| failed the configured number of times. This will be done in an endless 
| loop too, if the binding succeded every time.
| Should this behaviour be changed too? If the binding fails, YPERR_DOMAIN 
| is returned.
| 
| I aggree not to change de default behaviour of the lib, but allow to 
| change it unter program control. At least I need currently need the 
| possibility to catch the problem with ypbind to faile my request and 
| continue to work.
| 
| 
| My "current" patch will return YPERR_YPBIND only if there is no 
| previously setup entry in the _ypbindlist.
| Should I also terminate the processing after <n> retries, if there is an 
| entry in _ypbindlist present? I think yes.
| Should in such case the "still trying" message be written to stderr or 
| should the lib be silent. I think no printing should be done.
| 
| What about the following sollution:
| I reuse _yplib_nerrs for the new functionality.
|    _yplib_nerrs > 0 - current behaviour - wait endless and print a 
| message every _yplib_nerrs retries. The default value will be 5 as before.
|    _yplib_nerrs == 0 - wait endless without printing to stderr
|    (this is something that would already happen now when _yplib_nerrs 
| gets set to 0, because the print-check check starts with 1 and the first 
| printout will happen on the integer-wrap ...)
|    _yplib_nerrs < 0 - number of retries (as negative count) with ypbind 
| after that YPERR_YPBIND is returned. And the lib will return regardless 
| if the domainname has been queried before or not.
| I will add it to the rpcsvc/ypclnt.h headerfile as extern declaration 
| and add it the to ypclnt(3) manual.
| 
| This implies disabling the printing in the other routines (yp_first etc. 
| ) too on none-positive retry numbers.
| 
| Is this OK?

I think it is cleaner not to oveload the variable this way and use a new one.
Alterning the behavior of internal variables that are present in most reference
implementations is not a good idea, because if you take the code from NetBSD
to another OS it will not behave as expected. With the new function at least
you get a link error that tells you that you need to do something else.

christos