Date: Fri, 13 Sep 2002 08:40:48 -0700 From: Terry Lambert <tlambert2@mindspring.com> To: Jev <jev@ecad.org> Cc: freebsd-hackers@freebsd.org Subject: Re: gethostbyname_r() fbsd equiv? Message-ID: <3D820700.DB53B0F9@mindspring.com> References: <20020913145830.GB41842@ecad.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Jev wrote:
> Im trying to build some software on freebsd, which wants to use
> the thread safe gethostbyname_r(). Despite having very bad C skills im
> going to attempt to patch it. What would I use in place of
> gethostbyname_r() on freebsd?
Do you need the real gethostbyname_r(), or do you need the
bastardized Linux version? The real version has the prototype:
int gethostbyname_r(const char *name,
struct hostent *result,
struct hostent_data *buffer);
The bastardized Linux version has the prototype:
extern int gethostbyname_r(__const char *__restrict __name,
struct hostent *__restrict __result_buf,
char *__restrict __buf, size_t __buflen,
struct hostent **__restrict __result,
int *__restrict __h_errnop);
So the good one has 3 parameters, and the bad one has 5...
It's actually trivial to write this code, IMO. It's also not a
very useful exercise, since the resolver code should not be in
libc, it should be in a seperate library that is obtained from
the third party ISC BIND distribution, and reimported and updated
as ISC updates their code.
This is additionally complicated by the KAME changes to support
IPv6, which have not been standards-tracked into a mainstream
resolver library, and anr therefore pretty BSD-hacked-libc
resolver specific.
If you need a 3 parameter implementation, you should be able to
write it in under a day, even including makeing the gethostent
and sethostent code, etc., all reentrant (or at least based on
code that passes around a context).
If you just need to deal with the heap collisions from using
gethostbyname() in a threaded program, the most trivial way to
implement it is to implement a worker thread to do the lookups,
and write a gethostbyname_r() stub that rendesvous with the
worker thread to pass off work. This assumes that you don't
mind serializing lookup requests through a single thread, which
you shouldn't care about unless you are running -current with
Jon Mini's KSE based pthreads library, which can take advantage
of multiple CPUs, and which would suffer a performance penalty
by serializing requests through a single thread. Doing that
work should take you about two hours, if you are a competent
threads programmer, or just a competent programmer with a copy
of the O'Reilly book on pthreads.
If you just need *any* gethostbyname_r(), the SQUID distribution
has an asynchronous resolver library that implements one. That
library, though, is GPL'ed.
If you need to avoid the GPL because you don't want to have to
GPL the resulting code, then you won't be able to use the SQUID
library.
Finally, you could always pay someone to do the code for you; if
you did that, you would want to specify the interfaces and the
required behaviour ahead of time. My personal suggestion would be
to find the AIX or HP/UX manual pages, and use them to specify the
API, since they do a much better job than the Linux ones (the Linux
ones are mostly just the nasty 5 parameter prototypes, with no
explanation of the function of the parameters or the expected
behaviour in situ).
-- Terry
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3D820700.DB53B0F9>