"Daniel P. Berrange" <berra...@redhat.com> writes:
> On Wed, Jun 28, 2017 at 09:24:58AM -0500, Eric Blake wrote:
>> On 06/28/2017 08:23 AM, Daniel P. Berrange wrote:
>> > On Wed, Jun 28, 2017 at 09:08:49PM +0800, Mao Zhongyi wrote:
>> >> diff --git a/include/qemu/sockets.h b/include/qemu/sockets.h
>> >> index 5c326db..78e2b30 100644
>> >> --- a/include/qemu/sockets.h
>> >> +++ b/include/qemu/sockets.h
>> >
>> >> if (qemu_isdigit(buf[0])) {
>> >> - if (!inet_aton(buf, &saddr->sin_addr))
>> >> + if (!inet_aton(buf, &saddr->sin_addr)) {
>> >> + error_setg(errp, "host address '%s' is not a valid "
>> >> + "IPv4 address", buf);
>> >> return -1;
>> >> + }
>> >> } else {
>> >> - if ((he = gethostbyname(buf)) == NULL)
>> >> + he = gethostbyname(buf);
>> >> + if (he == NULL) {
>> >> + error_setg(errp, "can't resolve host address '%s': "
>> >> + "unknown host", buf);
>> >> return - 1;
>> >> + }
>> >
>> > gethostbyname sets 'h_errno' on failure, so you should pass that
>> > into error_setg_errno, instead of hardcoding 'unknown host' as a
>> > message
'unknown host' is misleading when h_errno != HOST_NOT_FOUND.
>> 'man gethostbyname' says it is deprecated, and that applications should
>> use getaddrinfo/getnameinfo instead. What's our story here?
>
> The real story is to get net/socket.c converted to QIOChannelSocket
> and kill this parse_host_port() method in sockets.c It is already
> broken by design since it takes a 'struct sockdddr_in' and thus
> can't do IPv6.
>
> This patch doesn't make the existing situation worse, so I think
> its fine to add this error reporting cleanup now, and not force
> immediate conversion to QIOChannelSocket today. The net/sockets.c
> code needs a further refactor before that conversion can be done
> in the right way - we've already reverted the wrong way twice ;-)
Until then, let's go with a generic error message, as I requested in my
review of v5. Just drop the misleading ": unknown host" part.
