Re: [PATCH v4 2/4] util/qemu-sockets.c: Split host:port parsing out of inet_parse

[Doug Evans <dje@google.com>]

[-- Attachment #1: Type: text/plain, Size: 3817 bytes --]

On Fri, Mar 5, 2021 at 1:51 PM Doug Evans <dje@google.com> wrote:

> On Fri, Mar 5, 2021 at 1:28 PM Samuel Thibault <samuel.thibault@gnu.org>
> wrote:
>
>> Daniel P. Berrangé, le mer. 03 mars 2021 18:11:41 +0000, a ecrit:
>> > On Wed, Mar 03, 2021 at 10:06:50AM -0800, Doug Evans wrote:
>> > > On Sun, Feb 28, 2021 at 1:40 PM Samuel Thibault <
>> samuel.thibault@gnu.org>
>> > > wrote:
>> > >
>> > > > > +  Examples:
>> > > > > +  hostfwd_add net0 tcp:127.0.0.1:10022-:22
>> > > > > +  hostfwd_add net0 tcp:[::1]:10022-[fe80::1:2:3:4]:22
>> > > >
>> > > > Yep, that looks good to me.
>> > >
>> > > Daniel, you wanted me to use inet_parse().
>> > > Is the above syntax ok with you?
>> > > You must have had some expectation that at least some of
>> > > the various flags that inet_parse() recognizes would be needed here.
>> >
>> > It feels like the ,ipv4=on|off,ipv6=on|off flags are relevant here,
>> > especially in the empty address case. eg
>> >
>> >    tcp::10022          - attempt to listen on both ipv4 + ipv6
>> >    tcp::10022,ipv4=off - listen on default address, but only for ipv6
>> >    tcp::10022,ipv6=off - listen on default address, but only for ipv4
>> >
>> > Basically this ends up bringing the hostfwd stuff into alignment with
>> > the way other backends deal with this
>>
>> I'm fine with this yes, better have a coherent user interface.
>>
>
> Cool. Here's the current help text I have:
>
> ---snip---
> #ifdef CONFIG_SLIRP
>     {
>         .name       = "hostfwd_add",
>         .args_type  = "arg1:s,arg2:s?",
>         .params     = "[netdev_id]
> [tcp|udp]:[hostaddr]:hostport-[guestaddr]:guestport",
>         .help       = "redirect TCP or UDP connections from host to guest
> (requires -net user)",
>         .cmd        = hmp_hostfwd_add,
>     },
> #endif
> SRST
> ``hostfwd_add``
>   Redirect TCP or UDP connections from host to guest (requires -net user).
>   IPv6 addresses are wrapped in square brackets, IPv4 addresses are not.
>
>   Examples:
>   hostfwd_add net0 tcp:127.0.0.1:10022-:22
>   hostfwd_add net0 tcp:[::1]:10022-[fe80::1:2:3:4]:22
>
>   Empty host addresses:
>   An empty address for the host, expressed as either "" or "[]", is
>   interpreted as listen at any address for both IPv4 and IPv6. To listen on
>   only IPv4 one can use "0.0.0.0". The equivalent IPv6 address, "[::]", is
>   interpreted as listen on both IPv4 and IPv6 addresses. To listen on only
>   IPv6 addresses, add ",ipv4=off" to the address. E.g.,
>   hostfwd_add net0 tcp::10022,ipv4=off-[fe80::1:2:3:4]:22
>   hostfwd_add net0 tcp:[]:10022,ipv4=off-[fe80::1:2:3:4]:22
>   hostfwd_add net0 tcp:[::]:10022,ipv4=off-[fe80::1:2:3:4]:22
>   And similarly for IPv4 only:
>   hostfwd_add net0 tcp::10022,ipv6=off-[fe80::1:2:3:4]:22
>   hostfwd_add net0 tcp:[]:10022,ipv6=off-[fe80::1:2:3:4]:22
>
>   Empty guest addresses:
>   An empty guest address for IPv4 is translated to the guest's address,
>   assuming that the guest is using DHCP to acquire its address.
>   Note that Libslirp currently only provides a "stateless" DHCPv6 server, a
>   consequence of which is that it cannot do the "addr-any" translation to
> the
>   guest address that is done for IPv4. In other words, the following is
>   currently not supported: hostfwd_add net0 tcp::10022-:22, the guest
>   address is required.
> ERST
> ---snip---
>
> Any corrections or suggestions?
>


For those following along, there are problems with the above help text
(e.g., it's wrong to say "tcp::10022-:22" is not supported, because it
obviously is! - the issue is more nuanced than that).
And I'm sure there are more that I have yet to find.
I'll give reviewers some time to comment on what's there now
before sending an updated proposed text.

[-- Attachment #2: Type: text/html, Size: 5525 bytes --]