Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

wyuenho@gmail.com (Jimmy Yuen Ho Wong) writes:

> branch: netsec
> commit 682578fcf74d4598e39eca81e09d81810d3fc28d
> Author: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
> Commit: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>
>     Add option to bypass NSM TLS checks on local networks
>     
>     * lisp/net/net-utils.el (nslookup-host-ipv4, nslookup-host-ipv6,
>       ipv6-expand): New functions to lookup IPv4 and IPv6 addresses from
>       DNS.

So this only works for people who have nslookup installed? Emacs
already uses getaddrinfo internally, would it help you if there was a
lisp-level interface to it?

Regards

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Jimmy Yuen Ho Wong]

> On 15 Jul 2018, at 12:46 pm, Robert Pluim <rpluim@gmail.com> wrote:
> 
> wyuenho@gmail.com (Jimmy Yuen Ho Wong) writes:
> 
>> branch: netsec
>> commit 682578fcf74d4598e39eca81e09d81810d3fc28d
>> Author: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>> Commit: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>> 
>>    Add option to bypass NSM TLS checks on local networks
>> 
>>    * lisp/net/net-utils.el (nslookup-host-ipv4, nslookup-host-ipv6,
>>      ipv6-expand): New functions to lookup IPv4 and IPv6 addresses from
>>      DNS.
> 
> So this only works for people who have nslookup installed? Emacs
> already uses getaddrinfo internally, would it help you if there was a
> lisp-level interface to it?
> 

Yes it would. I was asking for that exact same thing :) lend me a hand on this?

> Regards
> 
> Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Jimmy Yuen Ho Wong <wyuenho@gmail.com> writes:

>> On 15 Jul 2018, at 12:46 pm, Robert Pluim <rpluim@gmail.com> wrote:
>> 
>> wyuenho@gmail.com (Jimmy Yuen Ho Wong) writes:
>> 
>>> branch: netsec
>>> commit 682578fcf74d4598e39eca81e09d81810d3fc28d
>>> Author: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>>> Commit: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>>> 
>>>    Add option to bypass NSM TLS checks on local networks
>>> 
>>>    * lisp/net/net-utils.el (nslookup-host-ipv4, nslookup-host-ipv6,
>>>      ipv6-expand): New functions to lookup IPv4 and IPv6 addresses from
>>>      DNS.
>> 
>> So this only works for people who have nslookup installed? Emacs
>> already uses getaddrinfo internally, would it help you if there was a
>> lisp-level interface to it?
>> 
>
> Yes it would. I was asking for that exact same thing :) lend me a hand on this?

Iʼm terrible at choosing names, please suggest better ones
(hostname-lookup, gethostbyname,....?). Output currently looks like
this, including a port number, but thatʼs easily changed:

(get-address-info "www.slashdot.org" 'ipv4)
([216 105 38 15 0] [216 105 38 15 0] [216 105 38 15 0])

(get-address-info "google.com")
([172 217 19 238 0] [172 217 19 238 0] [172 217 19 238 0] [10752 5200 16391 2060 0 0 0 8206 0] [10752 5200 16391 2060 0 0 0 8206 0] [10752 5200 16391 2060 0 0 0 8206 0])

Eli, I see thereʼs a sys_getaddrinfo in w32.c, is something needed to get emacs
to use that on MS-Windows?

diff --git i/src/process.c w/src/process.c
index 279b74bc66..7d0bf74cbe 100644
--- i/src/process.c
+++ w/src/process.c
@@ -4531,6 +4531,55 @@ Data that is unavailable is returned as nil.  */)
 #endif
 }
 
+DEFUN ("get-address-info", Fget_address_info, Sget_address_info, 1, 2, 0,
+       doc: /* Look up ip address info of NAME.
+Optional parameter FAMILY controls whether to look up IPv4 or IPv6
+addresses.  The default of nil means look up both, symbol `ipv4' means
+IPv4 only, symbol `ipv6' mean IPv6 only.  Returns a list of addresses,
+or nil if none were found.  */)
+     (Lisp_Object name, Lisp_Object family)
+{
+  Lisp_Object addresses = Qnil;
+  struct addrinfo *res, *lres;
+  int ret;
+
+  struct addrinfo hints;
+  memset (&hints, 0, sizeof hints);
+  if (EQ (family, Qnil))
+    hints.ai_family = AF_UNSPEC;
+  if (EQ (family, Qipv4))
+    hints.ai_family = AF_INET;
+#ifdef AF_INET6
+  if (EQ (family, Qipv6))
+    hints.ai_family = AF_INET6;
+#endif
+  hints.ai_socktype = 0;
+
+  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);
+  if (ret)
+#ifdef HAVE_GAI_STRERROR
+    {
+      synchronize_system_messages_locale ();
+      char const *str = gai_strerror (ret);
+      if (! NILP (Vlocale_coding_system))
+        str = SSDATA (code_convert_string_norecord
+                      (build_string (str), Vlocale_coding_system, 0));
+      message ("\"%s\" \"%s\"", SSDATA (name), str);
+    }
+#else
+      message ("%s get-address-info error %d", SSDATA (name), ret);
+#endif
+  else
+    {
+      for (lres = res; lres; lres = lres->ai_next)
+        addresses = Fcons (conv_sockaddr_to_lisp (lres->ai_addr, lres->ai_addrlen), addresses);
+      addresses = Fnreverse (addresses);
+
+      freeaddrinfo (res);
+    }
+  return addresses;
+}
+
 /* Turn off input and output for process PROC.  */
 
 static void
@@ -8274,6 +8323,7 @@ returns non-`nil'.  */);
   defsubr (&Sset_network_process_option);
   defsubr (&Smake_network_process);
   defsubr (&Sformat_network_address);
+  defsubr (&Sget_address_info);
   defsubr (&Snetwork_interface_list);
   defsubr (&Snetwork_interface_info);
 #ifdef DATAGRAM_SOCKETS

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Eli Zaretskii]

> From: Robert Pluim <rpluim@gmail.com>
> Date: Mon, 16 Jul 2018 15:34:35 +0200
> Cc: emacs-devel@gnu.org
> 
> Eli, I see thereʼs a sys_getaddrinfo in w32.c, is something needed
> to get emacs to use that on MS-Windows?

No, you don't need anything special.  nt/inc/socket.h redirects
getaddrinfo into sys_getaddrinfo, and all our C sources see the
redirection.

> +DEFUN ("get-address-info", Fget_address_info, Sget_address_info, 1, 2, 0,
> +       doc: /* Look up ip address info of NAME.
> +Optional parameter FAMILY controls whether to look up IPv4 or IPv6
> +addresses.  The default of nil means look up both, symbol `ipv4' means
> +IPv4 only, symbol `ipv6' mean IPv6 only.  Returns a list of addresses,
> +or nil if none were found.  */)

This doc string doesn't tell that each address is a vector or a
string.

> +  if (EQ (family, Qipv4))
> +    hints.ai_family = AF_INET;
> +#ifdef AF_INET6
> +  if (EQ (family, Qipv6))
> +    hints.ai_family = AF_INET6;
> +#endif

Should we signal an error if 'ipv6' is requested on a system that
doesn't support that?

> +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);

You should encode NAME (using ENCODE_SYSTEM), because it could include
non-ASCII characters.  In general, any Lisp string should be encoded
before you can pass its data to a C library function.

Thanks.

P.S. This needs a NEWS entry, at the very least, and perhaps also an
update for the ELisp manual.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Jimmy Yuen Ho Wong]

On Mon, Jul 16, 2018 at 4:00 PM Eli Zaretskii <eliz@gnu.org> wrote:
>
> > From: Robert Pluim <rpluim@gmail.com>
> > Date: Mon, 16 Jul 2018 15:34:35 +0200
> > Cc: emacs-devel@gnu.org
> >
> > Eli, I see thereʼs a sys_getaddrinfo in w32.c, is something needed
> > to get emacs to use that on MS-Windows?
>
> No, you don't need anything special.  nt/inc/socket.h redirects
> getaddrinfo into sys_getaddrinfo, and all our C sources see the
> redirection.
>
> > +DEFUN ("get-address-info", Fget_address_info, Sget_address_info, 1, 2, 0,
> > +       doc: /* Look up ip address info of NAME.
> > +Optional parameter FAMILY controls whether to look up IPv4 or IPv6
> > +addresses.  The default of nil means look up both, symbol `ipv4' means
> > +IPv4 only, symbol `ipv6' mean IPv6 only.  Returns a list of addresses,
> > +or nil if none were found.  */)
>
> This doc string doesn't tell that each address is a vector or a
> string.
>
> > +  if (EQ (family, Qipv4))
> > +    hints.ai_family = AF_INET;
> > +#ifdef AF_INET6
> > +  if (EQ (family, Qipv6))
> > +    hints.ai_family = AF_INET6;
> > +#endif
>
> Should we signal an error if 'ipv6' is requested on a system that
> doesn't support that?
>
> > +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);
>
> You should encode NAME (using ENCODE_SYSTEM), because it could include
> non-ASCII characters.  In general, any Lisp string should be encoded
> before you can pass its data to a C library function.
>

Does getaddrinfo accept unicode hostnames? If not I think we can just
make sure NAME is all ASCII as per Punycode for i18nized host names.

> Thanks.
>
> P.S. This needs a NEWS entry, at the very least, and perhaps also an
> update for the ELisp manual.
>

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Jimmy Yuen Ho Wong]

On Mon, Jul 16, 2018 at 2:34 PM Robert Pluim <rpluim@gmail.com> wrote:
>
> Jimmy Yuen Ho Wong <wyuenho@gmail.com> writes:
>
> >> On 15 Jul 2018, at 12:46 pm, Robert Pluim <rpluim@gmail.com> wrote:
> >>
> >> wyuenho@gmail.com (Jimmy Yuen Ho Wong) writes:
> >>
> >>> branch: netsec
> >>> commit 682578fcf74d4598e39eca81e09d81810d3fc28d
> >>> Author: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
> >>> Commit: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
> >>>
> >>>    Add option to bypass NSM TLS checks on local networks
> >>>
> >>>    * lisp/net/net-utils.el (nslookup-host-ipv4, nslookup-host-ipv6,
> >>>      ipv6-expand): New functions to lookup IPv4 and IPv6 addresses from
> >>>      DNS.
> >>
> >> So this only works for people who have nslookup installed? Emacs
> >> already uses getaddrinfo internally, would it help you if there was a
> >> lisp-level interface to it?
> >>
> >
> > Yes it would. I was asking for that exact same thing :) lend me a hand on this?
>
> Iʼm terrible at choosing names, please suggest better ones
> (hostname-lookup, gethostbyname,....?). Output currently looks like
> this, including a port number, but thatʼs easily changed:
>

I wish I could think of a better name. All the good names are taken by
terrible implementations LOL

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Stefan Monnier]

>> Iʼm terrible at choosing names, please suggest better ones
>> (hostname-lookup, gethostbyname,....?). Output currently looks like
>> this, including a port number, but thatʼs easily changed:
> I wish I could think of a better name. All the good names are taken by
> terrible implementations LOL

All the C functions dealing with the network should use a common
namespace prefix, I think.  That could be "network-" or "inet-" or
various others.


        Stefan "just helping paint the shed"

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Robert Pluim <rpluim@gmail.com>
>> Date: Mon, 16 Jul 2018 15:34:35 +0200
>> Cc: emacs-devel@gnu.org
>> 
>> Eli, I see thereʼs a sys_getaddrinfo in w32.c, is something needed
>> to get emacs to use that on MS-Windows?
>
> No, you don't need anything special.  nt/inc/socket.h redirects
> getaddrinfo into sys_getaddrinfo, and all our C sources see the
> redirection.

Thanks. I always forget how the nt stuff works.

>> +DEFUN ("get-address-info", Fget_address_info, Sget_address_info, 1, 2, 0,
>> +       doc: /* Look up ip address info of NAME.
>> +Optional parameter FAMILY controls whether to look up IPv4 or IPv6
>> +addresses.  The default of nil means look up both, symbol `ipv4' means
>> +IPv4 only, symbol `ipv6' mean IPv6 only.  Returns a list of addresses,
>> +or nil if none were found.  */)
>
> This doc string doesn't tell that each address is a vector or a
> string.

Yes. Iʼm waiting for Jimmy to tell me if the format works for him,
then Iʼll document whatever we end up with (and it can currently only
return a vector, and includes a port, which is probably not needed).

>> +  if (EQ (family, Qipv4))
>> +    hints.ai_family = AF_INET;
>> +#ifdef AF_INET6
>> +  if (EQ (family, Qipv6))
>> +    hints.ai_family = AF_INET6;
>> +#endif
>
> Should we signal an error if 'ipv6' is requested on a system that
> doesn't support that?

Iʼd be more inclined to return nil in that case. The effect is the
same, and the caller doesnʼt need to do redundant error handling.

>> +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);
>
> You should encode NAME (using ENCODE_SYSTEM), because it could include
> non-ASCII characters.  In general, any Lisp string should be encoded
> before you can pass its data to a C library function.
>

My understanding is that this API only supports ASCII anyway. For
internationalized domain names you'd need to use puny-code (and we
donʼt currently use ENCODE_SYSTEM when calling getaddrinfo elsewhere).

> Thanks.
>
> P.S. This needs a NEWS entry, at the very least, and perhaps also an
> update for the ELisp manual.

Both, for sure.

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>>> Iʼm terrible at choosing names, please suggest better ones
>>> (hostname-lookup, gethostbyname,....?). Output currently looks like
>>> this, including a port number, but thatʼs easily changed:
>> I wish I could think of a better name. All the good names are taken by
>> terrible implementations LOL
>
> All the C functions dealing with the network should use a common
> namespace prefix, I think.  That could be "network-" or "inet-" or
> various others.
>

Emacs has a whole bunch of commands and variables starting with
'network-', that makes sense to me.

>         Stefan "just helping paint the shed"

network-lookup-info? network-lookup-hostname-info? There are so many
colours to choose from.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Andy Moreton]

On Mon 16 Jul 2018, Robert Pluim wrote:

> Jimmy Yuen Ho Wong <wyuenho@gmail.com> writes:
>
>>> On 15 Jul 2018, at 12:46 pm, Robert Pluim <rpluim@gmail.com> wrote:
>>> 
>>> wyuenho@gmail.com (Jimmy Yuen Ho Wong) writes:
>>> 
>>>> branch: netsec
>>>> commit 682578fcf74d4598e39eca81e09d81810d3fc28d
>>>> Author: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>>>> Commit: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
>>>> 
>>>>    Add option to bypass NSM TLS checks on local networks
>>>> 
>>>>    * lisp/net/net-utils.el (nslookup-host-ipv4, nslookup-host-ipv6,
>>>>      ipv6-expand): New functions to lookup IPv4 and IPv6 addresses from
>>>>      DNS.
>>> 
>>> So this only works for people who have nslookup installed? Emacs
>>> already uses getaddrinfo internally, would it help you if there was a
>>> lisp-level interface to it?
>>> 
>>
>> Yes it would. I was asking for that exact same thing :) lend me a hand on this?
>
> Iʼm terrible at choosing names, please suggest better ones
> (hostname-lookup, gethostbyname,....?). Output currently looks like
> this, including a port number, but thatʼs easily changed:

Given that we already have `network-interface-list' and
`network-interface-info' I think this should be named `network-address-info'.

Your patch works on Windows 10 (64bit mingw64 MSYS2):

ELISP> (get-address-info "www.slashdot.org" 'ipv4)
([216 105 38 15 0])

ELISP> (get-address-info "google.com")
([216 58 213 110 0])

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Jimmy Yuen Ho Wong]

>
> Given that we already have `network-interface-list' and
> `network-interface-info' I think this should be named `network-address-info'.
>

Sounds good to me!

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Jimmy Yuen Ho Wong]

>
> Yes. Iʼm waiting for Jimmy to tell me if the format works for him,
> then Iʼll document whatever we end up with (and it can currently only
> return a vector, and includes a port, which is probably not needed).
>

Oh you are waiting for me, you can just push to my branch or master
and I'll test it out.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Jimmy Yuen Ho Wong <wyuenho@gmail.com> writes:

>>
>> Yes. Iʼm waiting for Jimmy to tell me if the format works for him,
>> then Iʼll document whatever we end up with (and it can currently only
>> return a vector, and includes a port, which is probably not needed).
>>
>
> Oh you are waiting for me, you can just push to my branch or master
> and I'll test it out.

Iʼve pushed to the netsec branch.

Regards

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Paul Eggert]

Robert Pluim wrote:
> +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);

Why does the Lisp API always pass NULL here? Shouldn't there be some way to 
specify the service at the Lisp level?

More generally, there's a lot of code duplication between this new function and 
what's already in the implementation of make-network-process. Intead, 
make-network-process should call this new function (or some C variant of it).

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Eli Zaretskii]

> From: Jimmy Yuen Ho Wong <wyuenho@gmail.com>
> Date: Mon, 16 Jul 2018 16:24:45 +0100
> Cc: Robert Pluim <rpluim@gmail.com>, Emacs-Devel devel <emacs-devel@gnu.org>
> 
> > > +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);
> >
> > You should encode NAME (using ENCODE_SYSTEM), because it could include
> > non-ASCII characters.  In general, any Lisp string should be encoded
> > before you can pass its data to a C library function.
> >
> 
> Does getaddrinfo accept unicode hostnames? If not I think we can just
> make sure NAME is all ASCII as per Punycode for i18nized host names.

That's fine with me, but my point is that simply plugging the string
data into a libc function is usually not OK.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Eli Zaretskii]

> From: Robert Pluim <rpluim@gmail.com>
> Cc: emacs-devel@gnu.org
> Date: Mon, 16 Jul 2018 18:23:21 +0200
> 
> > No, you don't need anything special.  nt/inc/socket.h redirects
> > getaddrinfo into sys_getaddrinfo, and all our C sources see the
> > redirection.
> 
> Thanks. I always forget how the nt stuff works.

In general, all the sys_* stuff is invisible everywhere except in
w32.c.

> >> +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);
> >
> > You should encode NAME (using ENCODE_SYSTEM), because it could include
> > non-ASCII characters.  In general, any Lisp string should be encoded
> > before you can pass its data to a C library function.
> >
> 
> My understanding is that this API only supports ASCII anyway.

Then I think we should test that it's either a unibyte string or a
string whose size in bytes is equal to its size in characters, and
signal an error if that doesn't hold.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Eli Zaretskii]

> From: Robert Pluim <rpluim@gmail.com>
> Date: Mon, 16 Jul 2018 18:36:37 +0200
> Cc: emacs-devel@gnu.org
> 
> Emacs has a whole bunch of commands and variables starting with
> 'network-', that makes sense to me.
> 
> >         Stefan "just helping paint the shed"
> 
> network-lookup-info? network-lookup-hostname-info? There are so many
> colours to choose from.

FWIW, I'm okay with get-address-info, for 2 reasons:

  . we don't other related names start with anything like that;
  . other languages use names very close to getaddrinfo, so why cannot we?

But if you want to use some other name, feel free.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Stefan Monnier]

> network-lookup-info? network-lookup-hostname-info? There are so many
> colours to choose from.

I'll let you choose that part of the color (so long as it's black, of
course).


        Stefan

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Paul Eggert <eggert@cs.ucla.edu> writes:

> Robert Pluim wrote:
>> +  ret = getaddrinfo (SSDATA (name), NULL, &hints, &res);
>
> Why does the Lisp API always pass NULL here? Shouldn't there be some
> way to specify the service at the Lisp level?

I donʼt think thatʼs currently needed at the lisp level, and I didnʼt
want to overcomplicate the API.

> More generally, there's a lot of code duplication between this new
> function and what's already in the implementation of
> make-network-process. Intead, make-network-process should call this
> new function (or some C variant of it).

I guess that would be possible. Iʼll look into it.

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Eli Zaretskii <eliz@gnu.org> writes:
>
> Then I think we should test that it's either a unibyte string or a
> string whose size in bytes is equal to its size in characters, and
> signal an error if that doesn't hold.

So I tried using STRING_MULTIBYTE, but of course eww uses
puny-encode-domain on unicode hostnames, and that returns a multibyte
string that only contains ASCII characters. Rather than opening that
can of worms, I settled on checking

SBYTES (host) != SCHARS (host)

I couldn't find an "is this a pure-ASCII string" function.

Regards

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Eli Zaretskii]

> From: Robert Pluim <rpluim@gmail.com>
> Cc: emacs-devel@gnu.org
> Date: Tue, 17 Jul 2018 12:09:10 +0200
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> >
> > Then I think we should test that it's either a unibyte string or a
> > string whose size in bytes is equal to its size in characters, and
> > signal an error if that doesn't hold.
> 
> So I tried using STRING_MULTIBYTE, but of course eww uses
> puny-encode-domain on unicode hostnames, and that returns a multibyte
> string that only contains ASCII characters. Rather than opening that
> can of worms, I settled on checking
> 
> SBYTES (host) != SCHARS (host)

This will signal an error for unibyte strings, because there SBYTES is
always -1.  So I think you should do this instead:

  STRING_MULTIBYTE (host) && SBYTES (host) != SCHARS (host)

> I couldn't find an "is this a pure-ASCII string" function.

It is rarely needed, IME.

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Robert Pluim <rpluim@gmail.com>
>> Cc: emacs-devel@gnu.org
>> Date: Tue, 17 Jul 2018 12:09:10 +0200
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> >
>> > Then I think we should test that it's either a unibyte string or a
>> > string whose size in bytes is equal to its size in characters, and
>> > signal an error if that doesn't hold.
>> 
>> So I tried using STRING_MULTIBYTE, but of course eww uses
>> puny-encode-domain on unicode hostnames, and that returns a multibyte
>> string that only contains ASCII characters. Rather than opening that
>> can of worms, I settled on checking
>> 
>> SBYTES (host) != SCHARS (host)
>
> This will signal an error for unibyte strings, because there SBYTES is
> always -1.  So I think you should do this instead:
>

Ah, I was assuming they'd always be in sync.

>   STRING_MULTIBYTE (host) && SBYTES (host) != SCHARS (host)

OK, that works.

Thanks

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Robert Pluim <rpluim@gmail.com> writes:

> Eli Zaretskii <eliz@gnu.org> writes:
>
>>> From: Robert Pluim <rpluim@gmail.com>
>>> Cc: emacs-devel@gnu.org
>>> Date: Tue, 17 Jul 2018 12:09:10 +0200
>>> 
>>> Eli Zaretskii <eliz@gnu.org> writes:
>>> >
>>> > Then I think we should test that it's either a unibyte string or a
>>> > string whose size in bytes is equal to its size in characters, and
>>> > signal an error if that doesn't hold.
>>> 
>>> So I tried using STRING_MULTIBYTE, but of course eww uses
>>> puny-encode-domain on unicode hostnames, and that returns a multibyte
>>> string that only contains ASCII characters. Rather than opening that
>>> can of worms, I settled on checking
>>> 
>>> SBYTES (host) != SCHARS (host)
>>
>> This will signal an error for unibyte strings, because there SBYTES is
>> always -1.  So I think you should do this instead:
>>
>
> Ah, I was assuming they'd always be in sync.
>
>>   STRING_MULTIBYTE (host) && SBYTES (host) != SCHARS (host)
>
> OK, that works.

I couldn't get my original to fail with a unibyte string. Turns out
that even though size_byte is indeed -1 for unibyte strings, in that
case SBYTES returns the size field:

  ptrdiff_t nbytes = s->u.s.size_byte < 0 ? s->u.s.size : s->u.s.size_byte;
  
Having said that, testing for STRING_MULTIBYTE is harmless.

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Eli Zaretskii]

> From: Robert Pluim <rpluim@gmail.com>
> Date: Tue, 17 Jul 2018 18:17:57 +0200
> 
> I couldn't get my original to fail with a unibyte string. Turns out
> that even though size_byte is indeed -1 for unibyte strings, in that
> case SBYTES returns the size field:
> 
>   ptrdiff_t nbytes = s->u.s.size_byte < 0 ? s->u.s.size : s->u.s.size_byte;

Ah, yes.  I keep forgetting that.  Which is one more reason not to
rely on that, if you ask me ;-)

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Robert Pluim <rpluim@gmail.com>
>> Date: Tue, 17 Jul 2018 18:17:57 +0200
>> 
>> I couldn't get my original to fail with a unibyte string. Turns out
>> that even though size_byte is indeed -1 for unibyte strings, in that
>> case SBYTES returns the size field:
>> 
>>   ptrdiff_t nbytes = s->u.s.size_byte < 0 ? s->u.s.size : s->u.s.size_byte;
>
> Ah, yes.  I keep forgetting that.  Which is one more reason not to
> rely on that, if you ask me ;-)

STRING_MULTIBYTE it is then.

Robert

Re: netsec 682578f 4/6: Add option to bypass NSM TLS checks on local networks

[Robert Pluim]

Robert Pluim <rpluim@gmail.com> writes:

> Paul Eggert <eggert@cs.ucla.edu> writes:
>> More generally, there's a lot of code duplication between this new
>> function and what's already in the implementation of
>> make-network-process. Intead, make-network-process should call this
>> new function (or some C variant of it).

So I split the common functionality into an internal function, and
made the lisp function call that, and updated make-network-process to
call it. Also a lispref entry and some tests (that uncovered a rather
embarassing bug), all pushed to the netsec branch.

Regards

Robert