bug: --interface does not consume argument

[ciofeca]

If the argument of --interface is not a valid interface, it should abort with an error.

I did this (on a system without any eth8 network interface):

curl --interface eth8 192.168.5.5

I expected the following:

"Error: eth8 does not exist"

I instead got this:

"curl: (45) Could not resolve host: eth8"

curl/libcurl version

curl 7.55.1 (x86_64-pc-linux-gnu) libcurl/7.55.1 OpenSSL/1.0.2g zlib/1.2.11 libidn2/2.0.2 libpsl/0.18.0 (+libidn2/2.0.2) librtmp/2.3
Release-Date: 2017-08-14
Protocols: dict file ftp ftps gopher http https imap imaps ldap ldaps pop3 pop3s rtmp rtsp smb smbs smtp smtps telnet tftp
Features: AsynchDNS IDN IPv6 Largefile GSS-API Kerberos SPNEGO NTLM NTLM_WB SSL libz TLS-SRP UnixSockets HTTPS-proxy PSL

operating system

Ubuntu 17.04

[jay]

It does consume the argument:

"curl: (45) Could not resolve host: eth8"

Error code 45 is CURLE_INTERFACE_FAILED so that is the correct error code. The message
is because if the argument is not an interface then it falls back to treating it as host or ip.

curl/lib/connect.c

Lines 335 to 343 in c514af5

	if(!is_interface) {
	/*
	* This was not an interface, resolve the name as a host name
	* or IP number
	*
	* Temporarily force name resolution to use only the address type
	* of the connection. The resolve functions should really be changed
	* to take a type parameter instead.
	*/

If you turn on verbose you should see * Couldn't bind to 'eth8'

edit: we could make that bind error message what is shown by overwriting the generic resolve error that takes precedence

diff --git a/lib/connect.c b/lib/connect.c
index 16104bc..45e18bc 100644
--- a/lib/connect.c
+++ b/lib/connect.c
@@ -412,6 +412,10 @@ static CURLcode bindlocal(struct connectdata *conn,
     }
 
     if(done < 1) {
+      /* errorbuf is set false so failf will overwrite any message already in
+         the error buffer, so the user receives this error message instead of a
+         generic resolve error. */
+      data->state.errorbuf = FALSE;
       failf(data, "Couldn't bind to '%s'", dev);
       return CURLE_INTERFACE_FAILED;
     }

[jay]

It does consume the argument:

"curl: (45) Could not resolve host: eth8"

Error code 45 is CURLE_INTERFACE_FAILED so that is the correct error code. The message
is because if the argument is not an interface then it falls back to treating it as host or ip.

curl/lib/connect.c

Lines 335 to 343 in c514af5

	if(!is_interface) {
	/*
	* This was not an interface, resolve the name as a host name
	* or IP number
	*
	* Temporarily force name resolution to use only the address type
	* of the connection. The resolve functions should really be changed
	* to take a type parameter instead.
	*/

If you turn on verbose you should see * Couldn't bind to 'eth8'

edit: we could make that bind error message what is shown by overwriting the generic resolve error that takes precedence

diff --git a/lib/connect.c b/lib/connect.c
index 16104bc..45e18bc 100644
--- a/lib/connect.c
+++ b/lib/connect.c
@@ -412,6 +412,10 @@ static CURLcode bindlocal(struct connectdata *conn,
     }
 
     if(done < 1) {
+      /* errorbuf is set false so failf will overwrite any message already in
+         the error buffer, so the user receives this error message instead of a
+         generic resolve error. */
+      data->state.errorbuf = FALSE;
       failf(data, "Couldn't bind to '%s'", dev);
       return CURLE_INTERFACE_FAILED;
     }

[mkauf]

You get a better error message if you add the prefix "if!" to the interface name. This tells curl that it's really an interface. For more information, please see CURLOPT_INTERFACE.

Example:

$ curl --interface 'if!eth8' 192.168.5.5
curl: (45) Couldn't bind to interface 'eth8'

[mkauf]

You get a better error message if you add the prefix "if!" to the interface name. This tells curl that it's really an interface. For more information, please see CURLOPT_INTERFACE.

Example:

$ curl --interface 'if!eth8' 192.168.5.5
curl: (45) Couldn't bind to interface 'eth8'

[bagder]

@jay: sounds like a neat improvement

I see no bug or actual issue here. I'm closing...

[bagder]

@jay: sounds like a neat improvement

I see no bug or actual issue here. I'm closing...