iproute2: iplink_ stuff cleanup

iproute2: iplink_ stuff cleanup

[Yegor Yefremov]

Hello Stephen,

I'm still struggling to get ip/iplink_* routines to show up in
Android's ip binary
(https://groups.google.com/d/topic/android-building/yjV4iYnT1Zc/discussion).
I've looked at the algorithm that is used to access those functions
(like can_parse_opt, vlan_parse_opt etc.)

 snprintf(buf, sizeof(buf), LIBDIR "/ip/link_%s.so", id);
        dlh = dlopen(buf, RTLD_LAZY);
        if (dlh == NULL) {
                /* look in current binary, only open once */
                dlh = BODY;
                if (dlh == NULL) {
                        dlh = BODY = dlopen(NULL, RTLD_LAZY);
                        if (dlh == NULL)
                                return NULL;
                }
        }

        snprintf(buf, sizeof(buf), "%s_link_util", id);
        l = dlsym(dlh, buf);
        if (l == NULL)
                return NULL;

as far as I can see from the ip/Makefile there are no dynamic libs
like /ip/link_%s.so. Wouldn't it be simpler to let all iplink_*
objects to export their interfaces via header files and just make a
table in ip/iplink.c to hold protocol ID and pointer at link_utils
struct.

struct link_util can_link_util = {
        .id             = "can",
        .maxattr        = IFLA_CAN_MAX,
        .parse_opt      = can_parse_opt,
        .print_opt      = can_print_opt,
        .print_xstats   = can_print_xstats,
};

Am I missing something?

Regards,
Yegor

Re: iproute2: iplink_ stuff cleanup

[Stephen Hemminger]

On Thu, 15 Mar 2012 14:53:51 +0100
Yegor Yefremov <yegorslists@googlemail.com> wrote:

> Hello Stephen,
> 
> I'm still struggling to get ip/iplink_* routines to show up in
> Android's ip binary
> (https://groups.google.com/d/topic/android-building/yjV4iYnT1Zc/discussion).
> I've looked at the algorithm that is used to access those functions
> (like can_parse_opt, vlan_parse_opt etc.)
> 
>  snprintf(buf, sizeof(buf), LIBDIR "/ip/link_%s.so", id);
>         dlh = dlopen(buf, RTLD_LAZY);
>         if (dlh == NULL) {
>                 /* look in current binary, only open once */
>                 dlh = BODY;
>                 if (dlh == NULL) {
>                         dlh = BODY = dlopen(NULL, RTLD_LAZY);
>                         if (dlh == NULL)
>                                 return NULL;
>                 }
>         }
> 
>         snprintf(buf, sizeof(buf), "%s_link_util", id);
>         l = dlsym(dlh, buf);
>         if (l == NULL)
>                 return NULL;
> 
> as far as I can see from the ip/Makefile there are no dynamic libs
> like /ip/link_%s.so. Wouldn't it be simpler to let all iplink_*
> objects to export their interfaces via header files and just make a
> table in ip/iplink.c to hold protocol ID and pointer at link_utils
> struct.
> 
> struct link_util can_link_util = {
>         .id             = "can",
>         .maxattr        = IFLA_CAN_MAX,
>         .parse_opt      = can_parse_opt,
>         .print_opt      = can_print_opt,
>         .print_xstats   = can_print_xstats,
> };
> 
> Am I missing something?
> 
> Regards,
> Yegor

Ip utilities have dynamic extensibility, it is possible for someone
to add shared libraries for new functionality. This is a cool feature
but isn't used directly by the standard code. It does use it indirectly
by using dlopen() to find functions locally.

Think of it as introspection in C.

I won't take it out of the standard version, but you may have to find
another way to handle it on the Android universe.

Re: iproute2: iplink_ stuff cleanup

[Yegor Yefremov]

>> I'm still struggling to get ip/iplink_* routines to show up in
>> Android's ip binary
>> (https://groups.google.com/d/topic/android-building/yjV4iYnT1Zc/discussion).
>> I've looked at the algorithm that is used to access those functions
>> (like can_parse_opt, vlan_parse_opt etc.)
>>
>>  snprintf(buf, sizeof(buf), LIBDIR "/ip/link_%s.so", id);
>>         dlh = dlopen(buf, RTLD_LAZY);
>>         if (dlh == NULL) {
>>                 /* look in current binary, only open once */
>>                 dlh = BODY;
>>                 if (dlh == NULL) {
>>                         dlh = BODY = dlopen(NULL, RTLD_LAZY);
>>                         if (dlh == NULL)
>>                                 return NULL;
>>                 }
>>         }
>>
>>         snprintf(buf, sizeof(buf), "%s_link_util", id);
>>         l = dlsym(dlh, buf);
>>         if (l == NULL)
>>                 return NULL;
>>
>> as far as I can see from the ip/Makefile there are no dynamic libs
>> like /ip/link_%s.so. Wouldn't it be simpler to let all iplink_*
>> objects to export their interfaces via header files and just make a
>> table in ip/iplink.c to hold protocol ID and pointer at link_utils
>> struct.
>>
>> struct link_util can_link_util = {
>>         .id             = "can",
>>         .maxattr        = IFLA_CAN_MAX,
>>         .parse_opt      = can_parse_opt,
>>         .print_opt      = can_print_opt,
>>         .print_xstats   = can_print_xstats,
>> };
>>
>> Am I missing something?
>>
>> Regards,
>> Yegor
>
> Ip utilities have dynamic extensibility, it is possible for someone
> to add shared libraries for new functionality. This is a cool feature
> but isn't used directly by the standard code. It does use it indirectly
> by using dlopen() to find functions locally.
>
> Think of it as introspection in C.
>
> I won't take it out of the standard version, but you may have to find
> another way to handle it on the Android universe.

The solution turned out to be simple:
https://android-review.googlesource.com/#/c/34240/.
-Wl,--no-gc-sections did the job.

Can you suggest some more pro arguments for dynamic method of
exporting API routines?

Regards,
Yegor

Re: iproute2: iplink_ stuff cleanup

[Stephen Hemminger]

On Mon, 19 Mar 2012 23:22:59 +0100
Yegor Yefremov <yegorslists@googlemail.com> wrote:

> >> I'm still struggling to get ip/iplink_* routines to show up in
> >> Android's ip binary
> >> (https://groups.google.com/d/topic/android-building/yjV4iYnT1Zc/discussion).
> >> I've looked at the algorithm that is used to access those functions
> >> (like can_parse_opt, vlan_parse_opt etc.)
> >>
> >>  snprintf(buf, sizeof(buf), LIBDIR "/ip/link_%s.so", id);
> >>         dlh = dlopen(buf, RTLD_LAZY);
> >>         if (dlh == NULL) {
> >>                 /* look in current binary, only open once */
> >>                 dlh = BODY;
> >>                 if (dlh == NULL) {
> >>                         dlh = BODY = dlopen(NULL, RTLD_LAZY);
> >>                         if (dlh == NULL)
> >>                                 return NULL;
> >>                 }
> >>         }
> >>
> >>         snprintf(buf, sizeof(buf), "%s_link_util", id);
> >>         l = dlsym(dlh, buf);
> >>         if (l == NULL)
> >>                 return NULL;
> >>
> >> as far as I can see from the ip/Makefile there are no dynamic libs
> >> like /ip/link_%s.so. Wouldn't it be simpler to let all iplink_*
> >> objects to export their interfaces via header files and just make a
> >> table in ip/iplink.c to hold protocol ID and pointer at link_utils
> >> struct.
> >>
> >> struct link_util can_link_util = {
> >>         .id             = "can",
> >>         .maxattr        = IFLA_CAN_MAX,
> >>         .parse_opt      = can_parse_opt,
> >>         .print_opt      = can_print_opt,
> >>         .print_xstats   = can_print_xstats,
> >> };
> >>
> >> Am I missing something?
> >>
> >> Regards,
> >> Yegor
> >
> > Ip utilities have dynamic extensibility, it is possible for someone
> > to add shared libraries for new functionality. This is a cool feature
> > but isn't used directly by the standard code. It does use it indirectly
> > by using dlopen() to find functions locally.
> >
> > Think of it as introspection in C.
> >
> > I won't take it out of the standard version, but you may have to find
> > another way to handle it on the Android universe.
> 
> The solution turned out to be simple:
> https://android-review.googlesource.com/#/c/34240/.
> -Wl,--no-gc-sections did the job.
> 
> Can you suggest some more pro arguments for dynamic method of
> exporting API routines?

You could make it optional on your platform, but for mainline Linux
it has been around long enough that someone surely depends on it.