a defect between linker and module

[bycn82]

​Hi,

during the re-factoring the ipfw2 module, i want to add a logic to
determine whether the module is allowed to be loaded into system. so i
added a piece of code in MOD_LOAD.

but if i create a test module with below code in "modevent"

switch (type) {
case MOD_LOAD:
err = EEXISTS;
break;
case MOD_UNLOAD:

break;
default:

that means it will do nothing just return "load error" and the error code
is "file exists" or "module exists".

so when we loaded it , the module will not be loaded.  but when we run
kldstat, we can see the module named registered.

below is what I found.

when we kldload AAA module, actually it is calling functions in 3 layers
syscall ----- linker -------module

linker_load_file and linker_unload_file functions (in kern_linker.c)are
used to load/unload the module, it will register the name in an array named
linker_files. ( when we kldstat, actually is printing this array, i think )

and linker... will call module_register ( in kern_module.c ).

anway, the module was loaded in function module_register_init    line 112
of kern_modules.c

error = MOD_EVENT(mod, MOD_LOAD);
if (error) {
module_unload(mod); /* ignore error */
module_release(mod);
kprintf("module_register_init: MOD_LOAD (%s, %lx, %p) error %d\n",


so when the MOD_LOAD return error, it will call module_unload and
module_release.

but this function was invoked by the linker, and this function never return
any information, so the linker will never know what is going on inside the
function.

so the module is still registered inside the linker_files. i did not spend
much time on this, so that is all about what i found.


Regards,
Bill Yuan

On 8 February 2015 at 21:05, bycn82 <bycn82 at gmail.com> wrote:

> ​*Hi,*
>
> *during the re-factoring the ipfw2 module, i want to add a logic to
> determine whether the module is allowed to be loaded into system. so i
> added a piece of code in MOD_LOAD.*
>
> *but if i create a test module with below code in "modevent"*
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> *        switch (type) {                case MOD_LOAD:
>     err = EEXISTS;                        break;                case
> MOD_UNLOAD:                        break;                default:that means
> it will do nothing just return "load error" and the error code is "file
> exists" or "module exists".so when we loaded it , the module will not be
> loaded.  but when we run kldstat, we can see the module named
> registered.below is what I found.when we kldload AAA module, actually there
> are kldload syscall ----- linker -------modulelinker_load_file
> and linker_unload_file functions (in kern_linker.c)are used to load/unload
> the module, it will register the name in an array named linker_files. (
> when we kldstat, actually is printing this array, i think )and linker...
> will call module_register ( in kern_module.c ).  anway, the module was
> loaded in function module_register_init    line 112 of kern_modules.c
> error = MOD_EVENT(mod, MOD_LOAD);    if (error) { module_unload(mod); /*
> ignore error */ module_release(mod); kprintf("module_register_init:
> MOD_LOAD (%s, %lx, %p) error %d\n",  so when the MOD_LOAD return error, it
> will call module_unload and module_release.but this function was invoked by
> the linker, and this function never return any information, so the linker
> will never know what is going on inside the function.so the module is still
> registered inside the linker_files. i did not spend much time on this, so
> that is all about what i found. Regards,Bill Yuan*
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dragonflybsd.org/pipermail/users/attachments/20150208/48ef3be8/attachment-0008.html>