Message-ID: <20170622142524.5lg3p3blfumur6fs@redbean>
Date:   Thu, 22 Jun 2017 16:25:26 +0200
From:   Jessica Yu <jeyu@...hat.com>
To:     Wanlong Gao <gaowanlong@...wei.com>
Cc:     Xie XiuQi <xiexiuqi@...wei.com>, akpm@...ux-foundation.org,
        linux-kernel@...r.kernel.org, rusty@...tcorp.com.au,
        john.wanghui@...wei.com, wencongyang2@...wei.com,
        guijianfeng@...wei.com
Subject: Re: [PATCH] modpost: abort if a module name is too long

+++ Wanlong Gao [22/06/17 09:11 +0800]:
>
>
>On 2017/6/22 0:09, Jessica Yu wrote:
>> +++ Jessica Yu [06/06/17 20:41 -0700]:
>>> +++ Wanlong Gao [06/06/17 09:07 +0800]:
>>>>
>>>>
>>>> On 2017/6/5 10:09, Jessica Yu wrote:
>>>>> +++ Wanlong Gao [02/06/17 11:04 +0800]:
>>>>>>
>>>>>>
>>>>>> On 2017/6/2 7:23, Jessica Yu wrote:
>>>>>>> +++ Wanlong Gao [31/05/17 11:48 +0800]:
>>>>>>>>
>>>>>>>>
>>>>>>>> On 2017/5/31 11:30, Jessica Yu wrote:
>>>>>>>>> +++ Wanlong Gao [31/05/17 10:23 +0800]:
>>>>>>>>>> Hi Jessica,
>>>>>>>>>>
>>>>>>>>>> On 2017/5/29 17:10, Jessica Yu wrote:
>>>>>>>>>>> +++ Xie XiuQi [20/05/17 15:46 +0800]:
>>>>>>>>>>>> From: Wanlong Gao <gaowanlong@...wei.com>
>>>>>>>>>>>>
>>>>>>>>>>>> Module name has a limited length, but currently the build system
>>>>>>>>>>>> allows the build finishing even if the module name is too long.
>>>>>>>>>>>>
>>>>>>>>>>>> CC      /root/kprobe_example/abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz.mod.o
>>>>>>>>>>>> /root/kprobe_example/abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz.mod.c:9:2:
>>>>>>>>>>>> warning: initializer-string for array of chars is too long [enabled by default]
>>>>>>>>>>>> .name = KBUILD_MODNAME,
>>>>>>>>>>>> ^
>>>>>>>>>>>>
>>>>>>>>>>>> but it's merely a warning.
>>>>>>>>>>>>
>>>>>>>>>>>> This patch adds the check of the module name length in modpost and stops
>>>>>>>>>>>> the build properly.
>>>>>>>>>>>>
>>>>>>>>>>>> Signed-off-by: Wanlong Gao <gaowanlong@...wei.com>
>>>>>>>>>>>> Signed-off-by: Xie XiuQi <xiexiuqi@...wei.com>
>>>>>>>>>>>> ---
>>>>>>>>>>>> scripts/mod/modpost.c | 11 +++++++++++
>>>>>>>>>>>> 1 file changed, 11 insertions(+)
>>>>>>>>>>>>
>>>>>>>>>>>> diff --git a/scripts/mod/modpost.c b/scripts/mod/modpost.c
>>>>>>>>>>>> index 30d752a..db11c57 100644
>>>>>>>>>>>> --- a/scripts/mod/modpost.c
>>>>>>>>>>>> +++ b/scripts/mod/modpost.c
>>>>>>>>>>>> @@ -2166,6 +2166,17 @@ static int add_versions(struct buffer *b, struct module *mod)
>>>>>>>>>>>> {
>>>>>>>>>>>>    struct symbol *s, *exp;
>>>>>>>>>>>>    int err = 0;
>>>>>>>>>>>> +    const char *mod_name;
>>>>>>>>>>>> +
>>>>>>>>>>>> +    mod_name = strrchr(mod->name, '/');
>>>>>>>>>>>> +    if (mod_name == NULL)
>>>>>>>>>>>> +        mod_name = mod->name;
>>>>>>>>>>>> +    else
>>>>>>>>>>>> +        mod_name++;
>>>>>>>>>>>> +    if (strlen(mod_name) >= MODULE_NAME_LEN) {
>>>>>>>>>>>> +        merror("module name is too long [%s.ko]\n", mod->name);
>>>>>>>>>>>> +        return 1;
>>>>>>>>>>>> +    }
>>>>>>>>>>>
>>>>>>>>>>> Hi Xie,
>>>>>>>>>>>
>>>>>>>>>>> This check shouldn't be in add_versions() (which does something else entirely),
>>>>>>>>>>> it should probably be put in a separate helper function called from main. But
>>>>>>>>>>> I'm not a big fan of the extra string manipulation to do something this simple.
>>>>>>>>>>>
>>>>>>>>>>> I think this check can be vastly simplified, how about something like the
>>>>>>>>>>> following?
>>>>>>>>>>
>>>>>>>>>> This looks better, would you apply your following patch?
>>>>>>>>>>
>>>>>>>>>> Reviewed-by: Wanlong Gao <gaowanlong@...wei.com>
>>>>>>>>>> Tested-by: Wanlong Gao <gaowanlong@...wei.com>
>>>>>>>>>
>>>>>>>>> Sure, thanks for testing. I'll go ahead and format this into a proper
>>>>>>>>> patch and resend.
>>>>>>>>
>>>>>>>> Please wait, I just found that this patch makes the built module can't
>>>>>>>> be inserted by the following error:
>>>>>>>>
>>>>>>>> # insmod abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabc.ko
>>>>>>>> insmod: ERROR: could not insert module abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabc.ko: Invalid parameters
>>>>>>>>
>>>>>>>> # dmesg
>>>>>>>> abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabc: Unknown symbol __fentry__ (err -22)
>>>>>>>
>>>>>>> Hm, I am unable to reproduce this. It looks like __fentry__ is missing
>>>>>>> from your kernel, you may have a mismatch between the kernel config
>>>>>>> that you're running and the config you are using to build the module.
>>>>>>> In other words, it seems like you might have built the module with
>>>>>>> CONFIG_FTRACE but built the kernel without.
>>>>>>>
>>>>>>> Few questions -
>>>>>>>
>>>>>>> What is the output of running `grep __fentry__ /proc/kallsyms`?
>>>>>>>
>>>>>>
>>>>>> Sure it has.
>>>>>>
>>>>>>> Does your module correspond to the running kernel version?
>>>>>>
>>>>>> Sure.
>>>>>>
>>>>>>>
>>>>>>> Do you have CONFIG_FTRACE/FUNCTION_TRACER enabled in your running
>>>>>>> kernel?
>>>>>>>
>>>>>>
>>>>>> Sure.
>>>>>>
>>>>>>
>>>>>>> Is that the full dmesg output (are there any other error messages)?
>>>>>>
>>>>>> Even when I compiled the kernel with your patch, the kernel module load
>>>>>> failed at the boot time with the following error:
>>>>>>
>>>>>> [    1.656708] libcrc32c: no symbol version for __fentry__
>>>>>> [    1.656709] libcrc32c: Unknown symbol __fentry__ (err -22)
>>>>>>
>>>>>> But my above patch in add_versions() doesn't have such problem, I've no
>>>>>> idea why. Maybe your patch breaks some sections?
>>>>>
>>>>> Hm, I am still unable to reproduce this on my system with modversions
>>>>> enabled and the -rc2 kernel. But judging by the errno (-22) it looks
>>>>> like this is failing in check_version()/resolve_symbol() for you,
>>>>> which leads me to think that this is somehow messing with the
>>>>> __versions table generated by modpost (not sure why).
>>>>>
>>>>> Does the  ____versions[] array in the generated *.mod.c file for your
>>>>> test module look different with and without the patch? Also: what
>>>>> version of gcc and binutils are you using, and what kernel version are
>>>>> you testing on?
>>>>
>>>> The *.mod.c file are same except the added __modname_test section, the gcc
>>>> ,binutils and kernel are all from centos 7.2 (3.10.0-327).
>>>
>>> Thanks for the additional info. Just FYI, I'm going to be out this
>>> week and part of next week due to travelling, but I'll be able to take
>>> another look at this next Thurs/Fri. If we can't resolve the issue, we
>>> can just work on your original patch.
>>
>> Thanks for your patience, I've just moved abroad and getting to stable
>> internet has been a challenge :-/
>>
>> Here's my last attempt at fixing the BUILD_BUG_ON patch (I am not sure
>> why it seems to be messing with the __versions table on your setup,
>> perhaps it is related to .discard usage?).
>>
>> Do either of the patches below work on your setup? (try one or the
>> other and let me know if either of them work..)
>
>Sorry to say that neither ;< It seems not to add section in mod.c
>is more safe.

No problem, thanks for verifying! I originally liked the build bug
patch because it worked directly with the in-kernel module name,
instead of the filename (they can differ slightly, but the number of
chars should remain the same anyway..). In any case, could you modify
your original patch to put the modname check in a separate function,
maybe named check_modname_len(), and have it be called before
check_exports()?

Thanks!

Jessica

Powered by blists - more mailing lists