linux-kernel - Re: [PATCH v3 20/21] dyndbg: add user-flag, negating-flags, and filtering on flags

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <746984fb-00ee-9079-efac-50167f3c3e40@akamai.com>
Date:   Thu, 18 Jun 2020 14:17:07 -0400
From:   Jason Baron <jbaron@...mai.com>
To:     Petr Mladek <pmladek@...e.com>, Jim Cromie <jim.cromie@...il.com>
Cc:     linux-kernel@...r.kernel.org, akpm@...uxfoundation.org,
        gregkh@...uxfoundation.org, linux@...musvillemoes.dk,
        Jonathan Corbet <corbet@....net>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Will Deacon <will@...nel.org>,
        Orson Zhai <orson.zhai@...soc.com>, linux-doc@...r.kernel.org
Subject: Re: [PATCH v3 20/21] dyndbg: add user-flag, negating-flags, and
 filtering on flags



On 6/18/20 1:40 PM, Petr Mladek wrote:
> On Thu 2020-06-18 18:19:12, Petr Mladek wrote:
>> On Wed 2020-06-17 10:25:35, Jim Cromie wrote:
>>> 1. Add a user-flag [u] which works like the [pfmlt] flags, but has no
>>> effect on callsite behavior; it allows incremental marking of
>>> arbitrary sets of callsites.
>>>
>>> 2. Add [PFMLTU] flags, which negate their counterparts; P===!p etc.
>>> And in ddebug_read_flags():
>>>    current code does:	[pfmltu_] -> flags
>>>    copy it to:		[PFMLTU_] -> mask
>>>
>>> also disallow both of a pair: ie no 'pP', no true & false.
>>>
>>> 3. Add filtering ops into ddebug_change(), right after all the
>>> callsite-property selections are complete.  These filter on the
>>> callsite's current flagstate before applying modflags.
>>>
>>> Why ?
>>>
>>> The u-flag & filter flags
>>>
>>> The 'u' flag lets the user assemble an arbitary set of callsites.
>>> Then using filter flags, user can activate the 'u' callsite set.
>>>
>>>   #> echo 'file foo.c +u; file bar.c +u' > control   # and repeat
>>>   #> echo 'u+p' > control
>>>
>>> Of course, you can continue to just activate your set without ever
>>> marking it 1st, but you could trivially add the markup as you go, then
>>> be able to use it as a constraint later, to undo or modify your set.
>>>
>>>   #> echo 'file foo.c +up' >control
>>>   .. monitor, debug, finish ..
>>>   #> echo 'u-p' >control
>>>
>>>   # then later resume
>>>   #> echo 'u+p' >control
>>>
>>>   # disable some cluttering messages, and remove from u-set
>>>   #> echo 'file noisy.c function:jabber_* u-pu' >control
>>>
>>>   # for doc, recollection
>>>   grep =pu control > my-favorite-callsites
>>>
>>> Note:
>>>
>>> Your flagstate after boot is generally not all =_. -DDEBUG will arm
>>> compiled callsites by default, $builtinmod.dyndbg=+p bootargs can
>>> enable them early, and $module.dyndbg=+p bootargs will arm them when
>>> the module is loaded.  But you could manage them with u-flags:
>>>
>>>   #> echo '-t' >control		# clear t-flag to use it as 2ndary markup
>>>   #> echo 'p+ut' >control	# mark the boot-enabled set of callsites
>>>   #> echo '-p' >control		# clean your dmesg -w stream
>>>
>>>   ... monitor, debug ..
>>>   #> echo 'module of_interest $qterms +pu' >control	# build your set of useful debugs
>>>   #> echo 'module of_interest $qterms UT+pu' >control	# same, but dont alter ut marked set
>>
>> Does anyone requested this feature, please?
>>
>> For me, it is really hard to imagine people using these complex and hacky
>> steps.
> 
> I think that all this is motivated by adding support for module
> specific groups.
> 
> What about storing the group as yet another information for each
> message? I mean the same way as we store module name, file, line,
> function name.
> 
> Then we could add API to define group for a given message:
> 
>    pr_debug_group(group_id, fmt, ...);
> 
> the interface for the control file might be via new keyword "group".
> You could then do something like:
> 
>    echo module=drm group=0x3 +p >control
> 
> But more importantly you should add functions that might be called
> when the drm.debug parameter is changes. I have already mentioned
> it is another reply:
> 
>     dd_enable_module_group(module_name, group_id);
>     dd_disable_module_group(module_name, group_id);
> 
> 
> It will _not_ need any new flag or flag filtering.
> 
> Best Regards,
> Petr
> 

Yes, I'm wondering as well if people are really going to use the
new flags and filter flags - I mentioned that here:
https://lkml.org/lkml/2020/6/12/732

The grouping stuff is already being used by lots of modules so
that seems useful.

Thanks,

-Jason