Re: Re: Re: [PATCH 0/2] [BUGFIX] printk: Fix message continuation breakage involved with structured printk

[Yoshihiro YUNOMAE <yoshihiro.yunomae.ez@hitachi.com>]

(2013/12/24 12:00), Kay Sievers wrote:
> On Tue, Dec 24, 2013 at 3:50 AM, Yoshihiro YUNOMAE
> <yoshihiro.yunomae.ez@hitachi.com> wrote:
>> (2013/12/20 20:29), Kay Sievers wrote:
>>>
>>> On Fri, Dec 20, 2013 at 10:41 AM, Yoshihiro YUNOMAE
>>> <yoshihiro.yunomae.ez@hitachi.com> wrote:
>>>
>>>> This patch set fixes message continuation breakage involved with
>>>> structured
>>>> printk. A SCSI driver may output two continuation error messages like
>>>>       scmd_printk("foo");
>>>>       printf("bar\n");
>>>
>>>
>>> Which is the absolutely wrong thing to do. Structured logging and racy
>>> printk continuation must never be mixed. Userspace needs to be sure
>>> that dictionary entries are not subject to racy continuation hackery,
>>> and that these mwssages atomic, whole and intact.
>>
>>
>> I see.
>> As you say, user tools need to support messages output in multiple lines
>> for SMP environments even if this patch set is introduced.
>
> They cannot really, they can try to re-construct, but Information and
> context is sometimes lost with  the use of continuation lines.
> Structured logging need to be reliable and trustable, and it it is not
> "best effort", hence it cannot use continuation lines.
>
> Continuation lines are a nice debugging tool for humans only.
> Structured logging carries the human readable string but also
> machine-readable context and that alwasy needs to be safely
> machine-readable and recognizable.

I understand. I think if we make machine(user tools) handle important
messages, those messages need to be output in single line even if those
are not structured printk.

>>> Please do not mix the both and do not apply these patches.
>>
>> OK, I'll make important messages with KERN_CONT or no-prefix printk()
>> output those in single line.
>
> Yes, it should use the plain printk versions and not the one which add
> structured data.
>
> If structured logging is really wanted for more complex continuation
> lines, it might be the simplest to buffer the line locally, instead of
> the single printk-owned buffer, before the line is emitted to printk.

Yes, I'll do that.

Thanks,
Yoshihiro YUNOMAE

-- 
Yoshihiro YUNOMAE
Software Platform Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: yoshihiro.yunomae.ez@hitachi.com