Re: [PATCH] ttyprintk: Add TTY hangup callback.

From: "Samo Pogačnik" <samo_pogacnik@t-2.net>
To: Petr Mladek <pmladek@suse.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
	Jiri Slaby <jirislaby@kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Sergey Senozhatsky <sergey.senozhatsky@gmail.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	John Ogness <john.ogness@linutronix.de>,
	linux-kernel@vger.kernel.org,
	syzkaller-bugs <syzkaller-bugs@googlegroups.com>
Subject: Re: [PATCH] ttyprintk: Add TTY hangup callback.
Date: Mon, 26 Apr 2021 18:42:05 +0200	[thread overview]
Message-ID: <d4d0603716e5cb99a7a9a93d4f767278ac318557.camel@t-2.net> (raw)
In-Reply-To: <YIaPQzktArmoWbLr@alley>

Dne 26.04.2021 (pon) ob 12:00 +0200 je Petr Mladek napisal(a):
> On Sat 2021-04-24 11:57:47, Samo Pogačnik wrote:
> > Dne 24.04.2021 (sob) ob 10:16 +0900 je Tetsuo Handa napisal(a):
> > > On 2021/04/24 4:47, Samo Pogačnik wrote:
> > > > At any point the tpk_buffer is potentially multiplexed/interleaved by
> > > > parts
> > > > of
> > > > required output of any concurrent user, as buffs are being delivered by
> > > > the
> > > > scheduled writes.
> > > 
> > > As long as one line is printed by one printk() call,
> > > CONFIG_PRINTK_CALLER=y is
> > > helpful enough to distinguish multilplexed/interleaved messages. I
> > > consider
> > > that
> > > ttyprintk offers additional advantage over printk() for allow buffering
> > > one
> > > line
> > > of message from userspace.
> 
> It does not matter how much buffering games you play. As long as you
> use printk() to store single lines into the kernel logbuffer they
> alway could be interleaved with lines from other processes/CPUs.

Exactly. The only purpose of ttyprintk buffering is to mark any begining of
lines occurring within the userspace-string written into ttyprintk TTY. The
marked lines do not originate in the kernel source code, which is not obvious
otherwise (imho this is importannt). Even the CONFIG_PRINTK_CALLER=y does not
give this information, if the task ID printed does not live anymore.

> 
> > > > 
> > > > As per user buffers look promising with output formatting, the FDs
> > > > passing
> > > > between tasks lead to the same single buffer (Greg already mentioned
> > > > that).
> > > 
> > > Those programs which use FD passing know what they are doing. If they
> > > still
> > > want
> > > one line of message printed via ttyprintk interface, they must do their
> > > buffering
> > > before trying to write() to ttyprintk's file descriptor.
> 
> Lines might get interleaved when using printk().
> What is special about messages passed via ttyprintk()?
They do not originate in the kernel code.

> How many processes are using it?
In case of redirection any proces, that is writing to console.

> Do they print many lines?
?

> Is it really worth any added complexity?
No.

> 
> > On the other hand, my main concern is how to provide a reliable system wide
> > collection of all console output via ttyprintk console redirection, while
> > normal
> > operation of system console is preserved (except its output being detoured
> > via
> > printk and as such logged together with kernel output). Such logging is
> > particularly useful for after-the-fact inspection of system operation.
> 
> I am not sure if I understand the problem. But why does ttyprintk need
> any buffer at all. AFAIK, the use-case is to pass any written data into the
> kernel logbuffer via printk()?
(see above - it is not something the kernel is telling you)
> 
> Why tpk_write() does not call printk() directly?
(see above)
> 
> If you call printk() directly, the caller_id would be from the process
> that really wrote the data/message.
It can be a kernel-code originating message printk-ed on behalf of a user task
or a kernel-code originating message on behalf of a kernel task. Or it may be a
user-code originating message on behalf of its task, when printk-ed via
ttyprintk.

> 
> > That being said i am thinking about how to permanently enable this
> > redirection
> > as early as possible (i.e. via kernel command line option). I had a working
> > prototype for that some time ago (never posted). Would anybody like to see
> > such
> > functionality?
> 
> Please, do not add any complex code if it does not cause real life
> problems.
> 
Noted, thanks.

Best regards, Samo