Fwd: strncpy_from_user called from invalid context?

[Mulyadi Santosa]

Resending it to the new shiny mailing list address :D


---------- Forwarded message ----------
From: Mulyadi Santosa <mulyadi.santosa at gmail.com>
Date: Sun, Dec 19, 2010 at 12:36
Subject: Re: strncpy_from_user called from invalid context?
To: Greg Kerr <kerrgi at gmail.com>
Cc: kernelnewbies at nl.linux.org


Hi Greg....

let us see the basic situation first...since I am not an expert either...

On Sun, Dec 19, 2010 at 05:27, Greg Kerr <kerrgi at gmail.com> wrote:
> [  421.250737] BUG: sleeping function called from invalid context at
> arch/x86/mm/fault.c:1074
> [  421.250742] in_atomic(): 0, irqs_disabled(): 1, pid: 1718, name: dmesg
> [  421.250745] Pid: 1718, comm: dmesg Not tainted 2.6.35.9-64.fc14.x86_64 #1
> [  421.250748] Call Trace:
> [  421.250756]  [<ffffffff8103d1d5>] __might_sleep+0xeb/0xf0
> [  421.250761]  [<ffffffff8146cbf4>] do_page_fault+0x15c/0x265
> [  421.250765]  [<ffffffff8146a075>] page_fault+0x25/0x30
> [  421.250770]  [<ffffffff8121db83>] ? strncpy+0xc/0x27
> [  421.250774]  [<ffffffffa000f049>] j_sys_open+0x37/0xc6 [exe_probe]

I am not 100% sure if I "cut" the stack trace into the most relevant
path, so let's just assume it is indeed correct.

First, irqs_disabled = 1 makes me think that you probably hook into
atomic code path. Before answering that, may we know, where do you put
this hook anyway? at the sys_open() itself? or somewhere else?

Second complication is that fairly likely, it retrieves file name (I
hope I read your code correctly) from use memory area that might be
swapped out. If it is the problem, then mlock() it could be the
workaround. However, the solution might be non

All in all, it's better to also write down the way you register the
probe etc, that way we could have better understanding of your work,
thus might pinpoint another root of the problem itself.

PS: on a second thought, I am curious on why I see __do_fault in early
"stage" of your back trace. Could it be that your code might be called
during page fault itself? Thus it's something like: fault--> supposed
to be atomic from here on-->read a portion of a file--> your probe
called--> do something that blocks--> boom?

--
regards,

Mulyadi Santosa
Freelance Linux trainer and consultant

blog: the-hydra.blogspot.com
training: mulyaditraining.blogspot.com



-- 
regards,

Mulyadi Santosa
Freelance Linux trainer and consultant

blog: the-hydra.blogspot.com
training: mulyaditraining.blogspot.com