[BUG] kernel/printk: RT-adaption of kmsg_dump breaks mtdoops

[BUG] kernel/printk: RT-adaption of kmsg_dump breaks mtdoops

[Walter Stoll]

Effect observed
---------------

We want to store kernel oops logs on a NAND flash partition by means of the
MTD Oops/Panic console logger/driver. However, no log is generated.

We observed the effect with kernel version v5.4.138-rt62. However, we think
that the most recent kernel exhibits the same behavior because the structure of
the sources in question (see below) did not change.


Root cause
----------

Further investigations revealed, than the issue is due to the realtime adaption
of the kmsg_dump() function.

Non realtime version of kmsg_dump():
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/kernel/printk/printk.c?h=linux-5.4.y#n3149

Realtime version of kmsg_dump():
https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/tree/kernel/printk/printk.c?h=v5.4-rt#n2904

In our case, the the mtd_panic_write() function returns with error when the
kernel panics. As a result, no log is created. Call chain:
... panic() -> kmsg_dump() -> mtdoops_do_dump() -> mtdoops_write()
    mtd_panic_write()

Note that within the kmsg_dump() function, a locally create kmsg_dumper object
is passed to the dumper_local.dump(&dumper_local, reason) call. However the
callee mtdoops_do_dump() function expects this object being embedded in a
mtdoops_context object, see
https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/tree/drivers/mtd/mtdoops.c?h=v5.4-rt#n272

This is obviously not the case, therefore the error. Note that this could even
led to a security issue.


Bug fix proposal
----------------

We fixed the bug locally by applying the patch below. Note that we did not
modify the kmsg_dump() function. Instead we modified the mtdoops_do_dump()
function which seems for us the cleaner way.


diff --git a/drivers/mtd/mtdoops.c b/drivers/mtd/mtdoops.c
index 774970bfcf85..317ecf47fce8 100644
--- a/drivers/mtd/mtdoops.c
+++ b/drivers/mtd/mtdoops.c
@@ -269,8 +269,7 @@ static void find_next_position(struct mtdoops_context *cxt)
 static void mtdoops_do_dump(struct kmsg_dumper *dumper,
 			    enum kmsg_dump_reason reason)
 {
-	struct mtdoops_context *cxt = container_of(dumper,
-			struct mtdoops_context, dump);
+	struct mtdoops_context *cxt = &oops_cxt;
 
 	/* Only dump oopses if dump_oops is set */
 	if (reason == KMSG_DUMP_OOPS && !dump_oops)

Re: [BUG] kernel/printk: RT-adaption of kmsg_dump breaks mtdoops

[John Ogness]

Hello Walter,

On 2021-09-17, Walter Stoll <Walter.Stoll@duagon.com> wrote:
> Effect observed
> ---------------
>
> We want to store kernel oops logs on a NAND flash partition by means of the
> MTD Oops/Panic console logger/driver. However, no log is generated.
>
> We observed the effect with kernel version v5.4.138-rt62. However, we think
> that the most recent kernel exhibits the same behavior because the structure of
> the sources in question (see below) did not change.

This problem does not exist in v5.10-rt. The kmsg_dump API was changed
(later in mainline as well) to avoid this issue. For v5.10-rt the change
was buried in the all-in-one commit:

https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v5.10-rt&id=9835363e31e75d3287a81c6fc1b22c5c2b43d277

And later. the API change for mainline (5.12):

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f9f3f02db98bbe678a8e57fe9432b196174744a3

> Root cause
> ----------
>
> Further investigations revealed, than the issue is due to the realtime adaption
> of the kmsg_dump() function.
>
> Non realtime version of kmsg_dump():
> https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/kernel/printk/printk.c?h=linux-5.4.y#n3149
>
> Realtime version of kmsg_dump():
> https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/tree/kernel/printk/printk.c?h=v5.4-rt#n2904
>
> In our case, the the mtd_panic_write() function returns with error when the
> kernel panics. As a result, no log is created. Call chain:
> ... panic() -> kmsg_dump() -> mtdoops_do_dump() -> mtdoops_write()
>     mtd_panic_write()
>
> Note that within the kmsg_dump() function, a locally create kmsg_dumper object
> is passed to the dumper_local.dump(&dumper_local, reason) call. However the
> callee mtdoops_do_dump() function expects this object being embedded in a
> mtdoops_context object, see
> https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/tree/drivers/mtd/mtdoops.c?h=v5.4-rt#n272
>
> This is obviously not the case, therefore the error. Note that this could even
> led to a security issue.
>
>
> Bug fix proposal
> ----------------
>
> We fixed the bug locally by applying the patch below. Note that we did not
> modify the kmsg_dump() function. Instead we modified the mtdoops_do_dump()
> function which seems for us the cleaner way.

I agree that this is a better fix, rather than backporting the newer
API. For mtdoops, the only function registering a dumper is
mtdoops_notify_add() and it also directly references the global context
with:

struct mtdoops_context *cxt = &oops_cxt;

Reviewed-by: John Ogness <john.ogness@linutronix.de>

> diff --git a/drivers/mtd/mtdoops.c b/drivers/mtd/mtdoops.c
> index 774970bfcf85..317ecf47fce8 100644
> --- a/drivers/mtd/mtdoops.c
> +++ b/drivers/mtd/mtdoops.c
> @@ -269,8 +269,7 @@ static void find_next_position(struct mtdoops_context *cxt)
>  static void mtdoops_do_dump(struct kmsg_dumper *dumper,
>  			    enum kmsg_dump_reason reason)
>  {
> -	struct mtdoops_context *cxt = container_of(dumper,
> -			struct mtdoops_context, dump);
> +	struct mtdoops_context *cxt = &oops_cxt;
>  
>  	/* Only dump oopses if dump_oops is set */
>  	if (reason == KMSG_DUMP_OOPS && !dump_oops)

Thanks for your work on this!

John Ogness

AW: [BUG] kernel/printk: RT-adaption of kmsg_dump breaks mtdoops

[Walter Stoll]

> -----Ursprüngliche Nachricht-----
> Von: John Ogness <john.ogness@linutronix.de>
> Gesendet: Freitag, 17. September 2021 13:36
> An: Walter Stoll <Walter.Stoll@duagon.com>; linux-rt-users@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
> Betreff: Re: [BUG] kernel/printk: RT-adaption of kmsg_dump breaks mtdoops
> 
> Hello Walter,
> 
> On 2021-09-17, Walter Stoll <Walter.Stoll@duagon.com> wrote:
> > Effect observed
> > ---------------
> >
> > We want to store kernel oops logs on a NAND flash partition by means of the
> > MTD Oops/Panic console logger/driver. However, no log is generated.
> >
> > We observed the effect with kernel version v5.4.138-rt62. However, we think
> > that the most recent kernel exhibits the same behavior because the structure of
> > the sources in question (see below) did not change.
> 
> This problem does not exist in v5.10-rt. The kmsg_dump API was changed
> (later in mainline as well) to avoid this issue. For v5.10-rt the change
> was buried in the all-in-one commit:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v5.10-
> rt&id=9835363e31e75d3287a81c6fc1b22c5c2b43d277
> 
> And later. the API change for mainline (5.12):
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f9f3f02db98b
> be678a8e57fe9432b196174744a3
>

Ok, thanks a lot for your response. I was not aware of this.

> > Root cause
> > ----------
> >
> > Further investigations revealed, than the issue is due to the realtime adaption
> > of the kmsg_dump() function.
> >
> > Non realtime version of kmsg_dump():
> >
> https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/kernel/printk/printk
> .c?h=linux-5.4.y#n3149
> >
> > Realtime version of kmsg_dump():
> > https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-
> rt.git/tree/kernel/printk/printk.c?h=v5.4-rt#n2904
> >
> > In our case, the the mtd_panic_write() function returns with error when the
> > kernel panics. As a result, no log is created. Call chain:
> > ... panic() -> kmsg_dump() -> mtdoops_do_dump() -> mtdoops_write()
> >     mtd_panic_write()
> >
> > Note that within the kmsg_dump() function, a locally create kmsg_dumper object
> > is passed to the dumper_local.dump(&dumper_local, reason) call. However the
> > callee mtdoops_do_dump() function expects this object being embedded in a
> > mtdoops_context object, see
> > https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-
> rt.git/tree/drivers/mtd/mtdoops.c?h=v5.4-rt#n272
> >
> > This is obviously not the case, therefore the error. Note that this could even
> > led to a security issue.
> >
> >
> > Bug fix proposal
> > ----------------
> >
> > We fixed the bug locally by applying the patch below. Note that we did not
> > modify the kmsg_dump() function. Instead we modified the mtdoops_do_dump()
> > function which seems for us the cleaner way.
> 
> I agree that this is a better fix, rather than backporting the newer
> API. For mtdoops, the only function registering a dumper is
> mtdoops_notify_add() and it also directly references the global context
> with:
> 
> struct mtdoops_context *cxt = &oops_cxt;
> 
> Reviewed-by: John Ogness <john.ogness@linutronix.de>
> 
> > diff --git a/drivers/mtd/mtdoops.c b/drivers/mtd/mtdoops.c
> > index 774970bfcf85..317ecf47fce8 100644
> > --- a/drivers/mtd/mtdoops.c
> > +++ b/drivers/mtd/mtdoops.c
> > @@ -269,8 +269,7 @@ static void find_next_position(struct mtdoops_context *cxt)
> >  static void mtdoops_do_dump(struct kmsg_dumper *dumper,
> >  			    enum kmsg_dump_reason reason)
> >  {
> > -	struct mtdoops_context *cxt = container_of(dumper,
> > -			struct mtdoops_context, dump);
> > +	struct mtdoops_context *cxt = &oops_cxt;
> >
> >  	/* Only dump oopses if dump_oops is set */
> >  	if (reason == KMSG_DUMP_OOPS && !dump_oops)
> 
> Thanks for your work on this!
> 
> John Ogness

Regards
Walter