* [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer
@ 2026-11-19 8:45 Alexandre Becholey
2026-11-19 8:49 ` sashiko-bot
2026-11-19 12:02 ` Oliver Neukum
0 siblings, 2 replies; 6+ messages in thread
From: Alexandre Becholey @ 2026-11-19 8:45 UTC (permalink / raw)
To: Oliver Neukum, Greg Kroah-Hartman; +Cc: Alexandre Becholey, stable, linux-usb
mts_command_done() receives the data phase of a REQUEST SENSE into
srb->sense_buffer (SCSI_SENSE_BUFFERSIZE, 96 bytes), but with the
length of the command's own data buffer. A REQUEST SENSE sent through
SG_IO, allowed on a read-only /dev/sgN without any capability, can have
a much larger buffer (252 bytes with sg_requests), and a device that
returns that much writes past the end of the sense buffer.
The special case is not needed: for the error handler's own REQUEST
SENSE, scsi_eh_prep_cmnd() already points the data buffer at
sense_buffer with at most SCSI_SENSE_BUFFERSIZE. Handle REQUEST SENSE
like every other data-in command.
The bug was found, and the reproducer and this fix were written, by AI
agents (Claude Mythos and Claude Opus models). The analysis and the
patch were reviewed by a human.
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Alexandre Becholey <alex@becholey.ch>
---
An emulated ScanMaker X6 answering a 252-byte SG_IO REQUEST SENSE with
252 bytes triggers it (KASAN in dummy_timer, the HCD's copy into the
URB buffer; slub_debug redzone report with xHCI), on v7.3.1 and master
9d66f26fefac. No reports with the fix. User-space REQUEST SENSE now gets
the sense data in its own buffer; the error handler's REQUEST SENSE
returns the same data as before. Not tested on real hardware.
Reproducer available on request.
drivers/usb/image/microtek.c | 19 ++++++-------------
1 file changed, 6 insertions(+), 13 deletions(-)
diff --git a/drivers/usb/image/microtek.c b/drivers/usb/image/microtek.c
index 1f0b3c44d388..0b8d7f71c3c9 100644
--- a/drivers/usb/image/microtek.c
+++ b/drivers/usb/image/microtek.c
@@ -445,22 +445,15 @@ static void mts_command_done( struct urb *transfer )
return;
}
- if (context->srb->cmnd[0] == REQUEST_SENSE) {
+ if (context->data) {
mts_int_submit_urb(transfer,
context->data_pipe,
- context->srb->sense_buffer,
+ context->data,
context->data_length,
- mts_data_done);
- } else { if ( context->data ) {
- mts_int_submit_urb(transfer,
- context->data_pipe,
- context->data,
- context->data_length,
- scsi_sg_count(context->srb) > 1 ?
- mts_do_sg : mts_data_done);
- } else {
- mts_get_status(transfer);
- }
+ scsi_sg_count(context->srb) > 1 ?
+ mts_do_sg : mts_data_done);
+ } else {
+ mts_get_status(transfer);
}
}
--
2.53.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer
2026-11-19 8:45 [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer Alexandre Becholey
@ 2026-11-19 8:49 ` sashiko-bot
2026-11-19 12:02 ` Oliver Neukum
1 sibling, 0 replies; 6+ messages in thread
From: sashiko-bot @ 2026-11-19 8:49 UTC (permalink / raw)
To: Alexandre Becholey; +Cc: linux-usb
> mts_command_done() receives the data phase of a REQUEST SENSE into
> srb->sense_buffer (SCSI_SENSE_BUFFERSIZE, 96 bytes), but with the
> length of the command's own data buffer. A REQUEST SENSE sent through
> SG_IO, allowed on a read-only /dev/sgN without any capability, can have
> a much larger buffer (252 bytes with sg_requests), and a device that
> returns that much writes past the end of the sense buffer.
>
> The special case is not needed: for the error handler's own REQUEST
> SENSE, scsi_eh_prep_cmnd() already points the data buffer at
> sense_buffer with at most SCSI_SENSE_BUFFERSIZE. Handle REQUEST SENSE
> like every other data-in command.
>
> The bug was found, and the reproducer and this fix were written, by AI
> agents (Claude Mythos and Claude Opus models). The analysis and the
> patch were reviewed by a human.
>
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Cc: stable@vger.kernel.org
> Signed-off-by: Alexandre Becholey <alex@becholey.ch>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261119084526.535901-1-alex@becholey.ch?part=1
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer
2026-11-19 8:45 [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer Alexandre Becholey
2026-11-19 8:49 ` sashiko-bot
@ 2026-11-19 12:02 ` Oliver Neukum
2026-11-19 12:28 ` Alexandre Becholey
1 sibling, 1 reply; 6+ messages in thread
From: Oliver Neukum @ 2026-11-19 12:02 UTC (permalink / raw)
To: Alexandre Becholey, Oliver Neukum, Greg Kroah-Hartman; +Cc: stable, linux-usb
On 19.11.26 09:45, Alexandre Becholey wrote:
> The special case is not needed: for the error handler's own REQUEST
> SENSE, scsi_eh_prep_cmnd() already points the data buffer at
> sense_buffer with at most SCSI_SENSE_BUFFERSIZE. Handle REQUEST SENSE
> like every other data-in command.
Hi,
thank you for looking at this, but this fix seems to not take an
approach I can approve. Where sense_buffer is pointed to is
internal to the SCSI layer. We cannot use that in an LLD.
If you think that LLDs should not use sense_buffer then the clean
solution is to remove it from struct scsi_cmnd.
Regards
Oliver
NACKED-BY: Oliver Neukum <oneukum@suse.com>
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer
2026-11-19 12:02 ` Oliver Neukum
@ 2026-11-19 12:28 ` Alexandre Becholey
2026-11-19 12:51 ` Oliver Neukum
0 siblings, 1 reply; 6+ messages in thread
From: Alexandre Becholey @ 2026-11-19 12:28 UTC (permalink / raw)
To: Oliver Neukum; +Cc: Oliver Neukum, Greg Kroah-Hartman, stable, linux-usb
On Thursday, November 19th, 2026 at 1:02 PM, Oliver Neukum <oneukum@suse.com> wrote:
> On 19.11.26 09:45, Alexandre Becholey wrote:
>
> > The special case is not needed: for the error handler's own REQUEST
> > SENSE, scsi_eh_prep_cmnd() already points the data buffer at
> > sense_buffer with at most SCSI_SENSE_BUFFERSIZE. Handle REQUEST SENSE
> > like every other data-in command.
> Hi,
>
> thank you for looking at this, but this fix seems to not take an
> approach I can approve. Where sense_buffer is pointed to is
> internal to the SCSI layer. We cannot use that in an LLD.
>
> If you think that LLDs should not use sense_buffer then the clean
> solution is to remove it from struct scsi_cmnd.
>
> Regards
> Oliver
>
> NACKED-BY: Oliver Neukum <oneukum@suse.com>
>
Hi Oliver,
Would you take a v2 that keeps receiving the REQUEST SENSE data into
sense_buffer and only limits the transfer to SCSI_SENSE_BUFFERSIZE?
if (context->srb->cmnd[0] == REQUEST_SENSE) {
mts_int_submit_urb(transfer,
context->data_pipe,
context->srb->sense_buffer,
min_t(unsigned int, context->data_length,
SCSI_SENSE_BUFFERSIZE),
mts_data_done);
The device can then no longer write past the 96-byte buffer, and
nothing else changes.
Kind regards,
Alexandre
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer
2026-11-19 12:28 ` Alexandre Becholey
@ 2026-11-19 12:51 ` Oliver Neukum
2026-11-20 15:14 ` Alexandre Becholey
0 siblings, 1 reply; 6+ messages in thread
From: Oliver Neukum @ 2026-11-19 12:51 UTC (permalink / raw)
To: Alexandre Becholey, Oliver Neukum
Cc: Oliver Neukum, Greg Kroah-Hartman, stable, linux-usb
On 19.11.26 13:28, Alexandre Becholey wrote:
> On Thursday, November 19th, 2026 at 1:02 PM, Oliver Neukum <oneukum@suse.com> wrote:
>
>> On 19.11.26 09:45, Alexandre Becholey wrote:
>>
>>> The special case is not needed: for the error handler's own REQUEST
>>> SENSE, scsi_eh_prep_cmnd() already points the data buffer at
>>> sense_buffer with at most SCSI_SENSE_BUFFERSIZE. Handle REQUEST SENSE
>>> like every other data-in command.
>> Hi,
>>
>> thank you for looking at this, but this fix seems to not take an
>> approach I can approve. Where sense_buffer is pointed to is
>> internal to the SCSI layer. We cannot use that in an LLD.
>>
>> If you think that LLDs should not use sense_buffer then the clean
>> solution is to remove it from struct scsi_cmnd.
>>
>> Regards
>> Oliver
>>
>> NACKED-BY: Oliver Neukum <oneukum@suse.com>
>>
> Hi Oliver,
>
> Would you take a v2 that keeps receiving the REQUEST SENSE data into
> sense_buffer and only limits the transfer to SCSI_SENSE_BUFFERSIZE?
Hi,
sure that would be fine.
Yet, I am afraid I have to ask why you want to solve this issue
at the level of the LLD? This looks like a generic issue to me.
The size of sense_buffer is documented. I don't see how this
issue can be limited to microtek.
Regards
Oliver
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer
2026-11-19 12:51 ` Oliver Neukum
@ 2026-11-20 15:14 ` Alexandre Becholey
0 siblings, 0 replies; 6+ messages in thread
From: Alexandre Becholey @ 2026-11-20 15:14 UTC (permalink / raw)
To: Oliver Neukum; +Cc: Oliver Neukum, Greg Kroah-Hartman, stable, linux-usb
On Thursday, November 19th, 2026 at 1:51 PM, Oliver Neukum <oneukum@suse.com> wrote:
>
>
> On 19.11.26 13:28, Alexandre Becholey wrote:
> > On Thursday, November 19th, 2026 at 1:02 PM, Oliver Neukum <oneukum@suse.com> wrote:
> >
> >> On 19.11.26 09:45, Alexandre Becholey wrote:
> >>
> >>> The special case is not needed: for the error handler's own REQUEST
> >>> SENSE, scsi_eh_prep_cmnd() already points the data buffer at
> >>> sense_buffer with at most SCSI_SENSE_BUFFERSIZE. Handle REQUEST SENSE
> >>> like every other data-in command.
> >> Hi,
> >>
> >> thank you for looking at this, but this fix seems to not take an
> >> approach I can approve. Where sense_buffer is pointed to is
> >> internal to the SCSI layer. We cannot use that in an LLD.
> >>
> >> If you think that LLDs should not use sense_buffer then the clean
> >> solution is to remove it from struct scsi_cmnd.
> >>
> >> Regards
> >> Oliver
> >>
> >> NACKED-BY: Oliver Neukum <oneukum@suse.com>
> >>
> > Hi Oliver,
> >
> > Would you take a v2 that keeps receiving the REQUEST SENSE data into
> > sense_buffer and only limits the transfer to SCSI_SENSE_BUFFERSIZE?
>
> Hi,
>
> sure that would be fine.
Thanks, v2 will do that.
>
> Yet, I am afraid I have to ask why you want to solve this issue
> at the level of the LLD? This looks like a generic issue to me.
> The size of sense_buffer is documented. I don't see how this
> issue can be limited to microtek.
>
> Regards
> Oliver
>
>
The overflow comes from microtek receiving the data phase of a
REQUEST SENSE command into sense_buffer with the length of the
command's own data buffer. That length is chosen by whoever sends
the command (sg allows REQUEST SENSE on a read-only node) and has
nothing to do with sense_buffer, so I don't think the core can see
the problem.
I looked at the other drivers for that special-case REQUEST SENSE
(in drivers/scsi, drivers/ata and drivers/usb). They either
transfer its data into the command's data buffer like any other
data-in command, or emulate it there. I found no other driver that
copies a caller-sized transfer into sense_buffer, so I think the
fix belongs in microtek. If you see a place where the core should
enforce the size, I am happy to look at it.
Kind regards,
Alexandre
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2026-11-20 15:14 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-19 8:45 [PATCH] usb: image: microtek: don't receive REQUEST SENSE data into sense_buffer Alexandre Becholey
2026-11-19 8:49 ` sashiko-bot
2026-11-19 12:02 ` Oliver Neukum
2026-11-19 12:28 ` Alexandre Becholey
2026-11-19 12:51 ` Oliver Neukum
2026-11-20 15:14 ` Alexandre Becholey
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.