* [PATCH] block: avoid sign extend problem with default queue flags mask
@ 2022-09-30 15:03 Brian Foster
2022-09-30 18:45 ` Joel Savitz
` (2 more replies)
0 siblings, 3 replies; 5+ messages in thread
From: Brian Foster @ 2022-09-30 15:03 UTC (permalink / raw)
To: linux-block; +Cc: Nico Pache, Joel Savitz
request_queue->queue_flags is an 8-byte field. Most queue flag
modifications occur through bit field helpers, but default flags can
be logically OR'd via the QUEUE_FLAG_MQ_DEFAULT mask. If this mask
happens to include bit 31, the assignment can sign extend the field
and set all upper 32 bits.
This exact problem has been observed on a downstream kernel that
happens to use bit 31 for QUEUE_FLAG_NOWAIT. This is not an
immediate problem for current upstream because bit 31 is not
included in the default flag assignment (and is not used at all,
actually). Regardless, fix up the QUEUE_FLAG_MQ_DEFAULT mask
definition to avoid the landmine in the future.
Signed-off-by: Brian Foster <bfoster@redhat.com>
---
Just to elaborate, I ran a quick test to change QUEUE_FLAG_NOWAIT to use
bit 31. With that change but without this patch, I see the following
queue state:
# cat /sys/kernel/debug/block/vda/state
SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63
And then with the patch applied:
# cat /sys/kernel/debug/block/vda/state
SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT
Thanks.
Brian
include/linux/blkdev.h | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 84b13fdd34a7..28c3037cb25c 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -580,9 +580,9 @@ struct request_queue {
#define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */
#define QUEUE_FLAG_SQ_SCHED 30 /* single queue style io dispatch */
-#define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
- (1 << QUEUE_FLAG_SAME_COMP) | \
- (1 << QUEUE_FLAG_NOWAIT))
+#define QUEUE_FLAG_MQ_DEFAULT ((1ULL << QUEUE_FLAG_IO_STAT) | \
+ (1ULL << QUEUE_FLAG_SAME_COMP) | \
+ (1ULL << QUEUE_FLAG_NOWAIT))
void blk_queue_flag_set(unsigned int flag, struct request_queue *q);
void blk_queue_flag_clear(unsigned int flag, struct request_queue *q);
--
2.37.2
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH] block: avoid sign extend problem with default queue flags mask
2022-09-30 15:03 [PATCH] block: avoid sign extend problem with default queue flags mask Brian Foster
@ 2022-09-30 18:45 ` Joel Savitz
2022-09-30 18:49 ` Nico Pache
2022-09-30 21:33 ` Jens Axboe
2 siblings, 0 replies; 5+ messages in thread
From: Joel Savitz @ 2022-09-30 18:45 UTC (permalink / raw)
To: Brian Foster; +Cc: linux-block, Nico Pache
On Fri, Sep 30, 2022 at 6:03 PM Brian Foster <bfoster@redhat.com> wrote:
>
> request_queue->queue_flags is an 8-byte field. Most queue flag
> modifications occur through bit field helpers, but default flags can
> be logically OR'd via the QUEUE_FLAG_MQ_DEFAULT mask. If this mask
> happens to include bit 31, the assignment can sign extend the field
> and set all upper 32 bits.
>
> This exact problem has been observed on a downstream kernel that
> happens to use bit 31 for QUEUE_FLAG_NOWAIT. This is not an
> immediate problem for current upstream because bit 31 is not
> included in the default flag assignment (and is not used at all,
> actually). Regardless, fix up the QUEUE_FLAG_MQ_DEFAULT mask
> definition to avoid the landmine in the future.
>
> Signed-off-by: Brian Foster <bfoster@redhat.com>
> ---
>
> Just to elaborate, I ran a quick test to change QUEUE_FLAG_NOWAIT to use
> bit 31. With that change but without this patch, I see the following
> queue state:
>
> # cat /sys/kernel/debug/block/vda/state
> SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63
>
> And then with the patch applied:
>
> # cat /sys/kernel/debug/block/vda/state
> SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT
>
> Thanks.
>
> Brian
>
> include/linux/blkdev.h | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 84b13fdd34a7..28c3037cb25c 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -580,9 +580,9 @@ struct request_queue {
> #define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */
> #define QUEUE_FLAG_SQ_SCHED 30 /* single queue style io dispatch */
>
> -#define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
> - (1 << QUEUE_FLAG_SAME_COMP) | \
> - (1 << QUEUE_FLAG_NOWAIT))
> +#define QUEUE_FLAG_MQ_DEFAULT ((1ULL << QUEUE_FLAG_IO_STAT) | \
> + (1ULL << QUEUE_FLAG_SAME_COMP) | \
> + (1ULL << QUEUE_FLAG_NOWAIT))
>
> void blk_queue_flag_set(unsigned int flag, struct request_queue *q);
> void blk_queue_flag_clear(unsigned int flag, struct request_queue *q);
> --
> 2.37.2
>
Tested-by: Joel Savitz <jsavitz@redhat.com>
Reviewed-by: Joel Savitz <jsavitz@redhat.com>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] block: avoid sign extend problem with default queue flags mask
2022-09-30 15:03 [PATCH] block: avoid sign extend problem with default queue flags mask Brian Foster
2022-09-30 18:45 ` Joel Savitz
@ 2022-09-30 18:49 ` Nico Pache
2022-09-30 21:33 ` Jens Axboe
2 siblings, 0 replies; 5+ messages in thread
From: Nico Pache @ 2022-09-30 18:49 UTC (permalink / raw)
To: Brian Foster; +Cc: linux-block, Joel Savitz
On Fri, Sep 30, 2022 at 9:03 AM Brian Foster <bfoster@redhat.com> wrote:
>
> request_queue->queue_flags is an 8-byte field. Most queue flag
> modifications occur through bit field helpers, but default flags can
> be logically OR'd via the QUEUE_FLAG_MQ_DEFAULT mask. If this mask
> happens to include bit 31, the assignment can sign extend the field
> and set all upper 32 bits.
>
> This exact problem has been observed on a downstream kernel that
> happens to use bit 31 for QUEUE_FLAG_NOWAIT. This is not an
> immediate problem for current upstream because bit 31 is not
> included in the default flag assignment (and is not used at all,
> actually). Regardless, fix up the QUEUE_FLAG_MQ_DEFAULT mask
> definition to avoid the landmine in the future.
>
> Signed-off-by: Brian Foster <bfoster@redhat.com>
> ---
>
> Just to elaborate, I ran a quick test to change QUEUE_FLAG_NOWAIT to use
> bit 31. With that change but without this patch, I see the following
> queue state:
>
> # cat /sys/kernel/debug/block/vda/state
> SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63
>
> And then with the patch applied:
>
> # cat /sys/kernel/debug/block/vda/state
> SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT
>
> Thanks.
>
> Brian
>
> include/linux/blkdev.h | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 84b13fdd34a7..28c3037cb25c 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -580,9 +580,9 @@ struct request_queue {
> #define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */
> #define QUEUE_FLAG_SQ_SCHED 30 /* single queue style io dispatch */
>
> -#define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
> - (1 << QUEUE_FLAG_SAME_COMP) | \
> - (1 << QUEUE_FLAG_NOWAIT))
> +#define QUEUE_FLAG_MQ_DEFAULT ((1ULL << QUEUE_FLAG_IO_STAT) | \
> + (1ULL << QUEUE_FLAG_SAME_COMP) | \
> + (1ULL << QUEUE_FLAG_NOWAIT))
>
> void blk_queue_flag_set(unsigned int flag, struct request_queue *q);
> void blk_queue_flag_clear(unsigned int flag, struct request_queue *q);
> --
> 2.37.2
>
Looks good, thanks Brian!
Acked-by: Nico Pache <npache@redhat.com>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] block: avoid sign extend problem with default queue flags mask
2022-09-30 15:03 [PATCH] block: avoid sign extend problem with default queue flags mask Brian Foster
2022-09-30 18:45 ` Joel Savitz
2022-09-30 18:49 ` Nico Pache
@ 2022-09-30 21:33 ` Jens Axboe
2022-10-03 12:52 ` Brian Foster
2 siblings, 1 reply; 5+ messages in thread
From: Jens Axboe @ 2022-09-30 21:33 UTC (permalink / raw)
To: Brian Foster, linux-block; +Cc: Nico Pache, Joel Savitz
On 9/30/22 9:03 AM, Brian Foster wrote:
> request_queue->queue_flags is an 8-byte field. Most queue flag
> modifications occur through bit field helpers, but default flags can
> be logically OR'd via the QUEUE_FLAG_MQ_DEFAULT mask. If this mask
> happens to include bit 31, the assignment can sign extend the field
> and set all upper 32 bits.
>
> This exact problem has been observed on a downstream kernel that
> happens to use bit 31 for QUEUE_FLAG_NOWAIT. This is not an
> immediate problem for current upstream because bit 31 is not
> included in the default flag assignment (and is not used at all,
> actually). Regardless, fix up the QUEUE_FLAG_MQ_DEFAULT mask
> definition to avoid the landmine in the future.
>
> Signed-off-by: Brian Foster <bfoster@redhat.com>
> ---
>
> Just to elaborate, I ran a quick test to change QUEUE_FLAG_NOWAIT to use
> bit 31. With that change but without this patch, I see the following
> queue state:
>
> # cat /sys/kernel/debug/block/vda/state
> SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63
>
> And then with the patch applied:
>
> # cat /sys/kernel/debug/block/vda/state
> SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT
>
> Thanks.
>
> Brian
>
> include/linux/blkdev.h | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 84b13fdd34a7..28c3037cb25c 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -580,9 +580,9 @@ struct request_queue {
> #define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */
> #define QUEUE_FLAG_SQ_SCHED 30 /* single queue style io dispatch */
>
> -#define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
> - (1 << QUEUE_FLAG_SAME_COMP) | \
> - (1 << QUEUE_FLAG_NOWAIT))
> +#define QUEUE_FLAG_MQ_DEFAULT ((1ULL << QUEUE_FLAG_IO_STAT) | \
> + (1ULL << QUEUE_FLAG_SAME_COMP) | \
> + (1ULL << QUEUE_FLAG_NOWAIT))
Shouldn't this just be 1UL << foo? The queue_flags are not 8-bytes,
they are unsigned long. That happens to be 8-bytes on 64-bit archs,
but it's 4-bytes on 32-bit archs.
--
Jens Axboe
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] block: avoid sign extend problem with default queue flags mask
2022-09-30 21:33 ` Jens Axboe
@ 2022-10-03 12:52 ` Brian Foster
0 siblings, 0 replies; 5+ messages in thread
From: Brian Foster @ 2022-10-03 12:52 UTC (permalink / raw)
To: Jens Axboe; +Cc: linux-block, Nico Pache, Joel Savitz
On Fri, Sep 30, 2022 at 03:33:29PM -0600, Jens Axboe wrote:
> On 9/30/22 9:03 AM, Brian Foster wrote:
> > request_queue->queue_flags is an 8-byte field. Most queue flag
> > modifications occur through bit field helpers, but default flags can
> > be logically OR'd via the QUEUE_FLAG_MQ_DEFAULT mask. If this mask
> > happens to include bit 31, the assignment can sign extend the field
> > and set all upper 32 bits.
> >
> > This exact problem has been observed on a downstream kernel that
> > happens to use bit 31 for QUEUE_FLAG_NOWAIT. This is not an
> > immediate problem for current upstream because bit 31 is not
> > included in the default flag assignment (and is not used at all,
> > actually). Regardless, fix up the QUEUE_FLAG_MQ_DEFAULT mask
> > definition to avoid the landmine in the future.
> >
> > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > ---
> >
> > Just to elaborate, I ran a quick test to change QUEUE_FLAG_NOWAIT to use
> > bit 31. With that change but without this patch, I see the following
> > queue state:
> >
> > # cat /sys/kernel/debug/block/vda/state
> > SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63
> >
> > And then with the patch applied:
> >
> > # cat /sys/kernel/debug/block/vda/state
> > SAME_COMP|IO_STAT|INIT_DONE|WC|STATS|REGISTERED|30|NOWAIT
> >
> > Thanks.
> >
> > Brian
> >
> > include/linux/blkdev.h | 6 +++---
> > 1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> > index 84b13fdd34a7..28c3037cb25c 100644
> > --- a/include/linux/blkdev.h
> > +++ b/include/linux/blkdev.h
> > @@ -580,9 +580,9 @@ struct request_queue {
> > #define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */
> > #define QUEUE_FLAG_SQ_SCHED 30 /* single queue style io dispatch */
> >
> > -#define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \
> > - (1 << QUEUE_FLAG_SAME_COMP) | \
> > - (1 << QUEUE_FLAG_NOWAIT))
> > +#define QUEUE_FLAG_MQ_DEFAULT ((1ULL << QUEUE_FLAG_IO_STAT) | \
> > + (1ULL << QUEUE_FLAG_SAME_COMP) | \
> > + (1ULL << QUEUE_FLAG_NOWAIT))
>
> Shouldn't this just be 1UL << foo? The queue_flags are not 8-bytes,
> they are unsigned long. That happens to be 8-bytes on 64-bit archs,
> but it's 4-bytes on 32-bit archs.
>
Oops.. yes, that makes sense. I guess that means we shouldn't really
expect to see anything use the upper 32 bits. The extension still makes
the state output look wonky in the (1 << 31) case, so I'll send a v2
with that fixed..
Brian
> --
> Jens Axboe
>
>
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2022-10-03 12:52 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-09-30 15:03 [PATCH] block: avoid sign extend problem with default queue flags mask Brian Foster
2022-09-30 18:45 ` Joel Savitz
2022-09-30 18:49 ` Nico Pache
2022-09-30 21:33 ` Jens Axboe
2022-10-03 12:52 ` Brian Foster
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.