diff mbox

[BUGFIX] block: add missing group association in bio_split

Message ID 1462567512-3007-1-git-send-email-paolo.valente@linaro.org
State New
Headers show

Commit Message

Paolo Valente May 6, 2016, 8:45 p.m. UTC
When a bio is split, the newly created bio must be associated with the
same blkcg as the original bio (if BLK_CGROUP is enabled). If this
operation is not performed, then the new bio is not associated with
any group, and the group of the current task is returned when the
group of the bio is requested.

Depending on the frequency of splits, this may cause a large
percentage of the bios belonging to a given group to be treated as if
belonging to other groups (in most cases as if belonging to the root
group). The expected group isolation may thereby be then broken.

This commit adds the missing association in bio_split.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>

---
 block/bio.c | 5 +++++
 1 file changed, 5 insertions(+)

-- 
1.9.1

Comments

Jeff Moyer May 9, 2016, 2:35 p.m. UTC | #1
Jeff Moyer <jmoyer@redhat.com> writes:

> Paolo Valente <paolo.valente@linaro.org> writes:

>

>> @@ -1811,6 +1811,11 @@ struct bio *bio_split(struct bio *bio, int sectors,

>>  

>>  	bio_advance(bio, split->bi_iter.bi_size);

>>  

>> +#ifdef CONFIG_BLK_CGROUP

>> +	if (bio->bi_css)

>> +		bio_associate_blkcg(split, bio->bi_css);

>> +#endif

>> +

>>  	return split;

>>  }

>>  EXPORT_SYMBOL(bio_split);

>

> Get rid of the #ifdefery.  This should be just:

>

> bio_associate_blkcg(split, bio->bi_css);


Gah, I see that the bi_css member is only present for CONFIG_BLK_CGROUP.
I guess we'll have to live with the ifdef.

-Jeff
Paolo Valente May 9, 2016, 2:39 p.m. UTC | #2
Il 09/05/2016 16:35, Jeff Moyer ha scritto:
> Jeff Moyer <jmoyer@redhat.com> writes:

>

>> Paolo Valente <paolo.valente@linaro.org> writes:

>>

>>> @@ -1811,6 +1811,11 @@ struct bio *bio_split(struct bio *bio, int sectors,

>>>

>>>   	bio_advance(bio, split->bi_iter.bi_size);

>>>

>>> +#ifdef CONFIG_BLK_CGROUP

>>> +	if (bio->bi_css)

>>> +		bio_associate_blkcg(split, bio->bi_css);

>>> +#endif

>>> +

>>>   	return split;

>>>   }

>>>   EXPORT_SYMBOL(bio_split);

>>

>> Get rid of the #ifdefery.  This should be just:

>>

>> bio_associate_blkcg(split, bio->bi_css);

>

> Gah, I see that the bi_css member is only present for CONFIG_BLK_CGROUP.

> I guess we'll have to live with the ifdef.

>


We have already tried to remove it, but it seems it would require other 
major changes.

Thanks,
Paolo

> -Jeff

>
Jens Axboe May 9, 2016, 5:58 p.m. UTC | #3
On 05/09/2016 08:56 AM, Mark Brown wrote:
> On Mon, May 09, 2016 at 04:39:04PM +0200, Paolo wrote:

>> Il 09/05/2016 16:35, Jeff Moyer ha scritto:

>

>>> Gah, I see that the bi_css member is only present for CONFIG_BLK_CGROUP.

>>> I guess we'll have to live with the ifdef.

>

>> We have already tried to remove it, but it seems it would require other

>> major changes.

>

> It probably should get fixed, but separately to the bug fix.


It's a minor tweak, might as well submit them together. Because if you 
don't, it has a tendency to be forgotten once the original issue is 
resolved.

-- 
Jens Axboe
diff mbox

Patch

diff --git a/block/bio.c b/block/bio.c
index 807d25e..c4a3834 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -1811,6 +1811,11 @@  struct bio *bio_split(struct bio *bio, int sectors,
 
 	bio_advance(bio, split->bi_iter.bi_size);
 
+#ifdef CONFIG_BLK_CGROUP
+	if (bio->bi_css)
+		bio_associate_blkcg(split, bio->bi_css);
+#endif
+
 	return split;
 }
 EXPORT_SYMBOL(bio_split);