Mike Snitzer writes:
submit_bio_noacct_add_head() in block device layer when we want to
split bio and send remaining back to itself.
Ordering aside, you cannot split more than once. So your proposed fix
to insert at head isn't valid because you're still implicitly allocating
more than one bio from the bioset which could cause deadlock in a low
memory situation.
I had to deal with a comparable issue with DM core not too long ago, see
this commit:
commit ee1dfad5325ff1cfb2239e564cd411b3bfe8667a
Author: Mike Snitzer <snitzer@xxxxxxxxxx>
Date: Mon Sep 14 13:04:19 2020 -0400
dm: fix bio splitting and its bio completion order for regular IO
dm_queue_split() is removed because __split_and_process_bio() _must_
handle splitting bios to ensure proper bio submission and completion
ordering as a bio is split.
Otherwise, multiple recursive calls to ->submit_bio will cause multiple
split bios to be allocated from the same ->bio_split mempool at the same
time. This would result in deadlock in low memory conditions because no
progress could be made (only one bio is available in ->bio_split
mempool).
This fix has been verified to still fix the loss of performance, due
to excess splitting, that commit 120c9257f5f1 provided.
Fixes: 120c9257f5f1 ("Revert "dm: always call blk_queue_split() in dm_process_bio()"")
Cc: stable@xxxxxxxxxxxxxxx # 5.0+, requires custom backport due to 5.9 changes
Reported-by: Ming Lei <ming.lei@xxxxxxxxxx>
Signed-off-by: Mike Snitzer <snitzer@xxxxxxxxxx>
Basically you cannot split the same bio more than once without
recursing. Your elaborate documentation shows things going wrong quite
early in step 3. That additional split and recursing back to MD
shouldn't happen before the first bio split completes.
Seems the proper fix is to disallow max_sectors_kb to be imposed, via
blk_queue_split(), if MD has further splitting constraints, via
chunk_sectors, that negate max_sectors_kb anyway.
Mike
Hi Mike,
I think you're right that a driver should not split the same bio more
than once without recursing when using the same mempool.
If a driver only split bio once, the out-of-order issue no longer exists.
(Therefore, this problem won't occur on DM device.)
But the MD devices are using their private bioset (mddev->bio_set
or conf->bio_split) for splitting by themselves that are not the same
bioset used in blk_queue_split() (i.e. q->bio_split). The deadlock
you have mentioned might not happen to them.
I think there are two solutions:
1. In case MD devices want to change to use q->bio_split someday
without this out-of-order issue, make them do split once would be
a solution.
2. If MD devices should split the bio twice, so we can separately handle
limits in blk_queue_split() and each raid level's (raid0, raid5,
raid1, ...).
I will try to find another solution in this case.
My proposal is not suitable after I reconsider the problem:
If a bio is split into A part and B part.
+------|------+
| A | B |
+------|------+
I think a driver should make sure A part is always handled before B
part.
Inserting bio at head of current->bio_list and submitting bio in the
same
time while handling A part could make bios generated from A part be
handled before B part. This broke the order of those bios that generated
form A part.
(Maybe I should find a way to make B part at the head of
bio_list_on_stack[1]
while submitting it...)
Thanks for your comments.
I will try to figure out a better way to fix it in the next version.
Best regards,
Danny Shih
