Re: [Regression] md/raid1: write-intent logging/bitmap issue since fd3b6975e9c1 - v5.16-rc1

Re: [Regression] md/raid1: write-intent logging/bitmap issue since fd3b6975e9c1 - v5.16-rc1

[Linus Torvalds]

[-- Attachment #1: Type: text/plain, Size: 763 bytes --]

[ Jens wasn't cc'd for some reason but was the signer-off-on the patch
you bisected to. Added him to the cc. I'll bounce the original
separately, as I also don't see this on lore.kernel.org - it might not
have gotten there yet ]

On Mon, Jan 3, 2022 at 11:30 AM Norbert Warmuth <nwarmuth@t-online.de> wrote:
>
> Please verify and either revert or fixup fd3b6975e9c1 if my analysis is
> correct.

Can you check if moving the WriteMostly bit to the "do behind I/O?"
section fixes things for you?

IOW, something like the attached patch..

Warning: This is very much a "Money see, monkey do" patch. I'm not
really familiar with the raid1 code ]

But yeah, if you see corruption and there isn't an absolutely trivial
fix for this, we should revert.

           Linus

[-- Attachment #2: patch.diff --]
[-- Type: text/x-patch, Size: 816 bytes --]

 drivers/md/raid1.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index 7dc8026cf6ee..85505424f7a4 100644
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c
@@ -1496,12 +1496,13 @@ static void raid1_write_request(struct mddev *mddev, struct bio *bio,
 		if (!r1_bio->bios[i])
 			continue;
 
-		if (first_clone && test_bit(WriteMostly, &rdev->flags)) {
+		if (first_clone) {
 			/* do behind I/O ?
 			 * Not if there are too many, or cannot
 			 * allocate memory, or a reader on WriteMostly
 			 * is waiting for behind writes to flush */
 			if (bitmap &&
+			    test_bit(WriteMostly, &rdev->flags) &&
 			    (atomic_read(&bitmap->behind_writes)
 			     < mddev->bitmap_info.max_write_behind) &&
 			    !waitqueue_active(&bitmap->behind_wait)) {

Re: [Regression] md/raid1: write-intent logging/bitmap issue since fd3b6975e9c1 - v5.16-rc1

[Song Liu]

> On Jan 3, 2022, at 11:53 AM, Linus Torvalds <torvalds@linux-foundation.org> wrote:
> 
> [ Jens wasn't cc'd for some reason but was the signer-off-on the patch
> you bisected to. Added him to the cc. I'll bounce the original
> separately, as I also don't see this on lore.kernel.org - it might not
> have gotten there yet ]
> 
> On Mon, Jan 3, 2022 at 11:30 AM Norbert Warmuth <nwarmuth@t-online.de> wrote:
>> 
>> Please verify and either revert or fixup fd3b6975e9c1 if my analysis is
>> correct.
> 
> Can you check if moving the WriteMostly bit to the "do behind I/O?"
> section fixes things for you?
> 
> IOW, something like the attached patch..
> 
> Warning: This is very much a "Money see, monkey do" patch. I'm not
> really familiar with the raid1 code ]
> 
> But yeah, if you see corruption and there isn't an absolutely trivial
> fix for this, we should revert.
> 
>           Linus
> <patch.diff>

I can confirm this patch fixes the issue. I will send pull request with 
the fix shortly. 

Thanks,
Song

Re: [Regression] md/raid1: write-intent logging/bitmap issue since fd3b6975e9c1 - v5.16-rc1

[Jens Axboe]

On 1/3/22 11:53 AM, Linus Torvalds wrote:
> [ Jens wasn't cc'd for some reason but was the signer-off-on the patch
> you bisected to. Added him to the cc. I'll bounce the original
> separately, as I also don't see this on lore.kernel.org - it might not
> have gotten there yet ]

Thanks for adding me in, Song took care of it now. Will get sent out
shortly.

-- 
Jens Axboe