diff mbox series

block: only rewind to last split

Message ID 20180801093357.18417-1-yanaijie@huawei.com (mailing list archive)
State New, archived
Headers show
Series block: only rewind to last split | expand

Commit Message

Jason Yan Aug. 1, 2018, 9:33 a.m. UTC
Commit 63573e359d05 ("bio-integrity: Restore original iterator on verify
stage") made the bio-integrity verify function works fine when the bio
is smaller than the queue limit.

But since commit 54efd50bfd87 ("block: make generic_make_request handle
arbitrarily sized bios") the bio will split in block layer if the size
of bio is larger than the max length of the queue limit. This will make
the bio-integrity verify function do not work. We will see such error
log with splited bios:

[69163.386065] sda: ref tag error at location 0 (rcvd 1024)

Suppose we have a bio of size 1024k submitted and the driver can only
accept bio of max size 512k.

The original bio:

          bi_sector         1024k
     bio      +-------------------------------+

After the split we have two bios:

          bi_sector  512k
     split    +---------------+
                           bi_sector  512k
     bio      +---------------+---------------+

The new split bio is cloned from the first 512k of the original bio. And
the original bio is advanced to the next 512k(that is 512k~1024k). The
we submit these two requests to the driver.

After the these two bios are complete and we rewind the original bio in
bio_integrity_verify_fn(), the bio restored to the beginnig, but it does
not match the data we submitted to the driver:

          bi_sector         1024k
     bio      +-------------------------------+

When the split bio is complete the verify works fine. But when the whole
original bio is complete we just rewind the whole bio, which makes the
seed does not match the ref tag at all. So we see the above error log.

Fix this by clear the bi_iter.bi_done after split, so that we will only
rewind to the last sector we last start to split.

Fixes: 63573e359d05 ("bio-integrity: Restore original iterator on verify stage")
Signed-off-by: Jason Yan <yanaijie@huawei.com>
Reported-by: Zhang Liao <zhangliao2@huawei.com>
Tested-by: Li Chunjiang <lichunjiang2@huawei.com>
CC: Jens Axboe <axboe@kernel.dk>
CC: Kent Overstreet <kent.overstreet@gmail.com>
CC: Christoph Hellwig <hch@lst.de>
CC: Martin K. Petersen <martin.petersen@oracle.com>
CC: Miao Xie <miaoxie@huawei.com>
CC: Zhaohongjiang <zhaohongjiang@huawei.com>
---
 block/bio.c | 8 ++++++++
 1 file changed, 8 insertions(+)

Comments

Jason Yan Aug. 1, 2018, 9:32 a.m. UTC | #1
Sorry, please ignore this because I saw Greg Edwards's patch ("block: 
reset bi_iter.bi_done after splitting bio") already fixed this.

On 2018/8/1 17:33, Jason Yan wrote:
> Commit 63573e359d05 ("bio-integrity: Restore original iterator on verify
> stage") made the bio-integrity verify function works fine when the bio
> is smaller than the queue limit.
>
> But since commit 54efd50bfd87 ("block: make generic_make_request handle
> arbitrarily sized bios") the bio will split in block layer if the size
> of bio is larger than the max length of the queue limit. This will make
> the bio-integrity verify function do not work. We will see such error
> log with splited bios:
>
> [69163.386065] sda: ref tag error at location 0 (rcvd 1024)
>
> Suppose we have a bio of size 1024k submitted and the driver can only
> accept bio of max size 512k.
>
> The original bio:
>
>            bi_sector         1024k
>       bio      +-------------------------------+
>
> After the split we have two bios:
>
>            bi_sector  512k
>       split    +---------------+
>                             bi_sector  512k
>       bio      +---------------+---------------+
>
> The new split bio is cloned from the first 512k of the original bio. And
> the original bio is advanced to the next 512k(that is 512k~1024k). The
> we submit these two requests to the driver.
>
> After the these two bios are complete and we rewind the original bio in
> bio_integrity_verify_fn(), the bio restored to the beginnig, but it does
> not match the data we submitted to the driver:
>
>            bi_sector         1024k
>       bio      +-------------------------------+
>
> When the split bio is complete the verify works fine. But when the whole
> original bio is complete we just rewind the whole bio, which makes the
> seed does not match the ref tag at all. So we see the above error log.
>
> Fix this by clear the bi_iter.bi_done after split, so that we will only
> rewind to the last sector we last start to split.
>
> Fixes: 63573e359d05 ("bio-integrity: Restore original iterator on verify stage")
> Signed-off-by: Jason Yan <yanaijie@huawei.com>
> Reported-by: Zhang Liao <zhangliao2@huawei.com>
> Tested-by: Li Chunjiang <lichunjiang2@huawei.com>
> CC: Jens Axboe <axboe@kernel.dk>
> CC: Kent Overstreet <kent.overstreet@gmail.com>
> CC: Christoph Hellwig <hch@lst.de>
> CC: Martin K. Petersen <martin.petersen@oracle.com>
> CC: Miao Xie <miaoxie@huawei.com>
> CC: Zhaohongjiang <zhaohongjiang@huawei.com>
> ---
>   block/bio.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
>
> diff --git a/block/bio.c b/block/bio.c
> index 67eff5eddc49..afa4f408b9dc 100644
> --- a/block/bio.c
> +++ b/block/bio.c
> @@ -1867,6 +1867,14 @@ struct bio *bio_split(struct bio *bio, int sectors,
>
>   	bio_advance(bio, split->bi_iter.bi_size);
>
> +	/*
> +	 * Now bi_iter.bi_done is only used for bio-integrity to verify
> +	 * the data come up from the driver. If we don't clear bi_done
> +	 * here, the original bio will rewind to the beginning and the
> +	 * seed is error. Please check bio_integrity_verify_fn().
> +	 */
> +	bio->bi_iter.bi_done = 0;
> +
>   	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
>   		bio_set_flag(split, BIO_TRACE_COMPLETION);
>
>
diff mbox series

Patch

diff --git a/block/bio.c b/block/bio.c
index 67eff5eddc49..afa4f408b9dc 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -1867,6 +1867,14 @@  struct bio *bio_split(struct bio *bio, int sectors,
 
 	bio_advance(bio, split->bi_iter.bi_size);
 
+	/*
+	 * Now bi_iter.bi_done is only used for bio-integrity to verify
+	 * the data come up from the driver. If we don't clear bi_done
+	 * here, the original bio will rewind to the beginning and the
+	 * seed is error. Please check bio_integrity_verify_fn().
+	 */
+	bio->bi_iter.bi_done = 0;
+
 	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
 		bio_set_flag(split, BIO_TRACE_COMPLETION);