diff mbox

[Regression/Behavior,change] dm-flakey corrupt read bio, even the feature is drop_writes

Message ID 20160822150720.GA29950@redhat.com (mailing list archive)
State New, archived
Headers show

Commit Message

Mike Snitzer Aug. 22, 2016, 3:07 p.m. UTC
On Mon, Aug 22 2016 at  4:05am -0400,
Lukas Herbolt <lherbolt@redhat.com> wrote:

> Hello,
> 
> There is patch from Mike. It's part of current pull request to 4.8-rc1
> For more details check:
>  - https://www.redhat.com/archives/dm-devel/2016-July/msg00561.html
>  - https://www.redhat.com/archives/dm-devel/2016-August/msg00109.html
> 
> Lukas
> 
> On Mon, Aug 22, 2016 at 9:31 AM, Qu Wenruo <quwenruo@cn.fujitsu.com> wrote:
> > Hi, Mike and btrfs and dm guys
> >
> > When doing regression test on v4.8-rc1, we found that fstests/btrfs/056
> > always fails. With the following dmesg:
> > ---
> > Buffer I/O error on dev dm-0, logical block 1310704, async page read
> > Buffer I/O error on dev dm-0, logical block 16, async page read
> > Buffer I/O error on dev dm-0, logical block 16, async page read
> > ---
> >
> > And bisect leads to the following commits:
> > ---
> > commit 99f3c90d0d85708e7401a81ce3314e50bf7f2819
> > Author: Mike Snitzer <snitzer@redhat.com>
> > Date:   Fri Jul 29 13:19:55 2016 -0400
> >
> >     dm flakey: error READ bios during the down_interval
> > ---
> >
> > While according to the document of dm-flakey, it says that when using
> > drop_writes feature, read bios are not affected:
> > ---
> >   drop_writes:
> >         All write I/O is silently ignored.
> >         Read I/O is handled correctly.

I went back to the dm-flakey.c code at the time that the 'drop_writes'
feature was added via commit b26f5e3d.  It does confirm your
understanding of how reads should be handled if drop_writes is enabled.

Not sure why I thought differently.  Please try the following patch.

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Qu Wenruo Aug. 23, 2016, 8:40 a.m. UTC | #1
At 08/22/2016 11:07 PM, Mike Snitzer wrote:
> On Mon, Aug 22 2016 at  4:05am -0400,
> Lukas Herbolt <lherbolt@redhat.com> wrote:
>
>> Hello,
>>
>> There is patch from Mike. It's part of current pull request to 4.8-rc1
>> For more details check:
>>  - https://www.redhat.com/archives/dm-devel/2016-July/msg00561.html
>>  - https://www.redhat.com/archives/dm-devel/2016-August/msg00109.html
>>
>> Lukas
>>
>> On Mon, Aug 22, 2016 at 9:31 AM, Qu Wenruo <quwenruo@cn.fujitsu.com> wrote:
>>> Hi, Mike and btrfs and dm guys
>>>
>>> When doing regression test on v4.8-rc1, we found that fstests/btrfs/056
>>> always fails. With the following dmesg:
>>> ---
>>> Buffer I/O error on dev dm-0, logical block 1310704, async page read
>>> Buffer I/O error on dev dm-0, logical block 16, async page read
>>> Buffer I/O error on dev dm-0, logical block 16, async page read
>>> ---
>>>
>>> And bisect leads to the following commits:
>>> ---
>>> commit 99f3c90d0d85708e7401a81ce3314e50bf7f2819
>>> Author: Mike Snitzer <snitzer@redhat.com>
>>> Date:   Fri Jul 29 13:19:55 2016 -0400
>>>
>>>     dm flakey: error READ bios during the down_interval
>>> ---
>>>
>>> While according to the document of dm-flakey, it says that when using
>>> drop_writes feature, read bios are not affected:
>>> ---
>>>   drop_writes:
>>>         All write I/O is silently ignored.
>>>         Read I/O is handled correctly.
>
> I went back to the dm-flakey.c code at the time that the 'drop_writes'
> feature was added via commit b26f5e3d.  It does confirm your
> understanding of how reads should be handled if drop_writes is enabled.
>
> Not sure why I thought differently.  Please try the following patch.

In fact, I quite understand the idea to corrupt READ bio, if drop_writes 
and corrupt_bio_byte is given at the same time.

>
> diff --git a/drivers/md/dm-flakey.c b/drivers/md/dm-flakey.c
> index 97e446d..6a2e8dd 100644
> --- a/drivers/md/dm-flakey.c
> +++ b/drivers/md/dm-flakey.c
> @@ -289,15 +289,13 @@ static int flakey_map(struct dm_target *ti, struct bio *bio)
>  		pb->bio_submitted = true;
>
>  		/*
> -		 * Map reads as normal only if corrupt_bio_byte set.
> +		 * Error reads if neither corrupt_bio_byte or drop_writes are set.
> +		 * Otherwise, flakey_end_io() will decide if the reads should be modified.
>  		 */
>  		if (bio_data_dir(bio) == READ) {
> -			/* If flags were specified, only corrupt those that match. */
> -			if (fc->corrupt_bio_byte && (fc->corrupt_bio_rw == READ) &&
> -			    all_corrupt_bio_flags_match(bio, fc))
> -				goto map_bio;
> -			else
> +			if (!fc->corrupt_bio_byte && !test_bit(DROP_WRITES, &fc->flags))
>  				return -EIO;
> +			goto map_bio;
>  		}
>
>  		/*
> @@ -334,14 +332,21 @@ static int flakey_end_io(struct dm_target *ti, struct bio *bio, int error)
>  	struct flakey_c *fc = ti->private;
>  	struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
>
> -	/*
> -	 * Corrupt successful READs while in down state.
> -	 */
>  	if (!error && pb->bio_submitted && (bio_data_dir(bio) == READ)) {
> -		if (fc->corrupt_bio_byte)
> +		if (fc->corrupt_bio_byte && (fc->corrupt_bio_rw == READ) &&
> +		    all_corrupt_bio_flags_match(bio, fc)) {
> +			/*
> +			 * Corrupt successful matching READs while in down state.
> +			 */
>  			corrupt_bio_data(bio, fc);
> -		else
> +
> +		} else if (!test_bit(DROP_WRITES, &fc->flags)) {
> +			/*
> +			 * Error read during the down_interval if drop_writes
> +			 * wasn't configured.
> +			 */
>  			return -EIO;
> +		}
>  	}
>
>  	return error;
>
>
It looks good to me, as it can also handle corrupt_bio_byte and drop_writes.

Tested-by: Qu Wenruo <quwenruo@cn.fujitsu.com>

Thanks,
Qu


--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/md/dm-flakey.c b/drivers/md/dm-flakey.c
index 97e446d..6a2e8dd 100644
--- a/drivers/md/dm-flakey.c
+++ b/drivers/md/dm-flakey.c
@@ -289,15 +289,13 @@  static int flakey_map(struct dm_target *ti, struct bio *bio)
 		pb->bio_submitted = true;
 
 		/*
-		 * Map reads as normal only if corrupt_bio_byte set.
+		 * Error reads if neither corrupt_bio_byte or drop_writes are set.
+		 * Otherwise, flakey_end_io() will decide if the reads should be modified.
 		 */
 		if (bio_data_dir(bio) == READ) {
-			/* If flags were specified, only corrupt those that match. */
-			if (fc->corrupt_bio_byte && (fc->corrupt_bio_rw == READ) &&
-			    all_corrupt_bio_flags_match(bio, fc))
-				goto map_bio;
-			else
+			if (!fc->corrupt_bio_byte && !test_bit(DROP_WRITES, &fc->flags))
 				return -EIO;
+			goto map_bio;
 		}
 
 		/*
@@ -334,14 +332,21 @@  static int flakey_end_io(struct dm_target *ti, struct bio *bio, int error)
 	struct flakey_c *fc = ti->private;
 	struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
 
-	/*
-	 * Corrupt successful READs while in down state.
-	 */
 	if (!error && pb->bio_submitted && (bio_data_dir(bio) == READ)) {
-		if (fc->corrupt_bio_byte)
+		if (fc->corrupt_bio_byte && (fc->corrupt_bio_rw == READ) &&
+		    all_corrupt_bio_flags_match(bio, fc)) {
+			/*
+			 * Corrupt successful matching READs while in down state.
+			 */
 			corrupt_bio_data(bio, fc);
-		else
+
+		} else if (!test_bit(DROP_WRITES, &fc->flags)) {
+			/*
+			 * Error read during the down_interval if drop_writes
+			 * wasn't configured.
+			 */
 			return -EIO;
+		}
 	}
 
 	return error;