New kernel warning triggered by blktests

New kernel warning triggered by blktests

[Bart Van Assche]

Hi Changheun,

If I run blktest srp/001 then a kernel warning appears that I haven't
seen before. I think this is a side-effect of the patch that limits the
bio size. Please take a look.

Thanks,

Bart.

WARNING: CPU: 1 PID: 15449 at block/bio.c:1034
__bio_iov_iter_get_pages+0x324/0x350
Call Trace:
 bio_iov_iter_get_pages+0x6c/0x360
 __blkdev_direct_IO_simple+0x291/0x580
 blkdev_direct_IO+0xb5/0xc0
 generic_file_direct_write+0x10d/0x290
 __generic_file_write_iter+0x120/0x290
 blkdev_write_iter+0x16e/0x280
 new_sync_write+0x268/0x380
 vfs_write+0x3e0/0x4f0
 ksys_write+0xd9/0x180
 __x64_sys_write+0x43/0x50
 do_syscall_64+0x32/0x80
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Re: New kernel warning triggered by blktests

[Jens Axboe]

On 4/23/21 2:38 PM, Bart Van Assche wrote:
> Hi Changheun,
> 
> If I run blktest srp/001 then a kernel warning appears that I haven't
> seen before. I think this is a side-effect of the patch that limits the
> bio size. Please take a look.
> 
> Thanks,
> 
> Bart.
> 
> WARNING: CPU: 1 PID: 15449 at block/bio.c:1034
> __bio_iov_iter_get_pages+0x324/0x350
> Call Trace:
>  bio_iov_iter_get_pages+0x6c/0x360
>  __blkdev_direct_IO_simple+0x291/0x580
>  blkdev_direct_IO+0xb5/0xc0
>  generic_file_direct_write+0x10d/0x290
>  __generic_file_write_iter+0x120/0x290
>  blkdev_write_iter+0x16e/0x280
>  new_sync_write+0x268/0x380
>  vfs_write+0x3e0/0x4f0
>  ksys_write+0xd9/0x180
>  __x64_sys_write+0x43/0x50
>  do_syscall_64+0x32/0x80
>  entry_SYSCALL_64_after_hwframe+0x44/0xae

I wonder if this is a case of violating "must always be able to add a page"?
Bart, does the below change anything?


diff --git a/include/linux/bio.h b/include/linux/bio.h
index f1a99f0a240c..c6428c9f9bf7 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -121,7 +121,8 @@ static inline bool bio_full(struct bio *bio, unsigned len)
 	if (bio->bi_vcnt >= bio->bi_max_vecs)
 		return true;
 
-	if (bio->bi_iter.bi_size > bio_max_size(bio) - len)
+	if (bio->bi_iter.bi_size &&
+	    bio->bi_iter.bi_size > bio_max_size(bio) - len)
 		return true;
 
 	return false;

-- 
Jens Axboe

Re: New kernel warning triggered by blktests

[Bart Van Assche]

On 4/23/21 2:43 PM, Jens Axboe wrote:
> I wonder if this is a case of violating "must always be able to add a page"?
> Bart, does the below change anything?
> 
> diff --git a/include/linux/bio.h b/include/linux/bio.h
> index f1a99f0a240c..c6428c9f9bf7 100644
> --- a/include/linux/bio.h
> +++ b/include/linux/bio.h
> @@ -121,7 +121,8 @@ static inline bool bio_full(struct bio *bio, unsigned len)
>  	if (bio->bi_vcnt >= bio->bi_max_vecs)
>  		return true;
>  
> -	if (bio->bi_iter.bi_size > bio_max_size(bio) - len)
> +	if (bio->bi_iter.bi_size &&
> +	    bio->bi_iter.bi_size > bio_max_size(bio) - len)
>  		return true;
>  
>  	return false;

Hi Jens,

Thank you for having taken a look. If I apply the following debug patch:

--- a/block/bio.c
+++ b/block/bio.c
@@ -1031,8 +1031,13 @@ static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
 			if (same_page)
 				put_page(page);
 		} else {
-			if (WARN_ON_ONCE(bio_full(bio, len)))
+			if (WARN_ON_ONCE(bio_full(bio, len))) {
+				pr_info("bi_vcnt %u/%u; bi_size %u/%u; len %u\n",
+					bio->bi_vcnt, bio->bi_max_vecs,
+					bio->bi_iter.bi_size, bio_max_size(bio),
+					len);
                                 return -EINVAL;
+			}
 			__bio_add_page(bio, page, len, offset);
 		}
 		offset = 0;

then the following output appears:

	bi_vcnt 12/256; bi_size 126976/130560; len 4096

so I don't think that the above patch would help.

What is remarkable is that test srp/001 does not submit any I/O towards the
block device associated with the SRP initiator (other than a partition table
read). I think this that the following command from tests/srp/rc triggers
the kernel warning:

		dd if=/dev/zero of="${r}" bs=1M count=$((ramdisk_size>>20)) "${oflag[@]}" >/dev/null 2>&1 || return $?

That dd command writes to a null_blk instance. After having added another
debug print statement, the following appeared in the kernel log:

Apr 23 19:08:04 ubuntu-vm kernel: null_blk: module loaded
Apr 23 19:08:04 ubuntu-vm kernel: blk_queue_max_hw_sectors: max_hw_sectors = 255; max_sectors = 255; bio_max_bytes = 130560
Apr 23 19:08:04 ubuntu-vm kernel: blk_queue_max_hw_sectors: max_hw_sectors = 255; max_sectors = 255; bio_max_bytes = 130560

That's the same 130560 byte limit as in the previous print statement.

Bart.