[PATCH] btrfs-progs: volumes: check stripe end against device boundary

[PATCH] btrfs-progs: volumes: check stripe end against device boundary

[Qu Wenruo]

There is a report of kernel generationg READ bio access beyond device
boundary.

To make sure it's not caused by on-disk chunk items, add extra stripe
check against device total_bytes so that btrfs check can detect such
problem early on.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 volumes.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/volumes.c b/volumes.c
index 538f799e7211..0467dd63d5cb 100644
--- a/volumes.c
+++ b/volumes.c
@@ -1843,6 +1843,7 @@ int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
 	u64 type;
 	u32 chunk_ondisk_size;
 	u32 sectorsize = fs_info->sectorsize;
+	int i;
 
 	/*
 	 * Basic chunk item size check.  Note that btrfs_chunk already contains
@@ -1953,6 +1954,34 @@ int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
 		return -EIO;
 	}
 
+	/*
+	 * Stripe check against device boundary
+	 */
+	for (i = 0; i < num_stripes; i++) {
+		struct btrfs_device *dev;
+		u64 physical;
+		u64 devid;
+		u64 stripe_len;
+
+		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
+		physical = btrfs_stripe_offset_nr(leaf, chunk, i);
+		stripe_len = calc_stripe_length(type, length, num_stripes);
+		dev = btrfs_find_device(fs_info, devid, NULL, NULL);
+		/*
+		 * Device not found? Then we may be in the bootstrap process for
+		 * system chunks. Skip.
+		 */
+		if (!dev)
+			continue;
+		if (physical + stripe_len > dev->total_bytes) {
+			error(
+"chunk %llu stripe %d is beyond device boundary: devid %llu total_bytes %llu stripe start %llu len %llu",
+				logical, i, devid, dev->total_bytes, physical,
+				stripe_len);
+			return -EUCLEAN;
+		}
+	}
+
 	return 0;
 }
 
-- 
2.28.0

Re: [PATCH] btrfs-progs: volumes: check stripe end against device boundary

[Qu Wenruo]

On 2020/8/3 下午3:06, Qu Wenruo wrote:
> There is a report of kernel generationg READ bio access beyond device
> boundary.
> 
> To make sure it's not caused by on-disk chunk items, add extra stripe
> check against device total_bytes so that btrfs check can detect such
> problem early on.

To add more info on this, kernel can already detects such problem at
mount time, by verify_one_dev_extent() on each dev extent read and
compare it against existing chunks.

Thus there is no need to enhance kernel already.

Thanks,
Qu

> 
> Signed-off-by: Qu Wenruo <wqu@suse.com>
> ---
>  volumes.c | 29 +++++++++++++++++++++++++++++
>  1 file changed, 29 insertions(+)
> 
> diff --git a/volumes.c b/volumes.c
> index 538f799e7211..0467dd63d5cb 100644
> --- a/volumes.c
> +++ b/volumes.c
> @@ -1843,6 +1843,7 @@ int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
>  	u64 type;
>  	u32 chunk_ondisk_size;
>  	u32 sectorsize = fs_info->sectorsize;
> +	int i;
>  
>  	/*
>  	 * Basic chunk item size check.  Note that btrfs_chunk already contains
> @@ -1953,6 +1954,34 @@ int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
>  		return -EIO;
>  	}
>  
> +	/*
> +	 * Stripe check against device boundary
> +	 */
> +	for (i = 0; i < num_stripes; i++) {
> +		struct btrfs_device *dev;
> +		u64 physical;
> +		u64 devid;
> +		u64 stripe_len;
> +
> +		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
> +		physical = btrfs_stripe_offset_nr(leaf, chunk, i);
> +		stripe_len = calc_stripe_length(type, length, num_stripes);
> +		dev = btrfs_find_device(fs_info, devid, NULL, NULL);
> +		/*
> +		 * Device not found? Then we may be in the bootstrap process for
> +		 * system chunks. Skip.
> +		 */
> +		if (!dev)
> +			continue;
> +		if (physical + stripe_len > dev->total_bytes) {
> +			error(
> +"chunk %llu stripe %d is beyond device boundary: devid %llu total_bytes %llu stripe start %llu len %llu",
> +				logical, i, devid, dev->total_bytes, physical,
> +				stripe_len);
> +			return -EUCLEAN;
> +		}
> +	}
> +
>  	return 0;
>  }
>  
>