loop device changes the block size and causes misaligned accesses to the real device, which can't be processed

The loop device advertises a block size of 1024 even when configured over a cdrom.

Quoted 2 lines · Ruben Garcia

The loop device advertises a block size of 1024 even when configured over a cdrom.

This bug and many other loop bugs are fixed in loop-AES package.

http://loop-aes.sourceforge.net/loop-AES/loop-AES-v2.0d.tar.bz2

The loop device advertises a block size of 1024 even when configured over a cdrom.

When burning a ext2 on a cd, and mounting it directly, I get:

blocksize=2048;

when I losetup /dev/loop0 /dev/cdrom, and then try to mount, I get:

blocksize=1024; and then misaligned transfer; this results in not being able to read the superblock.

The loop device should be changed to export the same blocksize of the underlying device

Quoted 11 lines · Ruben Garcia

The loop device advertises a block size of 1024 even when configured over a cdrom.

When burning a ext2 on a cd, and mounting it directly, I get:

blocksize=2048;

when I losetup /dev/loop0 /dev/cdrom, and then try to mount, I get:

blocksize=1024; and then misaligned transfer; this results in not being able to read the superblock.

The loop device should be changed to export the same blocksize of the underlying device

Huh, if you look at loop.c it appears to do that already (line 735) but it doesn't. This patch makes it so.

drivers/block/loop.c renamed from drivers/block/loop.c-orig +3 -2

@@ -732,8 +732,6 @@
 	mapping_set_gfp_mask(inode->i_mapping,
 			     lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
 
-	set_blocksize(bdev, lo_blocksize);
-
 	lo->lo_bio = lo->lo_biotail = NULL;
 
 	/*

@@ -749,6 +747,7 @@
 	if (S_ISBLK(inode->i_mode)) {
 		request_queue_t *q = bdev_get_queue(lo_device);
 
+		blk_queue_hardsect_size(lo->lo_queue, q->hardsect_size);
 		blk_queue_max_sectors(lo->lo_queue, q->max_sectors);
 		blk_queue_max_phys_segments(lo->lo_queue,q->max_phys_segments);
 		blk_queue_max_hw_segments(lo->lo_queue, q->max_hw_segments);

@@ -757,6 +756,8 @@
 		blk_queue_merge_bvec(lo->lo_queue, q->merge_bvec_fn);
 	}
 
+	set_blocksize(bdev, lo_blocksize);
+
 	kernel_thread(loop_thread, lo, CLONE_KERNEL);
 	down(&lo->lo_sem);
 

HTH,

Ok, Ben's patch will make the loop device work as any other device, and then ext2 will complain that the hard blocksize is bigger than the blocksize used for ext2 (in my example of 1k for ext2) and fail to mount it.

This is better than getting misaligned transfers et all, and is consistent with using the real device.

On the other hand, it is much more useful being able to actually mount the ext2 fs, and I managed to do that with the loop-aes patch (Thanks Jari Ruusu)

I confirm this bug closed. Thanks to all