On Tue, Mar 11, 2025 at 09:09:08PM +0800, Li Wang wrote: > On Tue, Mar 11, 2025 at 11:33 AM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote: > > > On Tue, Mar 11, 2025 at 09:43:42AM +0800, Li Wang wrote: > > > On Mon, Mar 10, 2025 at 11:15 PM Luis Chamberlain <mcgrof@xxxxxxxxxx> > > wrote: > > > > > > > There's a fix for this already in next > > > > > > > > > > Oh? Which commit? > > > > Oh seems linux-next hasn't been updated in a few days, so you can try > > this patch: > > > > https://lore.kernel.org/all/20250307020403.3068567-1-mcgrof@xxxxxxxxxx/ > > > --- a/block/bdev.c > +++ b/block/bdev.c > @@ -181,6 +181,8 @@ EXPORT_SYMBOL(set_blocksize); > > int sb_set_blocksize(struct super_block *sb, int size) > { > + if (!(sb->s_type->fs_flags & FS_LBS) && size > PAGE_SIZE) > + return 0; > if (set_blocksize(sb->s_bdev_file, size)) > return 0; > ... > > > Thanks, but looking at the code change, seems filesystems with FS_LBS > (e.g., bcachefs, XFS) can still mount larger block sizes properly. IOW, > the test ioctl_loop06 still failed on RHEL9 (XFS) platform. The test does not fail because the filesystem being used, the test fails because it expects setting the block size > PAGE_SIZE will fail for the loop back device it is creating and then setting the block size for it. There are two tests which fail: * set block size > PAGE_SIZE with LOOP_SET_BLOCK_SIZE * set block size > PAGE_SIZE with LOOP_CONFIGURE It expects to fail. The new work enables the block layer to support block sizes > PAGE_SIZE on block devices, essentially that the logical or physical block size can be > PAGE_SIZE. That is supported now. > Is that expected? Or, should we adjust the testcase for FS with FS_LBS > as exception? Contrary to filesystems, block drivers use now implicitly use blk_validate_limits() through queue_limits_commit_update to validate queue limits, but PAGE_SIZE is no longer an issue. In the loop back driver case we have then: ioctl LOOP_CONFIGURE --> loop_configure() lim = queue_limits_start_update(lo->lo_queue); loop_update_limits(lo, &lim, config->block_size); ioctl LOOP_SET_BLOCK_SIZE --> loop_set_block_size() lim = queue_limits_start_update(lo->lo_queue); loop_update_limits(lo, &lim, arg); I don't see anything wrong with the loop driver supporting logical block sizes > PAGE_SIZE, from a quick look, but it does not matter. If tests exists which are verifying you *can't* do it, then we're bound to get other similar reports. queue_limits_commit_update() calls blk_validate_block_size. I just sent a fix. Thanks for the report. Luis
