Re: [PATCH v5 11/11] loop: Add LOOP_CONFIGURE ioctl

[Kristian Klausen]

On 13.05.2020 15.38, Martijn Coenen wrote:
> This allows userspace to completely setup a loop device with a single
> ioctl, removing the in-between state where the device can be partially
> configured - eg the loop device has a backing file associated with it,
> but is reading from the wrong offset.
>
> Besides removing the intermediate state, another big benefit of this
> ioctl is that LOOP_SET_STATUS can be slow; the main reason for this
> slowness is that LOOP_SET_STATUS(64) calls blk_mq_freeze_queue() to
> freeze the associated queue; this requires waiting for RCU
> synchronization, which I've measured can take about 15-20ms on this
> device on average.
>
> In addition to doing what LOOP_SET_STATUS can do, LOOP_CONFIGURE can
> also be used to:
> - Set the correct block size immediately by setting
>    loop_config.block_size (avoids LOOP_SET_BLOCK_SIZE)
> - Explicitly request direct I/O mode by setting LO_FLAGS_DIRECT_IO
>    in loop_config.info.lo_flags (avoids LOOP_SET_DIRECT_IO)
> - Explicitly request read-only mode by setting LO_FLAGS_READ_ONLY
>    in loop_config.info.lo_flags
>
> Here's setting up ~70 regular loop devices with an offset on an x86
> Android device, using LOOP_SET_FD and LOOP_SET_STATUS:
>
> vsoc_x86:/system/apex # time for i in `seq 30 100`;
> do losetup -r -o 4096 /dev/block/loop$i com.android.adbd.apex; done
>      0m03.40s real     0m00.02s user     0m00.03s system
>
> Here's configuring ~70 devices in the same way, but using a modified
> losetup that uses the new LOOP_CONFIGURE ioctl:
>
> vsoc_x86:/system/apex # time for i in `seq 30 100`;
> do losetup -r -o 4096 /dev/block/loop$i com.android.adbd.apex; done
>      0m01.94s real     0m00.01s user     0m00.01s system
>
> Reviewed-by: Christoph Hellwig <hch@xxxxxx>
> Signed-off-by: Martijn Coenen <maco@xxxxxxxxxxx>
> ---
>   drivers/block/loop.c      | 104 ++++++++++++++++++++++++++++----------
>   include/uapi/linux/loop.h |  21 ++++++++
>   2 files changed, 97 insertions(+), 28 deletions(-)
>
> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> index 13518ba191f5..a565c5aafa52 100644
> --- a/drivers/block/loop.c
> +++ b/drivers/block/loop.c
> @@ -228,6 +228,19 @@ static void __loop_update_dio(struct loop_device *lo, bool dio)
>   		blk_mq_unfreeze_queue(lo->lo_queue);
>   }
>   
> +/**
> + * loop_validate_block_size() - validates the passed in block size
> + * @bsize: size to validate
> + */
> +static int
> +loop_validate_block_size(unsigned short bsize)
> +{
> +	if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
> +		return -EINVAL;
> +
> +	return 0;
> +}
> +
>   /**
>    * loop_set_size() - sets device size and notifies userspace
>    * @lo: struct loop_device to set the size for
> @@ -1050,23 +1063,24 @@ loop_set_status_from_info(struct loop_device *lo,
>   	return 0;
>   }
>   
> -static int loop_set_fd(struct loop_device *lo, fmode_t mode,
> -		       struct block_device *bdev, unsigned int arg)
> +static int loop_configure(struct loop_device *lo, fmode_t mode,
> +			  struct block_device *bdev,
> +			  const struct loop_config *config)
>   {
>   	struct file	*file;
>   	struct inode	*inode;
>   	struct address_space *mapping;
>   	struct block_device *claimed_bdev = NULL;
> -	int		lo_flags = 0;
>   	int		error;
>   	loff_t		size;
>   	bool		partscan;
> +	unsigned short  bsize;
>   
>   	/* This is safe, since we have a reference from open(). */
>   	__module_get(THIS_MODULE);
>   
>   	error = -EBADF;
> -	file = fget(arg);
> +	file = fget(config->fd);
>   	if (!file)
>   		goto out;
>   
> @@ -1075,7 +1089,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
>   	 * here to avoid changing device under exclusive owner.
>   	 */
>   	if (!(mode & FMODE_EXCL)) {
> -		claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
> +		claimed_bdev = bd_start_claiming(bdev, loop_configure);
>   		if (IS_ERR(claimed_bdev)) {
>   			error = PTR_ERR(claimed_bdev);
>   			goto out_putf;
> @@ -1097,11 +1111,26 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
>   	mapping = file->f_mapping;
>   	inode = mapping->host;
>   
> +	size = get_loop_size(lo, file);
> +
> +	if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
> +		error = -EINVAL;
> +		goto out_unlock;
> +	}
> +
> +	if (config->block_size) {
> +		error = loop_validate_block_size(config->block_size);
> +		if (error)
> +			goto out_unlock;
> +	}
> +
> +	error = loop_set_status_from_info(lo, &config->info);

loop_set_status_from_info() doesn't call loop_config_discard() nor does 
loop_configure(), I assume it should be called?

I'm asking because I upgraded to util-linux 2.37 which use the new 
LOOP_CONFIGURE ioctl and fstrim is complaining that "the discard 
operation is not supported" and the missing call of 
loop_config_discard() stood out. Sadly I haven't been able to reproduce 
the issue reliably outside CI (yet).

> +	if (error)
> +		goto out_unlock;
> +
>   	if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
>   	    !file->f_op->write_iter)
> -		lo_flags |= LO_FLAGS_READ_ONLY;
> -
> -	size = get_loop_size(lo, file);
> +		lo->lo_flags |= LO_FLAGS_READ_ONLY;
>   
>   	error = loop_prepare_queue(lo);
>   	if (error)