Re: [PATCH 02/24] compat: scsi: sg: fix v3 compat read/write interface

[Christoph Hellwig <hch@infradead.org>]

On Wed, Dec 11, 2019 at 09:42:36PM +0100, Arnd Bergmann wrote:
> In the v5.4 merge window, a cleanup patch from Al Viro conflicted
> with my rework of the compat handling for sg.c read(). Linus Torvalds
> did a correct merge but pointed out that the resulting code is still
> unsatisfactory.
> 
> I later noticed that the sg_new_read() function still gets the compat
> mode wrong, when the 'count' argument is large enough to pass a
> compat_sg_io_hdr object, but not a nativ sg_io_hdr.
> 
> To address both of these, move the definition of compat_sg_io_hdr
> into a scsi/sg.h to make it visible to sg.c and rewrite the logic
> for reading req_pack_id as well as the size check to a simpler
> version that gets the expected results.
> 
> Fixes: c35a5cfb4150 ("scsi: sg: sg_read(): simplify reading ->pack_id of userland sg_io_hdr_t")
> Fixes: 98aaaec4a150 ("compat_ioctl: reimplement SG_IO handling")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  block/scsi_ioctl.c |  29 +----------
>  drivers/scsi/sg.c  | 125 +++++++++++++++++++++------------------------
>  include/scsi/sg.h  |  30 +++++++++++
>  3 files changed, 89 insertions(+), 95 deletions(-)
> 
> diff --git a/block/scsi_ioctl.c b/block/scsi_ioctl.c
> index 650bade5ea5a..b61dbf4d8443 100644
> --- a/block/scsi_ioctl.c
> +++ b/block/scsi_ioctl.c
> @@ -20,6 +20,7 @@
>  #include <scsi/scsi.h>
>  #include <scsi/scsi_ioctl.h>
>  #include <scsi/scsi_cmnd.h>
> +#include <scsi/sg.h>
>  
>  struct blk_cmd_filter {
>  	unsigned long read_ok[BLK_SCSI_CMD_PER_LONG];
> @@ -550,34 +551,6 @@ static inline int blk_send_start_stop(struct request_queue *q,
>  	return __blk_send_generic(q, bd_disk, GPCMD_START_STOP_UNIT, data);
>  }
>  
> -#ifdef CONFIG_COMPAT
> -struct compat_sg_io_hdr {
> -	compat_int_t interface_id;	/* [i] 'S' for SCSI generic (required) */
> -	compat_int_t dxfer_direction;	/* [i] data transfer direction  */
> -	unsigned char cmd_len;		/* [i] SCSI command length ( <= 16 bytes) */
> -	unsigned char mx_sb_len;	/* [i] max length to write to sbp */
> -	unsigned short iovec_count;	/* [i] 0 implies no scatter gather */
> -	compat_uint_t dxfer_len;	/* [i] byte count of data transfer */
> -	compat_uint_t dxferp;		/* [i], [*io] points to data transfer memory
> -						or scatter gather list */
> -	compat_uptr_t cmdp;		/* [i], [*i] points to command to perform */
> -	compat_uptr_t sbp;		/* [i], [*o] points to sense_buffer memory */
> -	compat_uint_t timeout;		/* [i] MAX_UINT->no timeout (unit: millisec) */
> -	compat_uint_t flags;		/* [i] 0 -> default, see SG_FLAG... */
> -	compat_int_t pack_id;		/* [i->o] unused internally (normally) */
> -	compat_uptr_t usr_ptr;		/* [i->o] unused internally */
> -	unsigned char status;		/* [o] scsi status */
> -	unsigned char masked_status;	/* [o] shifted, masked scsi status */
> -	unsigned char msg_status;	/* [o] messaging level data (optional) */
> -	unsigned char sb_len_wr;	/* [o] byte count actually written to sbp */
> -	unsigned short host_status;	/* [o] errors from host adapter */
> -	unsigned short driver_status;	/* [o] errors from software driver */
> -	compat_int_t resid;		/* [o] dxfer_len - actual_transferred */
> -	compat_uint_t duration;		/* [o] time taken by cmd (unit: millisec) */
> -	compat_uint_t info;		/* [o] auxiliary information */
> -};
> -#endif
> -
>  int put_sg_io_hdr(const struct sg_io_hdr *hdr, void __user *argp)
>  {
>  #ifdef CONFIG_COMPAT
> diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c
> index 160748ad9c0f..985546aac236 100644
> --- a/drivers/scsi/sg.c
> +++ b/drivers/scsi/sg.c
> @@ -198,6 +198,7 @@ static void sg_device_destroy(struct kref *kref);
>  
>  #define SZ_SG_HEADER sizeof(struct sg_header)
>  #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
> +#define SZ_COMPAT_SG_IO_HDR sizeof(struct compat_sg_io_hdr)

I'd rather not add more defines like this.  The raw sizeof is
much more readable and obvious.

>  
> +	if (count < SZ_SG_HEADER)
> +		goto unknown_id;
> +
> +	/* negative reply_len means v3 format, otherwise v1/v2 */
> +	if (get_user(reply_len, &old_hdr->reply_len))
> +		return -EFAULT;
> +	if (reply_len >= 0)
> +		return get_user(*pack_id, &old_hdr->pack_id);
> +
> +	if (in_compat_syscall() && count >= SZ_COMPAT_SG_IO_HDR) {
> +		struct compat_sg_io_hdr __user *hp = buf;
> +		return get_user(*pack_id, &hp->pack_id);
> +	}
> +
> +	if (count >= SZ_SG_IO_HDR) {
> +		struct sg_io_hdr __user *hp = buf;
> +		return get_user(*pack_id, &hp->pack_id);
> +	}
> +
> +unknown_id:
> +	/* no valid header was passed, so ignore the pack_id */
> +	*pack_id = -1;
> +	return 0;
> +}

I find the structure here a little confusing, as it doesn't follow
the normal flow.  What do you think of:

	if (count >= SZ_SG_HEADER) {
		if (get_user(reply_len, &old_hdr->reply_len))
			return -EFAULT;

		/* negative reply_len means v3 format, otherwise v1/v2 */
		if (reply_len >= 0)
			return get_user(*pack_id, &old_hdr->pack_id);

		if (in_compat_syscall() {
			if (count >= SZ_COMPAT_SG_IO_HDR) {
				struct compat_sg_io_hdr __user *hp = buf;

				return get_user(*pack_id, &hp->pack_id);
			}
		} else {
			if (count >= SZ_SG_IO_HDR) {
				struct sg_io_hdr __user *hp = buf;

				return get_user(*pack_id, &hp->pack_id);
			}
		}
	}

	/* no valid header was passed, so ignore the pack_id */
	*pack_id = -1;
	return 0;
}