queue-4.4/uas-fix-alignment-of-scatter-gather-segments.patch

   1 From 3ae62a42090f1ed48e2313ed256a1182a85fb575 Mon Sep 17 00:00:00 2001
   2 From: Oliver Neukum <oneukum@suse.com>
   3 Date: Tue, 30 Apr 2019 12:21:45 +0200
   4 Subject: UAS: fix alignment of scatter/gather segments
   5
   6 From: Oliver Neukum <oneukum@suse.com>
   7
   8 commit 3ae62a42090f1ed48e2313ed256a1182a85fb575 upstream.
   9
  10 This is the UAS version of
  11
  12 747668dbc061b3e62bc1982767a3a1f9815fcf0e
  13 usb-storage: Set virt_boundary_mask to avoid SG overflows
  14
  15 We are not as likely to be vulnerable as storage, as it is unlikelier
  16 that UAS is run over a controller without native support for SG,
  17 but the issue exists.
  18 The issue has been existing since the inception of the driver.
  19
  20 Fixes: 115bb1ffa54c ("USB: Add UAS driver")
  21 Signed-off-by: Oliver Neukum <oneukum@suse.com>
  22 Cc: stable <stable@vger.kernel.org>
  23 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  24
  25 ---
  26  drivers/usb/storage/uas.c |   38 ++++++++++++++++++++++++--------------
  27  1 file changed, 24 insertions(+), 14 deletions(-)
  28
  29 --- a/drivers/usb/storage/uas.c
  30 +++ b/drivers/usb/storage/uas.c
  31 @@ -772,23 +772,33 @@ static int uas_slave_alloc(struct scsi_d
  32  {
  33         struct uas_dev_info *devinfo =
  34                 (struct uas_dev_info *)sdev->host->hostdata;
  35 +       int maxp;
  36
  37         sdev->hostdata = devinfo;
  38
  39 -       /* USB has unusual DMA-alignment requirements: Although the
  40 -        * starting address of each scatter-gather element doesn't matter,
  41 -        * the length of each element except the last must be divisible
  42 -        * by the Bulk maxpacket value.  There's currently no way to
  43 -        * express this by block-layer constraints, so we'll cop out
  44 -        * and simply require addresses to be aligned at 512-byte
  45 -        * boundaries.  This is okay since most block I/O involves
  46 -        * hardware sectors that are multiples of 512 bytes in length,
  47 -        * and since host controllers up through USB 2.0 have maxpacket
  48 -        * values no larger than 512.
  49 -        *
  50 -        * But it doesn't suffice for Wireless USB, where Bulk maxpacket
  51 -        * values can be as large as 2048.  To make that work properly
  52 -        * will require changes to the block layer.
  53 +       /*
  54 +        * We have two requirements here. We must satisfy the requirements
  55 +        * of the physical HC and the demands of the protocol, as we
  56 +        * definitely want no additional memory allocation in this path
  57 +        * ruling out using bounce buffers.
  58 +        *
  59 +        * For a transmission on USB to continue we must never send
  60 +        * a package that is smaller than maxpacket. Hence the length of each
  61 +         * scatterlist element except the last must be divisible by the
  62 +         * Bulk maxpacket value.
  63 +        * If the HC does not ensure that through SG,
  64 +        * the upper layer must do that. We must assume nothing
  65 +        * about the capabilities off the HC, so we use the most
  66 +        * pessimistic requirement.
  67 +        */
  68 +
  69 +       maxp = usb_maxpacket(devinfo->udev, devinfo->data_in_pipe, 0);
  70 +       blk_queue_virt_boundary(sdev->request_queue, maxp - 1);
  71 +
  72 +       /*
  73 +        * The protocol has no requirements on alignment in the strict sense.
  74 +        * Controllers may or may not have alignment restrictions.
  75 +        * As this is not exported, we use an extremely conservative guess.
  76          */
  77         blk_queue_update_dma_alignment(sdev->request_queue, (512 - 1));
  78