git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From 26d56a9fcb2014b99e654127960aa0a48a391e3c Mon Sep 17 00:00:00 2001
   2 From: Owen Gu <guhuinan@xiaomi.com>
   3 Date: Thu, 20 Nov 2025 20:33:36 +0800
   4 Subject: usb: uas: fix urb unmapping issue when the uas device is remove during ongoing data transfer
   5
   6 From: Owen Gu <guhuinan@xiaomi.com>
   7
   8 commit 26d56a9fcb2014b99e654127960aa0a48a391e3c upstream.
   9
  10 When a UAS device is unplugged during data transfer, there is
  11 a probability of a system panic occurring. The root cause is
  12 an access to an invalid memory address during URB callback handling.
  13 Specifically, this happens when the dma_direct_unmap_sg() function
  14 is called within the usb_hcd_unmap_urb_for_dma() interface, but the
  15 sg->dma_address field is 0 and the sg data structure has already been
  16 freed.
  17
  18 The SCSI driver sends transfer commands by invoking uas_queuecommand_lck()
  19 in uas.c, using the uas_submit_urbs() function to submit requests to USB.
  20 Within the uas_submit_urbs() implementation, three URBs (sense_urb,
  21 data_urb, and cmd_urb) are sequentially submitted. Device removal may
  22 occur at any point during uas_submit_urbs execution, which may result
  23 in URB submission failure. However, some URBs might have been successfully
  24 submitted before the failure, and uas_submit_urbs will return the -ENODEV
  25 error code in this case. The current error handling directly calls
  26 scsi_done(). In the SCSI driver, this eventually triggers scsi_complete()
  27 to invoke scsi_end_request() for releasing the sgtable. The successfully
  28 submitted URBs, when being unlinked to giveback, call
  29 usb_hcd_unmap_urb_for_dma() in hcd.c, leading to exceptions during sg
  30 unmapping operations since the sg data structure has already been freed.
  31
  32 This patch modifies the error condition check in the uas_submit_urbs()
  33 function. When a UAS device is removed but one or more URBs have already
  34 been successfully submitted to USB, it avoids immediately invoking
  35 scsi_done() and save the cmnd to devinfo->cmnd array. If the successfully
  36 submitted URBs is completed before devinfo->resetting being set, then
  37 the scsi_done() function will be called within uas_try_complete() after
  38 all pending URB operations are finalized. Otherwise, the scsi_done()
  39 function will be called within uas_zap_pending(), which is executed after
  40 usb_kill_anchored_urbs().
  41
  42 The error handling only takes effect when uas_queuecommand_lck() calls
  43 uas_submit_urbs() and returns the error value -ENODEV . In this case,
  44 the device is disconnected, and the flow proceeds to uas_disconnect(),
  45 where uas_zap_pending() is invoked to call uas_try_complete().
  46
  47 Fixes: eb2a86ae8c54 ("USB: UAS: fix disconnect by unplugging a hub")
  48 Cc: stable <stable@kernel.org>
  49 Signed-off-by: Yu Chen <chenyu45@xiaomi.com>
  50 Signed-off-by: Owen Gu <guhuinan@xiaomi.com>
  51 Acked-by: Oliver Neukum <oneukum@suse.com>
  52 Link: https://patch.msgid.link/20251120123336.3328-1-guhuinan@xiaomi.com
  53 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  54 ---
  55  drivers/usb/storage/uas.c |    5 +++++
  56  1 file changed, 5 insertions(+)
  57
  58 --- a/drivers/usb/storage/uas.c
  59 +++ b/drivers/usb/storage/uas.c
  60 @@ -698,6 +698,10 @@ static int uas_queuecommand_lck(struct s
  61          * of queueing, no matter how fatal the error
  62          */
  63         if (err == -ENODEV) {
  64 +               if (cmdinfo->state & (COMMAND_INFLIGHT | DATA_IN_URB_INFLIGHT |
  65 +                               DATA_OUT_URB_INFLIGHT))
  66 +                       goto out;
  67 +
  68                 set_host_byte(cmnd, DID_NO_CONNECT);
  69                 scsi_done(cmnd);
  70                 goto zombie;
  71 @@ -711,6 +715,7 @@ static int uas_queuecommand_lck(struct s
  72                 uas_add_work(cmnd);
  73         }
  74
  75 +out:
  76         devinfo->cmnd[idx] = cmnd;
  77  zombie:
  78         spin_unlock_irqrestore(&devinfo->lock, flags);