From: Jan Höppner <hoeppner@linux.ibm.com>
Subject: s390/dasd: Use correct lock while counting channel queue length
Patch-mainline: v6.4-rc6
Git-commit: ccc45cb4e7271c74dbb27776ae8f73d84557f5c6
References: LTC#202775 bsc#1212443

Description:   s390/dasd: Use correct lock while counting channel queue length
Symptom:       Endless list iterations and eventual system hang.
Problem:       Instead of dasd_device lock, the dasd_block lock is used. Due to
               this, accessing the device request queue is unprotected and a
               simple iteration to count the length can be endless. This in
               turn will lead to a hang since the block lock is never released.
Solution:      Correctly use the device lock via get_ccwdev_lock(base->cdev).
Reproduction:  During I/O on a DASD disk, issue the dasdinfo command that
               utilizes the BIODASDINFO ioctl. Depending on the timing and the
               amount of I/O the issue will occur sooner or later.
               Exmaple:
               In one shell issue:
                    while true; do
                        dd if=/dev/dasdc of=/dev/null bs=1M;
                        blockdev --flushbufs /dev/dasdc;
                    done
               In a second shell issue:
                    while true; do
                        dasdinfo -a -d /dev/dasdc &> /dev/null;
                    done
               Running 'dasdinfo' as described above in several shells might
               trigger the issue earlier.

Upstream-Description:

              s390/dasd: Use correct lock while counting channel queue length

              The lock around counting the channel queue length in the BIODASDINFO
              ioctl was incorrectly changed to the dasd_block->queue_lock with commit
              583d6535cb9d ("dasd: remove dead code"). This can lead to endless list
              iterations and a subsequent crash.

              The queue_lock is supposed to be used only for queue lists belonging to
              dasd_block. For dasd_device related queue lists the ccwdev lock must be
              used.

              Fix the mentioned issues by correctly using the ccwdev lock instead of
              the queue lock.

              Fixes: 583d6535cb9d ("dasd: remove dead code")
              Cc: stable@vger.kernel.org # v5.0+
              Signed-off-by: Jan Höppner <hoeppner@linux.ibm.com>
              Reviewed-by: Stefan Haberland <sth@linux.ibm.com>
              Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
              Link: https://lore.kernel.org/r/20230609153750.1258763-2-sth@linux.ibm.com
              Signed-off-by: Jens Axboe <axboe@kernel.dk>


Signed-off-by: Jan Höppner <hoeppner@linux.ibm.com>
Acked-by: Miroslav Franc <mfranc@suse.cz>
---
 drivers/s390/block/dasd_ioctl.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- a/drivers/s390/block/dasd_ioctl.c
+++ b/drivers/s390/block/dasd_ioctl.c
@@ -463,10 +463,10 @@ static int dasd_ioctl_information(struct
 
 	memcpy(dasd_info->type, base->discipline->name, 4);
 
-	spin_lock_irqsave(&block->queue_lock, flags);
+	spin_lock_irqsave(get_ccwdev_lock(base->cdev), flags);
 	list_for_each(l, &base->ccw_queue)
 		dasd_info->chanq_len++;
-	spin_unlock_irqrestore(&block->queue_lock, flags);
+	spin_unlock_irqrestore(get_ccwdev_lock(base->cdev), flags);
 
 	rc = 0;
 	if (copy_to_user(argp, dasd_info,