On 29/10/2018 19:31, Manjunath Patil wrote:
info->nr_rings isn't adjusted in case of ENOMEM error from
negotiate_mq(). This leads to kernel panic in error path.
Typical call stack involving panic -
#8 page_fault at ffffffff8175936f
[exception RIP: blkif_free_ring+33]
RIP: ffffffffa0149491 RSP: ffff8804f7673c08 RFLAGS: 00010292
...
#9 blkif_free at ffffffffa0149aaa [xen_blkfront]
#10 talk_to_blkback at ffffffffa014c8cd [xen_blkfront]
#11 blkback_changed at ffffffffa014ea8b [xen_blkfront]
#12 xenbus_otherend_changed at ffffffff81424670
#13 backend_changed at ffffffff81426dc3
#14 xenwatch_thread at ffffffff81422f29
#15 kthread at ffffffff810abe6a
#16 ret_from_fork at ffffffff81754078
Though either of my changes avoid the panic, I included both the
changes. This issue got introduced with "7ed8ce1 xen-blkfront: move
negotiate_mq to cover all cases of new VBDs"
Please us the correct format for specifying another commit.
Signed-off-by: Manjunath Patil <manjunath.b.patil@xxxxxxxxxx>
Can you please add a "Fixes:" tag and "Cc: stable@xxxxxxxxxxxxxxx" ?
---
drivers/block/xen-blkfront.c | 7 +++++--
1 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 429d201..dc8fe25 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -1350,8 +1350,10 @@ static void blkif_free(struct blkfront_info *info, int
suspend)
if (info->rq)
blk_mq_stop_hw_queues(info->rq);
- for (i = 0; i < info->nr_rings; i++)
- blkif_free_ring(&info->rinfo[i]);
+ if (info->rinfo) {
+ for (i = 0; i < info->nr_rings; i++)
+ blkif_free_ring(&info->rinfo[i]);
+ }
I'd drop this change.
kfree(info->rinfo);
info->rinfo = NULL;
@@ -1919,6 +1921,7 @@ static int negotiate_mq(struct blkfront_info *info)
GFP_KERNEL);
if (!info->rinfo) {
xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info
structure");
+ info->nr_rings = 0;
return -ENOMEM;
}
Juergen
