On 03/25/2014 06:04 PM, Olivier Bonvalet wrote:
> Le mardi 25 mars 2014 à 17:46 -0500, Alex Elder a écrit :
>> On 03/25/2014 05:17 PM, Olivier Bonvalet wrote:
>>>
>>> I now have this one very often (here 5 minutes after the host boot) :
>>
>> I am fairly sure this indicates a use-after-free scenario,
>> likely caused by something getting deleted before every
>> user was done with it.
>>
>> I believe Ilya is done for the night; I'm going to spend some
>> time looking at this to try to determine the cause. If you
>> are willing I'd love to have you try whatever fix I come up
>> with. I'd rather find a fix than just collect more information,
>> but I may need to get more, we'll see.
>>
>> Thank you for all your reports, they help a lot.
>>
>> -Alex
>
> Ok. I can apply some patch to help debug that yes.
> I will try to reproduce on a different host, without customer data.
>
> But I think I will stop here for the night too.
>
> Thanks for your time,
> Olivier
Here's something that will provide a few more pieces of
information. If you're around and you're able to try it
out it might confirm something had likely been destroyed.
I'll keep sending stuff as I come up with it (even though
I realize you may not be around until morning).
-Alex
Index: b/drivers/block/rbd.c
===================================================================
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -2132,6 +2132,35 @@ static void rbd_img_obj_callback(struct
spin_lock_irq(&img_request->completion_lock);
if (which > img_request->next_completion)
goto out;
+ if (which != img_request->next_completion) {
+ printk("%s: bad image object request information:\n", __func__);
+ printk("obj_request %p\n", obj_request);
+ printk(" ->object_name <%s>\n", obj_request->object_name);
+ printk(" ->offset %llu\n", obj_request->offset);
+ printk(" ->length %llu\n", obj_request->length);
+ printk(" ->type 0x%x\n", (u32)obj_request->type);
+ printk(" ->flags 0x%lx\n", obj_request->flags);
+ printk(" ->img_request %p\n", obj_request->img_request);
+ printk(" ->which %u\n", obj_request->which);
+ printk(" ->xferred %llu\n", obj_request->xferred);
+ printk(" ->result %d\n", obj_request->result);
+ printk(" ->kref %d\n", atomic_read(&obj_request->kref));
+
+ printk("img_request %p\n", img_request);
+ printk(" ->snap 0x%016llx\n", img_request->snap_id);
+ printk(" ->offset %llu\n", img_request->offset);
+ printk(" ->length %llu\n", img_request->length);
+ printk(" ->flags 0x%lx\n", img_request->flags);
+ printk(" ->obj_request_count %u\n",
+ img_request->obj_request_count);
+ printk(" ->next_completion %u\n",
+ img_request->next_completion);
+ printk(" ->xferred %llu\n", img_request->xferred);
+ printk(" ->result %d\n", img_request->result);
+ printk(" ->obj_requests head %p\n",
+ img_request->obj_requests.next);
+ printk(" ->kref %d\n", atomic_read(&img_request->kref));
+ }
rbd_assert(which == img_request->next_completion);
for_each_obj_request_from(img_request, obj_request) {
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
