On Sat, Jul 2, 2011 at 12:59 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > On Sat, 2 Jul 2011, Andi Kleen wrote: > >> > The problem is that blk_peek_request() calls scsi_prep_fn(), which >> > does this: >> > >> > struct scsi_device *sdev = q->queuedata; >> > int ret = BLKPREP_KILL; >> > >> > if (req->cmd_type == REQ_TYPE_BLOCK_PC) >> > ret = scsi_setup_blk_pc_cmnd(sdev, req); >> > return scsi_prep_return(q, req, ret); >> > >> > It doesn't check to see if sdev is NULL, nor does >> > scsi_setup_blk_pc_cmnd(). That accounts for this error: >> >> I actually added a NULL check in scsi_setup_blk_pc_cmnd early on, >> but that just caused RCU CPU stalls afterwards and then eventually >> a hung system. > > The RCU problem is likely to be a separate issue. It might even be a > result of the use-after-free problem with the elevator. > > At any rate, it's clear that the crash in the refcounting log you > posted occurred because scsi_setup_blk_pc_cmnd() called > scsi_prep_state_check(), which tried to dereference the NULL pointer. > > Would you like to try this patch to see if it fixes the problem? As I > said before, I'm not certain it's the best thing to do, but it worked > on my system. > > Alan Stern > > > > > Index: usb-3.0/drivers/scsi/scsi_lib.c > =================================================================== > --- usb-3.0.orig/drivers/scsi/scsi_lib.c > +++ usb-3.0/drivers/scsi/scsi_lib.c > @@ -1247,6 +1247,8 @@ int scsi_prep_fn(struct request_queue *q > struct scsi_device *sdev = q->queuedata; > int ret = BLKPREP_KILL; > > + if (!sdev) > + return ret; > if (req->cmd_type == REQ_TYPE_BLOCK_PC) > ret = scsi_setup_blk_pc_cmnd(sdev, req); > return scsi_prep_return(q, req, ret); > Index: usb-3.0/drivers/scsi/scsi_sysfs.c > =================================================================== > --- usb-3.0.orig/drivers/scsi/scsi_sysfs.c > +++ usb-3.0/drivers/scsi/scsi_sysfs.c > @@ -322,6 +322,8 @@ static void scsi_device_dev_release_user > kfree(evt); > } > > + /* Freeing the queue signals to block that we're done */ > + scsi_free_queue(sdev->request_queue); > blk_put_queue(sdev->request_queue); > /* NULL queue means the device can't be used */ > sdev->request_queue = NULL; > @@ -936,8 +938,6 @@ void __scsi_remove_device(struct scsi_de > /* cause the request function to reject all I/O requests */ > sdev->request_queue->queuedata = NULL; > > - /* Freeing the queue signals to block that we're done */ > - scsi_free_queue(sdev->request_queue); > put_device(dev); > } This patch seems to resolve the block/scsi null-ptr de-references in our libsas/isci environment, we have yet to try James' alternative [1]. Do we potentially need both? Commit 86cbfb56 moved scsi_free_queue to __scsi_remove_device() but it seems only the "sdev->request_queue->queuedata = NULL" needed to be moved? The conversation appeared to be awaiting test results... [1]: http://marc.info/?l=linux-scsi&m=131007155700831&w=2 -- Dan -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
