Re: Slab corruption in 2.6.16-rc5-mm2

[Linus Torvalds <torvalds@osdl.org>]

On Mon, 6 Mar 2006, Jesper Juhl wrote:
> 
> Ok, booting a plain 2.6.16-rc5-mm2 kernel with the above being the
> only change made results in this :

Yeah. I'm not surprised. A real mode-sense shouldn't be even 64 bytes, 
much less 96, so it shouldn't have overflowed, and we had no indication of 
the corruption spreading past the one allocation anyway.

It did/does seem a bug, though, so worth checking.

So onward in our tireless battle. Does this patch make any difference for 
you? It does two things:

 - it clears the "->sense" buffer in blk_end_sync_rq() (since it won't be 
   valid any more: the request is gone)
 - it adds a BUG_ON() if we appear to have already done the sense fill on 
   SCSI IO completion, and do it again.

Now, I've not tried either of these, and the BUG_ON() in particular might 
be a false positive itself, but it might be worth testing.

		Linus

---
diff --git a/block/ll_rw_blk.c b/block/ll_rw_blk.c
index 03d9c82..4351d34 100644
--- a/block/ll_rw_blk.c
+++ b/block/ll_rw_blk.c
@@ -2637,6 +2637,7 @@ void blk_end_sync_rq(struct request *rq,
 	struct completion *waiting = rq->waiting;
 
 	rq->waiting = NULL;
+	rq->sense = NULL;
 	__blk_put_request(rq->q, rq);
 
 	/*
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 701a328..2b60769 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -961,6 +961,10 @@ void scsi_io_completion(struct scsi_cmnd
 		if (result) {
 			clear_errors = 0;
 			if (sense_valid && req->sense) {
+
+				/* Have we already filled the sense buffer? */
+				BUG_ON(req->sense_len);
+
 				/*
 				 * SG_IO wants current and deferred errors
 				 */
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/