On Aug 9, 2013, Ed Cashin wrote:

On Aug 8, 2013, at 9:05 PM, Kent Overstreet wrote:
...

quoted

It's in the for-jens branch now.

Just examining the patches, I like the way it cleans up the aoe code.  I
had a question about a new BUG added by the for-jens branch the
read-response handling path of the aoe driver.

The aoe driver in linux-bcache/for-jens commit 4c36c973a8f45 is
passing my tests.

Here is a patch against that branch illustrating my suggestion for
handling bad target responses gracefully.


commit 2c39f50b1ee02e2ac07fd072a883a91713da53cc
Author: Ed Cashin [off-list ref]
Date:   Tue Aug 13 10:50:28 2013 -0400

    aoe: bad AoE responses fail I/O without BUG
    
    Instead of having a BUG when the AoE target does something wrong,
    just fail the I/O and log the problem with rate limiting.

diff --git a/drivers/block/aoe/aoecmd.c b/drivers/block/aoe/aoecmd.c
index cacd48e..b9916a6 100644
--- a/drivers/block/aoe/aoecmd.c
+++ b/drivers/block/aoe/aoecmd.c

@@ -1096,7 +1096,6 @@ bvcpy(struct sk_buff *skb, struct bio *bio, struct bvec_iter iter, long cnt)
 	int soff = 0;
 	struct bio_vec bv;
 
-	BUG_ON(cnt > iter.bi_size);
 	iter.bi_size = cnt;
 
 	__bio_for_each_segment(bv, bio, iter, iter) {

@@ -1196,6 +1195,14 @@ noskb:		if (buf)
 			clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
 			break;
 		}
+		if (n > f->iter.bi_size) {
+			pr_err_ratelimited("%s e%ld.%d.  bytes=%ld need=%u\n",
+				"aoe: too-large data size in read from",
+				(long) d->aoemajor, d->aoeminor,
+				n, f->iter.bi_size);
+			clear_bit(BIO_UPTODATE, &buf->bio->bi_flags);
+			break;
+		}
 		bvcpy(skb, f->buf->bio, f->iter, n);
 	case ATA_CMD_PIO_WRITE:
 	case ATA_CMD_PIO_WRITE_EXT: