bio-integrity: use hardware sectors instead of block layer sectors

Smatch tagged this code as suspicious because we never use the "nr_sectors" variable. Looking at the code, I think we did intend to use "nr_sectors" instead of "sectors" when we call bio_integrity_mark_tail().

The difference between "sectors" and "nr_sectors" is that "sectors" is in terms of 512 byte sectors and "nr_sectors" is in terms of hardware sectors. They are only different for 4k sector devices.

---
I'm only guessing as to the intent and I can't test this myself. Please handle with care.

index 612a5c3..ce65453 100644
@@ -676,7 +676,7 @@ void bio_integrity_trim(struct bio *bio, unsigned int offset,
 	nr_sectors = bio_integrity_hw_sectors(bi, sectors);
 	bip->bip_sector = bip->bip_sector + offset;
 	bio_integrity_mark_head(bip, offset * bi->tuple_size);
-	bio_integrity_mark_tail(bip, sectors * bi->tuple_size);
+	bio_integrity_mark_tail(bip, nr_sectors * bi->tuple_size);
 }
 EXPORT_SYMBOL(bio_integrity_trim);
 

That code is so asking for the variable to be called "hw_sectors".

-- Jamie

Smatch tagged this code as suspicious because we never use the "nr_sectors" variable. Looking at the code, we did intend to use "nr_sectors" instead of "sectors" when we call bio_integrity_mark_tail().

The difference between "sectors" and "nr_sectors" is that "sectors" is in terms of 512 byte sectors and "nr_sectors" is in terms of hardware sectors. They are only different for 4k sector devices.

Also I changed the name because as Jamie Lokier points out, "that code is so asking for the variable to be called 'hw_sectors'."

index 612a5c3..d8cd1e2 100644
@@ -667,16 +667,16 @@ void bio_integrity_trim(struct bio *bio, unsigned int offset,
 {
 	struct bio_integrity_payload *bip = bio->bi_integrity;
 	struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
-	unsigned int nr_sectors;
+	unsigned int hw_sectors;
 
 	BUG_ON(bip = NULL);
 	BUG_ON(bi = NULL);
 	BUG_ON(!bio_flagged(bio, BIO_CLONED));
 
-	nr_sectors = bio_integrity_hw_sectors(bi, sectors);
+	hw_sectors = bio_integrity_hw_sectors(bi, sectors);
 	bip->bip_sector = bip->bip_sector + offset;
 	bio_integrity_mark_head(bip, offset * bi->tuple_size);
-	bio_integrity_mark_tail(bip, sectors * bi->tuple_size);
+	bio_integrity_mark_tail(bip, hw_sectors * bi->tuple_size);
 }
 EXPORT_SYMBOL(bio_integrity_trim);
 

Dan,

This one bitrotted in my mailbox for a while because it conflicted with something else I was working on and then it slipped through the cracks. Sorry about that.

Dan> The difference between "sectors" and "nr_sectors" is that "sectors" Dan> is in terms of 512 byte sectors and "nr_sectors" is in terms of Dan> hardware sectors. They are only different for 4k sector devices.

Dan> Also I changed the name because as Jamie Lokier points out, "that Dan> code is so asking for the variable to be called 'hw_sectors'."

The change is obviously functionally correct. But I object to the notion of hw_sectors. The 1:1 mapping of DIF tuples and logical blocks is even going away in SBC3. So let's not perpetuate that.

I have a patch in my queue that gets rid of all the hw_sector references in the integrity code. I'll make sure to include your fix.

So thanks for spotting this. I obviously haven't tested PI drives with a 4KB logical block size in combination with device mapper...

Martin K. Petersen	Oracle Linux Engineering

Hi,

I'm just curious: did anything come of this? Nowadays gcc 4.6 warns about the same unused nr_sector var smatch warned about, so new readers are coming to the same question of why this function doesn't use hardware sectors.