[PATCH 01/18] nvme: add missing unmaps in nvme_queue_rq

[PATCH 01/18] nvme: add missing unmaps in nvme_queue_rq

[Christoph Hellwig]

When we fail various metadata related operations in nvme_queue_rq we
need to unmap the data SGL.

Cc: stable@vger.kernel.org
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/nvme/host/pci.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 22d8375..2f05292 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -906,19 +906,28 @@ static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
 			goto retry_cmd;
 		}
 		if (blk_integrity_rq(req)) {
-			if (blk_rq_count_integrity_sg(req->q, req->bio) != 1)
+			if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) {
+				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
+						dma_dir);
 				goto error_cmd;
+			}
 
 			sg_init_table(iod->meta_sg, 1);
 			if (blk_rq_map_integrity_sg(
-					req->q, req->bio, iod->meta_sg) != 1)
+					req->q, req->bio, iod->meta_sg) != 1) {
+				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
+						dma_dir);
 				goto error_cmd;
+			}
 
 			if (rq_data_dir(req))
 				nvme_dif_remap(req, nvme_dif_prep);
 
-			if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir))
+			if (!dma_map_sg(nvmeq->q_dmadev, iod->meta_sg, 1, dma_dir)) {
+				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
+						dma_dir);
 				goto error_cmd;
+			}
 		}
 	}
 
-- 
1.9.1

Re: [PATCH 01/18] nvme: add missing unmaps in nvme_queue_rq

[Sagi Grimberg]

On 10/16/2015 8:58 AM, Christoph Hellwig wrote:
> When we fail various metadata related operations in nvme_queue_rq we
> need to unmap the data SGL.
>
> Cc: stable@vger.kernel.org
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>   drivers/nvme/host/pci.c | 15 ++++++++++++---
>   1 file changed, 12 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> index 22d8375..2f05292 100644
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -906,19 +906,28 @@ static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx,
>   			goto retry_cmd;
>   		}
>   		if (blk_integrity_rq(req)) {
> -			if (blk_rq_count_integrity_sg(req->q, req->bio) != 1)
> +			if (blk_rq_count_integrity_sg(req->q, req->bio) != 1) {
> +				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
> +						dma_dir);
>   				goto error_cmd;
> +			}
>
>   			sg_init_table(iod->meta_sg, 1);
>   			if (blk_rq_map_integrity_sg(
> -					req->q, req->bio, iod->meta_sg) != 1)
> +					req->q, req->bio, iod->meta_sg) != 1) {
> +				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
> +						dma_dir);
>   				goto error_cmd;
> +			}

This is not related to the patch itself. But this condition seems bogus
to me. We passed meta_sg that consists of a single entry. If we
happened to map more than a single entry we're already in trouble as we
overrun meta_sg (modified the iod->sg pointer).

I think a WARN_ON_ONCE statement is more suitable here (which should
probably come as a separate patch).

Other than that, looks good to me:

Reviewed-by: Sagi Grimberg <sagig@mellanox.com>

Re: [PATCH 01/18] nvme: add missing unmaps in nvme_queue_rq

[Busch, Keith]

On Tue, Oct 20, 2015 at 01:04:24PM +0300, Sagi Grimberg wrote:
> On 10/16/2015 8:58 AM, Christoph Hellwig wrote:
> >  			sg_init_table(iod->meta_sg, 1);
> >  			if (blk_rq_map_integrity_sg(
> >-					req->q, req->bio, iod->meta_sg) != 1)
> >+					req->q, req->bio, iod->meta_sg) != 1) {
> >+				dma_unmap_sg(dev->dev, iod->sg, iod->nents,
> >+						dma_dir);
> >  				goto error_cmd;
> >+			}
> 
> This is not related to the patch itself. But this condition seems bogus
> to me. We passed meta_sg that consists of a single entry. If we
> happened to map more than a single entry we're already in trouble as we
> overrun meta_sg (modified the iod->sg pointer).
> 
> I think a WARN_ON_ONCE statement is more suitable here (which should
> probably come as a separate patch).

We are in trouble if it maps more than 1, but I think the condition
here is intended to guard against 0 rather than > 1. We should already
be ensured it won't be > 1 from a previous check.

Based on the implementation of blk_rq_map_integrity_sg and the functions
earlier setup, I don't think we can ever see 0 here either.