NVMe: Fix memory leak on retried commands

Resources are reallocated for requeued commands, so unmap and release
the iod for the failed command.

It's a pretty bad memory leak and causes a kernel hang if you remove a
drive because of a busy dma pool. You'll get messages spewing like this:

  nvme 0000:xx:xx.x: dma_pool_destroy prp list 256, ffff880420dec000 busy

and lock up pci and the driver since removal never completes while
holding a lock.

Cc: [email protected]
Cc: <[email protected]> # 4.0.x-
Signed-off-by: Keith Busch <[email protected]>
Reviewed-by: Christoph Hellwig <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

Author: Keith Busch

drivers/block/nvme-core.c

@@ -604,19 +604,21 @@ static void req_completion(struct nvme_queue *nvmeq, void *ctx,
 	struct request *req = iod_get_private(iod);
 	struct nvme_cmd_info *cmd_rq = blk_mq_rq_to_pdu(req);
 	u16 status = le16_to_cpup(&cqe->status) >> 1;
+	bool requeue = false;
 	int error = 0;
 
 	if (unlikely(status)) {
 		if (!(status & NVME_SC_DNR || blk_noretry_request(req))
 		    && (jiffies - req->start_time) < req->timeout) {
 			unsigned long flags;
 
+			requeue = true;
 			blk_mq_requeue_request(req);
 			spin_lock_irqsave(req->q->queue_lock, flags);
 			if (!blk_queue_stopped(req->q))
 				blk_mq_kick_requeue_list(req->q);
 			spin_unlock_irqrestore(req->q->queue_lock, flags);
-			return;
+			goto release_iod;
 		}
 
 		if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
@@ -639,6 +641,7 @@ static void req_completion(struct nvme_queue *nvmeq, void *ctx,
 			"completing aborted command with status:%04x\n",
 			error);
 
+release_iod:
 	if (iod->nents) {
 		dma_unmap_sg(nvmeq->dev->dev, iod->sg, iod->nents,
 			rq_data_dir(req) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
@@ -651,7 +654,8 @@ static void req_completion(struct nvme_queue *nvmeq, void *ctx,
 	}
 	nvme_free_iod(nvmeq->dev, iod);
 
-	blk_mq_complete_request(req, error);
+	if (likely(!requeue))
+		blk_mq_complete_request(req, error);
 }
 
 /* length is in bytes.  gfp flags indicates whether we may sleep. */