nvmet-rdma: Fix a possible uninitialized variable dereference
authorSagi Grimberg <sagi@grimberg.me>
Thu, 9 Mar 2017 11:45:52 +0000 (13:45 +0200)
committerJens Axboe <axboe@fb.com>
Tue, 4 Apr 2017 15:48:23 +0000 (09:48 -0600)
When handling a new recv command, we grab a new rsp resource and
check for the queue state being live. In case the queue is not in
live state, we simply restore the rsp back to the free list. However
in this flow we didn't set rsp->queue yet, so we cannot dereference it.

Instead, make sure to initialize rsp->queue (and other rsp members)
as soon as possible so we won't reference uninitialized variables.

Reported-by: Yi Zhang <yizhan@redhat.com>
Reported-by: Raju Rangoju <rajur@chelsio.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Raju Rangoju <rajur@chelsio.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
drivers/nvme/target/rdma.c

index 9aa1da3..ecc4fe8 100644 (file)
@@ -703,11 +703,6 @@ static void nvmet_rdma_handle_command(struct nvmet_rdma_queue *queue,
 {
        u16 status;
 
-       cmd->queue = queue;
-       cmd->n_rdma = 0;
-       cmd->req.port = queue->port;
-
-
        ib_dma_sync_single_for_cpu(queue->dev->device,
                cmd->cmd->sge[0].addr, cmd->cmd->sge[0].length,
                DMA_FROM_DEVICE);
@@ -760,9 +755,12 @@ static void nvmet_rdma_recv_done(struct ib_cq *cq, struct ib_wc *wc)
 
        cmd->queue = queue;
        rsp = nvmet_rdma_get_rsp(queue);
+       rsp->queue = queue;
        rsp->cmd = cmd;
        rsp->flags = 0;
        rsp->req.cmd = cmd->nvme_cmd;
+       rsp->req.port = queue->port;
+       rsp->n_rdma = 0;
 
        if (unlikely(queue->state != NVMET_RDMA_Q_LIVE)) {
                unsigned long flags;