Message ID | 20200924092449.367288-1-nobuhiro1.iwamatsu@toshiba.co.jp |
---|---|
State | New |
Headers | show |
Series | [for,4.4,4.9,4.14] RDMA/ucma: ucma_context reference leak in error path | expand |
On Thu, Sep 24, 2020 at 06:24:49PM +0900, Nobuhiro Iwamatsu wrote: >From: Shamir Rabinovitch <shamir.rabinovitch@oracle.com> > >commit ef95a90ae6f4f21990e1f7ced6719784a409e811 upstream. > >Validating input parameters should be done before getting the cm_id >otherwise it can leak a cm_id reference. > >Fixes: 6a21dfc0d0db ("RDMA/ucma: Limit possible option size") >Signed-off-by: Shamir Rabinovitch <shamir.rabinovitch@oracle.com> >Reviewed-by: Leon Romanovsky <leonro@mellanox.com> >Signed-off-by: Jason Gunthorpe <jgg@mellanox.com> >[iwamatsu: Backported to 4.4, 4.9 and 4.14: adjust context] >Signed-off-by: Nobuhiro Iwamatsu (CIP) <nobuhiro1.iwamatsu@toshiba.co.jp> Queued up, thanks! -- Thanks, Sasha
diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 3e4d3d5560bf10..6315f77b4a58c1 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -1295,13 +1295,13 @@ static ssize_t ucma_set_option(struct ucma_file *file, const char __user *inbuf, if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; + if (unlikely(cmd.optlen > KMALLOC_MAX_SIZE)) + return -EINVAL; + ctx = ucma_get_ctx(file, cmd.id); if (IS_ERR(ctx)) return PTR_ERR(ctx); - if (unlikely(cmd.optlen > KMALLOC_MAX_SIZE)) - return -EINVAL; - optval = memdup_user((void __user *) (unsigned long) cmd.optval, cmd.optlen); if (IS_ERR(optval)) {