On 2019-07-25 1:11 p.m., Matthew Wilcox wrote: > On Thu, Jul 25, 2019 at 12:05:29PM -0700, Sagi Grimberg wrote: >> >>>>> NVMe-OF is configured using configfs. The target is specified by the >>>>> user writing a path to a configfs attribute. This is the way it works >>>>> today but with blkdev_get_by_path()[1]. For the passthru code, we need >>>>> to get a nvme_ctrl instead of a block_device, but the principal is the same. >>>> >>>> Why isn't a fd being passed in there instead of a random string? >>> >>> I suppose we could echo a string of the file descriptor number there, >>> and look up the fd in the process' file descriptor table ... >> >> Assuming that there is a open handle somewhere out there... Yes, that would be a step backwards from an interface. The user would then need a special process to open the fd and pass it through configfs. They couldn't just do it with basic bash commands. > Well, that's how we'd know that the application echoing /dev/nvme3 into > configfs actually has permission to access /dev/nvme3. It's the kernel that's accessing the device so it has permission. root permission is required to configure the kernel. > Think about > containers, for example. It's not exactly safe to mount configfs in a > non-root container since it can access any NVMe device in the system, > not just ones which it's been given permission to access. Right? I don't think it really makes any sense to talk about NVMe-of and containers. Though, if we did it would be solely on the configuration interface so that users inside a container might be able to configure a new target for resources they can see and they'd have to have their own view into configfs.... Logan
