Messages in this thread | | | From | Dan Williams <> | Date | Wed, 19 Dec 2018 17:41:56 -0800 | Subject | Re: [PATCH V2 1/1] device-dax: check for vma range while dax_mmap. |
| |
On Wed, Dec 12, 2018 at 10:12 PM Yi Zhang <yi.z.zhang@linux.intel.com> wrote: > > On 2018-12-10 at 16:10:31 -0800, Dan Williams wrote: > > On Tue, Aug 21, 2018 at 12:38 AM Yi Zhang <yi.z.zhang@linux.intel.com> wrote: > > > > > > On 2018-08-20 at 12:50:31 -0700, Dave Jiang wrote: > > > > > > > > > > > > On 08/20/2018 10:53 AM, Verma, Vishal L wrote: > > > > > > > > > > On Mon, 2018-08-13 at 20:02 +0800, Zhang Yi wrote: > > > > >> This patch prevents a user mapping an illegal vma range that is larger > > > > >> than a dax device physical resource. > > > > >> > > > > >> When qemu maps the dax device for virtual nvdimm's backend device, the > > > > >> v-nvdimm label area is defined at the end of mapped range. By using an > > > > >> illegal size that exceeds the range of the device dax, it will trigger a > > > > >> fault with qemu. > > > > >> > > > > >> Signed-off-by: Zhang Yi <yi.z.zhang@linux.intel.com> > > > > >> --- > > > > >> drivers/dax/device.c | 29 +++++++++++++++++++++++++++++ > > > > >> 1 file changed, 29 insertions(+) > > > > >> > > > > > > > > > > Looks good to me: > > > > > Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> > > > > > > > > Applied. > > > Thanks Dava and Vishal's kindly review. Thank you. > > > > So, it turns out this patch did not get merged for 4.20. I fumbled it > > when returning from vacation. However, I'm not sure it is needed. As > > long as attempts to access the out-of-range capacity results in SIGBUS > > then the implementation is correct. This is similar to the case where > > a file is truncated after the vma is established. That size is > > validated at fault time. > The problem is that we didn't get the fault at we initial the mapping > until attempt to access it, then qemu will failed unexpect without any > output, I think is is better to mention user that we are starting at a > illegal size, but not faulting at an uncertained time.
That can always happen with mmap'd files. There is no guarantee that a file range an application successfully mmap'd can be faulted in without triggering a SIGBUS later. So this change would make device-dax semantics stricter than regular file semantics. For example the following program prints "map: pass" and then terminates with SIGBUS. The "test_data" file is a zero sized file.
int main(void) { int fd = open("test_data", O_RDWR); void *addr;
if (fd < 0) return -1;
addr = mmap(NULL, 1 << 20, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); printf("map: %s\n", addr == MAP_FAILED ? "fail" : "pass");
*(char *) addr = 0;
return 0; }
| |