On Mon, Jul 29, 2019 at 02:02:54PM +0530, Souptick Joarder wrote: > On Mon, Jul 29, 2019 at 1:35 PM Souptick Joarder wrote: > > > > On Sun, Jul 28, 2019 at 11:36 PM Marek Marczykowski-Górecki > > wrote: > > > > > > On Fri, Feb 15, 2019 at 08:18:31AM +0530, Souptick Joarder wrote: > > > > Convert to use vm_map_pages() to map range of kernel > > > > memory to user vma. > > > > > > > > map->count is passed to vm_map_pages() and internal API > > > > verify map->count against count ( count = vma_pages(vma)) > > > > for page array boundary overrun condition. > > > > > > This commit breaks gntdev driver. If vma->vm_pgoff > 0, vm_map_pages > > > will: > > > - use map->pages starting at vma->vm_pgoff instead of 0 > > > > The actual code ignores vma->vm_pgoff > 0 scenario and mapped > > the entire map->pages[i]. Why the entire map->pages[i] needs to be mapped > > if vma->vm_pgoff > 0 (in original code) ? vma->vm_pgoff is used as index passed to gntdev_find_map_index. It's basically (ab)using this parameter for "which grant reference to map". > > are you referring to set vma->vm_pgoff = 0 irrespective of value passed > > from user space ? If yes, using vm_map_pages_zero() is an alternate > > option. Yes, that should work. > > > - verify map->count against vma_pages()+vma->vm_pgoff instead of just > > > vma_pages(). > > > > In original code -> > > > > diff --git a/drivers/xen/gntdev.c b/drivers/xen/gntdev.c > > index 559d4b7f807d..469dfbd6cf90 100644 > > --- a/drivers/xen/gntdev.c > > +++ b/drivers/xen/gntdev.c > > @@ -1084,7 +1084,7 @@ static int gntdev_mmap(struct file *flip, struct > > vm_area_struct *vma) > > int index = vma->vm_pgoff; > > int count = vma_pages(vma); > > > > Count is user passed value. > > > > struct gntdev_grant_map *map; > > - int i, err = -EINVAL; > > + int err = -EINVAL; > > if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED)) > > return -EINVAL; > > @@ -1145,12 +1145,9 @@ static int gntdev_mmap(struct file *flip, > > struct vm_area_struct *vma) > > goto out_put_map; > > if (!use_ptemod) { > > - for (i = 0; i < count; i++) { > > - err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE, > > - map->pages[i]); > > > > and when count > i , we end up with trying to map memory outside > > boundary of map->pages[i], which was not correct. > > typo. > s/count > i / count > map->count gntdev_find_map_index verifies it. Specifically, it looks for a map matching both index and count. > > > > - if (err) > > - goto out_put_map; > > - } > > + err = vm_map_pages(vma, map->pages, map->count); > > + if (err) > > + goto out_put_map; > > > > With this commit, inside __vm_map_pages(), we have addressed this scenario. > > > > +static int __vm_map_pages(struct vm_area_struct *vma, struct page **pages, > > + unsigned long num, unsigned long offset) > > +{ > > + unsigned long count = vma_pages(vma); > > + unsigned long uaddr = vma->vm_start; > > + int ret, i; > > + > > + /* Fail if the user requested offset is beyond the end of the object */ > > + if (offset > num) > > + return -ENXIO; > > + > > + /* Fail if the user requested size exceeds available object size */ > > + if (count > num - offset) > > + return -ENXIO; > > > > By checking count > num -offset. (considering vma->vm_pgoff != 0 as well). > > So we will never cross the boundary of map->pages[i]. > > > > > > > > > > In practice, this breaks using a single gntdev FD for mapping multiple > > > grants. > > > > How ? gntdev uses vma->vm_pgoff to select which grant entry should be mapped. map struct returned by gntdev_find_map_index() describes just the pages to be mapped. Specifically map->pages[0] should be mapped at vma->vm_start, not vma->vm_start+vma->vm_pgoff*PAGE_SIZE. When trying to map grant with index (aka vma->vm_pgoff) > 1, __vm_map_pages() will refuse to map it because it will expect map->count to be at least vma_pages(vma)+vma->vm_pgoff, while it is exactly vma_pages(vma). > > > It looks like vm_map_pages() is not a good fit for this code and IMO it > > > should be reverted. > > > > Did you hit any issue around this code in real time ? Yes, relevant strace output: [pid 857] ioctl(7, IOCTL_GNTDEV_MAP_GRANT_REF, 0x7ffd3407b6d0) = 0 [pid 857] mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_SHARED, 7, 0) = 0x777f1211b000 [pid 857] ioctl(7, IOCTL_GNTDEV_SET_UNMAP_NOTIFY, 0x7ffd3407b710) = 0 [pid 857] ioctl(7, IOCTL_GNTDEV_MAP_GRANT_REF, 0x7ffd3407b6d0) = 0 [pid 857] mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_SHARED, 7, 0x1000) = -1 ENXIO (No such device or address) details here: https://github.com/QubesOS/qubes-issues/issues/5199 > > > > > > > > > > > Signed-off-by: Souptick Joarder > > > > Reviewed-by: Boris Ostrovsky > > > > --- > > > > drivers/xen/gntdev.c | 11 ++++------- > > > > 1 file changed, 4 insertions(+), 7 deletions(-) > > > > > > > > diff --git a/drivers/xen/gntdev.c b/drivers/xen/gntdev.c > > > > index 5efc5ee..5d64262 100644 > > > > --- a/drivers/xen/gntdev.c > > > > +++ b/drivers/xen/gntdev.c > > > > @@ -1084,7 +1084,7 @@ static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma) > > > > int index = vma->vm_pgoff; > > > > int count = vma_pages(vma); > > > > struct gntdev_grant_map *map; > > > > - int i, err = -EINVAL; > > > > + int err = -EINVAL; > > > > > > > > if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED)) > > > > return -EINVAL; > > > > @@ -1145,12 +1145,9 @@ static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma) > > > > goto out_put_map; > > > > > > > > if (!use_ptemod) { > > > > - for (i = 0; i < count; i++) { > > > > - err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE, > > > > - map->pages[i]); > > > > - if (err) > > > > - goto out_put_map; > > > > - } > > > > + err = vm_map_pages(vma, map->pages, map->count); > > > > + if (err) > > > > + goto out_put_map; > > > > } else { > > > > #ifdef CONFIG_X86 > > > > /* > > > > > > -- > > > Best Regards, > > > Marek Marczykowski-Górecki > > > Invisible Things Lab > > > A: Because it messes up the order in which people normally read text. > > > Q: Why is top-posting such a bad thing? -- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing?