On čtvrtek 5. října 2023 15:56:47 CEST Matthew Wilcox (Oracle) wrote:

If the shared memory object is larger than the DRM object that it backs, we can overrun the page array. Limit the number of pages we install from each folio to prevent this.

Signed-off-by: Matthew Wilcox (Oracle) willy@infradead.org Reported-by: Oleksandr Natalenko oleksandr@natalenko.name Tested-by: Oleksandr Natalenko oleksandr@natalenko.name Link: https://lore.kernel.org/lkml/13360591.uLZWGnKmhe@natalenko.name/ Fixes: 3291e09a4638 ("drm: convert drm_gem_put_pages() to use a folio_batch") Cc: stable@vger.kernel.org # 6.5.x

---

drivers/gpu/drm/drm_gem.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c index 6129b89bb366..44a948b80ee1 100644 --- a/drivers/gpu/drm/drm_gem.c +++ b/drivers/gpu/drm/drm_gem.c @@ -540,7 +540,7 @@ struct page **drm_gem_get_pages(struct drm_gem_object *obj) struct page **pages; struct folio *folio; struct folio_batch fbatch;

• int i, j, npages;

• long i, j, npages;

if (WARN_ON(!obj->filp)) return ERR_PTR(-EINVAL); @@ -564,11 +564,13 @@ struct page **drm_gem_get_pages(struct drm_gem_object *obj) i = 0; while (i < npages) {

• long nr;
  

  folio = shmem_read_folio_gfp(mapping, i, mapping_gfp_mask(mapping)); if (IS_ERR(folio)) goto fail;

• for (j = 0; j < folio_nr_pages(folio); j++, i++)
  

• nr = min(npages - i, folio_nr_pages(folio));
  

• for (j = 0; j < nr; j++, i++)
  pages[i] = folio_file_page(folio, i);
  

/* Make sure shmem keeps __GFP_DMA32 allocated pages in the

Gentle ping. It would be nice to have this picked so that it gets into the stable kernel rather sooner than later.

Thanks.