Hi Nam,

+CC stable( heads up as this is a regression affecting 5.15.y and probably others, Greg: this was reproducible upstream so reported everything w.r.t upstream code but initially found on 5.15.y)

On 19/04/24 20:29, Nam Cao wrote:

On 2024-04-18 Harshit Mogalapalli wrote:

...

While fuzzing 5.15.y kernel with Syzkaller, we noticed a INFO: task hung bug in fb_deferred_io_work()

I think the problem is because of improper offset address calculation. The kernel calculate address offset with: offset = vmf->address - vmf->vma->vm_start

Now the problem is that your C program mmap the framebuffer at 2 different offsets: mmap(ptr, 4096, PROT_WRITE, MAP_FIXED | MAP_SHARED, fd, 0xff000); mmap(ptr, 4096, PROT_WRITE, MAP_FIXED | MAP_SHARED, fd, 0);

but the kernel doesn't take these different offsets into account. So, 2 different pages are mistakenly recognized as the same page.

Can you try the following patch?

This patch works well against the reproducer, this simplified repro and the longer repro which syzkaller generated couldn't trigger any hang with the below patch applied.

Thanks, Harshit

Best regards, Nam

diff --git a/drivers/video/fbdev/core/fb_defio.c b/drivers/video/fbdev/core/fb_defio.c index dae96c9f61cf..d5d6cd9e8b29 100644 --- a/drivers/video/fbdev/core/fb_defio.c +++ b/drivers/video/fbdev/core/fb_defio.c @@ -196,7 +196,8 @@ static vm_fault_t fb_deferred_io_track_page(struct fb_info *info, unsigned long */ static vm_fault_t fb_deferred_io_page_mkwrite(struct fb_info *info, struct vm_fault *vmf) {

• unsigned long offset = vmf->address - vmf->vma->vm_start;

• unsigned long offset = vmf->address - vmf->vma->vm_start

• 	+ (vmf->vma->vm_pgoff << PAGE_SHIFT);
  

  struct page *page = vmf->page;

file_update_time(vmf->vma->vm_file);