On Tue, Dec 17, 2024 at 07:40:15PM -0500, Steven Rostedt wrote:
> On Wed, 18 Dec 2024 07:43:46 +0800
> Edward Adam Davis <eada...@qq.com> wrote:
>
> > >
> > > A proper fix is being discussed here:
> > First, my fix is the first one.
>
> Yes I saw that.
>
> > Second, the root cause of the problem is an overflow when calculating
> > nr_pages.
> > >
> > > https://lore.kernel.org/linux-trace-kernel/20241216164931.57323-1-aha310...@gmail.com/
> > >
> > > Thank you,
> > >
> > > -- Steve
> > >
> > The calculation of nr_pages below overflows because the pgoff value is 8,
> > the nr_subbufs value is 3, and the subbuf_order value is 0.
>
> So basically you are saying that passing in the the mmap with the pgoff is
> what's causing it.
>
> > > > nr_pages = ((nr_subbufs + 1) << subbuf_order) - pgoff; /* +
> > > > meta-page */
> > > >
> > > > nr_vma_pages = vma_pages(vma);
>
>
> Thanks, I believe I now have a reproducer. And yes, I'll take your patch.
> (If Vincent is OK with it).
I wanted to look at the reproducer sent by Jeongjung yesterday but got
preempted. My bad.
To avoid repeating the (nr_subbufs + 1) << subbuf_order How about?
- nr_pages = ((nr_subbufs + 1) << subbuf_order) - pgoff; /* + meta-page */
+ nr_pages = ((nr_subbufs + 1) << subbuf_order); /* + meta-page */
+
+ if (pgoff > nr_pages)
+ return -EINVAL;
+
+ nr_pages -= pgoff;
And probably also
Fixes: 117c39200d9d ("ring-buffer: Introducing ring-buffer mapping functions")
>
> Here's the reproducer:
>
> ------------------------8<-------------------------
> #include <fcntl.h>
> #include <stdlib.h>
> #include <unistd.h>
> #include <asm/types.h>
> #include <sys/mman.h>
>
> int main(int argc, char **argv)
> {
> int page_size = getpagesize();
> int fd;
> void *meta;
>
> system("echo 1 > /sys/kernel/tracing/buffer_size_kb");
> fd = open("/sys/kernel/tracing/per_cpu/cpu0/trace_pipe_raw", O_RDONLY);
>
> meta = mmap(NULL, page_size, PROT_READ, MAP_SHARED, fd, page_size * 5);
> }
> ------------------------>8-------------------------
>
> Thanks,
>
>
> -- Steve
