Re: [PATCH 5.15 056/917] fuse: fix page stealing

On Tue, Nov 23, 2021 at 1:22 PM Miklos Szeredi [off-list ref] wrote:

On Tue, Nov 23, 2021 at 7:29 PM Justin Forbes [off-list ref] wrote:

quoted

On Mon, Nov 15, 2021 at 7:04 PM Greg Kroah-Hartman
[off-list ref] wrote:

quoted

From: Miklos Szeredi <redacted>

commit 712a951025c0667ff00b25afc360f74e639dfabe upstream.

It is possible to trigger a crash by splicing anon pipe bufs to the fuse
device.

The reason for this is that anon_pipe_buf_release() will reuse buf->page if
the refcount is 1, but that page might have already been stolen and its
flags modified (e.g. PG_lru added).

This happens in the unlikely case of fuse_dev_splice_write() getting around
to calling pipe_buf_release() after a page has been stolen, added to the
page cache and removed from the page cache.

Fix by calling pipe_buf_release() right after the page was inserted into
the page cache.  In this case the page has an elevated refcount so any
release function will know that the page isn't reusable.

Reported-by: Frank Dinoff <redacted>
Link: https://lore.kernel.org/r/CAAmZXrsGg2xsP1CK+cbuEMumtrqdvD-NKnWzhNcvn71RV3c1yw@mail.gmail.com/ (local)
Fixes: dd3bb14f44a6 ("fuse: support splice() writing to fuse device")
Cc: <redacted> # v2.6.35
Signed-off-by: Miklos Szeredi <redacted>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

It appears this patch causes a rather serious regression in flatpacks
using portals to access files.  Reverting this patch restores expected
behavior. I have asked users in the Fedora bug to test with 5.16-rc2
to see if we are just missing a dependent patch in stable, or if this
is broken there as well, but no response yet.:

https://bugzilla.redhat.com/show_bug.cgi?id=2025285
https://github.com/flatpak/flatpak/issues/4595

Hi,

Thanks for the report.  Can someone with the reproducer try the attached patch?

I think the race there is unlikely but possible.

Thanks, did a scratch build for that and dropped it in the bug. Only
one user has reported back, but the report was that it did not fix the
issue.  I have also gotten confirmation now that the issue is occuring
with 5.16-rc2.

Thanks,
Justin