Re: [PATCH mm-unstable fix] mm: userfaultfd: check for start + len overflow in validate_range: fix

[Ryan Roberts]

On 14/07/2023 19:29, Axel Rasmussen wrote:
> This commit removed an extra check for zero-length ranges, and folded it
> into the common validate_range() helper used by all UFFD ioctls.
> 
> It failed to notice though that UFFDIO_COPY *only* called validate_range
> on the dst range, not the src range. So removing this check actually let
> us proceed with zero-length source ranges, eventually hitting a BUG
> further down in the call stack.
> 
> The correct fix seems clear: call validate_range() on the src range too.
> 
> Other ioctls are not affected by this, as they only have one range, not
> two (src + dst).
> 
> Reported-by: syzbot+42309678e0bc7b32f8e9@xxxxxxxxxxxxxxxxxxxxxxxxx
> Closes: https://syzkaller.appspot.com/bug?extid=42309678e0bc7b32f8e9
> Signed-off-by: Axel Rasmussen <axelrasmussen@xxxxxxxxxx>
> ---
>  fs/userfaultfd.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> index 53a7220c4679..36d233759233 100644
> --- a/fs/userfaultfd.c
> +++ b/fs/userfaultfd.c
> @@ -1759,6 +1759,9 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
>  			   sizeof(uffdio_copy)-sizeof(__s64)))
>  		goto out;
>  
> +	ret = validate_range(ctx->mm, uffdio_copy.src, uffdio_copy.len);
> +	if (ret)
> +		goto out;
>  	ret = validate_range(ctx->mm, uffdio_copy.dst, uffdio_copy.len);
>  	if (ret)
>  		goto out;


Hi Axel,

I've just noticed that this patch, now in mm-unstable, regresses the mkdirty mm
selftest:

# [INFO] detected THP size: 2048 KiB
TAP version 13
1..6
# [INFO] PTRACE write access
ok 1 SIGSEGV generated, page not modified
# [INFO] PTRACE write access to THP
ok 2 SIGSEGV generated, page not modified
# [INFO] Page migration
ok 3 SIGSEGV generated, page not modified
# [INFO] Page migration of THP
ok 4 SIGSEGV generated, page not modified
# [INFO] PTE-mapping a THP
ok 5 SIGSEGV generated, page not modified
# [INFO] UFFDIO_COPY
not ok 6 UFFDIO_COPY failed
Bail out! 1 out of 6 tests failed
# Totals: pass:5 fail:1 xfail:0 xpass:0 skip:0 error:0

Whereas all 6 tests pass against v6.5-rc4.

I'm afraid I don't know the test well and haven't looked at what the issue might
be, but noticed and thought I should point it out.

bisect log:

git bisect start
# bad: [ad3232df3e410acc2229c9195479c5596c1d1f96] mm/memory_hotplug: embed
vmem_altmap details in memory block
git bisect bad ad3232df3e410acc2229c9195479c5596c1d1f96
# good: [5d0c230f1de8c7515b6567d9afba1f196fb4e2f4] Linux 6.5-rc4
git bisect good 5d0c230f1de8c7515b6567d9afba1f196fb4e2f4
# bad: [aa5712770e3f0edb31ae879cd6452d5c2111d4fb] mm: fix obsolete function name
above debug_pagealloc_enabled_static()
git bisect bad aa5712770e3f0edb31ae879cd6452d5c2111d4fb
# bad: [bef1ff8723df303a06cdaffe64d95db2f7e7d4f6] mm: userfaultfd: support
UFFDIO_POISON for hugetlbfs
git bisect bad bef1ff8723df303a06cdaffe64d95db2f7e7d4f6
# good: [4f4469463e8571012d2602b39f14ed3e3dbd972a] selftests/mm: add gup test
matrix in run_vmtests.sh
git bisect good 4f4469463e8571012d2602b39f14ed3e3dbd972a
# good: [5c0d69839ef4f560679919f3483a592741df74f8] memcg: drop kmem.limit_in_bytes
git bisect good 5c0d69839ef4f560679919f3483a592741df74f8
# good: [b75f155a299729dc62de0ce6a9400d076298aa4c] mm: compaction: skip the
memory hole rapidly when isolating free pages
git bisect good b75f155a299729dc62de0ce6a9400d076298aa4c
# good: [12b13121d9f4487301dd9fb765265b642b2f6d5d] mm/memcg: minor cleanup for
MEM_CGROUP_ID_MAX
git bisect good 12b13121d9f4487301dd9fb765265b642b2f6d5d
# bad: [c9c368e75919c105aae072896e58d0ad4639e505] mm: userfaultfd: check for
start + len overflow in validate_range: fix
git bisect bad c9c368e75919c105aae072896e58d0ad4639e505
# good: [9e707995021bbdfc67ad83a985f7796bba580bed]
mm-make-pte_marker_swapin_error-more-general-fix
git bisect good 9e707995021bbdfc67ad83a985f7796bba580bed
# good: [46b66377b696c43c89ed4b1cb3f56b64e8fd475b] mm: userfaultfd: check for
start + len overflow in validate_range
git bisect good 46b66377b696c43c89ed4b1cb3f56b64e8fd475b
# first bad commit: [c9c368e75919c105aae072896e58d0ad4639e505] mm: userfaultfd:
check for start + len overflow in validate_range: fix

Thanks,
Ryan