Thread (2 messages) flat view 2 messages, 2 authors, 13d ago

Re: [PATCH] rust: net: netlink: validate attribute length before casting to `c_int`

From: "Alexandre Courbot" <acourbot@nvidia.com>
Date: 2026-09-14 01:10:51
Also in: lkml, rust-for-linux

On Sun Sep 13, 2026 at 11:17 AM JST, Sagar Taunk wrote:
quoted hunk ↗ jump to hunk
As pointed out by Sashiko, `put()` trusted an unchecked `as` cast from
`usize` to `c_int`. When the length exceeds `i32::MAX`, that cast
wraps around to a negative value.

`nla_put()`'s `skb_tailroom()` check treats the length as signed, so
the wrapped negative value slips past it. Further down, `__nla_reserve()`
and `skb_put` reinterpret the same value as unsigned, turning it into an
enormous length and triggering a kernel panic via `skb_over_panic()`.

So, validate the cast with `c_int::try_from()` instead, and return
`EMSGSIZE` when the length doesn't fit.

Signed-off-by: Sagar Taunk <redacted>
---
 rust/kernel/net/netlink.rs | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)
diff --git a/rust/kernel/net/netlink.rs b/rust/kernel/net/netlink.rs
index 3c2b142a7402..f929f63b32c1 100644
--- a/rust/kernel/net/netlink.rs
+++ b/rust/kernel/net/netlink.rs
@@ -88,11 +88,19 @@ fn put<T>(&mut self, attrtype: c_int, value: &T) -> Result
         T: ?Sized + IntoBytes + Immutable,
     {
         let skb = self.skb.skb.as_ptr();
-        let len = size_of_val(value);
-        let ptr = core::ptr::from_ref(value).cast::<c_void>();
-        // SAFETY: `skb` is valid by `NetlinkSkBuff` type invariants, and the provided value is
-        // readable and initialized for its `size_of` bytes.
-        to_result(unsafe { bindings::nla_put(skb, attrtype, len as c_int, ptr) })
+        let bytes = value.as_bytes();
+        // `nla_put()` takes attrlen as a plain `c_int`. If `bytes.len()`
+        // doesn't fit, an `as` cast would wrap around a negative value.
+        // Which then, would feed a huge unsigned length to `__nla_reserve()`
+        // and `skb_put()` causing it to panic via `skb_over_panic()`. So,
+        // the following check will reject it instead.
+        let len = c_int::try_from(bytes.len()).map_err(|_| EMSGSIZE)?;
I understand that you wanted to reuse `bytes` here, but the original use
of `size_of_val` is clearer and keeping it would also reduce the diff,
so let's call `c_int::try_from` on that.

Also the comment is unneeded. The change of type is justified by the
fact that `nla_put` requires a `c_int`. Any further explanation adds
confusion, and the mentioned `__nla_reserve` doesn't even appear in the
chunk. This reads like what an LLM would produce, i.e. tediously.
+        let ptr = bytes.as_ptr().cast::<c_void>();
Same here, the original line was fine so no need to change it. So it
looks like `bytes` is not needed at all.

As a general rule, and to make reviewing easier, it is a good idea to
keep the diffs are small as possible. Here the commit message says this
patch validates a value, so it should limit itself to that.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help