[PATCH 1/2] rust: pin-init: internal: init: remove `#[disable_initialized_field_access]`

[PATCH 1/2] rust: pin-init: internal: init: remove `#[disable_initialized_field_access]`

[Benno Lossin]

Gary noticed [1] that the initializer macros as well as the `[Pin]Init`
traits cannot support packed struct, since they use operations that
require aligned pointers. This means that any code using packed structs
and pin-init is unsound.

Thus remove the `#[disable_initialized_field_access]` attribute from
`init!`, which is the only safe way to create an initializer of a packed
struct.

In the future, we can add support for packed structs by changing the
trait infrastructure to include `UnalignedInit` or hopefully another
mechanism.

Reported-by: Gary Guo <gary@garyguo.net>
Link: https://rust-for-linux.zulipchat.com/#narrow/channel/561532-pin-init/topic/initialized.20field.20accessor.20detection/with/576210658 [1]
Fixes: ceca298c53f9 ("rust: pin-init: internal: init: add escape hatch for referencing initialized fields")
Signed-off-by: Benno Lossin <lossin@kernel.org>
---
This commit does not need backporting, as ceca298c53f9 is not yet in any
stable tree.

However, the unsoundness still affects several stable trees, because it
was unknowingly fixed in commit 42415d163e5d ("rust: pin-init: add
references to previously initialized fields"). Before then, packed
structs compiled without any issues with pin-init and thus all prior
kernel versions with pin-init that do not contain that commit are
affected.

We introduced pin-init in 90e53c5e70a6 ("rust: add pin-init API core"),
which was included in 6.4. The affected stable trees that are still
maintained are: 6.17, 6.16, 6.12, and 6.6. Note that 6.18 and 6.19
already contain 42415d163e5d, so they are unaffected.

I will prepare a separate patch series to backport 42415d163e5d to each
of the affected trees, including the second patch of this series that
documents the fact that field accessors are load-bearing for soundness.

@asahi folks, let me know if I should prioritize a solution for packed
structs. Otherwise I'd like not support it at the moment, as that
requires some deeper changes to the internals of pin-init. I'm tracking
the status of packed structs in:

    https://github.com/Rust-for-Linux/pin-init/issues/112
---
 rust/pin-init/internal/src/init.rs | 39 ++++++------------------------
 1 file changed, 8 insertions(+), 31 deletions(-)

diff --git a/rust/pin-init/internal/src/init.rs b/rust/pin-init/internal/src/init.rs
index 42936f915a07..da53adc44ecf 100644
--- a/rust/pin-init/internal/src/init.rs
+++ b/rust/pin-init/internal/src/init.rs
@@ -62,7 +62,6 @@ fn ident(&self) -> Option<&Ident> {
 
 enum InitializerAttribute {
     DefaultError(DefaultErrorAttribute),
-    DisableInitializedFieldAccess,
 }
 
 struct DefaultErrorAttribute {
@@ -86,6 +85,7 @@ pub(crate) fn expand(
     let error = error.map_or_else(
         || {
             if let Some(default_error) = attrs.iter().fold(None, |acc, attr| {
+                #[expect(irrefutable_let_patterns)]
                 if let InitializerAttribute::DefaultError(DefaultErrorAttribute { ty }) = attr {
                     Some(ty.clone())
                 } else {
@@ -145,15 +145,7 @@ fn assert_zeroable<T: ?::core::marker::Sized>(_: *mut T)
     };
     // `mixed_site` ensures that the data is not accessible to the user-controlled code.
     let data = Ident::new("__data", Span::mixed_site());
-    let init_fields = init_fields(
-        &fields,
-        pinned,
-        !attrs
-            .iter()
-            .any(|attr| matches!(attr, InitializerAttribute::DisableInitializedFieldAccess)),
-        &data,
-        &slot,
-    );
+    let init_fields = init_fields(&fields, pinned, &data, &slot);
     let field_check = make_field_check(&fields, init_kind, &path);
     Ok(quote! {{
         // We do not want to allow arbitrary returns, so we declare this type as the `Ok` return
@@ -236,7 +228,6 @@ fn get_init_kind(rest: Option<(Token![..], Expr)>, dcx: &mut DiagCtxt) -> InitKi
 fn init_fields(
     fields: &Punctuated<InitializerField, Token![,]>,
     pinned: bool,
-    generate_initialized_accessors: bool,
     data: &Ident,
     slot: &Ident,
 ) -> TokenStream {
@@ -272,13 +263,6 @@ fn init_fields(
                         unsafe { &mut (*#slot).#ident }
                     }
                 };
-                let accessor = generate_initialized_accessors.then(|| {
-                    quote! {
-                        #(#cfgs)*
-                        #[allow(unused_variables)]
-                        let #ident = #accessor;
-                    }
-                });
                 quote! {
                     #(#attrs)*
                     {
@@ -286,7 +270,9 @@ fn init_fields(
                         // SAFETY: TODO
                         unsafe { #write(::core::ptr::addr_of_mut!((*#slot).#ident), #value_ident) };
                     }
-                    #accessor
+                    #(#cfgs)*
+                    #[allow(unused_variables)]
+                    let #ident = #accessor;
                 }
             }
             InitializerKind::Init { ident, value, .. } => {
@@ -326,20 +312,15 @@ fn init_fields(
                         },
                     )
                 };
-                let accessor = generate_initialized_accessors.then(|| {
-                    quote! {
-                        #(#cfgs)*
-                        #[allow(unused_variables)]
-                        let #ident = #accessor;
-                    }
-                });
                 quote! {
                     #(#attrs)*
                     {
                         let #init = #value;
                         #value_init
                     }
-                    #accessor
+                    #(#cfgs)*
+                    #[allow(unused_variables)]
+                    let #ident = #accessor;
                 }
             }
             InitializerKind::Code { block: value, .. } => quote! {
@@ -466,10 +447,6 @@ fn parse(input: syn::parse::ParseStream<'_>) -> syn::Result<Self> {
                 if a.path().is_ident("default_error") {
                     a.parse_args::<DefaultErrorAttribute>()
                         .map(InitializerAttribute::DefaultError)
-                } else if a.path().is_ident("disable_initialized_field_access") {
-                    a.meta
-                        .require_path_only()
-                        .map(|_| InitializerAttribute::DisableInitializedFieldAccess)
                 } else {
                     Err(syn::Error::new_spanned(a, "unknown initializer attribute"))
                 }

base-commit: 6de23f81a5e08be8fbf5e8d7e9febc72a5b5f27f
-- 
2.53.0

Re: [PATCH 1/2] rust: pin-init: internal: init: remove `#[disable_initialized_field_access]`

[Gary Guo]

On Sat Feb 28, 2026 at 11:37 AM GMT, Benno Lossin wrote:
> Gary noticed [1] that the initializer macros as well as the `[Pin]Init`
> traits cannot support packed struct, since they use operations that

This is better read as "unaligned field", rather than "packed struct". The
packed struct can be initialized as a whole by moving it into place.

After all, a value of a packed struct is itself a valid initializer and is
sound.

If you have

#[derive(Zeroable)]
struct Foo {
   x: u8,
   y: u16,
   z: u8,
}

it is also valid to have `pin_init!(Foo { x: 1, ..Zeroable::zeroed() })`

> require aligned pointers. This means that any code using packed structs
> and pin-init is unsound.
> 
> Thus remove the `#[disable_initialized_field_access]` attribute from
> `init!`, which is the only safe way to create an initializer of a packed
> struct.
> 
> In the future, we can add support for packed structs by changing the
> trait infrastructure to include `UnalignedInit` or hopefully another
> mechanism.

I guess the comment can be reworded a bit, to hint that this won't be added
unless there's a compelling need, e.g.

    If support for in-place initializing packed structs is required in the
    future, we can ...

Personally I think this won't be neeeded, as there are many ways around
supporting unaligned fields, e.g. by creating and moving the entire packed
structs. Another way is to use `Unaligned<T>` type for fields that needs
pin-init, similar to what zerocopy have.

Best,
Gary

> 
> Reported-by: Gary Guo <gary@garyguo.net>
> Link: https://rust-for-linux.zulipchat.com/#narrow/channel/561532-pin-init/topic/initialized.20field.20accessor.20detection/with/576210658 [1]
> Fixes: ceca298c53f9 ("rust: pin-init: internal: init: add escape hatch for referencing initialized fields")
> Signed-off-by: Benno Lossin <lossin@kernel.org>
> ---
> This commit does not need backporting, as ceca298c53f9 is not yet in any
> stable tree.
> 
> However, the unsoundness still affects several stable trees, because it
> was unknowingly fixed in commit 42415d163e5d ("rust: pin-init: add
> references to previously initialized fields"). Before then, packed
> structs compiled without any issues with pin-init and thus all prior
> kernel versions with pin-init that do not contain that commit are
> affected.
> 
> We introduced pin-init in 90e53c5e70a6 ("rust: add pin-init API core"),
> which was included in 6.4. The affected stable trees that are still
> maintained are: 6.17, 6.16, 6.12, and 6.6. Note that 6.18 and 6.19
> already contain 42415d163e5d, so they are unaffected.
> 
> I will prepare a separate patch series to backport 42415d163e5d to each
> of the affected trees, including the second patch of this series that
> documents the fact that field accessors are load-bearing for soundness.
> 
> @asahi folks, let me know if I should prioritize a solution for packed
> structs. Otherwise I'd like not support it at the moment, as that
> requires some deeper changes to the internals of pin-init. I'm tracking
> the status of packed structs in:
> 
>     https://github.com/Rust-for-Linux/pin-init/issues/112
> ---
>  rust/pin-init/internal/src/init.rs | 39 ++++++------------------------
>  1 file changed, 8 insertions(+), 31 deletions(-)

Re: [PATCH 1/2] rust: pin-init: internal: init: remove `#[disable_initialized_field_access]`

[Miguel Ojeda]

On Sat, Feb 28, 2026 at 12:37 PM Benno Lossin <lossin@kernel.org> wrote:
>
> The affected stable trees that are still
> maintained are: 6.17, 6.16, 6.12, and 6.6.

6.17 and 6.16 are not maintained, so we can ignore those, i.e. only
6.12 and 6.6 will need it.

Cheers,
Miguel

Re: [PATCH 1/2] rust: pin-init: internal: init: remove `#[disable_initialized_field_access]`

[Janne Grunau]

On Sat, Feb 28, 2026 at 12:37:04PM +0100, Benno Lossin wrote:
> Gary noticed [1] that the initializer macros as well as the `[Pin]Init`
> traits cannot support packed struct, since they use operations that
> require aligned pointers. This means that any code using packed structs
> and pin-init is unsound.
> 
> Thus remove the `#[disable_initialized_field_access]` attribute from
> `init!`, which is the only safe way to create an initializer of a packed
> struct.
> 
> In the future, we can add support for packed structs by changing the
> trait infrastructure to include `UnalignedInit` or hopefully another
> mechanism.
> 
> Reported-by: Gary Guo <gary@garyguo.net>
> Link: https://rust-for-linux.zulipchat.com/#narrow/channel/561532-pin-init/topic/initialized.20field.20accessor.20detection/with/576210658 [1]
> Fixes: ceca298c53f9 ("rust: pin-init: internal: init: add escape hatch for referencing initialized fields")
> Signed-off-by: Benno Lossin <lossin@kernel.org>
> ---
> This commit does not need backporting, as ceca298c53f9 is not yet in any
> stable tree.
> 
> However, the unsoundness still affects several stable trees, because it
> was unknowingly fixed in commit 42415d163e5d ("rust: pin-init: add
> references to previously initialized fields"). Before then, packed
> structs compiled without any issues with pin-init and thus all prior
> kernel versions with pin-init that do not contain that commit are
> affected.
> 
> We introduced pin-init in 90e53c5e70a6 ("rust: add pin-init API core"),
> which was included in 6.4. The affected stable trees that are still
> maintained are: 6.17, 6.16, 6.12, and 6.6. Note that 6.18 and 6.19
> already contain 42415d163e5d, so they are unaffected.
> 
> I will prepare a separate patch series to backport 42415d163e5d to each
> of the affected trees, including the second patch of this series that
> documents the fact that field accessors are load-bearing for soundness.
> 
> @asahi folks, let me know if I should prioritize a solution for packed
> structs. Otherwise I'd like not support it at the moment, as that
> requires some deeper changes to the internals of pin-init. I'm tracking
> the status of packed structs in:

I have worked around this in the downstream AOP audio driver now. I did
not do that initially since it looked more involved and we were planning
to bring support back.
These structs describe messages for communication with a coprocessor via
shared memory. They are derived by observing messages by tracing. So
there is only limited understanding how the messages are formated. Their
layout has for obvious reason match exactly so `#[repr(C, packed)]` is
the obvious choice.
One of the structs had a size of N * 4 - 1 which results in a alignment
of 1. Fortunately the struct could be padded to multiple of 4.
Nevertheless it was required to replace a few u32 with an unaligned
version.

I'm not sure if there is a need to support unaligned fields in pin-init.
The workarounds in the asahi GPU and AOP audio drivers are acceptable
and could stay indefinitely.

Janne

>     https://github.com/Rust-for-Linux/pin-init/issues/112

Re: [PATCH 1/2] rust: pin-init: internal: init: remove `#[disable_initialized_field_access]`

[Benno Lossin]

On Sun Mar 1, 2026 at 6:12 PM CET, Janne Grunau wrote:
> On Sat, Feb 28, 2026 at 12:37:04PM +0100, Benno Lossin wrote:
>> @asahi folks, let me know if I should prioritize a solution for packed
>> structs. Otherwise I'd like not support it at the moment, as that
>> requires some deeper changes to the internals of pin-init. I'm tracking
>> the status of packed structs in:
>
> I have worked around this in the downstream AOP audio driver now. I did
> not do that initially since it looked more involved and we were planning
> to bring support back.
> These structs describe messages for communication with a coprocessor via
> shared memory. They are derived by observing messages by tracing. So
> there is only limited understanding how the messages are formated. Their
> layout has for obvious reason match exactly so `#[repr(C, packed)]` is
> the obvious choice.
> One of the structs had a size of N * 4 - 1 which results in a alignment
> of 1. Fortunately the struct could be padded to multiple of 4.
> Nevertheless it was required to replace a few u32 with an unaligned
> version.
>
> I'm not sure if there is a need to support unaligned fields in pin-init.
> The workarounds in the asahi GPU and AOP audio drivers are acceptable
> and could stay indefinitely.

Thanks for the quick work! Yeah that sounds like a good solution for the
problem. Adding unaligned pointer support to initialization is quite a
pain. If it comes up we can still do it of course (but I'd only do it if
we can't find a workaround).

Cheers,
Benno