mirror of
https://github.com/torvalds/linux.git
synced 2025-11-02 01:29:02 +02:00
We're generally not proponents of rewrites (nasty uncomfortable things that make you late for dinner!). So why rewrite Binder? Binder has been evolving over the past 15+ years to meet the evolving needs of Android. Its responsibilities, expectations, and complexity have grown considerably during that time. While we expect Binder to continue to evolve along with Android, there are a number of factors that currently constrain our ability to develop/maintain it. Briefly those are: 1. Complexity: Binder is at the intersection of everything in Android and fulfills many responsibilities beyond IPC. It has become many things to many people, and due to its many features and their interactions with each other, its complexity is quite high. In just 6kLOC it must deliver transactions to the right threads. It must correctly parse and translate the contents of transactions, which can contain several objects of different types (e.g., pointers, fds) that can interact with each other. It controls the size of thread pools in userspace, and ensures that transactions are assigned to threads in ways that avoid deadlocks where the threadpool has run out of threads. It must track refcounts of objects that are shared by several processes by forwarding refcount changes between the processes correctly. It must handle numerous error scenarios and it combines/nests 13 different locks, 7 reference counters, and atomic variables. Finally, It must do all of this as fast and efficiently as possible. Minor performance regressions can cause a noticeably degraded user experience. 2. Things to improve: Thousand-line functions [1], error-prone error handling [2], and confusing structure can occur as a code base grows organically. After more than a decade of development, this codebase could use an overhaul. [1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/android/binder.c?h=v6.5#n2896 [2]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/android/binder.c?h=v6.5#n3658 3. Security critical: Binder is a critical part of Android's sandboxing strategy. Even Android's most de-privileged sandboxes (e.g. the Chrome renderer, or SW Codec) have direct access to Binder. More than just about any other component, it's important that Binder provide robust security, and itself be robust against security vulnerabilities. It's #1 (high complexity) that has made continuing to evolve Binder and resolving #2 (tech debt) exceptionally difficult without causing #3 (security issues). For Binder to continue to meet Android's needs, we need better ways to manage (and reduce!) complexity without increasing the risk. The biggest change is obviously the choice of programming language. We decided to use Rust because it directly addresses a number of the challenges within Binder that we have faced during the last years. It prevents mistakes with ref counting, locking, bounds checking, and also does a lot to reduce the complexity of error handling. Additionally, we've been able to use the more expressive type system to encode the ownership semantics of the various structs and pointers, which takes the complexity of managing object lifetimes out of the hands of the programmer, reducing the risk of use-after-frees and similar problems. Rust has many different pointer types that it uses to encode ownership semantics into the type system, and this is probably one of the most important aspects of how it helps in Binder. The Binder driver has a lot of different objects that have complex ownership semantics; some pointers own a refcount, some pointers have exclusive ownership, and some pointers just reference the object and it is kept alive in some other manner. With Rust, we can use a different pointer type for each kind of pointer, which enables the compiler to enforce that the ownership semantics are implemented correctly. Another useful feature is Rust's error handling. Rust allows for more simplified error handling with features such as destructors, and you get compilation failures if errors are not properly handled. This means that even though Rust requires you to spend more lines of code than C on things such as writing down invariants that are left implicit in C, the Rust driver is still slightly smaller than C binder: Rust is 5.5kLOC and C is 5.8kLOC. (These numbers are excluding blank lines, comments, binderfs, and any debugging facilities in C that are not yet implemented in the Rust driver. The numbers include abstractions in rust/kernel/ that are unlikely to be used by other drivers than Binder.) Although this rewrite completely rethinks how the code is structured and how assumptions are enforced, we do not fundamentally change *how* the driver does the things it does. A lot of careful thought has gone into the existing design. The rewrite is aimed rather at improving code health, structure, readability, robustness, security, maintainability and extensibility. We also include more inline documentation, and improve how assumptions in the code are enforced. Furthermore, all unsafe code is annotated with a SAFETY comment that explains why it is correct. We have left the binderfs filesystem component in C. Rewriting it in Rust would be a large amount of work and requires a lot of bindings to the file system interfaces. Binderfs has not historically had the same challenges with security and complexity, so rewriting binderfs seems to have lower value than the rest of Binder. Correctness and feature parity ------------------------------ Rust binder passes all tests that validate the correctness of Binder in the Android Open Source Project. We can boot a device, and run a variety of apps and functionality without issues. We have performed this both on the Cuttlefish Android emulator device, and on a Pixel 6 Pro. As for feature parity, Rust binder currently implements all features that C binder supports, with the exception of some debugging facilities. The missing debugging facilities will be added before we submit the Rust implementation upstream. Tracepoints ----------- I did not include all of the tracepoints as I felt that the mechansim for making C access fields of Rust structs should be discussed on list separately. I also did not include the support for building Rust Binder as a module since that requires exporting a bunch of additional symbols on the C side. Original RFC Link with old benchmark numbers: https://lore.kernel.org/r/20231101-rust-binder-v1-0-08ba9197f637@google.com Co-developed-by: Wedson Almeida Filho <wedsonaf@gmail.com> Signed-off-by: Wedson Almeida Filho <wedsonaf@gmail.com> Co-developed-by: Matt Gilbride <mattgilbride@google.com> Signed-off-by: Matt Gilbride <mattgilbride@google.com> Acked-by: Carlos Llamas <cmllamas@google.com> Acked-by: Paul Moore <paul@paul-moore.com> Signed-off-by: Alice Ryhl <aliceryhl@google.com> Link: https://lore.kernel.org/r/20250919-rust-binder-v2-1-a384b09f28dd@google.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
204 lines
9.1 KiB
Rust
204 lines
9.1 KiB
Rust
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
// Copyright (C) 2025 Google LLC.
|
|
|
|
//! Logic for closing files in a deferred manner.
|
|
//!
|
|
//! This file could make sense to have in `kernel::fs`, but it was rejected for being too
|
|
//! Binder-specific.
|
|
|
|
use core::mem::MaybeUninit;
|
|
use kernel::{
|
|
alloc::{AllocError, Flags},
|
|
bindings,
|
|
prelude::*,
|
|
};
|
|
|
|
/// Helper used for closing file descriptors in a way that is safe even if the file is currently
|
|
/// held using `fdget`.
|
|
///
|
|
/// Additional motivation can be found in commit 80cd795630d6 ("binder: fix use-after-free due to
|
|
/// ksys_close() during fdget()") and in the comments on `binder_do_fd_close`.
|
|
pub(crate) struct DeferredFdCloser {
|
|
inner: KBox<DeferredFdCloserInner>,
|
|
}
|
|
|
|
/// SAFETY: This just holds an allocation with no real content, so there's no safety issue with
|
|
/// moving it across threads.
|
|
unsafe impl Send for DeferredFdCloser {}
|
|
/// SAFETY: This just holds an allocation with no real content, so there's no safety issue with
|
|
/// moving it across threads.
|
|
unsafe impl Sync for DeferredFdCloser {}
|
|
|
|
/// # Invariants
|
|
///
|
|
/// If the `file` pointer is non-null, then it points at a `struct file` and owns a refcount to
|
|
/// that file.
|
|
#[repr(C)]
|
|
struct DeferredFdCloserInner {
|
|
twork: MaybeUninit<bindings::callback_head>,
|
|
file: *mut bindings::file,
|
|
}
|
|
|
|
impl DeferredFdCloser {
|
|
/// Create a new [`DeferredFdCloser`].
|
|
pub(crate) fn new(flags: Flags) -> Result<Self, AllocError> {
|
|
Ok(Self {
|
|
// INVARIANT: The `file` pointer is null, so the type invariant does not apply.
|
|
inner: KBox::new(
|
|
DeferredFdCloserInner {
|
|
twork: MaybeUninit::uninit(),
|
|
file: core::ptr::null_mut(),
|
|
},
|
|
flags,
|
|
)?,
|
|
})
|
|
}
|
|
|
|
/// Schedule a task work that closes the file descriptor when this task returns to userspace.
|
|
///
|
|
/// Fails if this is called from a context where we cannot run work when returning to
|
|
/// userspace. (E.g., from a kthread.)
|
|
pub(crate) fn close_fd(self, fd: u32) -> Result<(), DeferredFdCloseError> {
|
|
use bindings::task_work_notify_mode_TWA_RESUME as TWA_RESUME;
|
|
|
|
// In this method, we schedule the task work before closing the file. This is because
|
|
// scheduling a task work is fallible, and we need to know whether it will fail before we
|
|
// attempt to close the file.
|
|
|
|
// Task works are not available on kthreads.
|
|
let current = kernel::current!();
|
|
|
|
// Check if this is a kthread.
|
|
// SAFETY: Reading `flags` from a task is always okay.
|
|
if unsafe { ((*current.as_ptr()).flags & bindings::PF_KTHREAD) != 0 } {
|
|
return Err(DeferredFdCloseError::TaskWorkUnavailable);
|
|
}
|
|
|
|
// Transfer ownership of the box's allocation to a raw pointer. This disables the
|
|
// destructor, so we must manually convert it back to a KBox to drop it.
|
|
//
|
|
// Until we convert it back to a `KBox`, there are no aliasing requirements on this
|
|
// pointer.
|
|
let inner = KBox::into_raw(self.inner);
|
|
|
|
// The `callback_head` field is first in the struct, so this cast correctly gives us a
|
|
// pointer to the field.
|
|
let callback_head = inner.cast::<bindings::callback_head>();
|
|
// SAFETY: This pointer offset operation does not go out-of-bounds.
|
|
let file_field = unsafe { core::ptr::addr_of_mut!((*inner).file) };
|
|
|
|
let current = current.as_ptr();
|
|
|
|
// SAFETY: This function currently has exclusive access to the `DeferredFdCloserInner`, so
|
|
// it is okay for us to perform unsynchronized writes to its `callback_head` field.
|
|
unsafe { bindings::init_task_work(callback_head, Some(Self::do_close_fd)) };
|
|
|
|
// SAFETY: This inserts the `DeferredFdCloserInner` into the task workqueue for the current
|
|
// task. If this operation is successful, then this transfers exclusive ownership of the
|
|
// `callback_head` field to the C side until it calls `do_close_fd`, and we don't touch or
|
|
// invalidate the field during that time.
|
|
//
|
|
// When the C side calls `do_close_fd`, the safety requirements of that method are
|
|
// satisfied because when a task work is executed, the callback is given ownership of the
|
|
// pointer.
|
|
//
|
|
// The file pointer is currently null. If it is changed to be non-null before `do_close_fd`
|
|
// is called, then that change happens due to the write at the end of this function, and
|
|
// that write has a safety comment that explains why the refcount can be dropped when
|
|
// `do_close_fd` runs.
|
|
let res = unsafe { bindings::task_work_add(current, callback_head, TWA_RESUME) };
|
|
|
|
if res != 0 {
|
|
// SAFETY: Scheduling the task work failed, so we still have ownership of the box, so
|
|
// we may destroy it.
|
|
unsafe { drop(KBox::from_raw(inner)) };
|
|
|
|
return Err(DeferredFdCloseError::TaskWorkUnavailable);
|
|
}
|
|
|
|
// This removes the fd from the fd table in `current`. The file is not fully closed until
|
|
// `filp_close` is called. We are given ownership of one refcount to the file.
|
|
//
|
|
// SAFETY: This is safe no matter what `fd` is. If the `fd` is valid (that is, if the
|
|
// pointer is non-null), then we call `filp_close` on the returned pointer as required by
|
|
// `file_close_fd`.
|
|
let file = unsafe { bindings::file_close_fd(fd) };
|
|
if file.is_null() {
|
|
// We don't clean up the task work since that might be expensive if the task work queue
|
|
// is long. Just let it execute and let it clean up for itself.
|
|
return Err(DeferredFdCloseError::BadFd);
|
|
}
|
|
|
|
// Acquire a second refcount to the file.
|
|
//
|
|
// SAFETY: The `file` pointer points at a file with a non-zero refcount.
|
|
unsafe { bindings::get_file(file) };
|
|
|
|
// This method closes the fd, consuming one of our two refcounts. There could be active
|
|
// light refcounts created from that fd, so we must ensure that the file has a positive
|
|
// refcount for the duration of those active light refcounts. We do that by holding on to
|
|
// the second refcount until the current task returns to userspace.
|
|
//
|
|
// SAFETY: The `file` pointer is valid. Passing `current->files` as the file table to close
|
|
// it in is correct, since we just got the `fd` from `file_close_fd` which also uses
|
|
// `current->files`.
|
|
//
|
|
// Note: fl_owner_t is currently a void pointer.
|
|
unsafe { bindings::filp_close(file, (*current).files as bindings::fl_owner_t) };
|
|
|
|
// We update the file pointer that the task work is supposed to fput. This transfers
|
|
// ownership of our last refcount.
|
|
//
|
|
// INVARIANT: This changes the `file` field of a `DeferredFdCloserInner` from null to
|
|
// non-null. This doesn't break the type invariant for `DeferredFdCloserInner` because we
|
|
// still own a refcount to the file, so we can pass ownership of that refcount to the
|
|
// `DeferredFdCloserInner`.
|
|
//
|
|
// When `do_close_fd` runs, it must be safe for it to `fput` the refcount. However, this is
|
|
// the case because all light refcounts that are associated with the fd we closed
|
|
// previously must be dropped when `do_close_fd`, since light refcounts must be dropped
|
|
// before returning to userspace.
|
|
//
|
|
// SAFETY: Task works are executed on the current thread right before we return to
|
|
// userspace, so this write is guaranteed to happen before `do_close_fd` is called, which
|
|
// means that a race is not possible here.
|
|
unsafe { *file_field = file };
|
|
|
|
Ok(())
|
|
}
|
|
|
|
/// # Safety
|
|
///
|
|
/// The provided pointer must point at the `twork` field of a `DeferredFdCloserInner` stored in
|
|
/// a `KBox`, and the caller must pass exclusive ownership of that `KBox`. Furthermore, if the
|
|
/// file pointer is non-null, then it must be okay to release the refcount by calling `fput`.
|
|
unsafe extern "C" fn do_close_fd(inner: *mut bindings::callback_head) {
|
|
// SAFETY: The caller just passed us ownership of this box.
|
|
let inner = unsafe { KBox::from_raw(inner.cast::<DeferredFdCloserInner>()) };
|
|
if !inner.file.is_null() {
|
|
// SAFETY: By the type invariants, we own a refcount to this file, and the caller
|
|
// guarantees that dropping the refcount now is okay.
|
|
unsafe { bindings::fput(inner.file) };
|
|
}
|
|
// The allocation is freed when `inner` goes out of scope.
|
|
}
|
|
}
|
|
|
|
/// Represents a failure to close an fd in a deferred manner.
|
|
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
|
|
pub(crate) enum DeferredFdCloseError {
|
|
/// Closing the fd failed because we were unable to schedule a task work.
|
|
TaskWorkUnavailable,
|
|
/// Closing the fd failed because the fd does not exist.
|
|
BadFd,
|
|
}
|
|
|
|
impl From<DeferredFdCloseError> for Error {
|
|
fn from(err: DeferredFdCloseError) -> Error {
|
|
match err {
|
|
DeferredFdCloseError::TaskWorkUnavailable => ESRCH,
|
|
DeferredFdCloseError::BadFd => EBADF,
|
|
}
|
|
}
|
|
}
|