From 246f239d9f40f633160f0c18f87a20922d4e77bb Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Wed, 17 Apr 2024 14:06:37 +0200 Subject: Merging debian version 1.65.0+dfsg1-2. Signed-off-by: Daniel Baumann --- library/std/src/sys/hermit/condvar.rs | 90 ------- library/std/src/sys/hermit/fs.rs | 18 +- library/std/src/sys/hermit/futex.rs | 39 +++ library/std/src/sys/hermit/mod.rs | 17 +- library/std/src/sys/hermit/mutex.rs | 216 ---------------- library/std/src/sys/hermit/net.rs | 2 - library/std/src/sys/hermit/rwlock.rs | 144 ----------- library/std/src/sys/itron/mutex.rs | 6 - library/std/src/sys/sgx/abi/thread.rs | 8 +- library/std/src/sys/sgx/abi/usercalls/alloc.rs | 165 ++++++++++-- library/std/src/sys/sgx/abi/usercalls/mod.rs | 8 +- library/std/src/sys/sgx/abi/usercalls/raw.rs | 24 +- library/std/src/sys/sgx/abi/usercalls/tests.rs | 34 ++- library/std/src/sys/sgx/mod.rs | 2 +- library/std/src/sys/sgx/mutex.rs | 3 - library/std/src/sys/solid/fs.rs | 24 +- library/std/src/sys/solid/mod.rs | 2 +- library/std/src/sys/unix/fd.rs | 11 +- library/std/src/sys/unix/fs.rs | 24 +- library/std/src/sys/unix/locks/fuchsia_mutex.rs | 5 +- library/std/src/sys/unix/locks/futex_mutex.rs | 5 +- library/std/src/sys/unix/locks/futex_rwlock.rs | 2 +- library/std/src/sys/unix/locks/pthread_condvar.rs | 2 +- library/std/src/sys/unix/locks/pthread_mutex.rs | 2 +- library/std/src/sys/unix/mod.rs | 42 ++- library/std/src/sys/unix/net.rs | 22 ++ library/std/src/sys/unix/os_str.rs | 40 +-- library/std/src/sys/unix/os_str/tests.rs | 8 + library/std/src/sys/unix/process/process_common.rs | 59 ++++- .../src/sys/unix/process/process_common/tests.rs | 24 ++ library/std/src/sys/unix/process/process_unix.rs | 4 +- library/std/src/sys/unix/rand.rs | 18 +- library/std/src/sys/unix/thread.rs | 32 ++- library/std/src/sys/unix/thread_parker.rs | 281 --------------------- library/std/src/sys/unix/thread_parker/mod.rs | 21 ++ library/std/src/sys/unix/thread_parker/netbsd.rs | 113 +++++++++ library/std/src/sys/unix/thread_parker/pthread.rs | 271 ++++++++++++++++++++ library/std/src/sys/unsupported/alloc.rs | 7 +- library/std/src/sys/unsupported/common.rs | 2 +- library/std/src/sys/unsupported/fs.rs | 4 +- library/std/src/sys/unsupported/locks/mutex.rs | 3 - library/std/src/sys/unsupported/process.rs | 3 + library/std/src/sys/wasi/fs.rs | 6 +- library/std/src/sys/wasi/stdio.rs | 23 +- library/std/src/sys/windows/alloc.rs | 5 +- library/std/src/sys/windows/c.rs | 35 ++- library/std/src/sys/windows/cmath.rs | 2 +- library/std/src/sys/windows/compat.rs | 232 ++++++++--------- library/std/src/sys/windows/fs.rs | 119 +++++---- library/std/src/sys/windows/handle.rs | 12 +- library/std/src/sys/windows/locks/mutex.rs | 2 - library/std/src/sys/windows/mod.rs | 28 +- library/std/src/sys/windows/os.rs | 6 +- library/std/src/sys/windows/os_str.rs | 4 +- library/std/src/sys/windows/path/tests.rs | 2 +- library/std/src/sys/windows/rand.rs | 121 +++++++-- library/std/src/sys/windows/stdio.rs | 41 ++- library/std/src/sys/windows/thread_local_dtor.rs | 4 + library/std/src/sys/windows/thread_parker.rs | 22 +- 59 files changed, 1347 insertions(+), 1124 deletions(-) delete mode 100644 library/std/src/sys/hermit/condvar.rs create mode 100644 library/std/src/sys/hermit/futex.rs delete mode 100644 library/std/src/sys/hermit/mutex.rs delete mode 100644 library/std/src/sys/hermit/rwlock.rs delete mode 100644 library/std/src/sys/unix/thread_parker.rs create mode 100644 library/std/src/sys/unix/thread_parker/mod.rs create mode 100644 library/std/src/sys/unix/thread_parker/netbsd.rs create mode 100644 library/std/src/sys/unix/thread_parker/pthread.rs (limited to 'library/std/src/sys') diff --git a/library/std/src/sys/hermit/condvar.rs b/library/std/src/sys/hermit/condvar.rs deleted file mode 100644 index 22059ca0d..000000000 --- a/library/std/src/sys/hermit/condvar.rs +++ /dev/null @@ -1,90 +0,0 @@ -use crate::ffi::c_void; -use crate::ptr; -use crate::sync::atomic::{AtomicUsize, Ordering::SeqCst}; -use crate::sys::hermit::abi; -use crate::sys::locks::Mutex; -use crate::sys_common::lazy_box::{LazyBox, LazyInit}; -use crate::time::Duration; - -// The implementation is inspired by Andrew D. Birrell's paper -// "Implementing Condition Variables with Semaphores" - -pub struct Condvar { - counter: AtomicUsize, - sem1: *const c_void, - sem2: *const c_void, -} - -pub(crate) type MovableCondvar = LazyBox; - -impl LazyInit for Condvar { - fn init() -> Box { - Box::new(Self::new()) - } -} - -unsafe impl Send for Condvar {} -unsafe impl Sync for Condvar {} - -impl Condvar { - pub fn new() -> Self { - let mut condvar = - Self { counter: AtomicUsize::new(0), sem1: ptr::null(), sem2: ptr::null() }; - unsafe { - let _ = abi::sem_init(&mut condvar.sem1, 0); - let _ = abi::sem_init(&mut condvar.sem2, 0); - } - condvar - } - - pub unsafe fn notify_one(&self) { - if self.counter.load(SeqCst) > 0 { - self.counter.fetch_sub(1, SeqCst); - abi::sem_post(self.sem1); - abi::sem_timedwait(self.sem2, 0); - } - } - - pub unsafe fn notify_all(&self) { - let counter = self.counter.swap(0, SeqCst); - for _ in 0..counter { - abi::sem_post(self.sem1); - } - for _ in 0..counter { - abi::sem_timedwait(self.sem2, 0); - } - } - - pub unsafe fn wait(&self, mutex: &Mutex) { - self.counter.fetch_add(1, SeqCst); - mutex.unlock(); - abi::sem_timedwait(self.sem1, 0); - abi::sem_post(self.sem2); - mutex.lock(); - } - - pub unsafe fn wait_timeout(&self, mutex: &Mutex, dur: Duration) -> bool { - self.counter.fetch_add(1, SeqCst); - mutex.unlock(); - let millis = dur.as_millis().min(u32::MAX as u128) as u32; - - let res = if millis > 0 { - abi::sem_timedwait(self.sem1, millis) - } else { - abi::sem_trywait(self.sem1) - }; - - abi::sem_post(self.sem2); - mutex.lock(); - res == 0 - } -} - -impl Drop for Condvar { - fn drop(&mut self) { - unsafe { - let _ = abi::sem_destroy(self.sem1); - let _ = abi::sem_destroy(self.sem2); - } - } -} diff --git a/library/std/src/sys/hermit/fs.rs b/library/std/src/sys/hermit/fs.rs index fa9a7fb19..f921839cf 100644 --- a/library/std/src/sys/hermit/fs.rs +++ b/library/std/src/sys/hermit/fs.rs @@ -2,7 +2,7 @@ use crate::ffi::{CStr, CString, OsString}; use crate::fmt; use crate::hash::{Hash, Hasher}; use crate::io::{self, Error, ErrorKind}; -use crate::io::{IoSlice, IoSliceMut, ReadBuf, SeekFrom}; +use crate::io::{BorrowedCursor, IoSlice, IoSliceMut, SeekFrom}; use crate::os::unix::ffi::OsStrExt; use crate::path::{Path, PathBuf}; use crate::sys::cvt; @@ -41,6 +41,9 @@ pub struct OpenOptions { mode: i32, } +#[derive(Copy, Clone, Debug, Default)] +pub struct FileTimes {} + pub struct FilePermissions(!); pub struct FileType(!); @@ -110,6 +113,11 @@ impl fmt::Debug for FilePermissions { } } +impl FileTimes { + pub fn set_accessed(&mut self, _t: SystemTime) {} + pub fn set_modified(&mut self, _t: SystemTime) {} +} + impl FileType { pub fn is_dir(&self) -> bool { self.0 @@ -312,8 +320,8 @@ impl File { false } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { - crate::io::default_read_buf(|buf| self.read(buf), buf) + pub fn read_buf(&self, cursor: BorrowedCursor<'_>) -> io::Result<()> { + crate::io::default_read_buf(|buf| self.read(buf), cursor) } pub fn write(&self, buf: &[u8]) -> io::Result { @@ -344,6 +352,10 @@ impl File { pub fn set_permissions(&self, _perm: FilePermissions) -> io::Result<()> { Err(Error::from_raw_os_error(22)) } + + pub fn set_times(&self, _times: FileTimes) -> io::Result<()> { + Err(Error::from_raw_os_error(22)) + } } impl DirBuilder { diff --git a/library/std/src/sys/hermit/futex.rs b/library/std/src/sys/hermit/futex.rs new file mode 100644 index 000000000..b64c174b0 --- /dev/null +++ b/library/std/src/sys/hermit/futex.rs @@ -0,0 +1,39 @@ +use super::abi; +use crate::ptr::null; +use crate::sync::atomic::AtomicU32; +use crate::time::Duration; + +pub fn futex_wait(futex: &AtomicU32, expected: u32, timeout: Option) -> bool { + // Calculate the timeout as a relative timespec. + // + // Overflows are rounded up to an infinite timeout (None). + let timespec = timeout.and_then(|dur| { + Some(abi::timespec { + tv_sec: dur.as_secs().try_into().ok()?, + tv_nsec: dur.subsec_nanos().into(), + }) + }); + + let r = unsafe { + abi::futex_wait( + futex.as_mut_ptr(), + expected, + timespec.as_ref().map_or(null(), |t| t as *const abi::timespec), + abi::FUTEX_RELATIVE_TIMEOUT, + ) + }; + + r != -abi::errno::ETIMEDOUT +} + +#[inline] +pub fn futex_wake(futex: &AtomicU32) -> bool { + unsafe { abi::futex_wake(futex.as_mut_ptr(), 1) > 0 } +} + +#[inline] +pub fn futex_wake_all(futex: &AtomicU32) { + unsafe { + abi::futex_wake(futex.as_mut_ptr(), i32::MAX); + } +} diff --git a/library/std/src/sys/hermit/mod.rs b/library/std/src/sys/hermit/mod.rs index 60b7a973c..827d82900 100644 --- a/library/std/src/sys/hermit/mod.rs +++ b/library/std/src/sys/hermit/mod.rs @@ -25,6 +25,7 @@ pub mod cmath; pub mod env; pub mod fd; pub mod fs; +pub mod futex; #[path = "../unsupported/io.rs"] pub mod io; pub mod memchr; @@ -45,14 +46,14 @@ pub mod thread_local_dtor; pub mod thread_local_key; pub mod time; -mod condvar; -mod mutex; -mod rwlock; - +#[path = "../unix/locks"] pub mod locks { - pub use super::condvar::*; - pub use super::mutex::*; - pub use super::rwlock::*; + mod futex_condvar; + mod futex_mutex; + mod futex_rwlock; + pub(crate) use futex_condvar::MovableCondvar; + pub(crate) use futex_mutex::{MovableMutex, Mutex}; + pub(crate) use futex_rwlock::{MovableRwLock, RwLock}; } use crate::io::ErrorKind; @@ -98,7 +99,7 @@ pub extern "C" fn __rust_abort() { // SAFETY: must be called only once during runtime initialization. // NOTE: this is not guaranteed to run, for example when Rust code is called externally. -pub unsafe fn init(argc: isize, argv: *const *const u8) { +pub unsafe fn init(argc: isize, argv: *const *const u8, _sigpipe: u8) { let _ = net::init(); args::init(argc, argv); } diff --git a/library/std/src/sys/hermit/mutex.rs b/library/std/src/sys/hermit/mutex.rs deleted file mode 100644 index eb15a04ff..000000000 --- a/library/std/src/sys/hermit/mutex.rs +++ /dev/null @@ -1,216 +0,0 @@ -use crate::cell::UnsafeCell; -use crate::collections::VecDeque; -use crate::hint; -use crate::ops::{Deref, DerefMut, Drop}; -use crate::ptr; -use crate::sync::atomic::{AtomicUsize, Ordering}; -use crate::sys::hermit::abi; - -/// This type provides a lock based on busy waiting to realize mutual exclusion -/// -/// # Description -/// -/// This structure behaves a lot like a common mutex. There are some differences: -/// -/// - By using busy waiting, it can be used outside the runtime. -/// - It is a so called ticket lock and is completely fair. -#[cfg_attr(target_arch = "x86_64", repr(align(128)))] -#[cfg_attr(not(target_arch = "x86_64"), repr(align(64)))] -struct Spinlock { - queue: AtomicUsize, - dequeue: AtomicUsize, - data: UnsafeCell, -} - -unsafe impl Sync for Spinlock {} -unsafe impl Send for Spinlock {} - -/// A guard to which the protected data can be accessed -/// -/// When the guard falls out of scope it will release the lock. -struct SpinlockGuard<'a, T: ?Sized + 'a> { - dequeue: &'a AtomicUsize, - data: &'a mut T, -} - -impl Spinlock { - pub const fn new(user_data: T) -> Spinlock { - Spinlock { - queue: AtomicUsize::new(0), - dequeue: AtomicUsize::new(1), - data: UnsafeCell::new(user_data), - } - } - - #[inline] - fn obtain_lock(&self) { - let ticket = self.queue.fetch_add(1, Ordering::SeqCst) + 1; - let mut counter: u16 = 0; - while self.dequeue.load(Ordering::SeqCst) != ticket { - counter += 1; - if counter < 100 { - hint::spin_loop(); - } else { - counter = 0; - unsafe { - abi::yield_now(); - } - } - } - } - - #[inline] - pub unsafe fn lock(&self) -> SpinlockGuard<'_, T> { - self.obtain_lock(); - SpinlockGuard { dequeue: &self.dequeue, data: &mut *self.data.get() } - } -} - -impl Default for Spinlock { - fn default() -> Spinlock { - Spinlock::new(Default::default()) - } -} - -impl<'a, T: ?Sized> Deref for SpinlockGuard<'a, T> { - type Target = T; - fn deref(&self) -> &T { - &*self.data - } -} - -impl<'a, T: ?Sized> DerefMut for SpinlockGuard<'a, T> { - fn deref_mut(&mut self) -> &mut T { - &mut *self.data - } -} - -impl<'a, T: ?Sized> Drop for SpinlockGuard<'a, T> { - /// The dropping of the SpinlockGuard will release the lock it was created from. - fn drop(&mut self) { - self.dequeue.fetch_add(1, Ordering::SeqCst); - } -} - -/// Realize a priority queue for tasks -struct PriorityQueue { - queues: [Option>; abi::NO_PRIORITIES], - prio_bitmap: u64, -} - -impl PriorityQueue { - pub const fn new() -> PriorityQueue { - PriorityQueue { - queues: [ - None, None, None, None, None, None, None, None, None, None, None, None, None, None, - None, None, None, None, None, None, None, None, None, None, None, None, None, None, - None, None, None, - ], - prio_bitmap: 0, - } - } - - /// Add a task id by its priority to the queue - pub fn push(&mut self, prio: abi::Priority, id: abi::Tid) { - let i: usize = prio.into().into(); - self.prio_bitmap |= (1 << i) as u64; - if let Some(queue) = &mut self.queues[i] { - queue.push_back(id); - } else { - let mut queue = VecDeque::new(); - queue.push_back(id); - self.queues[i] = Some(queue); - } - } - - fn pop_from_queue(&mut self, queue_index: usize) -> Option { - if let Some(queue) = &mut self.queues[queue_index] { - let id = queue.pop_front(); - - if queue.is_empty() { - self.prio_bitmap &= !(1 << queue_index as u64); - } - - id - } else { - None - } - } - - /// Pop the task handle with the highest priority from the queue - pub fn pop(&mut self) -> Option { - for i in 0..abi::NO_PRIORITIES { - if self.prio_bitmap & (1 << i) != 0 { - return self.pop_from_queue(i); - } - } - - None - } -} - -struct MutexInner { - locked: bool, - blocked_task: PriorityQueue, -} - -impl MutexInner { - pub const fn new() -> MutexInner { - MutexInner { locked: false, blocked_task: PriorityQueue::new() } - } -} - -pub struct Mutex { - inner: Spinlock, -} - -pub type MovableMutex = Mutex; - -unsafe impl Send for Mutex {} -unsafe impl Sync for Mutex {} - -impl Mutex { - pub const fn new() -> Mutex { - Mutex { inner: Spinlock::new(MutexInner::new()) } - } - - #[inline] - pub unsafe fn init(&mut self) {} - - #[inline] - pub unsafe fn lock(&self) { - loop { - let mut guard = self.inner.lock(); - if guard.locked == false { - guard.locked = true; - return; - } else { - let prio = abi::get_priority(); - let id = abi::getpid(); - - guard.blocked_task.push(prio, id); - abi::block_current_task(); - drop(guard); - abi::yield_now(); - } - } - } - - #[inline] - pub unsafe fn unlock(&self) { - let mut guard = self.inner.lock(); - guard.locked = false; - if let Some(tid) = guard.blocked_task.pop() { - abi::wakeup_task(tid); - } - } - - #[inline] - pub unsafe fn try_lock(&self) -> bool { - let mut guard = self.inner.lock(); - if guard.locked == false { - guard.locked = true; - } - guard.locked - } -} diff --git a/library/std/src/sys/hermit/net.rs b/library/std/src/sys/hermit/net.rs index 745476171..8a13879d8 100644 --- a/library/std/src/sys/hermit/net.rs +++ b/library/std/src/sys/hermit/net.rs @@ -487,6 +487,4 @@ pub mod netc { #[derive(Copy, Clone)] pub struct sockaddr {} - - pub type socklen_t = usize; } diff --git a/library/std/src/sys/hermit/rwlock.rs b/library/std/src/sys/hermit/rwlock.rs deleted file mode 100644 index 9701bab1f..000000000 --- a/library/std/src/sys/hermit/rwlock.rs +++ /dev/null @@ -1,144 +0,0 @@ -use crate::cell::UnsafeCell; -use crate::sys::locks::{MovableCondvar, Mutex}; -use crate::sys_common::lazy_box::{LazyBox, LazyInit}; - -pub struct RwLock { - lock: Mutex, - cond: MovableCondvar, - state: UnsafeCell, -} - -pub type MovableRwLock = RwLock; - -enum State { - Unlocked, - Reading(usize), - Writing, -} - -unsafe impl Send for RwLock {} -unsafe impl Sync for RwLock {} - -// This rwlock implementation is a relatively simple implementation which has a -// condition variable for readers/writers as well as a mutex protecting the -// internal state of the lock. A current downside of the implementation is that -// unlocking the lock will notify *all* waiters rather than just readers or just -// writers. This can cause lots of "thundering stampede" problems. While -// hopefully correct this implementation is very likely to want to be changed in -// the future. - -impl RwLock { - pub const fn new() -> RwLock { - RwLock { - lock: Mutex::new(), - cond: MovableCondvar::new(), - state: UnsafeCell::new(State::Unlocked), - } - } - - #[inline] - pub unsafe fn read(&self) { - self.lock.lock(); - while !(*self.state.get()).inc_readers() { - self.cond.wait(&self.lock); - } - self.lock.unlock(); - } - - #[inline] - pub unsafe fn try_read(&self) -> bool { - self.lock.lock(); - let ok = (*self.state.get()).inc_readers(); - self.lock.unlock(); - return ok; - } - - #[inline] - pub unsafe fn write(&self) { - self.lock.lock(); - while !(*self.state.get()).inc_writers() { - self.cond.wait(&self.lock); - } - self.lock.unlock(); - } - - #[inline] - pub unsafe fn try_write(&self) -> bool { - self.lock.lock(); - let ok = (*self.state.get()).inc_writers(); - self.lock.unlock(); - return ok; - } - - #[inline] - pub unsafe fn read_unlock(&self) { - self.lock.lock(); - let notify = (*self.state.get()).dec_readers(); - self.lock.unlock(); - if notify { - // FIXME: should only wake up one of these some of the time - self.cond.notify_all(); - } - } - - #[inline] - pub unsafe fn write_unlock(&self) { - self.lock.lock(); - (*self.state.get()).dec_writers(); - self.lock.unlock(); - // FIXME: should only wake up one of these some of the time - self.cond.notify_all(); - } -} - -impl State { - fn inc_readers(&mut self) -> bool { - match *self { - State::Unlocked => { - *self = State::Reading(1); - true - } - State::Reading(ref mut cnt) => { - *cnt += 1; - true - } - State::Writing => false, - } - } - - fn inc_writers(&mut self) -> bool { - match *self { - State::Unlocked => { - *self = State::Writing; - true - } - State::Reading(_) | State::Writing => false, - } - } - - fn dec_readers(&mut self) -> bool { - let zero = match *self { - State::Reading(ref mut cnt) => { - *cnt -= 1; - *cnt == 0 - } - State::Unlocked | State::Writing => invalid(), - }; - if zero { - *self = State::Unlocked; - } - zero - } - - fn dec_writers(&mut self) { - match *self { - State::Writing => {} - State::Unlocked | State::Reading(_) => invalid(), - } - *self = State::Unlocked; - } -} - -fn invalid() -> ! { - panic!("inconsistent rwlock"); -} diff --git a/library/std/src/sys/itron/mutex.rs b/library/std/src/sys/itron/mutex.rs index 715e94c3b..085662e6d 100644 --- a/library/std/src/sys/itron/mutex.rs +++ b/library/std/src/sys/itron/mutex.rs @@ -31,12 +31,6 @@ impl Mutex { Mutex { mtx: SpinIdOnceCell::new() } } - pub unsafe fn init(&mut self) { - // Initialize `self.mtx` eagerly - let id = new_mtx().unwrap_or_else(|e| fail(e, &"acre_mtx")); - unsafe { self.mtx.set_unchecked((id, ())) }; - } - /// Get the inner mutex's ID, which is lazily created. fn raw(&self) -> abi::ID { match self.mtx.get_or_try_init(|| new_mtx().map(|id| (id, ()))) { diff --git a/library/std/src/sys/sgx/abi/thread.rs b/library/std/src/sys/sgx/abi/thread.rs index ef55b821a..2b23e368c 100644 --- a/library/std/src/sys/sgx/abi/thread.rs +++ b/library/std/src/sys/sgx/abi/thread.rs @@ -7,7 +7,11 @@ use fortanix_sgx_abi::Tcs; #[unstable(feature = "sgx_platform", issue = "56975")] pub fn current() -> Tcs { extern "C" { - fn get_tcs_addr() -> Tcs; + fn get_tcs_addr() -> *mut u8; + } + let addr = unsafe { get_tcs_addr() }; + match Tcs::new(addr) { + Some(tcs) => tcs, + None => rtabort!("TCS must not be placed at address zero (this is a linker error)"), } - unsafe { get_tcs_addr() } } diff --git a/library/std/src/sys/sgx/abi/usercalls/alloc.rs b/library/std/src/sys/sgx/abi/usercalls/alloc.rs index ea24fedd0..5409bd177 100644 --- a/library/std/src/sys/sgx/abi/usercalls/alloc.rs +++ b/library/std/src/sys/sgx/abi/usercalls/alloc.rs @@ -56,6 +56,8 @@ unsafe impl UserSafeSized for Usercall {} #[unstable(feature = "sgx_platform", issue = "56975")] unsafe impl UserSafeSized for Return {} #[unstable(feature = "sgx_platform", issue = "56975")] +unsafe impl UserSafeSized for Cancel {} +#[unstable(feature = "sgx_platform", issue = "56975")] unsafe impl UserSafeSized for [T; 2] {} /// A type that can be represented in memory as one or more `UserSafeSized`s. @@ -115,7 +117,7 @@ pub unsafe trait UserSafe { /// * the pointer is null. /// * the pointed-to range is not in user memory. unsafe fn check_ptr(ptr: *const Self) { - let is_aligned = |p| -> bool { 0 == (p as usize) & (Self::align_of() - 1) }; + let is_aligned = |p: *const u8| -> bool { p.is_aligned_to(Self::align_of()) }; assert!(is_aligned(ptr as *const u8)); assert!(is_user_range(ptr as _, mem::size_of_val(unsafe { &*ptr }))); @@ -305,6 +307,34 @@ where } } +// Split a memory region ptr..ptr + len into three parts: +// +--------+ +// | small0 | Chunk smaller than 8 bytes +// +--------+ +// | big | Chunk 8-byte aligned, and size a multiple of 8 bytes +// +--------+ +// | small1 | Chunk smaller than 8 bytes +// +--------+ +fn region_as_aligned_chunks(ptr: *const u8, len: usize) -> (usize, usize, usize) { + let small0_size = if ptr as usize % 8 == 0 { 0 } else { 8 - ptr as usize % 8 }; + let small1_size = (len - small0_size as usize) % 8; + let big_size = len - small0_size as usize - small1_size as usize; + + (small0_size, big_size, small1_size) +} + +unsafe fn copy_quadwords(src: *const u8, dst: *mut u8, len: usize) { + unsafe { + asm!( + "rep movsq (%rsi), (%rdi)", + inout("rcx") len / 8 => _, + inout("rdi") dst => _, + inout("rsi") src => _, + options(att_syntax, nostack, preserves_flags) + ); + } +} + /// Copies `len` bytes of data from enclave pointer `src` to userspace `dst` /// /// This function mitigates stale data vulnerabilities by ensuring all writes to untrusted memory are either: @@ -343,17 +373,6 @@ pub(crate) unsafe fn copy_to_userspace(src: *const u8, dst: *mut u8, len: usize) } } - unsafe fn copy_aligned_quadwords_to_userspace(src: *const u8, dst: *mut u8, len: usize) { - unsafe { - asm!( - "rep movsq (%rsi), (%rdi)", - inout("rcx") len / 8 => _, - inout("rdi") dst => _, - inout("rsi") src => _, - options(att_syntax, nostack, preserves_flags) - ); - } - } assert!(!src.is_null()); assert!(!dst.is_null()); assert!(is_enclave_range(src, len)); @@ -367,10 +386,10 @@ pub(crate) unsafe fn copy_to_userspace(src: *const u8, dst: *mut u8, len: usize) unsafe { copy_bytewise_to_userspace(src, dst, len); } - } else if len % 8 == 0 && dst as usize % 8 == 0 { + } else if len % 8 == 0 && dst.is_aligned_to(8) { // Copying 8-byte aligned quadwords: copy quad word per quad word unsafe { - copy_aligned_quadwords_to_userspace(src, dst, len); + copy_quadwords(src, dst, len); } } else { // Split copies into three parts: @@ -381,20 +400,16 @@ pub(crate) unsafe fn copy_to_userspace(src: *const u8, dst: *mut u8, len: usize) // +--------+ // | small1 | Chunk smaller than 8 bytes // +--------+ + let (small0_size, big_size, small1_size) = region_as_aligned_chunks(dst, len); unsafe { // Copy small0 - let small0_size = (8 - dst as usize % 8) as u8; - let small0_src = src; - let small0_dst = dst; - copy_bytewise_to_userspace(small0_src as _, small0_dst, small0_size as _); + copy_bytewise_to_userspace(src, dst, small0_size as _); // Copy big - let small1_size = ((len - small0_size as usize) % 8) as u8; - let big_size = len - small0_size as usize - small1_size as usize; let big_src = src.offset(small0_size as _); let big_dst = dst.offset(small0_size as _); - copy_aligned_quadwords_to_userspace(big_src as _, big_dst, big_size); + copy_quadwords(big_src as _, big_dst, big_size); // Copy small1 let small1_src = src.offset(big_size as isize + small0_size as isize); @@ -404,6 +419,106 @@ pub(crate) unsafe fn copy_to_userspace(src: *const u8, dst: *mut u8, len: usize) } } +/// Copies `len` bytes of data from userspace pointer `src` to enclave pointer `dst` +/// +/// This function mitigates AEPIC leak vulnerabilities by ensuring all reads from untrusted memory are 8-byte aligned +/// +/// # Panics +/// This function panics if: +/// +/// * The `src` pointer is null +/// * The `dst` pointer is null +/// * The `src` memory range is not in user memory +/// * The `dst` memory range is not in enclave memory +/// +/// # References +/// - https://www.intel.com/content/www/us/en/security-center/advisory/intel-sa-00657.html +/// - https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/advisory-guidance/stale-data-read-from-xapic.html +pub(crate) unsafe fn copy_from_userspace(src: *const u8, dst: *mut u8, len: usize) { + // Copies memory region `src..src + len` to the enclave at `dst`. The source memory region + // is: + // - strictly less than 8 bytes in size and may be + // - located at a misaligned memory location + fn copy_misaligned_chunk_to_enclave(src: *const u8, dst: *mut u8, len: usize) { + let mut tmp_buff = [0u8; 16]; + + unsafe { + // Compute an aligned memory region to read from + // +--------+ <-- aligned_src + aligned_len (8B-aligned) + // | pad1 | + // +--------+ <-- src + len (misaligned) + // | | + // | | + // | | + // +--------+ <-- src (misaligned) + // | pad0 | + // +--------+ <-- aligned_src (8B-aligned) + let pad0_size = src as usize % 8; + let aligned_src = src.sub(pad0_size); + + let pad1_size = 8 - (src.add(len) as usize % 8); + let aligned_len = pad0_size + len + pad1_size; + + debug_assert!(len < 8); + debug_assert_eq!(aligned_src as usize % 8, 0); + debug_assert_eq!(aligned_len % 8, 0); + debug_assert!(aligned_len <= 16); + + // Copy the aligned buffer to a temporary buffer + // Note: copying from a slightly different memory location is a bit odd. In this case it + // can't lead to page faults or inadvertent copying from the enclave as we only ensured + // that the `src` pointer is aligned at an 8 byte boundary. As pages are 4096 bytes + // aligned, `aligned_src` must be on the same page as `src`. A similar argument can be made + // for `src + len` + copy_quadwords(aligned_src as _, tmp_buff.as_mut_ptr(), aligned_len); + + // Copy the correct parts of the temporary buffer to the destination + ptr::copy(tmp_buff.as_ptr().add(pad0_size), dst, len); + } + } + + assert!(!src.is_null()); + assert!(!dst.is_null()); + assert!(is_user_range(src, len)); + assert!(is_enclave_range(dst, len)); + assert!(!(src as usize).overflowing_add(len + 8).1); + assert!(!(dst as usize).overflowing_add(len + 8).1); + + if len < 8 { + copy_misaligned_chunk_to_enclave(src, dst, len); + } else if len % 8 == 0 && src as usize % 8 == 0 { + // Copying 8-byte aligned quadwords: copy quad word per quad word + unsafe { + copy_quadwords(src, dst, len); + } + } else { + // Split copies into three parts: + // +--------+ + // | small0 | Chunk smaller than 8 bytes + // +--------+ + // | big | Chunk 8-byte aligned, and size a multiple of 8 bytes + // +--------+ + // | small1 | Chunk smaller than 8 bytes + // +--------+ + let (small0_size, big_size, small1_size) = region_as_aligned_chunks(dst, len); + + unsafe { + // Copy small0 + copy_misaligned_chunk_to_enclave(src, dst, small0_size); + + // Copy big + let big_src = src.add(small0_size); + let big_dst = dst.add(small0_size); + copy_quadwords(big_src, big_dst, big_size); + + // Copy small1 + let small1_src = src.add(big_size + small0_size); + let small1_dst = dst.add(big_size + small0_size); + copy_misaligned_chunk_to_enclave(small1_src, small1_dst, small1_size); + } + } +} + #[unstable(feature = "sgx_platform", issue = "56975")] impl UserRef where @@ -468,7 +583,7 @@ where pub fn copy_to_enclave(&self, dest: &mut T) { unsafe { assert_eq!(mem::size_of_val(dest), mem::size_of_val(&*self.0.get())); - ptr::copy( + copy_from_userspace( self.0.get() as *const T as *const u8, dest as *mut T as *mut u8, mem::size_of_val(dest), @@ -494,7 +609,11 @@ where { /// Copies the value from user memory into enclave memory. pub fn to_enclave(&self) -> T { - unsafe { ptr::read(self.0.get()) } + unsafe { + let mut data: T = mem::MaybeUninit::uninit().assume_init(); + copy_from_userspace(self.0.get() as _, &mut data as *mut T as _, mem::size_of::()); + data + } } } diff --git a/library/std/src/sys/sgx/abi/usercalls/mod.rs b/library/std/src/sys/sgx/abi/usercalls/mod.rs index 79d1db5e1..e19e84326 100644 --- a/library/std/src/sys/sgx/abi/usercalls/mod.rs +++ b/library/std/src/sys/sgx/abi/usercalls/mod.rs @@ -292,12 +292,17 @@ fn check_os_error(err: Result) -> i32 { } } -trait FromSgxResult { +/// Translate the raw result of an SGX usercall. +#[unstable(feature = "sgx_platform", issue = "56975")] +pub trait FromSgxResult { + /// Return type type Return; + /// Translate the raw result of an SGX usercall. fn from_sgx_result(self) -> IoResult; } +#[unstable(feature = "sgx_platform", issue = "56975")] impl FromSgxResult for (Result, T) { type Return = T; @@ -310,6 +315,7 @@ impl FromSgxResult for (Result, T) { } } +#[unstable(feature = "sgx_platform", issue = "56975")] impl FromSgxResult for Result { type Return = (); diff --git a/library/std/src/sys/sgx/abi/usercalls/raw.rs b/library/std/src/sys/sgx/abi/usercalls/raw.rs index 4267b96cc..10c1456d4 100644 --- a/library/std/src/sys/sgx/abi/usercalls/raw.rs +++ b/library/std/src/sys/sgx/abi/usercalls/raw.rs @@ -37,14 +37,23 @@ pub unsafe fn do_usercall( (a, b) } -type Register = u64; +/// A value passed or returned in a CPU register. +#[unstable(feature = "sgx_platform", issue = "56975")] +pub type Register = u64; -trait RegisterArgument { +/// Translate a type from/to Register to be used as an argument. +#[unstable(feature = "sgx_platform", issue = "56975")] +pub trait RegisterArgument { + /// Translate a Register to Self. fn from_register(_: Register) -> Self; + /// Translate self to a Register. fn into_register(self) -> Register; } -trait ReturnValue { +/// Translate a pair of Registers to the raw usercall return value. +#[unstable(feature = "sgx_platform", issue = "56975")] +pub trait ReturnValue { + /// Translate a pair of Registers to the raw usercall return value. fn from_registers(call: &'static str, regs: (Register, Register)) -> Self; } @@ -68,6 +77,7 @@ macro_rules! define_usercalls { macro_rules! define_ra { (< $i:ident > $t:ty) => { + #[unstable(feature = "sgx_platform", issue = "56975")] impl<$i> RegisterArgument for $t { fn from_register(a: Register) -> Self { a as _ @@ -78,6 +88,7 @@ macro_rules! define_ra { } }; ($i:ty as $t:ty) => { + #[unstable(feature = "sgx_platform", issue = "56975")] impl RegisterArgument for $t { fn from_register(a: Register) -> Self { a as $i as _ @@ -88,6 +99,7 @@ macro_rules! define_ra { } }; ($t:ty) => { + #[unstable(feature = "sgx_platform", issue = "56975")] impl RegisterArgument for $t { fn from_register(a: Register) -> Self { a as _ @@ -112,6 +124,7 @@ define_ra!(usize as isize); define_ra!( *const T); define_ra!( *mut T); +#[unstable(feature = "sgx_platform", issue = "56975")] impl RegisterArgument for bool { fn from_register(a: Register) -> bool { if a != 0 { true } else { false } @@ -121,6 +134,7 @@ impl RegisterArgument for bool { } } +#[unstable(feature = "sgx_platform", issue = "56975")] impl RegisterArgument for Option> { fn from_register(a: Register) -> Option> { NonNull::new(a as _) @@ -130,12 +144,14 @@ impl RegisterArgument for Option> { } } +#[unstable(feature = "sgx_platform", issue = "56975")] impl ReturnValue for ! { fn from_registers(call: &'static str, _regs: (Register, Register)) -> Self { rtabort!("Usercall {call}: did not expect to be re-entered"); } } +#[unstable(feature = "sgx_platform", issue = "56975")] impl ReturnValue for () { fn from_registers(call: &'static str, usercall_retval: (Register, Register)) -> Self { rtassert!(usercall_retval.0 == 0); @@ -144,6 +160,7 @@ impl ReturnValue for () { } } +#[unstable(feature = "sgx_platform", issue = "56975")] impl ReturnValue for T { fn from_registers(call: &'static str, usercall_retval: (Register, Register)) -> Self { rtassert!(usercall_retval.1 == 0); @@ -151,6 +168,7 @@ impl ReturnValue for T { } } +#[unstable(feature = "sgx_platform", issue = "56975")] impl ReturnValue for (T, U) { fn from_registers(_call: &'static str, regs: (Register, Register)) -> Self { (T::from_register(regs.0), U::from_register(regs.1)) diff --git a/library/std/src/sys/sgx/abi/usercalls/tests.rs b/library/std/src/sys/sgx/abi/usercalls/tests.rs index cbf7d7d54..58b8eb215 100644 --- a/library/std/src/sys/sgx/abi/usercalls/tests.rs +++ b/library/std/src/sys/sgx/abi/usercalls/tests.rs @@ -1,8 +1,8 @@ -use super::alloc::copy_to_userspace; use super::alloc::User; +use super::alloc::{copy_from_userspace, copy_to_userspace}; #[test] -fn test_copy_function() { +fn test_copy_to_userspace_function() { let mut src = [0u8; 100]; let mut dst = User::<[u8]>::uninitialized(100); @@ -17,12 +17,38 @@ fn test_copy_function() { dst.copy_from_enclave(&[0u8; 100]); // Copy src[0..size] to dst + offset - unsafe { copy_to_userspace(src.as_ptr(), dst.as_mut_ptr().offset(offset), size) }; + unsafe { copy_to_userspace(src.as_ptr(), dst.as_mut_ptr().add(offset), size) }; // Verify copy for byte in 0..size { unsafe { - assert_eq!(*dst.as_ptr().offset(offset + byte as isize), src[byte as usize]); + assert_eq!(*dst.as_ptr().add(offset + byte), src[byte as usize]); + } + } + } + } +} + +#[test] +fn test_copy_from_userspace_function() { + let mut dst = [0u8; 100]; + let mut src = User::<[u8]>::uninitialized(100); + + src.copy_from_enclave(&[0u8; 100]); + + for size in 0..48 { + // For all possible alignment + for offset in 0..8 { + // overwrite complete dst + dst = [0u8; 100]; + + // Copy src[0..size] to dst + offset + unsafe { copy_from_userspace(src.as_ptr().offset(offset), dst.as_mut_ptr(), size) }; + + // Verify copy + for byte in 0..size { + unsafe { + assert_eq!(dst[byte as usize], *src.as_ptr().offset(offset + byte as isize)); } } } diff --git a/library/std/src/sys/sgx/mod.rs b/library/std/src/sys/sgx/mod.rs index 696400670..b1d32929e 100644 --- a/library/std/src/sys/sgx/mod.rs +++ b/library/std/src/sys/sgx/mod.rs @@ -47,7 +47,7 @@ pub mod locks { // SAFETY: must be called only once during runtime initialization. // NOTE: this is not guaranteed to run, for example when Rust code is called externally. -pub unsafe fn init(argc: isize, argv: *const *const u8) { +pub unsafe fn init(argc: isize, argv: *const *const u8, _sigpipe: u8) { unsafe { args::init(argc, argv); } diff --git a/library/std/src/sys/sgx/mutex.rs b/library/std/src/sys/sgx/mutex.rs index 513cd77fd..aa747d56b 100644 --- a/library/std/src/sys/sgx/mutex.rs +++ b/library/std/src/sys/sgx/mutex.rs @@ -20,9 +20,6 @@ impl Mutex { Mutex { inner: SpinMutex::new(WaitVariable::new(false)) } } - #[inline] - pub unsafe fn init(&mut self) {} - #[inline] pub unsafe fn lock(&self) { let mut guard = self.inner.lock(); diff --git a/library/std/src/sys/solid/fs.rs b/library/std/src/sys/solid/fs.rs index a2cbee4dc..969222253 100644 --- a/library/std/src/sys/solid/fs.rs +++ b/library/std/src/sys/solid/fs.rs @@ -2,7 +2,7 @@ use super::{abi, error}; use crate::{ ffi::{CStr, CString, OsStr, OsString}, fmt, - io::{self, IoSlice, IoSliceMut, ReadBuf, SeekFrom}, + io::{self, BorrowedCursor, IoSlice, IoSliceMut, SeekFrom}, mem::MaybeUninit, os::raw::{c_int, c_short}, os::solid::ffi::OsStrExt, @@ -77,6 +77,9 @@ pub struct OpenOptions { custom_flags: i32, } +#[derive(Copy, Clone, Debug, Default)] +pub struct FileTimes {} + #[derive(Clone, PartialEq, Eq, Debug)] pub struct FilePermissions(c_short); @@ -126,6 +129,11 @@ impl FilePermissions { } } +impl FileTimes { + pub fn set_accessed(&mut self, _t: SystemTime) {} + pub fn set_modified(&mut self, _t: SystemTime) {} +} + impl FileType { pub fn is_dir(&self) -> bool { self.is(abi::S_IFDIR) @@ -358,13 +366,13 @@ impl File { } } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { + pub fn read_buf(&self, mut cursor: BorrowedCursor<'_>) -> io::Result<()> { unsafe { - let len = buf.remaining(); + let len = cursor.capacity(); let mut out_num_bytes = MaybeUninit::uninit(); error::SolidError::err_if_negative(abi::SOLID_FS_Read( self.fd.raw(), - buf.unfilled_mut().as_mut_ptr() as *mut u8, + cursor.as_mut().as_mut_ptr() as *mut u8, len, out_num_bytes.as_mut_ptr(), )) @@ -376,9 +384,7 @@ impl File { // Safety: `num_bytes_read` bytes were written to the unfilled // portion of the buffer - buf.assume_init(num_bytes_read); - - buf.add_filled(num_bytes_read); + cursor.advance(num_bytes_read); Ok(()) } @@ -452,6 +458,10 @@ impl File { pub fn set_permissions(&self, _perm: FilePermissions) -> io::Result<()> { unsupported() } + + pub fn set_times(&self, _times: FileTimes) -> io::Result<()> { + unsupported() + } } impl Drop for File { diff --git a/library/std/src/sys/solid/mod.rs b/library/std/src/sys/solid/mod.rs index 778a589d1..5867979a2 100644 --- a/library/std/src/sys/solid/mod.rs +++ b/library/std/src/sys/solid/mod.rs @@ -56,7 +56,7 @@ pub mod locks { // SAFETY: must be called only once during runtime initialization. // NOTE: this is not guaranteed to run, for example when Rust code is called externally. -pub unsafe fn init(_argc: isize, _argv: *const *const u8) {} +pub unsafe fn init(_argc: isize, _argv: *const *const u8, _sigpipe: u8) {} // SAFETY: must be called only once during runtime cleanup. pub unsafe fn cleanup() {} diff --git a/library/std/src/sys/unix/fd.rs b/library/std/src/sys/unix/fd.rs index 30812dabb..dbaa3c33e 100644 --- a/library/std/src/sys/unix/fd.rs +++ b/library/std/src/sys/unix/fd.rs @@ -4,7 +4,7 @@ mod tests; use crate::cmp; -use crate::io::{self, IoSlice, IoSliceMut, Read, ReadBuf}; +use crate::io::{self, BorrowedCursor, IoSlice, IoSliceMut, Read}; use crate::os::unix::io::{AsFd, AsRawFd, BorrowedFd, FromRawFd, IntoRawFd, OwnedFd, RawFd}; use crate::sys::cvt; use crate::sys_common::{AsInner, FromInner, IntoInner}; @@ -131,20 +131,19 @@ impl FileDesc { } } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { + pub fn read_buf(&self, mut cursor: BorrowedCursor<'_>) -> io::Result<()> { let ret = cvt(unsafe { libc::read( self.as_raw_fd(), - buf.unfilled_mut().as_mut_ptr() as *mut libc::c_void, - cmp::min(buf.remaining(), READ_LIMIT), + cursor.as_mut().as_mut_ptr() as *mut libc::c_void, + cmp::min(cursor.capacity(), READ_LIMIT), ) })?; // Safety: `ret` bytes were written to the initialized portion of the buffer unsafe { - buf.assume_init(ret as usize); + cursor.advance(ret as usize); } - buf.add_filled(ret as usize); Ok(()) } diff --git a/library/std/src/sys/unix/fs.rs b/library/std/src/sys/unix/fs.rs index b5cc8038c..cc347e358 100644 --- a/library/std/src/sys/unix/fs.rs +++ b/library/std/src/sys/unix/fs.rs @@ -2,7 +2,7 @@ use crate::os::unix::prelude::*; use crate::ffi::{CStr, CString, OsStr, OsString}; use crate::fmt; -use crate::io::{self, Error, IoSlice, IoSliceMut, ReadBuf, SeekFrom}; +use crate::io::{self, BorrowedCursor, Error, IoSlice, IoSliceMut, SeekFrom}; use crate::mem; use crate::os::unix::io::{AsFd, AsRawFd, BorrowedFd, FromRawFd, IntoRawFd}; use crate::path::{Path, PathBuf}; @@ -544,11 +544,11 @@ impl Default for FileTimes { fn default() -> Self { // Redox doesn't appear to support `UTIME_OMIT`, so we stub it out here, and always return // an error in `set_times`. - // ESP-IDF does not support `futimens` at all and the behavior for that OS is therefore + // ESP-IDF and HorizonOS do not support `futimens` at all and the behavior for those OS is therefore // the same as for Redox. - #[cfg(any(target_os = "redox", target_os = "espidf"))] + #[cfg(any(target_os = "redox", target_os = "espidf", target_os = "horizon"))] let omit = libc::timespec { tv_sec: 0, tv_nsec: 0 }; - #[cfg(not(any(target_os = "redox", target_os = "espidf")))] + #[cfg(not(any(target_os = "redox", target_os = "espidf", target_os = "horizon")))] let omit = libc::timespec { tv_sec: 0, tv_nsec: libc::UTIME_OMIT as _ }; Self([omit; 2]) } @@ -687,7 +687,11 @@ impl Iterator for ReadDir { impl Drop for Dir { fn drop(&mut self) { let r = unsafe { libc::closedir(self.0) }; - debug_assert_eq!(r, 0); + assert!( + r == 0 || crate::io::Error::last_os_error().kind() == crate::io::ErrorKind::Interrupted, + "unexpected error during closedir: {:?}", + crate::io::Error::last_os_error() + ); } } @@ -825,6 +829,7 @@ impl DirEntry { target_os = "fuchsia", target_os = "redox" )))] + #[cfg_attr(miri, allow(unused))] fn name_cstr(&self) -> &CStr { unsafe { CStr::from_ptr(self.entry.d_name.as_ptr()) } } @@ -836,6 +841,7 @@ impl DirEntry { target_os = "fuchsia", target_os = "redox" ))] + #[cfg_attr(miri, allow(unused))] fn name_cstr(&self) -> &CStr { &self.name } @@ -1031,8 +1037,8 @@ impl File { self.0.read_at(buf, offset) } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { - self.0.read_buf(buf) + pub fn read_buf(&self, cursor: BorrowedCursor<'_>) -> io::Result<()> { + self.0.read_buf(cursor) } pub fn write(&self, buf: &[u8]) -> io::Result { @@ -1079,9 +1085,9 @@ impl File { pub fn set_times(&self, times: FileTimes) -> io::Result<()> { cfg_if::cfg_if! { - if #[cfg(any(target_os = "redox", target_os = "espidf"))] { + if #[cfg(any(target_os = "redox", target_os = "espidf", target_os = "horizon"))] { // Redox doesn't appear to support `UTIME_OMIT`. - // ESP-IDF does not support `futimens` at all and the behavior for that OS is therefore + // ESP-IDF and HorizonOS do not support `futimens` at all and the behavior for those OS is therefore // the same as for Redox. drop(times); Err(io::const_io_error!( diff --git a/library/std/src/sys/unix/locks/fuchsia_mutex.rs b/library/std/src/sys/unix/locks/fuchsia_mutex.rs index ce427599c..117611ce4 100644 --- a/library/std/src/sys/unix/locks/fuchsia_mutex.rs +++ b/library/std/src/sys/unix/locks/fuchsia_mutex.rs @@ -85,9 +85,6 @@ impl Mutex { Mutex { futex: AtomicU32::new(UNLOCKED) } } - #[inline] - pub unsafe fn init(&mut self) {} - #[inline] pub unsafe fn try_lock(&self) -> bool { let thread_self = zx_thread_self(); @@ -138,7 +135,7 @@ impl Mutex { } } - // The state has changed or a wakeup occured, try to lock the mutex. + // The state has changed or a wakeup occurred, try to lock the mutex. match self.futex.compare_exchange(UNLOCKED, owned_state, Acquire, Relaxed) { Ok(_) => return, Err(updated) => state = updated, diff --git a/library/std/src/sys/unix/locks/futex_mutex.rs b/library/std/src/sys/unix/locks/futex_mutex.rs index 99ba86e5f..33b13dad4 100644 --- a/library/std/src/sys/unix/locks/futex_mutex.rs +++ b/library/std/src/sys/unix/locks/futex_mutex.rs @@ -19,9 +19,6 @@ impl Mutex { Self { futex: AtomicU32::new(0) } } - #[inline] - pub unsafe fn init(&mut self) {} - #[inline] pub unsafe fn try_lock(&self) -> bool { self.futex.compare_exchange(0, 1, Acquire, Relaxed).is_ok() @@ -53,7 +50,7 @@ impl Mutex { // We avoid an unnecessary write if it as already set to 2, // to be friendlier for the caches. if state != 2 && self.futex.swap(2, Acquire) == 0 { - // We changed it from 0 to 2, so we just succesfully locked it. + // We changed it from 0 to 2, so we just successfully locked it. return; } diff --git a/library/std/src/sys/unix/locks/futex_rwlock.rs b/library/std/src/sys/unix/locks/futex_rwlock.rs index b3bbbf743..0cc92244e 100644 --- a/library/std/src/sys/unix/locks/futex_rwlock.rs +++ b/library/std/src/sys/unix/locks/futex_rwlock.rs @@ -54,7 +54,7 @@ fn is_read_lockable(state: u32) -> bool { // We don't allow read-locking if there's readers waiting, even if the lock is unlocked // and there's no writers waiting. The only situation when this happens is after unlocking, // at which point the unlocking thread might be waking up writers, which have priority over readers. - // The unlocking thread will clear the readers waiting bit and wake up readers, if necssary. + // The unlocking thread will clear the readers waiting bit and wake up readers, if necessary. state & MASK < MAX_READERS && !has_readers_waiting(state) && !has_writers_waiting(state) } diff --git a/library/std/src/sys/unix/locks/pthread_condvar.rs b/library/std/src/sys/unix/locks/pthread_condvar.rs index abf27e7db..4741c0c67 100644 --- a/library/std/src/sys/unix/locks/pthread_condvar.rs +++ b/library/std/src/sys/unix/locks/pthread_condvar.rs @@ -172,7 +172,7 @@ impl Condvar { let mut sys_now = libc::timeval { tv_sec: 0, tv_usec: 0 }; let stable_now = Instant::now(); let r = libc::gettimeofday(&mut sys_now, ptr::null_mut()); - debug_assert_eq!(r, 0); + assert_eq!(r, 0, "unexpected error: {:?}", crate::io::Error::last_os_error()); let nsec = dur.subsec_nanos() as libc::c_long + (sys_now.tv_usec * 1000) as libc::c_long; let extra = (nsec / 1_000_000_000) as libc::time_t; diff --git a/library/std/src/sys/unix/locks/pthread_mutex.rs b/library/std/src/sys/unix/locks/pthread_mutex.rs index 98afee69b..5964935dd 100644 --- a/library/std/src/sys/unix/locks/pthread_mutex.rs +++ b/library/std/src/sys/unix/locks/pthread_mutex.rs @@ -52,7 +52,7 @@ impl Mutex { Mutex { inner: UnsafeCell::new(libc::PTHREAD_MUTEX_INITIALIZER) } } #[inline] - pub unsafe fn init(&mut self) { + unsafe fn init(&mut self) { // Issue #33770 // // A pthread mutex initialized with PTHREAD_MUTEX_INITIALIZER will have diff --git a/library/std/src/sys/unix/mod.rs b/library/std/src/sys/unix/mod.rs index 3d0d91460..c84e292ea 100644 --- a/library/std/src/sys/unix/mod.rs +++ b/library/std/src/sys/unix/mod.rs @@ -44,12 +44,13 @@ pub mod thread_parker; pub mod time; #[cfg(target_os = "espidf")] -pub fn init(argc: isize, argv: *const *const u8) {} +pub fn init(argc: isize, argv: *const *const u8, _sigpipe: u8) {} #[cfg(not(target_os = "espidf"))] // SAFETY: must be called only once during runtime initialization. // NOTE: this is not guaranteed to run, for example when Rust code is called externally. -pub unsafe fn init(argc: isize, argv: *const *const u8) { +// See `fn init()` in `library/std/src/rt.rs` for docs on `sigpipe`. +pub unsafe fn init(argc: isize, argv: *const *const u8, sigpipe: u8) { // The standard streams might be closed on application startup. To prevent // std::io::{stdin, stdout,stderr} objects from using other unrelated file // resources opened later, we reopen standards streams when they are closed. @@ -61,8 +62,9 @@ pub unsafe fn init(argc: isize, argv: *const *const u8) { // want! // // Hence, we set SIGPIPE to ignore when the program starts up in order - // to prevent this problem. - reset_sigpipe(); + // to prevent this problem. Add `#[unix_sigpipe = "..."]` above `fn main()` to + // alter this behavior. + reset_sigpipe(sigpipe); stack_overflow::init(); args::init(argc, argv); @@ -151,9 +153,31 @@ pub unsafe fn init(argc: isize, argv: *const *const u8) { } } - unsafe fn reset_sigpipe() { + unsafe fn reset_sigpipe(#[allow(unused_variables)] sigpipe: u8) { #[cfg(not(any(target_os = "emscripten", target_os = "fuchsia", target_os = "horizon")))] - rtassert!(signal(libc::SIGPIPE, libc::SIG_IGN) != libc::SIG_ERR); + { + // We don't want to add this as a public type to libstd, nor do we + // want to `include!` a file from the compiler (which would break + // Miri and xargo for example), so we choose to duplicate these + // constants from `compiler/rustc_session/src/config/sigpipe.rs`. + // See the other file for docs. NOTE: Make sure to keep them in + // sync! + mod sigpipe { + pub const INHERIT: u8 = 1; + pub const SIG_IGN: u8 = 2; + pub const SIG_DFL: u8 = 3; + } + + let handler = match sigpipe { + sigpipe::INHERIT => None, + sigpipe::SIG_IGN => Some(libc::SIG_IGN), + sigpipe::SIG_DFL => Some(libc::SIG_DFL), + _ => unreachable!(), + }; + if let Some(handler) = handler { + rtassert!(signal(libc::SIGPIPE, handler) != libc::SIG_ERR); + } + } } } @@ -295,8 +319,10 @@ pub fn abort_internal() -> ! { cfg_if::cfg_if! { if #[cfg(target_os = "android")] { - #[link(name = "dl")] - #[link(name = "log")] + #[link(name = "dl", kind = "static", modifiers = "-bundle", + cfg(target_feature = "crt-static"))] + #[link(name = "dl", cfg(not(target_feature = "crt-static")))] + #[link(name = "log", cfg(not(target_feature = "crt-static")))] extern "C" {} } else if #[cfg(target_os = "freebsd")] { #[link(name = "execinfo")] diff --git a/library/std/src/sys/unix/net.rs b/library/std/src/sys/unix/net.rs index 462a45b01..b84bf8f92 100644 --- a/library/std/src/sys/unix/net.rs +++ b/library/std/src/sys/unix/net.rs @@ -392,6 +392,17 @@ impl Socket { Ok(raw != 0) } + #[cfg(any(target_os = "android", target_os = "linux",))] + pub fn set_quickack(&self, quickack: bool) -> io::Result<()> { + setsockopt(self, libc::IPPROTO_TCP, libc::TCP_QUICKACK, quickack as c_int) + } + + #[cfg(any(target_os = "android", target_os = "linux",))] + pub fn quickack(&self) -> io::Result { + let raw: c_int = getsockopt(self, libc::IPPROTO_TCP, libc::TCP_QUICKACK)?; + Ok(raw != 0) + } + #[cfg(any(target_os = "android", target_os = "linux",))] pub fn set_passcred(&self, passcred: bool) -> io::Result<()> { setsockopt(self, libc::SOL_SOCKET, libc::SO_PASSCRED, passcred as libc::c_int) @@ -427,6 +438,17 @@ impl Socket { self.0.set_nonblocking(nonblocking) } + #[cfg(any(target_os = "linux", target_os = "freebsd", target_os = "openbsd"))] + pub fn set_mark(&self, mark: u32) -> io::Result<()> { + #[cfg(target_os = "linux")] + let option = libc::SO_MARK; + #[cfg(target_os = "freebsd")] + let option = libc::SO_USER_COOKIE; + #[cfg(target_os = "openbsd")] + let option = libc::SO_RTABLE; + setsockopt(self, libc::SOL_SOCKET, option, mark as libc::c_int) + } + pub fn take_error(&self) -> io::Result> { let raw: c_int = getsockopt(self, libc::SOL_SOCKET, libc::SO_ERROR)?; if raw == 0 { Ok(None) } else { Ok(Some(io::Error::from_raw_os_error(raw as i32))) } diff --git a/library/std/src/sys/unix/os_str.rs b/library/std/src/sys/unix/os_str.rs index ccbc18224..017e2af29 100644 --- a/library/std/src/sys/unix/os_str.rs +++ b/library/std/src/sys/unix/os_str.rs @@ -11,7 +11,7 @@ use crate::str; use crate::sync::Arc; use crate::sys_common::{AsInner, IntoInner}; -use core::str::lossy::{Utf8Lossy, Utf8LossyChunk}; +use core::str::Utf8Chunks; #[cfg(test)] #[path = "../unix/os_str/tests.rs"] @@ -29,26 +29,32 @@ pub struct Slice { } impl fmt::Debug for Slice { - fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { - // Writes out a valid unicode string with the correct escape sequences - - formatter.write_str("\"")?; - for Utf8LossyChunk { valid, broken } in Utf8Lossy::from_bytes(&self.inner).chunks() { - for c in valid.chars().flat_map(|c| c.escape_debug()) { - formatter.write_char(c)? - } - - for b in broken { - write!(formatter, "\\x{:02X}", b)?; - } - } - formatter.write_str("\"") + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + fmt::Debug::fmt(&Utf8Chunks::new(&self.inner).debug(), f) } } impl fmt::Display for Slice { - fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { - fmt::Display::fmt(&Utf8Lossy::from_bytes(&self.inner), formatter) + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + // If we're the empty string then our iterator won't actually yield + // anything, so perform the formatting manually + if self.inner.is_empty() { + return "".fmt(f); + } + + for chunk in Utf8Chunks::new(&self.inner) { + let valid = chunk.valid(); + // If we successfully decoded the whole chunk as a valid string then + // we can return a direct formatting of the string which will also + // respect various formatting flags if possible. + if chunk.invalid().is_empty() { + return valid.fmt(f); + } + + f.write_str(valid)?; + f.write_char(char::REPLACEMENT_CHARACTER)?; + } + Ok(()) } } diff --git a/library/std/src/sys/unix/os_str/tests.rs b/library/std/src/sys/unix/os_str/tests.rs index 213277f01..22ba0c923 100644 --- a/library/std/src/sys/unix/os_str/tests.rs +++ b/library/std/src/sys/unix/os_str/tests.rs @@ -8,3 +8,11 @@ fn slice_debug_output() { assert_eq!(output, expected); } + +#[test] +fn display() { + assert_eq!( + "Hello\u{FFFD}\u{FFFD} There\u{FFFD} Goodbye", + Slice::from_u8_slice(b"Hello\xC0\x80 There\xE6\x83 Goodbye").to_string(), + ); +} diff --git a/library/std/src/sys/unix/process/process_common.rs b/library/std/src/sys/unix/process/process_common.rs index bca1b65a7..2834ee0ac 100644 --- a/library/std/src/sys/unix/process/process_common.rs +++ b/library/std/src/sys/unix/process/process_common.rs @@ -45,11 +45,31 @@ cfg_if::cfg_if! { } #[allow(dead_code)] pub unsafe fn sigaddset(set: *mut libc::sigset_t, signum: libc::c_int) -> libc::c_int { - use crate::{slice, mem}; + use crate::{ + mem::{align_of, size_of}, + slice, + }; + use libc::{c_ulong, sigset_t}; + + // The implementations from bionic (android libc) type pun `sigset_t` as an + // array of `c_ulong`. This works, but lets add a smoke check to make sure + // that doesn't change. + const _: () = assert!( + align_of::() == align_of::() + && (size_of::() % size_of::()) == 0 + ); - let raw = slice::from_raw_parts_mut(set as *mut u8, mem::size_of::()); let bit = (signum - 1) as usize; - raw[bit / 8] |= 1 << (bit % 8); + if set.is_null() || bit >= (8 * size_of::()) { + crate::sys::unix::os::set_errno(libc::EINVAL); + return -1; + } + let raw = slice::from_raw_parts_mut( + set as *mut c_ulong, + size_of::() / size_of::(), + ); + const LONG_BIT: usize = size_of::() * 8; + raw[bit / LONG_BIT] |= 1 << (bit % LONG_BIT); return 0; } } else { @@ -72,6 +92,7 @@ pub struct Command { argv: Argv, env: CommandEnv, + program_kind: ProgramKind, cwd: Option, uid: Option, gid: Option, @@ -128,15 +149,40 @@ pub enum Stdio { Fd(FileDesc), } +#[derive(Copy, Clone, Debug, Eq, PartialEq)] +pub enum ProgramKind { + /// A program that would be looked up on the PATH (e.g. `ls`) + PathLookup, + /// A relative path (e.g. `my-dir/foo`, `../foo`, `./foo`) + Relative, + /// An absolute path. + Absolute, +} + +impl ProgramKind { + fn new(program: &OsStr) -> Self { + if program.bytes().starts_with(b"/") { + Self::Absolute + } else if program.bytes().contains(&b'/') { + // If the program has more than one component in it, it is a relative path. + Self::Relative + } else { + Self::PathLookup + } + } +} + impl Command { #[cfg(not(target_os = "linux"))] pub fn new(program: &OsStr) -> Command { let mut saw_nul = false; + let program_kind = ProgramKind::new(program.as_ref()); let program = os2c(program, &mut saw_nul); Command { argv: Argv(vec![program.as_ptr(), ptr::null()]), args: vec![program.clone()], program, + program_kind, env: Default::default(), cwd: None, uid: None, @@ -154,11 +200,13 @@ impl Command { #[cfg(target_os = "linux")] pub fn new(program: &OsStr) -> Command { let mut saw_nul = false; + let program_kind = ProgramKind::new(program.as_ref()); let program = os2c(program, &mut saw_nul); Command { argv: Argv(vec![program.as_ptr(), ptr::null()]), args: vec![program.clone()], program, + program_kind, env: Default::default(), cwd: None, uid: None, @@ -234,6 +282,11 @@ impl Command { OsStr::from_bytes(self.program.as_bytes()) } + #[allow(dead_code)] + pub fn get_program_kind(&self) -> ProgramKind { + self.program_kind + } + pub fn get_args(&self) -> CommandArgs<'_> { let mut iter = self.args.iter(); iter.next(); diff --git a/library/std/src/sys/unix/process/process_common/tests.rs b/library/std/src/sys/unix/process/process_common/tests.rs index 1956b3692..d176b3401 100644 --- a/library/std/src/sys/unix/process/process_common/tests.rs +++ b/library/std/src/sys/unix/process/process_common/tests.rs @@ -122,3 +122,27 @@ fn test_process_group_no_posix_spawn() { t!(cat.wait()); } } + +#[test] +fn test_program_kind() { + let vectors = &[ + ("foo", ProgramKind::PathLookup), + ("foo.out", ProgramKind::PathLookup), + ("./foo", ProgramKind::Relative), + ("../foo", ProgramKind::Relative), + ("dir/foo", ProgramKind::Relative), + // Note that paths on Unix can't contain / in them, so this is actually the directory "fo\\" + // followed by the file "o". + ("fo\\/o", ProgramKind::Relative), + ("/foo", ProgramKind::Absolute), + ("/dir/../foo", ProgramKind::Absolute), + ]; + + for (program, expected_kind) in vectors { + assert_eq!( + ProgramKind::new(program.as_ref()), + *expected_kind, + "actual != expected program kind for input {program}", + ); + } +} diff --git a/library/std/src/sys/unix/process/process_unix.rs b/library/std/src/sys/unix/process/process_unix.rs index 75bb92437..26ae62817 100644 --- a/library/std/src/sys/unix/process/process_unix.rs +++ b/library/std/src/sys/unix/process/process_unix.rs @@ -453,7 +453,9 @@ impl Command { // successfully launch the program, but erroneously return // ENOENT when used with posix_spawn_file_actions_addchdir_np // which was introduced in macOS 10.15. - return Ok(None); + if self.get_program_kind() == ProgramKind::Relative { + return Ok(None); + } } match posix_spawn_file_actions_addchdir_np.get() { Some(f) => Some((f, cwd)), diff --git a/library/std/src/sys/unix/rand.rs b/library/std/src/sys/unix/rand.rs index bf4920488..a6fe07873 100644 --- a/library/std/src/sys/unix/rand.rs +++ b/library/std/src/sys/unix/rand.rs @@ -1,13 +1,13 @@ -use crate::mem; -use crate::slice; - pub fn hashmap_random_keys() -> (u64, u64) { - let mut v = (0, 0); - unsafe { - let view = slice::from_raw_parts_mut(&mut v as *mut _ as *mut u8, mem::size_of_val(&v)); - imp::fill_bytes(view); - } - v + const KEY_LEN: usize = core::mem::size_of::(); + + let mut v = [0u8; KEY_LEN * 2]; + imp::fill_bytes(&mut v); + + let key1 = v[0..KEY_LEN].try_into().unwrap(); + let key2 = v[KEY_LEN..].try_into().unwrap(); + + (u64::from_ne_bytes(key1), u64::from_ne_bytes(key2)) } #[cfg(all( diff --git a/library/std/src/sys/unix/thread.rs b/library/std/src/sys/unix/thread.rs index 36a3fa602..f6b627afc 100644 --- a/library/std/src/sys/unix/thread.rs +++ b/library/std/src/sys/unix/thread.rs @@ -116,11 +116,9 @@ impl Thread { debug_assert_eq!(ret, 0); } - #[cfg(any(target_os = "linux", target_os = "android"))] + #[cfg(target_os = "android")] pub fn set_name(name: &CStr) { const PR_SET_NAME: libc::c_int = 15; - // pthread wrapper only appeared in glibc 2.12, so we use syscall - // directly. unsafe { libc::prctl( PR_SET_NAME, @@ -132,6 +130,17 @@ impl Thread { } } + #[cfg(target_os = "linux")] + pub fn set_name(name: &CStr) { + const TASK_COMM_LEN: usize = 16; + + unsafe { + // Available since glibc 2.12, musl 1.1.16, and uClibc 1.0.20. + let name = truncate_cstr(name, TASK_COMM_LEN); + libc::pthread_setname_np(libc::pthread_self(), name.as_ptr()); + } + } + #[cfg(any(target_os = "freebsd", target_os = "dragonfly", target_os = "openbsd"))] pub fn set_name(name: &CStr) { unsafe { @@ -142,6 +151,7 @@ impl Thread { #[cfg(any(target_os = "macos", target_os = "ios", target_os = "watchos"))] pub fn set_name(name: &CStr) { unsafe { + let name = truncate_cstr(name, libc::MAXTHREADNAMESIZE); libc::pthread_setname_np(name.as_ptr()); } } @@ -271,6 +281,20 @@ impl Drop for Thread { } } +#[cfg(any(target_os = "linux", target_os = "macos", target_os = "ios", target_os = "watchos"))] +fn truncate_cstr(cstr: &CStr, max_with_nul: usize) -> crate::borrow::Cow<'_, CStr> { + use crate::{borrow::Cow, ffi::CString}; + + if cstr.to_bytes_with_nul().len() > max_with_nul { + let bytes = cstr.to_bytes()[..max_with_nul - 1].to_vec(); + // SAFETY: the non-nul bytes came straight from a CStr. + // (CString will add the terminating nul.) + Cow::Owned(unsafe { CString::from_vec_unchecked(bytes) }) + } else { + Cow::Borrowed(cstr) + } +} + pub fn available_parallelism() -> io::Result { cfg_if::cfg_if! { if #[cfg(any( @@ -423,7 +447,7 @@ mod cgroups { Some(b"") => Cgroup::V2, Some(controllers) if from_utf8(controllers) - .is_ok_and(|c| c.split(",").any(|c| c == "cpu")) => + .is_ok_and(|c| c.split(',').any(|c| c == "cpu")) => { Cgroup::V1 } diff --git a/library/std/src/sys/unix/thread_parker.rs b/library/std/src/sys/unix/thread_parker.rs deleted file mode 100644 index ca1a7138f..000000000 --- a/library/std/src/sys/unix/thread_parker.rs +++ /dev/null @@ -1,281 +0,0 @@ -//! Thread parking without `futex` using the `pthread` synchronization primitives. - -#![cfg(not(any( - target_os = "linux", - target_os = "android", - all(target_os = "emscripten", target_feature = "atomics"), - target_os = "freebsd", - target_os = "openbsd", - target_os = "dragonfly", - target_os = "fuchsia", -)))] - -use crate::cell::UnsafeCell; -use crate::marker::PhantomPinned; -use crate::pin::Pin; -use crate::ptr::addr_of_mut; -use crate::sync::atomic::AtomicUsize; -use crate::sync::atomic::Ordering::SeqCst; -use crate::time::Duration; - -const EMPTY: usize = 0; -const PARKED: usize = 1; -const NOTIFIED: usize = 2; - -unsafe fn lock(lock: *mut libc::pthread_mutex_t) { - let r = libc::pthread_mutex_lock(lock); - debug_assert_eq!(r, 0); -} - -unsafe fn unlock(lock: *mut libc::pthread_mutex_t) { - let r = libc::pthread_mutex_unlock(lock); - debug_assert_eq!(r, 0); -} - -unsafe fn notify_one(cond: *mut libc::pthread_cond_t) { - let r = libc::pthread_cond_signal(cond); - debug_assert_eq!(r, 0); -} - -unsafe fn wait(cond: *mut libc::pthread_cond_t, lock: *mut libc::pthread_mutex_t) { - let r = libc::pthread_cond_wait(cond, lock); - debug_assert_eq!(r, 0); -} - -const TIMESPEC_MAX: libc::timespec = - libc::timespec { tv_sec: ::MAX, tv_nsec: 1_000_000_000 - 1 }; - -unsafe fn wait_timeout( - cond: *mut libc::pthread_cond_t, - lock: *mut libc::pthread_mutex_t, - dur: Duration, -) { - // Use the system clock on systems that do not support pthread_condattr_setclock. - // This unfortunately results in problems when the system time changes. - #[cfg(any( - target_os = "macos", - target_os = "ios", - target_os = "watchos", - target_os = "espidf" - ))] - let (now, dur) = { - use super::time::SystemTime; - use crate::cmp::min; - - // OSX implementation of `pthread_cond_timedwait` is buggy - // with super long durations. When duration is greater than - // 0x100_0000_0000_0000 seconds, `pthread_cond_timedwait` - // in macOS Sierra return error 316. - // - // This program demonstrates the issue: - // https://gist.github.com/stepancheg/198db4623a20aad2ad7cddb8fda4a63c - // - // To work around this issue, and possible bugs of other OSes, timeout - // is clamped to 1000 years, which is allowable per the API of `park_timeout` - // because of spurious wakeups. - let dur = min(dur, Duration::from_secs(1000 * 365 * 86400)); - let now = SystemTime::now().t; - (now, dur) - }; - // Use the monotonic clock on other systems. - #[cfg(not(any( - target_os = "macos", - target_os = "ios", - target_os = "watchos", - target_os = "espidf" - )))] - let (now, dur) = { - use super::time::Timespec; - - (Timespec::now(libc::CLOCK_MONOTONIC), dur) - }; - - let timeout = - now.checked_add_duration(&dur).and_then(|t| t.to_timespec()).unwrap_or(TIMESPEC_MAX); - let r = libc::pthread_cond_timedwait(cond, lock, &timeout); - debug_assert!(r == libc::ETIMEDOUT || r == 0); -} - -pub struct Parker { - state: AtomicUsize, - lock: UnsafeCell, - cvar: UnsafeCell, - // The `pthread` primitives require a stable address, so make this struct `!Unpin`. - _pinned: PhantomPinned, -} - -impl Parker { - /// Construct the UNIX parker in-place. - /// - /// # Safety - /// The constructed parker must never be moved. - pub unsafe fn new(parker: *mut Parker) { - // Use the default mutex implementation to allow for simpler initialization. - // This could lead to undefined behaviour when deadlocking. This is avoided - // by not deadlocking. Note in particular the unlocking operation before any - // panic, as code after the panic could try to park again. - addr_of_mut!((*parker).state).write(AtomicUsize::new(EMPTY)); - addr_of_mut!((*parker).lock).write(UnsafeCell::new(libc::PTHREAD_MUTEX_INITIALIZER)); - - cfg_if::cfg_if! { - if #[cfg(any( - target_os = "macos", - target_os = "ios", - target_os = "watchos", - target_os = "l4re", - target_os = "android", - target_os = "redox" - ))] { - addr_of_mut!((*parker).cvar).write(UnsafeCell::new(libc::PTHREAD_COND_INITIALIZER)); - } else if #[cfg(any(target_os = "espidf", target_os = "horizon"))] { - let r = libc::pthread_cond_init(addr_of_mut!((*parker).cvar).cast(), crate::ptr::null()); - assert_eq!(r, 0); - } else { - use crate::mem::MaybeUninit; - let mut attr = MaybeUninit::::uninit(); - let r = libc::pthread_condattr_init(attr.as_mut_ptr()); - assert_eq!(r, 0); - let r = libc::pthread_condattr_setclock(attr.as_mut_ptr(), libc::CLOCK_MONOTONIC); - assert_eq!(r, 0); - let r = libc::pthread_cond_init(addr_of_mut!((*parker).cvar).cast(), attr.as_ptr()); - assert_eq!(r, 0); - let r = libc::pthread_condattr_destroy(attr.as_mut_ptr()); - assert_eq!(r, 0); - } - } - } - - // This implementation doesn't require `unsafe`, but other implementations - // may assume this is only called by the thread that owns the Parker. - pub unsafe fn park(self: Pin<&Self>) { - // If we were previously notified then we consume this notification and - // return quickly. - if self.state.compare_exchange(NOTIFIED, EMPTY, SeqCst, SeqCst).is_ok() { - return; - } - - // Otherwise we need to coordinate going to sleep - lock(self.lock.get()); - match self.state.compare_exchange(EMPTY, PARKED, SeqCst, SeqCst) { - Ok(_) => {} - Err(NOTIFIED) => { - // We must read here, even though we know it will be `NOTIFIED`. - // This is because `unpark` may have been called again since we read - // `NOTIFIED` in the `compare_exchange` above. We must perform an - // acquire operation that synchronizes with that `unpark` to observe - // any writes it made before the call to unpark. To do that we must - // read from the write it made to `state`. - let old = self.state.swap(EMPTY, SeqCst); - - unlock(self.lock.get()); - - assert_eq!(old, NOTIFIED, "park state changed unexpectedly"); - return; - } // should consume this notification, so prohibit spurious wakeups in next park. - Err(_) => { - unlock(self.lock.get()); - - panic!("inconsistent park state") - } - } - - loop { - wait(self.cvar.get(), self.lock.get()); - - match self.state.compare_exchange(NOTIFIED, EMPTY, SeqCst, SeqCst) { - Ok(_) => break, // got a notification - Err(_) => {} // spurious wakeup, go back to sleep - } - } - - unlock(self.lock.get()); - } - - // This implementation doesn't require `unsafe`, but other implementations - // may assume this is only called by the thread that owns the Parker. Use - // `Pin` to guarantee a stable address for the mutex and condition variable. - pub unsafe fn park_timeout(self: Pin<&Self>, dur: Duration) { - // Like `park` above we have a fast path for an already-notified thread, and - // afterwards we start coordinating for a sleep. - // return quickly. - if self.state.compare_exchange(NOTIFIED, EMPTY, SeqCst, SeqCst).is_ok() { - return; - } - - lock(self.lock.get()); - match self.state.compare_exchange(EMPTY, PARKED, SeqCst, SeqCst) { - Ok(_) => {} - Err(NOTIFIED) => { - // We must read again here, see `park`. - let old = self.state.swap(EMPTY, SeqCst); - unlock(self.lock.get()); - - assert_eq!(old, NOTIFIED, "park state changed unexpectedly"); - return; - } // should consume this notification, so prohibit spurious wakeups in next park. - Err(_) => { - unlock(self.lock.get()); - panic!("inconsistent park_timeout state") - } - } - - // Wait with a timeout, and if we spuriously wake up or otherwise wake up - // from a notification we just want to unconditionally set the state back to - // empty, either consuming a notification or un-flagging ourselves as - // parked. - wait_timeout(self.cvar.get(), self.lock.get(), dur); - - match self.state.swap(EMPTY, SeqCst) { - NOTIFIED => unlock(self.lock.get()), // got a notification, hurray! - PARKED => unlock(self.lock.get()), // no notification, alas - n => { - unlock(self.lock.get()); - panic!("inconsistent park_timeout state: {n}") - } - } - } - - pub fn unpark(self: Pin<&Self>) { - // To ensure the unparked thread will observe any writes we made - // before this call, we must perform a release operation that `park` - // can synchronize with. To do that we must write `NOTIFIED` even if - // `state` is already `NOTIFIED`. That is why this must be a swap - // rather than a compare-and-swap that returns if it reads `NOTIFIED` - // on failure. - match self.state.swap(NOTIFIED, SeqCst) { - EMPTY => return, // no one was waiting - NOTIFIED => return, // already unparked - PARKED => {} // gotta go wake someone up - _ => panic!("inconsistent state in unpark"), - } - - // There is a period between when the parked thread sets `state` to - // `PARKED` (or last checked `state` in the case of a spurious wake - // up) and when it actually waits on `cvar`. If we were to notify - // during this period it would be ignored and then when the parked - // thread went to sleep it would never wake up. Fortunately, it has - // `lock` locked at this stage so we can acquire `lock` to wait until - // it is ready to receive the notification. - // - // Releasing `lock` before the call to `notify_one` means that when the - // parked thread wakes it doesn't get woken only to have to wait for us - // to release `lock`. - unsafe { - lock(self.lock.get()); - unlock(self.lock.get()); - notify_one(self.cvar.get()); - } - } -} - -impl Drop for Parker { - fn drop(&mut self) { - unsafe { - libc::pthread_cond_destroy(self.cvar.get_mut()); - libc::pthread_mutex_destroy(self.lock.get_mut()); - } - } -} - -unsafe impl Sync for Parker {} -unsafe impl Send for Parker {} diff --git a/library/std/src/sys/unix/thread_parker/mod.rs b/library/std/src/sys/unix/thread_parker/mod.rs new file mode 100644 index 000000000..e2453580d --- /dev/null +++ b/library/std/src/sys/unix/thread_parker/mod.rs @@ -0,0 +1,21 @@ +//! Thread parking on systems without futex support. + +#![cfg(not(any( + target_os = "linux", + target_os = "android", + all(target_os = "emscripten", target_feature = "atomics"), + target_os = "freebsd", + target_os = "openbsd", + target_os = "dragonfly", + target_os = "fuchsia", +)))] + +cfg_if::cfg_if! { + if #[cfg(target_os = "netbsd")] { + mod netbsd; + pub use netbsd::Parker; + } else { + mod pthread; + pub use pthread::Parker; + } +} diff --git a/library/std/src/sys/unix/thread_parker/netbsd.rs b/library/std/src/sys/unix/thread_parker/netbsd.rs new file mode 100644 index 000000000..7657605b5 --- /dev/null +++ b/library/std/src/sys/unix/thread_parker/netbsd.rs @@ -0,0 +1,113 @@ +use crate::ffi::{c_int, c_void}; +use crate::pin::Pin; +use crate::ptr::{null, null_mut}; +use crate::sync::atomic::{ + AtomicU64, + Ordering::{Acquire, Relaxed, Release}, +}; +use crate::time::Duration; +use libc::{_lwp_self, clockid_t, lwpid_t, time_t, timespec, CLOCK_MONOTONIC}; + +extern "C" { + fn ___lwp_park60( + clock_id: clockid_t, + flags: c_int, + ts: *mut timespec, + unpark: lwpid_t, + hint: *const c_void, + unparkhint: *const c_void, + ) -> c_int; + fn _lwp_unpark(lwp: lwpid_t, hint: *const c_void) -> c_int; +} + +/// The thread is not parked and the token is not available. +/// +/// Zero cannot be a valid LWP id, since it is used as empty value for the unpark +/// argument in _lwp_park. +const EMPTY: u64 = 0; +/// The token is available. Do not park anymore. +const NOTIFIED: u64 = u64::MAX; + +pub struct Parker { + /// The parker state. Contains either one of the two state values above or the LWP + /// id of the parked thread. + state: AtomicU64, +} + +impl Parker { + pub unsafe fn new(parker: *mut Parker) { + parker.write(Parker { state: AtomicU64::new(EMPTY) }) + } + + // Does not actually need `unsafe` or `Pin`, but the pthread implementation does. + pub unsafe fn park(self: Pin<&Self>) { + // If the token has already been made available, we can skip + // a bit of work, so check for it here. + if self.state.load(Acquire) != NOTIFIED { + let parked = _lwp_self() as u64; + let hint = self.state.as_mut_ptr().cast(); + if self.state.compare_exchange(EMPTY, parked, Relaxed, Acquire).is_ok() { + // Loop to guard against spurious wakeups. + loop { + ___lwp_park60(0, 0, null_mut(), 0, hint, null()); + if self.state.load(Acquire) == NOTIFIED { + break; + } + } + } + } + + // At this point, the change to NOTIFIED has always been observed with acquire + // ordering, so we can just use a relaxed store here (instead of a swap). + self.state.store(EMPTY, Relaxed); + } + + // Does not actually need `unsafe` or `Pin`, but the pthread implementation does. + pub unsafe fn park_timeout(self: Pin<&Self>, dur: Duration) { + if self.state.load(Acquire) != NOTIFIED { + let parked = _lwp_self() as u64; + let hint = self.state.as_mut_ptr().cast(); + let mut timeout = timespec { + // Saturate so that the operation will definitely time out + // (even if it is after the heat death of the universe). + tv_sec: dur.as_secs().try_into().ok().unwrap_or(time_t::MAX), + tv_nsec: dur.subsec_nanos().into(), + }; + + if self.state.compare_exchange(EMPTY, parked, Relaxed, Acquire).is_ok() { + // Timeout needs to be mutable since it is modified on NetBSD 9.0 and + // above. + ___lwp_park60(CLOCK_MONOTONIC, 0, &mut timeout, 0, hint, null()); + // Use a swap to get acquire ordering even if the token was set after + // the timeout occurred. + self.state.swap(EMPTY, Acquire); + return; + } + } + + self.state.store(EMPTY, Relaxed); + } + + // Does not actually need `Pin`, but the pthread implementation does. + pub fn unpark(self: Pin<&Self>) { + let state = self.state.swap(NOTIFIED, Release); + if !matches!(state, EMPTY | NOTIFIED) { + let lwp = state as lwpid_t; + let hint = self.state.as_mut_ptr().cast(); + + // If the parking thread terminated and did not actually park, this will + // probably return an error, which is OK. In the worst case, another + // thread has received the same LWP id. It will then receive a spurious + // wakeup, but those are allowable per the API contract. The same reasoning + // applies if a timeout occurred before this call, but the state was not + // yet reset. + + // SAFETY: + // The syscall has no invariants to hold. Only unsafe because it is an + // extern function. + unsafe { + _lwp_unpark(lwp, hint); + } + } + } +} diff --git a/library/std/src/sys/unix/thread_parker/pthread.rs b/library/std/src/sys/unix/thread_parker/pthread.rs new file mode 100644 index 000000000..3dfc0026e --- /dev/null +++ b/library/std/src/sys/unix/thread_parker/pthread.rs @@ -0,0 +1,271 @@ +//! Thread parking without `futex` using the `pthread` synchronization primitives. + +use crate::cell::UnsafeCell; +use crate::marker::PhantomPinned; +use crate::pin::Pin; +use crate::ptr::addr_of_mut; +use crate::sync::atomic::AtomicUsize; +use crate::sync::atomic::Ordering::SeqCst; +use crate::time::Duration; + +const EMPTY: usize = 0; +const PARKED: usize = 1; +const NOTIFIED: usize = 2; + +unsafe fn lock(lock: *mut libc::pthread_mutex_t) { + let r = libc::pthread_mutex_lock(lock); + debug_assert_eq!(r, 0); +} + +unsafe fn unlock(lock: *mut libc::pthread_mutex_t) { + let r = libc::pthread_mutex_unlock(lock); + debug_assert_eq!(r, 0); +} + +unsafe fn notify_one(cond: *mut libc::pthread_cond_t) { + let r = libc::pthread_cond_signal(cond); + debug_assert_eq!(r, 0); +} + +unsafe fn wait(cond: *mut libc::pthread_cond_t, lock: *mut libc::pthread_mutex_t) { + let r = libc::pthread_cond_wait(cond, lock); + debug_assert_eq!(r, 0); +} + +const TIMESPEC_MAX: libc::timespec = + libc::timespec { tv_sec: ::MAX, tv_nsec: 1_000_000_000 - 1 }; + +unsafe fn wait_timeout( + cond: *mut libc::pthread_cond_t, + lock: *mut libc::pthread_mutex_t, + dur: Duration, +) { + // Use the system clock on systems that do not support pthread_condattr_setclock. + // This unfortunately results in problems when the system time changes. + #[cfg(any( + target_os = "macos", + target_os = "ios", + target_os = "watchos", + target_os = "espidf" + ))] + let (now, dur) = { + use crate::cmp::min; + use crate::sys::time::SystemTime; + + // OSX implementation of `pthread_cond_timedwait` is buggy + // with super long durations. When duration is greater than + // 0x100_0000_0000_0000 seconds, `pthread_cond_timedwait` + // in macOS Sierra return error 316. + // + // This program demonstrates the issue: + // https://gist.github.com/stepancheg/198db4623a20aad2ad7cddb8fda4a63c + // + // To work around this issue, and possible bugs of other OSes, timeout + // is clamped to 1000 years, which is allowable per the API of `park_timeout` + // because of spurious wakeups. + let dur = min(dur, Duration::from_secs(1000 * 365 * 86400)); + let now = SystemTime::now().t; + (now, dur) + }; + // Use the monotonic clock on other systems. + #[cfg(not(any( + target_os = "macos", + target_os = "ios", + target_os = "watchos", + target_os = "espidf" + )))] + let (now, dur) = { + use crate::sys::time::Timespec; + + (Timespec::now(libc::CLOCK_MONOTONIC), dur) + }; + + let timeout = + now.checked_add_duration(&dur).and_then(|t| t.to_timespec()).unwrap_or(TIMESPEC_MAX); + let r = libc::pthread_cond_timedwait(cond, lock, &timeout); + debug_assert!(r == libc::ETIMEDOUT || r == 0); +} + +pub struct Parker { + state: AtomicUsize, + lock: UnsafeCell, + cvar: UnsafeCell, + // The `pthread` primitives require a stable address, so make this struct `!Unpin`. + _pinned: PhantomPinned, +} + +impl Parker { + /// Construct the UNIX parker in-place. + /// + /// # Safety + /// The constructed parker must never be moved. + pub unsafe fn new(parker: *mut Parker) { + // Use the default mutex implementation to allow for simpler initialization. + // This could lead to undefined behaviour when deadlocking. This is avoided + // by not deadlocking. Note in particular the unlocking operation before any + // panic, as code after the panic could try to park again. + addr_of_mut!((*parker).state).write(AtomicUsize::new(EMPTY)); + addr_of_mut!((*parker).lock).write(UnsafeCell::new(libc::PTHREAD_MUTEX_INITIALIZER)); + + cfg_if::cfg_if! { + if #[cfg(any( + target_os = "macos", + target_os = "ios", + target_os = "watchos", + target_os = "l4re", + target_os = "android", + target_os = "redox" + ))] { + addr_of_mut!((*parker).cvar).write(UnsafeCell::new(libc::PTHREAD_COND_INITIALIZER)); + } else if #[cfg(any(target_os = "espidf", target_os = "horizon"))] { + let r = libc::pthread_cond_init(addr_of_mut!((*parker).cvar).cast(), crate::ptr::null()); + assert_eq!(r, 0); + } else { + use crate::mem::MaybeUninit; + let mut attr = MaybeUninit::::uninit(); + let r = libc::pthread_condattr_init(attr.as_mut_ptr()); + assert_eq!(r, 0); + let r = libc::pthread_condattr_setclock(attr.as_mut_ptr(), libc::CLOCK_MONOTONIC); + assert_eq!(r, 0); + let r = libc::pthread_cond_init(addr_of_mut!((*parker).cvar).cast(), attr.as_ptr()); + assert_eq!(r, 0); + let r = libc::pthread_condattr_destroy(attr.as_mut_ptr()); + assert_eq!(r, 0); + } + } + } + + // This implementation doesn't require `unsafe`, but other implementations + // may assume this is only called by the thread that owns the Parker. + pub unsafe fn park(self: Pin<&Self>) { + // If we were previously notified then we consume this notification and + // return quickly. + if self.state.compare_exchange(NOTIFIED, EMPTY, SeqCst, SeqCst).is_ok() { + return; + } + + // Otherwise we need to coordinate going to sleep + lock(self.lock.get()); + match self.state.compare_exchange(EMPTY, PARKED, SeqCst, SeqCst) { + Ok(_) => {} + Err(NOTIFIED) => { + // We must read here, even though we know it will be `NOTIFIED`. + // This is because `unpark` may have been called again since we read + // `NOTIFIED` in the `compare_exchange` above. We must perform an + // acquire operation that synchronizes with that `unpark` to observe + // any writes it made before the call to unpark. To do that we must + // read from the write it made to `state`. + let old = self.state.swap(EMPTY, SeqCst); + + unlock(self.lock.get()); + + assert_eq!(old, NOTIFIED, "park state changed unexpectedly"); + return; + } // should consume this notification, so prohibit spurious wakeups in next park. + Err(_) => { + unlock(self.lock.get()); + + panic!("inconsistent park state") + } + } + + loop { + wait(self.cvar.get(), self.lock.get()); + + match self.state.compare_exchange(NOTIFIED, EMPTY, SeqCst, SeqCst) { + Ok(_) => break, // got a notification + Err(_) => {} // spurious wakeup, go back to sleep + } + } + + unlock(self.lock.get()); + } + + // This implementation doesn't require `unsafe`, but other implementations + // may assume this is only called by the thread that owns the Parker. Use + // `Pin` to guarantee a stable address for the mutex and condition variable. + pub unsafe fn park_timeout(self: Pin<&Self>, dur: Duration) { + // Like `park` above we have a fast path for an already-notified thread, and + // afterwards we start coordinating for a sleep. + // return quickly. + if self.state.compare_exchange(NOTIFIED, EMPTY, SeqCst, SeqCst).is_ok() { + return; + } + + lock(self.lock.get()); + match self.state.compare_exchange(EMPTY, PARKED, SeqCst, SeqCst) { + Ok(_) => {} + Err(NOTIFIED) => { + // We must read again here, see `park`. + let old = self.state.swap(EMPTY, SeqCst); + unlock(self.lock.get()); + + assert_eq!(old, NOTIFIED, "park state changed unexpectedly"); + return; + } // should consume this notification, so prohibit spurious wakeups in next park. + Err(_) => { + unlock(self.lock.get()); + panic!("inconsistent park_timeout state") + } + } + + // Wait with a timeout, and if we spuriously wake up or otherwise wake up + // from a notification we just want to unconditionally set the state back to + // empty, either consuming a notification or un-flagging ourselves as + // parked. + wait_timeout(self.cvar.get(), self.lock.get(), dur); + + match self.state.swap(EMPTY, SeqCst) { + NOTIFIED => unlock(self.lock.get()), // got a notification, hurray! + PARKED => unlock(self.lock.get()), // no notification, alas + n => { + unlock(self.lock.get()); + panic!("inconsistent park_timeout state: {n}") + } + } + } + + pub fn unpark(self: Pin<&Self>) { + // To ensure the unparked thread will observe any writes we made + // before this call, we must perform a release operation that `park` + // can synchronize with. To do that we must write `NOTIFIED` even if + // `state` is already `NOTIFIED`. That is why this must be a swap + // rather than a compare-and-swap that returns if it reads `NOTIFIED` + // on failure. + match self.state.swap(NOTIFIED, SeqCst) { + EMPTY => return, // no one was waiting + NOTIFIED => return, // already unparked + PARKED => {} // gotta go wake someone up + _ => panic!("inconsistent state in unpark"), + } + + // There is a period between when the parked thread sets `state` to + // `PARKED` (or last checked `state` in the case of a spurious wake + // up) and when it actually waits on `cvar`. If we were to notify + // during this period it would be ignored and then when the parked + // thread went to sleep it would never wake up. Fortunately, it has + // `lock` locked at this stage so we can acquire `lock` to wait until + // it is ready to receive the notification. + // + // Releasing `lock` before the call to `notify_one` means that when the + // parked thread wakes it doesn't get woken only to have to wait for us + // to release `lock`. + unsafe { + lock(self.lock.get()); + unlock(self.lock.get()); + notify_one(self.cvar.get()); + } + } +} + +impl Drop for Parker { + fn drop(&mut self) { + unsafe { + libc::pthread_cond_destroy(self.cvar.get_mut()); + libc::pthread_mutex_destroy(self.lock.get_mut()); + } + } +} + +unsafe impl Sync for Parker {} +unsafe impl Send for Parker {} diff --git a/library/std/src/sys/unsupported/alloc.rs b/library/std/src/sys/unsupported/alloc.rs index 8d5d0a2f5..d715ae454 100644 --- a/library/std/src/sys/unsupported/alloc.rs +++ b/library/std/src/sys/unsupported/alloc.rs @@ -1,15 +1,16 @@ use crate::alloc::{GlobalAlloc, Layout, System}; +use crate::ptr::null_mut; #[stable(feature = "alloc_system_type", since = "1.28.0")] unsafe impl GlobalAlloc for System { #[inline] unsafe fn alloc(&self, _layout: Layout) -> *mut u8 { - 0 as *mut u8 + null_mut() } #[inline] unsafe fn alloc_zeroed(&self, _layout: Layout) -> *mut u8 { - 0 as *mut u8 + null_mut() } #[inline] @@ -17,6 +18,6 @@ unsafe impl GlobalAlloc for System { #[inline] unsafe fn realloc(&self, _ptr: *mut u8, _layout: Layout, _new_size: usize) -> *mut u8 { - 0 as *mut u8 + null_mut() } } diff --git a/library/std/src/sys/unsupported/common.rs b/library/std/src/sys/unsupported/common.rs index 4c9ade4a8..5cd9e57de 100644 --- a/library/std/src/sys/unsupported/common.rs +++ b/library/std/src/sys/unsupported/common.rs @@ -6,7 +6,7 @@ pub mod memchr { // SAFETY: must be called only once during runtime initialization. // NOTE: this is not guaranteed to run, for example when Rust code is called externally. -pub unsafe fn init(_argc: isize, _argv: *const *const u8) {} +pub unsafe fn init(_argc: isize, _argv: *const *const u8, _sigpipe: u8) {} // SAFETY: must be called only once during runtime cleanup. // NOTE: this is not guaranteed to run, for example when the program aborts. diff --git a/library/std/src/sys/unsupported/fs.rs b/library/std/src/sys/unsupported/fs.rs index 0e1a6257e..6ac1b5d2b 100644 --- a/library/std/src/sys/unsupported/fs.rs +++ b/library/std/src/sys/unsupported/fs.rs @@ -1,7 +1,7 @@ use crate::ffi::OsString; use crate::fmt; use crate::hash::{Hash, Hasher}; -use crate::io::{self, IoSlice, IoSliceMut, ReadBuf, SeekFrom}; +use crate::io::{self, BorrowedCursor, IoSlice, IoSliceMut, SeekFrom}; use crate::path::{Path, PathBuf}; use crate::sys::time::SystemTime; use crate::sys::unsupported; @@ -214,7 +214,7 @@ impl File { self.0 } - pub fn read_buf(&self, _buf: &mut ReadBuf<'_>) -> io::Result<()> { + pub fn read_buf(&self, _cursor: BorrowedCursor<'_>) -> io::Result<()> { self.0 } diff --git a/library/std/src/sys/unsupported/locks/mutex.rs b/library/std/src/sys/unsupported/locks/mutex.rs index d7cb12e0c..2be0b34b9 100644 --- a/library/std/src/sys/unsupported/locks/mutex.rs +++ b/library/std/src/sys/unsupported/locks/mutex.rs @@ -16,9 +16,6 @@ impl Mutex { Mutex { locked: Cell::new(false) } } - #[inline] - pub unsafe fn init(&mut self) {} - #[inline] pub unsafe fn lock(&self) { assert_eq!(self.locked.replace(true), false, "cannot recursively acquire mutex"); diff --git a/library/std/src/sys/unsupported/process.rs b/library/std/src/sys/unsupported/process.rs index 42a1ff730..633f17c05 100644 --- a/library/std/src/sys/unsupported/process.rs +++ b/library/std/src/sys/unsupported/process.rs @@ -200,6 +200,9 @@ impl<'a> Iterator for CommandArgs<'a> { fn next(&mut self) -> Option<&'a OsStr> { None } + fn size_hint(&self) -> (usize, Option) { + (0, Some(0)) + } } impl<'a> ExactSizeIterator for CommandArgs<'a> {} diff --git a/library/std/src/sys/wasi/fs.rs b/library/std/src/sys/wasi/fs.rs index 6614ae397..510cf36b1 100644 --- a/library/std/src/sys/wasi/fs.rs +++ b/library/std/src/sys/wasi/fs.rs @@ -3,7 +3,7 @@ use super::fd::WasiFd; use crate::ffi::{CStr, CString, OsStr, OsString}; use crate::fmt; -use crate::io::{self, IoSlice, IoSliceMut, ReadBuf, SeekFrom}; +use crate::io::{self, BorrowedCursor, IoSlice, IoSliceMut, SeekFrom}; use crate::iter; use crate::mem::{self, ManuallyDrop}; use crate::os::raw::c_int; @@ -439,8 +439,8 @@ impl File { true } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { - crate::io::default_read_buf(|buf| self.read(buf), buf) + pub fn read_buf(&self, cursor: BorrowedCursor<'_>) -> io::Result<()> { + crate::io::default_read_buf(|buf| self.read(buf), cursor) } pub fn write(&self, buf: &[u8]) -> io::Result { diff --git a/library/std/src/sys/wasi/stdio.rs b/library/std/src/sys/wasi/stdio.rs index 4cc0e4ed5..d2081771b 100644 --- a/library/std/src/sys/wasi/stdio.rs +++ b/library/std/src/sys/wasi/stdio.rs @@ -4,7 +4,7 @@ use super::fd::WasiFd; use crate::io::{self, IoSlice, IoSliceMut}; use crate::mem::ManuallyDrop; use crate::os::raw; -use crate::os::wasi::io::{AsRawFd, FromRawFd}; +use crate::os::wasi::io::{AsFd, AsRawFd, BorrowedFd, FromRawFd}; pub struct Stdin; pub struct Stdout; @@ -23,6 +23,13 @@ impl AsRawFd for Stdin { } } +impl AsFd for Stdin { + #[inline] + fn as_fd(&self) -> BorrowedFd<'_> { + unsafe { BorrowedFd::borrow_raw(0) } + } +} + impl io::Read for Stdin { fn read(&mut self, data: &mut [u8]) -> io::Result { self.read_vectored(&mut [IoSliceMut::new(data)]) @@ -51,6 +58,13 @@ impl AsRawFd for Stdout { } } +impl AsFd for Stdout { + #[inline] + fn as_fd(&self) -> BorrowedFd<'_> { + unsafe { BorrowedFd::borrow_raw(1) } + } +} + impl io::Write for Stdout { fn write(&mut self, data: &[u8]) -> io::Result { self.write_vectored(&[IoSlice::new(data)]) @@ -82,6 +96,13 @@ impl AsRawFd for Stderr { } } +impl AsFd for Stderr { + #[inline] + fn as_fd(&self) -> BorrowedFd<'_> { + unsafe { BorrowedFd::borrow_raw(2) } + } +} + impl io::Write for Stderr { fn write(&mut self, data: &[u8]) -> io::Result { self.write_vectored(&[IoSlice::new(data)]) diff --git a/library/std/src/sys/windows/alloc.rs b/library/std/src/sys/windows/alloc.rs index fdc81cdea..d53ea1600 100644 --- a/library/std/src/sys/windows/alloc.rs +++ b/library/std/src/sys/windows/alloc.rs @@ -16,6 +16,7 @@ mod tests; // Flag to indicate that the memory returned by `HeapAlloc` should be zeroed. const HEAP_ZERO_MEMORY: c::DWORD = 0x00000008; +#[link(name = "kernel32")] extern "system" { // Get a handle to the default heap of the current process, or null if the operation fails. // @@ -168,7 +169,7 @@ unsafe fn allocate(layout: Layout, zeroed: bool) -> *mut u8 { // SAFETY: Because the size and alignment of a header is <= `MIN_ALIGN` and `aligned` // is aligned to at least `MIN_ALIGN` and has at least `MIN_ALIGN` bytes of padding before // it, it is safe to write a header directly before it. - unsafe { ptr::write((aligned as *mut Header).offset(-1), Header(ptr)) }; + unsafe { ptr::write((aligned as *mut Header).sub(1), Header(ptr)) }; // SAFETY: The returned pointer does not point to the to the start of an allocated block, // but there is a header readable directly before it containing the location of the start @@ -213,7 +214,7 @@ unsafe impl GlobalAlloc for System { // SAFETY: Because of the contract of `System`, `ptr` is guaranteed to be non-null // and have a header readable directly before it. - unsafe { ptr::read((ptr as *mut Header).offset(-1)).0 } + unsafe { ptr::read((ptr as *mut Header).sub(1)).0 } } }; diff --git a/library/std/src/sys/windows/c.rs b/library/std/src/sys/windows/c.rs index 478068c73..89d0ab59b 100644 --- a/library/std/src/sys/windows/c.rs +++ b/library/std/src/sys/windows/c.rs @@ -66,6 +66,7 @@ pub type LPSYSTEM_INFO = *mut SYSTEM_INFO; pub type LPWSABUF = *mut WSABUF; pub type LPWSAOVERLAPPED = *mut c_void; pub type LPWSAOVERLAPPED_COMPLETION_ROUTINE = *mut c_void; +pub type BCRYPT_ALG_HANDLE = LPVOID; pub type PCONDITION_VARIABLE = *mut CONDITION_VARIABLE; pub type PLARGE_INTEGER = *mut c_longlong; @@ -278,6 +279,7 @@ pub const STATUS_INVALID_PARAMETER: NTSTATUS = 0xc000000d_u32 as _; pub const STATUS_PENDING: NTSTATUS = 0x103 as _; pub const STATUS_END_OF_FILE: NTSTATUS = 0xC0000011_u32 as _; pub const STATUS_NOT_IMPLEMENTED: NTSTATUS = 0xC0000002_u32 as _; +pub const STATUS_NOT_SUPPORTED: NTSTATUS = 0xC00000BB_u32 as _; // Equivalent to the `NT_SUCCESS` C preprocessor macro. // See: https://docs.microsoft.com/en-us/windows-hardware/drivers/kernel/using-ntstatus-values @@ -285,7 +287,8 @@ pub fn nt_success(status: NTSTATUS) -> bool { status >= 0 } -pub const BCRYPT_USE_SYSTEM_PREFERRED_RNG: DWORD = 0x00000002; +// "RNG\0" +pub const BCRYPT_RNG_ALGORITHM: &[u16] = &[b'R' as u16, b'N' as u16, b'G' as u16, 0]; #[repr(C)] pub struct UNICODE_STRING { @@ -454,6 +457,12 @@ pub enum FILE_INFO_BY_HANDLE_CLASS { MaximumFileInfoByHandlesClass, } +#[repr(C)] +pub struct FILE_ATTRIBUTE_TAG_INFO { + pub FileAttributes: DWORD, + pub ReparseTag: DWORD, +} + #[repr(C)] pub struct FILE_DISPOSITION_INFO { pub DeleteFile: BOOLEAN, @@ -501,6 +510,8 @@ pub struct FILE_END_OF_FILE_INFO { pub EndOfFile: LARGE_INTEGER, } +/// NB: Use carefully! In general using this as a reference is likely to get the +/// provenance wrong for the `rest` field! #[repr(C)] pub struct REPARSE_DATA_BUFFER { pub ReparseTag: c_uint, @@ -509,6 +520,8 @@ pub struct REPARSE_DATA_BUFFER { pub rest: (), } +/// NB: Use carefully! In general using this as a reference is likely to get the +/// provenance wrong for the `PathBuffer` field! #[repr(C)] pub struct SYMBOLIC_LINK_REPARSE_BUFFER { pub SubstituteNameOffset: c_ushort, @@ -519,6 +532,8 @@ pub struct SYMBOLIC_LINK_REPARSE_BUFFER { pub PathBuffer: WCHAR, } +/// NB: Use carefully! In general using this as a reference is likely to get the +/// provenance wrong for the `PathBuffer` field! #[repr(C)] pub struct MOUNT_POINT_REPARSE_BUFFER { pub SubstituteNameOffset: c_ushort, @@ -1217,11 +1232,18 @@ extern "system" { // >= Vista / Server 2008 // https://docs.microsoft.com/en-us/windows/win32/api/bcrypt/nf-bcrypt-bcryptgenrandom pub fn BCryptGenRandom( - hAlgorithm: LPVOID, + hAlgorithm: BCRYPT_ALG_HANDLE, pBuffer: *mut u8, cbBuffer: ULONG, dwFlags: ULONG, ) -> NTSTATUS; + pub fn BCryptOpenAlgorithmProvider( + phalgorithm: *mut BCRYPT_ALG_HANDLE, + pszAlgId: LPCWSTR, + pszimplementation: LPCWSTR, + dwflags: ULONG, + ) -> NTSTATUS; + pub fn BCryptCloseAlgorithmProvider(hAlgorithm: BCRYPT_ALG_HANDLE, dwFlags: ULONG) -> NTSTATUS; } // Functions that aren't available on every version of Windows that we support, @@ -1251,17 +1273,14 @@ compat_fn_with_fallback! { } compat_fn_optional! { - pub static SYNCH_API: &CStr = ansi_str!("api-ms-win-core-synch-l1-2-0"); - - // >= Windows 8 / Server 2012 - // https://docs.microsoft.com/en-us/windows/win32/api/synchapi/nf-synchapi-waitonaddress + crate::sys::compat::load_synch_functions(); pub fn WaitOnAddress( Address: LPVOID, CompareAddress: LPVOID, AddressSize: SIZE_T, dwMilliseconds: DWORD - ) -> BOOL; - pub fn WakeByAddressSingle(Address: LPVOID) -> (); + ); + pub fn WakeByAddressSingle(Address: LPVOID); } compat_fn_with_fallback! { diff --git a/library/std/src/sys/windows/cmath.rs b/library/std/src/sys/windows/cmath.rs index 1a5421fac..43ab8c7ee 100644 --- a/library/std/src/sys/windows/cmath.rs +++ b/library/std/src/sys/windows/cmath.rs @@ -44,7 +44,7 @@ mod shims { } // On 32-bit x86 MSVC these functions aren't defined, so we just define shims -// which promote everything fo f64, perform the calculation, and then demote +// which promote everything to f64, perform the calculation, and then demote // back to f32. While not precisely correct should be "correct enough" for now. #[cfg(all(target_env = "msvc", target_arch = "x86"))] mod shims { diff --git a/library/std/src/sys/windows/compat.rs b/library/std/src/sys/windows/compat.rs index ccc90177a..7dff81ecb 100644 --- a/library/std/src/sys/windows/compat.rs +++ b/library/std/src/sys/windows/compat.rs @@ -7,52 +7,66 @@ //! `GetModuleHandle` and `GetProcAddress` to look up DLL entry points at //! runtime. //! -//! This implementation uses a static initializer to look up the DLL entry -//! points. The CRT (C runtime) executes static initializers before `main` -//! is called (for binaries) and before `DllMain` is called (for DLLs). -//! This is the ideal time to look up DLL imports, because we are guaranteed -//! that no other threads will attempt to call these entry points. Thus, -//! we can look up the imports and store them in `static mut` fields -//! without any synchronization. +//! This is implemented simply by storing a function pointer in an atomic. +//! Loading and calling this function will have little or no overhead +//! compared with calling any other dynamically imported function. //! -//! This has an additional advantage: Because the DLL import lookup happens -//! at module initialization, the cost of these lookups is deterministic, -//! and is removed from the code paths that actually call the DLL imports. -//! That is, there is no unpredictable "cache miss" that occurs when calling -//! a DLL import. For applications that benefit from predictable delays, -//! this is a benefit. This also eliminates the comparison-and-branch -//! from the hot path. -//! -//! Currently, the standard library uses only a small number of dynamic -//! DLL imports. If this number grows substantially, then the cost of -//! performing all of the lookups at initialization time might become -//! substantial. -//! -//! The mechanism of registering a static initializer with the CRT is -//! documented in -//! [CRT Initialization](https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-initialization?view=msvc-160). -//! It works by contributing a global symbol to the `.CRT$XCU` section. -//! The linker builds a table of all static initializer functions. -//! The CRT startup code then iterates that table, calling each -//! initializer function. -//! -//! # **WARNING!!* -//! The environment that a static initializer function runs in is highly -//! constrained. There are **many** restrictions on what static initializers -//! can safely do. Static initializer functions **MUST NOT** do any of the -//! following (this list is not comprehensive): -//! * touch any other static field that is used by a different static -//! initializer, because the order that static initializers run in -//! is not defined. -//! * call `LoadLibrary` or any other function that acquires the DLL -//! loader lock. -//! * call any Rust function or CRT function that touches any static -//! (global) state. +//! The stored function pointer starts out as an importer function which will +//! swap itself with the real function when it's called for the first time. If +//! the real function can't be imported then a fallback function is used in its +//! place. While this is low cost for the happy path (where the function is +//! already loaded) it does mean there's some overhead the first time the +//! function is called. In the worst case, multiple threads may all end up +//! importing the same function unnecessarily. use crate::ffi::{c_void, CStr}; use crate::ptr::NonNull; +use crate::sync::atomic::Ordering; use crate::sys::c; +// This uses a static initializer to preload some imported functions. +// The CRT (C runtime) executes static initializers before `main` +// is called (for binaries) and before `DllMain` is called (for DLLs). +// +// It works by contributing a global symbol to the `.CRT$XCT` section. +// The linker builds a table of all static initializer functions. +// The CRT startup code then iterates that table, calling each +// initializer function. +// +// NOTE: User code should instead use .CRT$XCU to reliably run after std's initializer. +// If you're reading this and would like a guarantee here, please +// file an issue for discussion; currently we don't guarantee any functionality +// before main. +// See https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-initialization?view=msvc-170 +#[used] +#[link_section = ".CRT$XCT"] +static INIT_TABLE_ENTRY: unsafe extern "C" fn() = init; + +/// Preload some imported functions. +/// +/// Note that any functions included here will be unconditionally loaded in +/// the final binary, regardless of whether or not they're actually used. +/// +/// Therefore, this should be limited to `compat_fn_optional` functions which +/// must be preloaded or any functions where lazier loading demonstrates a +/// negative performance impact in practical situations. +/// +/// Currently we only preload `WaitOnAddress` and `WakeByAddressSingle`. +unsafe extern "C" fn init() { + // In an exe this code is executed before main() so is single threaded. + // In a DLL the system's loader lock will be held thereby synchronizing + // access. So the same best practices apply here as they do to running in DllMain: + // https://docs.microsoft.com/en-us/windows/win32/dlls/dynamic-link-library-best-practices + // + // DO NOT do anything interesting or complicated in this function! DO NOT call + // any Rust functions or CRT functions if those functions touch any global state, + // because this function runs during global initialization. For example, DO NOT + // do any dynamic allocation, don't call LoadLibrary, etc. + + // Attempt to preload the synch functions. + load_synch_functions(); +} + /// Helper macro for creating CStrs from literals and symbol names. macro_rules! ansi_str { (sym $ident:ident) => {{ @@ -85,39 +99,6 @@ pub(crate) const fn const_cstr_from_bytes(bytes: &'static [u8]) -> &'static CStr unsafe { crate::ffi::CStr::from_bytes_with_nul_unchecked(bytes) } } -#[used] -#[link_section = ".CRT$XCU"] -static INIT_TABLE_ENTRY: unsafe extern "C" fn() = init; - -/// This is where the magic preloading of symbols happens. -/// -/// Note that any functions included here will be unconditionally included in -/// the final binary, regardless of whether or not they're actually used. -/// -/// Therefore, this is limited to `compat_fn_optional` functions which must be -/// preloaded and any functions which may be more time sensitive, even for the first call. -unsafe extern "C" fn init() { - // There is no locking here. This code is executed before main() is entered, and - // is guaranteed to be single-threaded. - // - // DO NOT do anything interesting or complicated in this function! DO NOT call - // any Rust functions or CRT functions if those functions touch any global state, - // because this function runs during global initialization. For example, DO NOT - // do any dynamic allocation, don't call LoadLibrary, etc. - - if let Some(synch) = Module::new(c::SYNCH_API) { - // These are optional and so we must manually attempt to load them - // before they can be used. - c::WaitOnAddress::preload(synch); - c::WakeByAddressSingle::preload(synch); - } - - if let Some(kernel32) = Module::new(c::KERNEL32) { - // Preloading this means getting a precise time will be as fast as possible. - c::GetSystemTimePreciseAsFileTime::preload(kernel32); - } -} - /// Represents a loaded module. /// /// Note that the modules std depends on must not be unloaded. @@ -151,7 +132,7 @@ impl Module { macro_rules! compat_fn_with_fallback { (pub static $module:ident: &CStr = $name:expr; $( $(#[$meta:meta])* - pub fn $symbol:ident($($argname:ident: $argtype:ty),*) -> $rettype:ty $fallback_body:block + $vis:vis fn $symbol:ident($($argname:ident: $argtype:ty),*) -> $rettype:ty $fallback_body:block )*) => ( pub static $module: &CStr = $name; $( @@ -196,11 +177,6 @@ macro_rules! compat_fn_with_fallback { $fallback_body } - #[allow(unused)] - pub(in crate::sys) fn preload(module: Module) { - load_from_module(Some(module)); - } - #[inline(always)] pub unsafe fn call($($argname: $argtype),*) -> $rettype { let func: F = mem::transmute(PTR.load(Ordering::Relaxed)); @@ -208,66 +184,60 @@ macro_rules! compat_fn_with_fallback { } } $(#[$meta])* - pub use $symbol::call as $symbol; + $vis use $symbol::call as $symbol; )*) } -/// A function that either exists or doesn't. +/// Optionally loaded functions. /// -/// NOTE: Optional functions must be preloaded in the `init` function above, or they will always be None. +/// Actual loading of the function defers to $load_functions. macro_rules! compat_fn_optional { - (pub static $module:ident: &CStr = $name:expr; $( - $(#[$meta:meta])* - pub fn $symbol:ident($($argname:ident: $argtype:ty),*) -> $rettype:ty; - )*) => ( - pub static $module: &CStr = $name; + ($load_functions:expr; $( - $(#[$meta])* - pub mod $symbol { - #[allow(unused_imports)] - use super::*; - use crate::mem; - use crate::sync::atomic::{AtomicPtr, Ordering}; - use crate::sys::compat::Module; - use crate::ptr::{self, NonNull}; - - type F = unsafe extern "system" fn($($argtype),*) -> $rettype; - - /// `PTR` will either be `null()` or set to the loaded function. - static PTR: AtomicPtr = AtomicPtr::new(ptr::null_mut()); - - /// Only allow access to the function if it has loaded successfully. - #[inline(always)] - #[cfg(not(miri))] - pub fn option() -> Option { - unsafe { - NonNull::new(PTR.load(Ordering::Relaxed)).map(|f| mem::transmute(f)) + $(#[$meta:meta])* + $vis:vis fn $symbol:ident($($argname:ident: $argtype:ty),*) $(-> $rettype:ty)?; + )+) => ( + $( + pub mod $symbol { + use super::*; + use crate::ffi::c_void; + use crate::mem; + use crate::ptr::{self, NonNull}; + use crate::sync::atomic::{AtomicPtr, Ordering}; + + pub(in crate::sys) static PTR: AtomicPtr = AtomicPtr::new(ptr::null_mut()); + + type F = unsafe extern "system" fn($($argtype),*) $(-> $rettype)?; + + #[inline(always)] + pub fn option() -> Option { + // Miri does not understand the way we do preloading + // therefore load the function here instead. + #[cfg(miri)] $load_functions; + NonNull::new(PTR.load(Ordering::Relaxed)).map(|f| unsafe { mem::transmute(f) }) } } + )+ + ) +} - // Miri does not understand the way we do preloading - // therefore load the function here instead. - #[cfg(miri)] - pub fn option() -> Option { - let mut func = NonNull::new(PTR.load(Ordering::Relaxed)); - if func.is_none() { - unsafe { Module::new($module).map(preload) }; - func = NonNull::new(PTR.load(Ordering::Relaxed)); - } - unsafe { - func.map(|f| mem::transmute(f)) - } - } +/// Load all needed functions from "api-ms-win-core-synch-l1-2-0". +pub(super) fn load_synch_functions() { + fn try_load() -> Option<()> { + const MODULE_NAME: &CStr = ansi_str!("api-ms-win-core-synch-l1-2-0"); + const WAIT_ON_ADDRESS: &CStr = ansi_str!("WaitOnAddress"); + const WAKE_BY_ADDRESS_SINGLE: &CStr = ansi_str!("WakeByAddressSingle"); + + // Try loading the library and all the required functions. + // If any step fails, then they all fail. + let library = unsafe { Module::new(MODULE_NAME) }?; + let wait_on_address = library.proc_address(WAIT_ON_ADDRESS)?; + let wake_by_address_single = library.proc_address(WAKE_BY_ADDRESS_SINGLE)?; + + c::WaitOnAddress::PTR.store(wait_on_address.as_ptr(), Ordering::Relaxed); + c::WakeByAddressSingle::PTR.store(wake_by_address_single.as_ptr(), Ordering::Relaxed); + Some(()) + } - #[allow(unused)] - pub(in crate::sys) fn preload(module: Module) { - unsafe { - static SYMBOL_NAME: &CStr = ansi_str!(sym $symbol); - if let Some(f) = module.proc_address(SYMBOL_NAME) { - PTR.store(f.as_ptr(), Ordering::Relaxed); - } - } - } - } - )*) + try_load(); } diff --git a/library/std/src/sys/windows/fs.rs b/library/std/src/sys/windows/fs.rs index aed082b3e..155d0297e 100644 --- a/library/std/src/sys/windows/fs.rs +++ b/library/std/src/sys/windows/fs.rs @@ -2,8 +2,8 @@ use crate::os::windows::prelude::*; use crate::ffi::OsString; use crate::fmt; -use crate::io::{self, Error, IoSlice, IoSliceMut, ReadBuf, SeekFrom}; -use crate::mem; +use crate::io::{self, BorrowedCursor, Error, IoSlice, IoSliceMut, SeekFrom}; +use crate::mem::{self, MaybeUninit}; use crate::os::windows::io::{AsHandle, BorrowedHandle}; use crate::path::{Path, PathBuf}; use crate::ptr; @@ -11,7 +11,7 @@ use crate::slice; use crate::sync::Arc; use crate::sys::handle::Handle; use crate::sys::time::SystemTime; -use crate::sys::{c, cvt}; +use crate::sys::{c, cvt, Align8}; use crate::sys_common::{AsInner, FromInner, IntoInner}; use crate::thread; @@ -326,9 +326,15 @@ impl File { cvt(c::GetFileInformationByHandle(self.handle.as_raw_handle(), &mut info))?; let mut reparse_tag = 0; if info.dwFileAttributes & c::FILE_ATTRIBUTE_REPARSE_POINT != 0 { - let mut b = [0; c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE]; - if let Ok((_, buf)) = self.reparse_point(&mut b) { - reparse_tag = buf.ReparseTag; + let mut attr_tag: c::FILE_ATTRIBUTE_TAG_INFO = mem::zeroed(); + cvt(c::GetFileInformationByHandleEx( + self.handle.as_raw_handle(), + c::FileAttributeTagInfo, + ptr::addr_of_mut!(attr_tag).cast(), + mem::size_of::().try_into().unwrap(), + ))?; + if attr_tag.FileAttributes & c::FILE_ATTRIBUTE_REPARSE_POINT != 0 { + reparse_tag = attr_tag.ReparseTag; } } Ok(FileAttr { @@ -389,9 +395,15 @@ impl File { attr.file_size = info.AllocationSize as u64; attr.number_of_links = Some(info.NumberOfLinks); if attr.file_type().is_reparse_point() { - let mut b = [0; c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE]; - if let Ok((_, buf)) = self.reparse_point(&mut b) { - attr.reparse_tag = buf.ReparseTag; + let mut attr_tag: c::FILE_ATTRIBUTE_TAG_INFO = mem::zeroed(); + cvt(c::GetFileInformationByHandleEx( + self.handle.as_raw_handle(), + c::FileAttributeTagInfo, + ptr::addr_of_mut!(attr_tag).cast(), + mem::size_of::().try_into().unwrap(), + ))?; + if attr_tag.FileAttributes & c::FILE_ATTRIBUTE_REPARSE_POINT != 0 { + attr.reparse_tag = attr_tag.ReparseTag; } } Ok(attr) @@ -415,8 +427,8 @@ impl File { self.handle.read_at(buf, offset) } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { - self.handle.read_buf(buf) + pub fn read_buf(&self, cursor: BorrowedCursor<'_>) -> io::Result<()> { + self.handle.read_buf(cursor) } pub fn write(&self, buf: &[u8]) -> io::Result { @@ -458,38 +470,46 @@ impl File { Ok(Self { handle: self.handle.try_clone()? }) } - fn reparse_point<'a>( + // NB: returned pointer is derived from `space`, and has provenance to + // match. A raw pointer is returned rather than a reference in order to + // avoid narrowing provenance to the actual `REPARSE_DATA_BUFFER`. + fn reparse_point( &self, - space: &'a mut [u8; c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE], - ) -> io::Result<(c::DWORD, &'a c::REPARSE_DATA_BUFFER)> { + space: &mut Align8<[MaybeUninit]>, + ) -> io::Result<(c::DWORD, *const c::REPARSE_DATA_BUFFER)> { unsafe { let mut bytes = 0; cvt({ + // Grab this in advance to avoid it invalidating the pointer + // we get from `space.0.as_mut_ptr()`. + let len = space.0.len(); c::DeviceIoControl( self.handle.as_raw_handle(), c::FSCTL_GET_REPARSE_POINT, ptr::null_mut(), 0, - space.as_mut_ptr() as *mut _, - space.len() as c::DWORD, + space.0.as_mut_ptr().cast(), + len as c::DWORD, &mut bytes, ptr::null_mut(), ) })?; - Ok((bytes, &*(space.as_ptr() as *const c::REPARSE_DATA_BUFFER))) + const _: () = assert!(core::mem::align_of::() <= 8); + Ok((bytes, space.0.as_ptr().cast::())) } } fn readlink(&self) -> io::Result { - let mut space = [0u8; c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE]; + let mut space = Align8([MaybeUninit::::uninit(); c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE]); let (_bytes, buf) = self.reparse_point(&mut space)?; unsafe { - let (path_buffer, subst_off, subst_len, relative) = match buf.ReparseTag { + let (path_buffer, subst_off, subst_len, relative) = match (*buf).ReparseTag { c::IO_REPARSE_TAG_SYMLINK => { let info: *const c::SYMBOLIC_LINK_REPARSE_BUFFER = - &buf.rest as *const _ as *const _; + ptr::addr_of!((*buf).rest).cast(); + assert!(info.is_aligned()); ( - &(*info).PathBuffer as *const _ as *const u16, + ptr::addr_of!((*info).PathBuffer).cast::(), (*info).SubstituteNameOffset / 2, (*info).SubstituteNameLength / 2, (*info).Flags & c::SYMLINK_FLAG_RELATIVE != 0, @@ -497,9 +517,10 @@ impl File { } c::IO_REPARSE_TAG_MOUNT_POINT => { let info: *const c::MOUNT_POINT_REPARSE_BUFFER = - &buf.rest as *const _ as *const _; + ptr::addr_of!((*buf).rest).cast(); + assert!(info.is_aligned()); ( - &(*info).PathBuffer as *const _ as *const u16, + ptr::addr_of!((*info).PathBuffer).cast::(), (*info).SubstituteNameOffset / 2, (*info).SubstituteNameLength / 2, false, @@ -512,7 +533,7 @@ impl File { )); } }; - let subst_ptr = path_buffer.offset(subst_off as isize); + let subst_ptr = path_buffer.add(subst_off.into()); let mut subst = slice::from_raw_parts(subst_ptr, subst_len as usize); // Absolute paths start with an NT internal namespace prefix `\??\` // We should not let it leak through. @@ -649,27 +670,31 @@ impl File { /// A buffer for holding directory entries. struct DirBuff { - buffer: Vec, + buffer: Box; Self::BUFFER_SIZE]>>, } impl DirBuff { + const BUFFER_SIZE: usize = 1024; fn new() -> Self { - const BUFFER_SIZE: usize = 1024; - Self { buffer: vec![0_u8; BUFFER_SIZE] } + Self { + // Safety: `Align8<[MaybeUninit; N]>` does not need + // initialization. + buffer: unsafe { Box::new_uninit().assume_init() }, + } } fn capacity(&self) -> usize { - self.buffer.len() + self.buffer.0.len() } fn as_mut_ptr(&mut self) -> *mut u8 { - self.buffer.as_mut_ptr().cast() + self.buffer.0.as_mut_ptr().cast() } /// Returns a `DirBuffIter`. fn iter(&self) -> DirBuffIter<'_> { DirBuffIter::new(self) } } -impl AsRef<[u8]> for DirBuff { - fn as_ref(&self) -> &[u8] { - &self.buffer +impl AsRef<[MaybeUninit]> for DirBuff { + fn as_ref(&self) -> &[MaybeUninit] { + &self.buffer.0 } } @@ -677,7 +702,7 @@ impl AsRef<[u8]> for DirBuff { /// /// Currently only returns file names (UTF-16 encoded). struct DirBuffIter<'a> { - buffer: Option<&'a [u8]>, + buffer: Option<&'a [MaybeUninit]>, cursor: usize, } impl<'a> DirBuffIter<'a> { @@ -692,14 +717,21 @@ impl<'a> Iterator for DirBuffIter<'a> { let buffer = &self.buffer?[self.cursor..]; // Get the name and next entry from the buffer. - // SAFETY: The buffer contains a `FILE_ID_BOTH_DIR_INFO` struct but the - // last field (the file name) is unsized. So an offset has to be - // used to get the file name slice. + // SAFETY: + // - The buffer contains a `FILE_ID_BOTH_DIR_INFO` struct but the last + // field (the file name) is unsized. So an offset has to be used to + // get the file name slice. + // - The OS has guaranteed initialization of the fields of + // `FILE_ID_BOTH_DIR_INFO` and the trailing filename (for at least + // `FileNameLength` bytes) let (name, is_directory, next_entry) = unsafe { let info = buffer.as_ptr().cast::(); + // Guaranteed to be aligned in documentation for + // https://docs.microsoft.com/en-us/windows/win32/api/winbase/ns-winbase-file_id_both_dir_info + assert!(info.is_aligned()); let next_entry = (*info).NextEntryOffset as usize; let name = crate::slice::from_raw_parts( - (*info).FileName.as_ptr().cast::(), + ptr::addr_of!((*info).FileName).cast::(), (*info).FileNameLength as usize / size_of::(), ); let is_directory = ((*info).FileAttributes & c::FILE_ATTRIBUTE_DIRECTORY) != 0; @@ -1337,18 +1369,19 @@ fn symlink_junction_inner(original: &Path, junction: &Path) -> io::Result<()> { let h = f.as_inner().as_raw_handle(); unsafe { - let mut data = [0u8; c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE]; - let db = data.as_mut_ptr() as *mut c::REPARSE_MOUNTPOINT_DATA_BUFFER; - let buf = &mut (*db).ReparseTarget as *mut c::WCHAR; + let mut data = Align8([MaybeUninit::::uninit(); c::MAXIMUM_REPARSE_DATA_BUFFER_SIZE]); + let data_ptr = data.0.as_mut_ptr(); + let db = data_ptr.cast::(); + let buf = ptr::addr_of_mut!((*db).ReparseTarget).cast::(); let mut i = 0; // FIXME: this conversion is very hacky let v = br"\??\"; let v = v.iter().map(|x| *x as u16); for c in v.chain(original.as_os_str().encode_wide()) { - *buf.offset(i) = c; + *buf.add(i) = c; i += 1; } - *buf.offset(i) = 0; + *buf.add(i) = 0; i += 1; (*db).ReparseTag = c::IO_REPARSE_TAG_MOUNT_POINT; (*db).ReparseTargetMaximumLength = (i * 2) as c::WORD; @@ -1359,7 +1392,7 @@ fn symlink_junction_inner(original: &Path, junction: &Path) -> io::Result<()> { cvt(c::DeviceIoControl( h as *mut _, c::FSCTL_SET_REPARSE_POINT, - data.as_ptr() as *mut _, + data_ptr.cast(), (*db).ReparseDataLength + 8, ptr::null_mut(), 0, diff --git a/library/std/src/sys/windows/handle.rs b/library/std/src/sys/windows/handle.rs index e24b09cc9..ae33d48c6 100644 --- a/library/std/src/sys/windows/handle.rs +++ b/library/std/src/sys/windows/handle.rs @@ -4,7 +4,7 @@ mod tests; use crate::cmp; -use crate::io::{self, ErrorKind, IoSlice, IoSliceMut, Read, ReadBuf}; +use crate::io::{self, BorrowedCursor, ErrorKind, IoSlice, IoSliceMut, Read}; use crate::mem; use crate::os::windows::io::{ AsHandle, AsRawHandle, BorrowedHandle, FromRawHandle, IntoRawHandle, OwnedHandle, RawHandle, @@ -112,18 +112,16 @@ impl Handle { } } - pub fn read_buf(&self, buf: &mut ReadBuf<'_>) -> io::Result<()> { - let res = unsafe { - self.synchronous_read(buf.unfilled_mut().as_mut_ptr(), buf.remaining(), None) - }; + pub fn read_buf(&self, mut cursor: BorrowedCursor<'_>) -> io::Result<()> { + let res = + unsafe { self.synchronous_read(cursor.as_mut().as_mut_ptr(), cursor.capacity(), None) }; match res { Ok(read) => { // Safety: `read` bytes were written to the initialized portion of the buffer unsafe { - buf.assume_init(read as usize); + cursor.advance(read as usize); } - buf.add_filled(read as usize); Ok(()) } diff --git a/library/std/src/sys/windows/locks/mutex.rs b/library/std/src/sys/windows/locks/mutex.rs index f91e8f9f5..91207f5f4 100644 --- a/library/std/src/sys/windows/locks/mutex.rs +++ b/library/std/src/sys/windows/locks/mutex.rs @@ -37,8 +37,6 @@ impl Mutex { pub const fn new() -> Mutex { Mutex { srwlock: UnsafeCell::new(c::SRWLOCK_INIT) } } - #[inline] - pub unsafe fn init(&mut self) {} #[inline] pub unsafe fn lock(&self) { diff --git a/library/std/src/sys/windows/mod.rs b/library/std/src/sys/windows/mod.rs index b3f6d2d0a..eab9b9612 100644 --- a/library/std/src/sys/windows/mod.rs +++ b/library/std/src/sys/windows/mod.rs @@ -2,6 +2,7 @@ use crate::ffi::{CStr, OsStr, OsString}; use crate::io::ErrorKind; +use crate::mem::MaybeUninit; use crate::os::windows::ffi::{OsStrExt, OsStringExt}; use crate::path::PathBuf; use crate::time::Duration; @@ -47,7 +48,7 @@ cfg_if::cfg_if! { // SAFETY: must be called only once during runtime initialization. // NOTE: this is not guaranteed to run, for example when Rust code is called externally. -pub unsafe fn init(_argc: isize, _argv: *const *const u8) { +pub unsafe fn init(_argc: isize, _argv: *const *const u8, _sigpipe: u8) { stack_overflow::init(); // Normally, `thread::spawn` will call `Thread::set_name` but since this thread already @@ -204,8 +205,8 @@ where // This initial size also works around `GetFullPathNameW` returning // incorrect size hints for some short paths: // https://github.com/dylni/normpath/issues/5 - let mut stack_buf = [0u16; 512]; - let mut heap_buf = Vec::new(); + let mut stack_buf: [MaybeUninit; 512] = MaybeUninit::uninit_array(); + let mut heap_buf: Vec> = Vec::new(); unsafe { let mut n = stack_buf.len(); loop { @@ -214,6 +215,11 @@ where } else { let extra = n - heap_buf.len(); heap_buf.reserve(extra); + // We used `reserve` and not `reserve_exact`, so in theory we + // may have gotten more than requested. If so, we'd like to use + // it... so long as we won't cause overflow. + n = heap_buf.capacity().min(c::DWORD::MAX as usize); + // Safety: MaybeUninit does not need initialization heap_buf.set_len(n); &mut heap_buf[..] }; @@ -228,13 +234,13 @@ where // error" is still 0 then we interpret it as a 0 length buffer and // not an actual error. c::SetLastError(0); - let k = match f1(buf.as_mut_ptr(), n as c::DWORD) { + let k = match f1(buf.as_mut_ptr().cast::(), n as c::DWORD) { 0 if c::GetLastError() == 0 => 0, 0 => return Err(crate::io::Error::last_os_error()), n => n, } as usize; if k == n && c::GetLastError() == c::ERROR_INSUFFICIENT_BUFFER { - n *= 2; + n = n.saturating_mul(2).min(c::DWORD::MAX as usize); } else if k > n { n = k; } else if k == n { @@ -244,7 +250,9 @@ where // Therefore k never equals n. unreachable!(); } else { - return Ok(f2(&buf[..k])); + // Safety: First `k` values are initialized. + let slice: &[u16] = MaybeUninit::slice_assume_init_ref(&buf[..k]); + return Ok(f2(slice)); } } } @@ -321,3 +329,11 @@ pub fn abort_internal() -> ! { } crate::intrinsics::abort(); } + +/// Align the inner value to 8 bytes. +/// +/// This is enough for almost all of the buffers we're likely to work with in +/// the Windows APIs we use. +#[repr(C, align(8))] +#[derive(Copy, Clone)] +pub(crate) struct Align8(pub T); diff --git a/library/std/src/sys/windows/os.rs b/library/std/src/sys/windows/os.rs index bcac996c0..352337ba3 100644 --- a/library/std/src/sys/windows/os.rs +++ b/library/std/src/sys/windows/os.rs @@ -99,11 +99,11 @@ impl Iterator for Env { } let p = self.cur as *const u16; let mut len = 0; - while *p.offset(len) != 0 { + while *p.add(len) != 0 { len += 1; } - let s = slice::from_raw_parts(p, len as usize); - self.cur = self.cur.offset(len + 1); + let s = slice::from_raw_parts(p, len); + self.cur = self.cur.add(len + 1); // Windows allows environment variables to start with an equals // symbol (in any other position, this is the separator between diff --git a/library/std/src/sys/windows/os_str.rs b/library/std/src/sys/windows/os_str.rs index 11883f150..4bdd8c505 100644 --- a/library/std/src/sys/windows/os_str.rs +++ b/library/std/src/sys/windows/os_str.rs @@ -164,9 +164,7 @@ impl Slice { } pub fn to_owned(&self) -> Buf { - let mut buf = Wtf8Buf::with_capacity(self.inner.len()); - buf.push_wtf8(&self.inner); - Buf { inner: buf } + Buf { inner: self.inner.to_owned() } } pub fn clone_into(&self, buf: &mut Buf) { diff --git a/library/std/src/sys/windows/path/tests.rs b/library/std/src/sys/windows/path/tests.rs index 6eab38cab..623c62361 100644 --- a/library/std/src/sys/windows/path/tests.rs +++ b/library/std/src/sys/windows/path/tests.rs @@ -105,7 +105,7 @@ fn test_parse_prefix_verbatim_device() { assert_eq!(prefix, parse_prefix(r"\\?/C:\windows\system32\notepad.exe")); } -// See #93586 for more infomation. +// See #93586 for more information. #[test] fn test_windows_prefix_components() { use crate::path::Path; diff --git a/library/std/src/sys/windows/rand.rs b/library/std/src/sys/windows/rand.rs index f8fd93a73..d6cd8f802 100644 --- a/library/std/src/sys/windows/rand.rs +++ b/library/std/src/sys/windows/rand.rs @@ -1,35 +1,126 @@ -use crate::io; +//! # Random key generation +//! +//! This module wraps the RNG provided by the OS. There are a few different +//! ways to interface with the OS RNG so it's worth exploring each of the options. +//! Note that at the time of writing these all go through the (undocumented) +//! `bcryptPrimitives.dll` but they use different route to get there. +//! +//! Originally we were using [`RtlGenRandom`], however that function is +//! deprecated and warns it "may be altered or unavailable in subsequent versions". +//! +//! So we switched to [`BCryptGenRandom`] with the `BCRYPT_USE_SYSTEM_PREFERRED_RNG` +//! flag to query and find the system configured RNG. However, this change caused a small +//! but significant number of users to experience panics caused by a failure of +//! this function. See [#94098]. +//! +//! The current version changes this to use the `BCRYPT_RNG_ALG_HANDLE` +//! [Pseudo-handle], which gets the default RNG algorithm without querying the +//! system preference thus hopefully avoiding the previous issue. +//! This is only supported on Windows 10+ so a fallback is used for older versions. +//! +//! [#94098]: https://github.com/rust-lang/rust/issues/94098 +//! [`RtlGenRandom`]: https://docs.microsoft.com/en-us/windows/win32/api/ntsecapi/nf-ntsecapi-rtlgenrandom +//! [`BCryptGenRandom`]: https://docs.microsoft.com/en-us/windows/win32/api/bcrypt/nf-bcrypt-bcryptgenrandom +//! [Pseudo-handle]: https://docs.microsoft.com/en-us/windows/win32/seccng/cng-algorithm-pseudo-handles use crate::mem; use crate::ptr; use crate::sys::c; +/// Generates high quality secure random keys for use by [`HashMap`]. +/// +/// This is used to seed the default [`RandomState`]. +/// +/// [`HashMap`]: crate::collections::HashMap +/// [`RandomState`]: crate::collections::hash_map::RandomState pub fn hashmap_random_keys() -> (u64, u64) { - let mut v = (0, 0); - let ret = unsafe { - c::BCryptGenRandom( - ptr::null_mut(), - &mut v as *mut _ as *mut u8, - mem::size_of_val(&v) as c::ULONG, - c::BCRYPT_USE_SYSTEM_PREFERRED_RNG, - ) - }; - if ret != 0 { fallback_rng() } else { v } + Rng::open().and_then(|rng| rng.gen_random_keys()).unwrap_or_else(fallback_rng) +} + +struct Rng(c::BCRYPT_ALG_HANDLE); +impl Rng { + #[cfg(miri)] + fn open() -> Result { + const BCRYPT_RNG_ALG_HANDLE: c::BCRYPT_ALG_HANDLE = ptr::invalid_mut(0x81); + let _ = ( + c::BCryptOpenAlgorithmProvider, + c::BCryptCloseAlgorithmProvider, + c::BCRYPT_RNG_ALGORITHM, + c::STATUS_NOT_SUPPORTED, + ); + Ok(Self(BCRYPT_RNG_ALG_HANDLE)) + } + #[cfg(not(miri))] + // Open a handle to the RNG algorithm. + fn open() -> Result { + use crate::sync::atomic::AtomicPtr; + use crate::sync::atomic::Ordering::{Acquire, Release}; + const ERROR_VALUE: c::LPVOID = ptr::invalid_mut(usize::MAX); + + // An atomic is used so we don't need to reopen the handle every time. + static HANDLE: AtomicPtr = AtomicPtr::new(ptr::null_mut()); + + let mut handle = HANDLE.load(Acquire); + // We use a sentinel value to designate an error occurred last time. + if handle == ERROR_VALUE { + Err(c::STATUS_NOT_SUPPORTED) + } else if handle.is_null() { + let status = unsafe { + c::BCryptOpenAlgorithmProvider( + &mut handle, + c::BCRYPT_RNG_ALGORITHM.as_ptr(), + ptr::null(), + 0, + ) + }; + if c::nt_success(status) { + // If another thread opens a handle first then use that handle instead. + let result = HANDLE.compare_exchange(ptr::null_mut(), handle, Release, Acquire); + if let Err(previous_handle) = result { + // Close our handle and return the previous one. + unsafe { c::BCryptCloseAlgorithmProvider(handle, 0) }; + handle = previous_handle; + } + Ok(Self(handle)) + } else { + HANDLE.store(ERROR_VALUE, Release); + Err(status) + } + } else { + Ok(Self(handle)) + } + } + + fn gen_random_keys(self) -> Result<(u64, u64), c::NTSTATUS> { + let mut v = (0, 0); + let status = unsafe { + let size = mem::size_of_val(&v).try_into().unwrap(); + c::BCryptGenRandom(self.0, ptr::addr_of_mut!(v).cast(), size, 0) + }; + if c::nt_success(status) { Ok(v) } else { Err(status) } + } } /// Generate random numbers using the fallback RNG function (RtlGenRandom) #[cfg(not(target_vendor = "uwp"))] #[inline(never)] -fn fallback_rng() -> (u64, u64) { +fn fallback_rng(rng_status: c::NTSTATUS) -> (u64, u64) { let mut v = (0, 0); let ret = unsafe { c::RtlGenRandom(&mut v as *mut _ as *mut u8, mem::size_of_val(&v) as c::ULONG) }; - if ret != 0 { v } else { panic!("fallback RNG broken: {}", io::Error::last_os_error()) } + if ret != 0 { + v + } else { + panic!( + "RNG broken: {rng_status:#x}, fallback RNG broken: {}", + crate::io::Error::last_os_error() + ) + } } /// We can't use RtlGenRandom with UWP, so there is no fallback #[cfg(target_vendor = "uwp")] #[inline(never)] -fn fallback_rng() -> (u64, u64) { - panic!("fallback RNG broken: RtlGenRandom() not supported on UWP"); +fn fallback_rng(rng_status: c::NTSTATUS) -> (u64, u64) { + panic!("RNG broken: {rng_status:#x} fallback RNG broken: RtlGenRandom() not supported on UWP"); } diff --git a/library/std/src/sys/windows/stdio.rs b/library/std/src/sys/windows/stdio.rs index a001d6b98..70c9b14a0 100644 --- a/library/std/src/sys/windows/stdio.rs +++ b/library/std/src/sys/windows/stdio.rs @@ -3,6 +3,7 @@ use crate::char::decode_utf16; use crate::cmp; use crate::io; +use crate::mem::MaybeUninit; use crate::os::windows::io::{FromRawHandle, IntoRawHandle}; use crate::ptr; use crate::str; @@ -169,13 +170,14 @@ fn write( } fn write_valid_utf8_to_console(handle: c::HANDLE, utf8: &str) -> io::Result { - let mut utf16 = [0u16; MAX_BUFFER_SIZE / 2]; + let mut utf16 = [MaybeUninit::::uninit(); MAX_BUFFER_SIZE / 2]; let mut len_utf16 = 0; for (chr, dest) in utf8.encode_utf16().zip(utf16.iter_mut()) { - *dest = chr; + *dest = MaybeUninit::new(chr); len_utf16 += 1; } - let utf16 = &utf16[..len_utf16]; + // Safety: We've initialized `len_utf16` values. + let utf16: &[u16] = unsafe { MaybeUninit::slice_assume_init_ref(&utf16[..len_utf16]) }; let mut written = write_u16s(handle, &utf16)?; @@ -250,11 +252,14 @@ impl io::Read for Stdin { return Ok(bytes_copied); } else if buf.len() - bytes_copied < 4 { // Not enough space to get a UTF-8 byte. We will use the incomplete UTF8. - let mut utf16_buf = [0u16; 1]; + let mut utf16_buf = [MaybeUninit::new(0); 1]; // Read one u16 character. let read = read_u16s_fixup_surrogates(handle, &mut utf16_buf, 1, &mut self.surrogate)?; // Read bytes, using the (now-empty) self.incomplete_utf8 as extra space. - let read_bytes = utf16_to_utf8(&utf16_buf[..read], &mut self.incomplete_utf8.bytes)?; + let read_bytes = utf16_to_utf8( + unsafe { MaybeUninit::slice_assume_init_ref(&utf16_buf[..read]) }, + &mut self.incomplete_utf8.bytes, + )?; // Read in the bytes from incomplete_utf8 until the buffer is full. self.incomplete_utf8.len = read_bytes as u8; @@ -262,15 +267,18 @@ impl io::Read for Stdin { bytes_copied += self.incomplete_utf8.read(&mut buf[bytes_copied..]); Ok(bytes_copied) } else { - let mut utf16_buf = [0u16; MAX_BUFFER_SIZE / 2]; + let mut utf16_buf = [MaybeUninit::::uninit(); MAX_BUFFER_SIZE / 2]; + // In the worst case, a UTF-8 string can take 3 bytes for every `u16` of a UTF-16. So // we can read at most a third of `buf.len()` chars and uphold the guarantee no data gets // lost. let amount = cmp::min(buf.len() / 3, utf16_buf.len()); let read = read_u16s_fixup_surrogates(handle, &mut utf16_buf, amount, &mut self.surrogate)?; - - match utf16_to_utf8(&utf16_buf[..read], buf) { + // Safety `read_u16s_fixup_surrogates` returns the number of items + // initialized. + let utf16s = unsafe { MaybeUninit::slice_assume_init_ref(&utf16_buf[..read]) }; + match utf16_to_utf8(utf16s, buf) { Ok(value) => return Ok(bytes_copied + value), Err(e) => return Err(e), } @@ -283,14 +291,14 @@ impl io::Read for Stdin { // This is a best effort, and might not work if we are not the only reader on Stdin. fn read_u16s_fixup_surrogates( handle: c::HANDLE, - buf: &mut [u16], + buf: &mut [MaybeUninit], mut amount: usize, surrogate: &mut u16, ) -> io::Result { // Insert possibly remaining unpaired surrogate from last read. let mut start = 0; if *surrogate != 0 { - buf[0] = *surrogate; + buf[0] = MaybeUninit::new(*surrogate); *surrogate = 0; start = 1; if amount == 1 { @@ -303,7 +311,10 @@ fn read_u16s_fixup_surrogates( let mut amount = read_u16s(handle, &mut buf[start..amount])? + start; if amount > 0 { - let last_char = buf[amount - 1]; + // Safety: The returned `amount` is the number of values initialized, + // and it is not 0, so we know that `buf[amount - 1]` have been + // initialized. + let last_char = unsafe { buf[amount - 1].assume_init() }; if last_char >= 0xD800 && last_char <= 0xDBFF { // high surrogate *surrogate = last_char; @@ -313,7 +324,8 @@ fn read_u16s_fixup_surrogates( Ok(amount) } -fn read_u16s(handle: c::HANDLE, buf: &mut [u16]) -> io::Result { +// Returns `Ok(n)` if it initialized `n` values in `buf`. +fn read_u16s(handle: c::HANDLE, buf: &mut [MaybeUninit]) -> io::Result { // Configure the `pInputControl` parameter to not only return on `\r\n` but also Ctrl-Z, the // traditional DOS method to indicate end of character stream / user input (SUB). // See #38274 and https://stackoverflow.com/questions/43836040/win-api-readconsole. @@ -346,8 +358,9 @@ fn read_u16s(handle: c::HANDLE, buf: &mut [u16]) -> io::Result { } break; } - - if amount > 0 && buf[amount as usize - 1] == CTRL_Z { + // Safety: if `amount > 0`, then that many bytes were written, so + // `buf[amount as usize - 1]` has been initialized. + if amount > 0 && unsafe { buf[amount as usize - 1].assume_init() } == CTRL_Z { amount -= 1; } Ok(amount as usize) diff --git a/library/std/src/sys/windows/thread_local_dtor.rs b/library/std/src/sys/windows/thread_local_dtor.rs index 25d1c6e8e..9707a95df 100644 --- a/library/std/src/sys/windows/thread_local_dtor.rs +++ b/library/std/src/sys/windows/thread_local_dtor.rs @@ -8,10 +8,14 @@ #[thread_local] static mut DESTRUCTORS: Vec<(*mut u8, unsafe extern "C" fn(*mut u8))> = Vec::new(); +// Ensure this can never be inlined because otherwise this may break in dylibs. +// See #44391. +#[inline(never)] pub unsafe fn register_dtor(t: *mut u8, dtor: unsafe extern "C" fn(*mut u8)) { DESTRUCTORS.push((t, dtor)); } +#[inline(never)] // See comment above /// Runs destructors. This should not be called until thread exit. pub unsafe fn run_keyless_dtors() { // Drop all the destructors. diff --git a/library/std/src/sys/windows/thread_parker.rs b/library/std/src/sys/windows/thread_parker.rs index d876e0f6f..2f7ae863b 100644 --- a/library/std/src/sys/windows/thread_parker.rs +++ b/library/std/src/sys/windows/thread_parker.rs @@ -197,19 +197,17 @@ impl Parker { // purpose, to make sure every unpark() has a release-acquire ordering // with park(). if self.state.swap(NOTIFIED, Release) == PARKED { - if let Some(wake_by_address_single) = c::WakeByAddressSingle::option() { - unsafe { + unsafe { + if let Some(wake_by_address_single) = c::WakeByAddressSingle::option() { wake_by_address_single(self.ptr()); - } - } else { - // If we run NtReleaseKeyedEvent before the waiting thread runs - // NtWaitForKeyedEvent, this (shortly) blocks until we can wake it up. - // If the waiting thread wakes up before we run NtReleaseKeyedEvent - // (e.g. due to a timeout), this blocks until we do wake up a thread. - // To prevent this thread from blocking indefinitely in that case, - // park_impl() will, after seeing the state set to NOTIFIED after - // waking up, call NtWaitForKeyedEvent again to unblock us. - unsafe { + } else { + // If we run NtReleaseKeyedEvent before the waiting thread runs + // NtWaitForKeyedEvent, this (shortly) blocks until we can wake it up. + // If the waiting thread wakes up before we run NtReleaseKeyedEvent + // (e.g. due to a timeout), this blocks until we do wake up a thread. + // To prevent this thread from blocking indefinitely in that case, + // park_impl() will, after seeing the state set to NOTIFIED after + // waking up, call NtWaitForKeyedEvent again to unblock us. c::NtReleaseKeyedEvent(keyed_event_handle(), self.ptr(), 0, ptr::null_mut()); } } -- cgit v1.2.3