kernel/task.rs
1// SPDX-License-Identifier: GPL-2.0
2
3//! Tasks (threads and processes).
4//!
5//! C header: [`include/linux/sched.h`](srctree/include/linux/sched.h).
6
7use crate::{
8 bindings,
9 ffi::{c_int, c_long, c_uint},
10 mm::MmWithUser,
11 pid_namespace::PidNamespace,
12 sync::aref::ARef,
13 types::{NotThreadSafe, Opaque},
14};
15use core::{
16 cmp::{Eq, PartialEq},
17 ops::Deref,
18 ptr,
19};
20
21/// A sentinel value used for infinite timeouts.
22pub const MAX_SCHEDULE_TIMEOUT: c_long = c_long::MAX;
23
24/// Bitmask for tasks that are sleeping in an interruptible state.
25pub const TASK_INTERRUPTIBLE: c_int = bindings::TASK_INTERRUPTIBLE as c_int;
26/// Bitmask for tasks that are sleeping in an uninterruptible state.
27pub const TASK_UNINTERRUPTIBLE: c_int = bindings::TASK_UNINTERRUPTIBLE as c_int;
28/// Bitmask for tasks that are sleeping in a freezable state.
29pub const TASK_FREEZABLE: c_int = bindings::TASK_FREEZABLE as c_int;
30/// Convenience constant for waking up tasks regardless of whether they are in interruptible or
31/// uninterruptible sleep.
32pub const TASK_NORMAL: c_uint = bindings::TASK_NORMAL as c_uint;
33
34/// Returns the currently running task.
35#[macro_export]
36macro_rules! current {
37 () => {
38 // SAFETY: This expression creates a temporary value that is dropped at the end of the
39 // caller's scope. The following mechanisms ensure that the resulting `&CurrentTask` cannot
40 // leave current task context:
41 //
42 // * To return to userspace, the caller must leave the current scope.
43 // * Operations such as `begin_new_exec()` are necessarily unsafe and the caller of
44 // `begin_new_exec()` is responsible for safety.
45 // * Rust abstractions for things such as a `kthread_use_mm()` scope must require the
46 // closure to be `Send`, so the `NotThreadSafe` field of `CurrentTask` ensures that the
47 // `&CurrentTask` cannot cross the scope in either direction.
48 unsafe { &*$crate::task::Task::current() }
49 };
50}
51
52/// Wraps the kernel's `struct task_struct`.
53///
54/// # Invariants
55///
56/// All instances are valid tasks created by the C portion of the kernel.
57///
58/// Instances of this type are always refcounted, that is, a call to `get_task_struct` ensures
59/// that the allocation remains valid at least until the matching call to `put_task_struct`.
60///
61/// # Examples
62///
63/// The following is an example of getting the PID of the current thread with zero additional cost
64/// when compared to the C version:
65///
66/// ```
67/// let pid = current!().pid();
68/// ```
69///
70/// Getting the PID of the current process, also zero additional cost:
71///
72/// ```
73/// let pid = current!().group_leader().pid();
74/// ```
75///
76/// Getting the current task and storing it in some struct. The reference count is automatically
77/// incremented when creating `State` and decremented when it is dropped:
78///
79/// ```
80/// use kernel::{task::Task, sync::aref::ARef};
81///
82/// struct State {
83/// creator: ARef<Task>,
84/// index: u32,
85/// }
86///
87/// impl State {
88/// fn new() -> Self {
89/// Self {
90/// creator: ARef::from(&**current!()),
91/// index: 0,
92/// }
93/// }
94/// }
95/// ```
96#[repr(transparent)]
97pub struct Task(pub(crate) Opaque<bindings::task_struct>);
98
99// SAFETY: By design, the only way to access a `Task` is via the `current` function or via an
100// `ARef<Task>` obtained through the `AlwaysRefCounted` impl. This means that the only situation in
101// which a `Task` can be accessed mutably is when the refcount drops to zero and the destructor
102// runs. It is safe for that to happen on any thread, so it is ok for this type to be `Send`.
103unsafe impl Send for Task {}
104
105// SAFETY: It's OK to access `Task` through shared references from other threads because we're
106// either accessing properties that don't change (e.g., `pid`, `group_leader`) or that are properly
107// synchronised by C code (e.g., `signal_pending`).
108unsafe impl Sync for Task {}
109
110/// Represents the [`Task`] in the `current` global.
111///
112/// This type exists to provide more efficient operations that are only valid on the current task.
113/// For example, to retrieve the pid-namespace of a task, you must use rcu protection unless it is
114/// the current task.
115///
116/// # Invariants
117///
118/// Each value of this type must only be accessed from the task context it was created within.
119///
120/// Of course, every thread is in a different task context, but for the purposes of this invariant,
121/// these operations also permanently leave the task context:
122///
123/// * Returning to userspace from system call context.
124/// * Calling `release_task()`.
125/// * Calling `begin_new_exec()` in a binary format loader.
126///
127/// Other operations temporarily create a new sub-context:
128///
129/// * Calling `kthread_use_mm()` creates a new context, and `kthread_unuse_mm()` returns to the
130/// old context.
131///
132/// This means that a `CurrentTask` obtained before a `kthread_use_mm()` call may be used again
133/// once `kthread_unuse_mm()` is called, but it must not be used between these two calls.
134/// Conversely, a `CurrentTask` obtained between a `kthread_use_mm()`/`kthread_unuse_mm()` pair
135/// must not be used after `kthread_unuse_mm()`.
136#[repr(transparent)]
137pub struct CurrentTask(Task, NotThreadSafe);
138
139// Make all `Task` methods available on `CurrentTask`.
140impl Deref for CurrentTask {
141 type Target = Task;
142 #[inline]
143 fn deref(&self) -> &Task {
144 &self.0
145 }
146}
147
148/// The type of process identifiers (PIDs).
149pub type Pid = bindings::pid_t;
150
151/// The type of user identifiers (UIDs).
152#[derive(Copy, Clone)]
153pub struct Kuid {
154 kuid: bindings::kuid_t,
155}
156
157impl Task {
158 /// Returns a raw pointer to the current task.
159 ///
160 /// It is up to the user to use the pointer correctly.
161 #[inline]
162 pub fn current_raw() -> *mut bindings::task_struct {
163 // SAFETY: Getting the current pointer is always safe.
164 unsafe { bindings::get_current() }
165 }
166
167 /// Returns a task reference for the currently executing task/thread.
168 ///
169 /// The recommended way to get the current task/thread is to use the
170 /// [`current`] macro because it is safe.
171 ///
172 /// # Safety
173 ///
174 /// Callers must ensure that the returned object is only used to access a [`CurrentTask`]
175 /// within the task context that was active when this function was called. For more details,
176 /// see the invariants section for [`CurrentTask`].
177 #[inline]
178 pub unsafe fn current() -> impl Deref<Target = CurrentTask> {
179 struct TaskRef {
180 task: *const CurrentTask,
181 }
182
183 impl Deref for TaskRef {
184 type Target = CurrentTask;
185
186 fn deref(&self) -> &Self::Target {
187 // SAFETY: The returned reference borrows from this `TaskRef`, so it cannot outlive
188 // the `TaskRef`, which the caller of `Task::current()` has promised will not
189 // outlive the task/thread for which `self.task` is the `current` pointer. Thus, it
190 // is okay to return a `CurrentTask` reference here.
191 unsafe { &*self.task }
192 }
193 }
194
195 TaskRef {
196 // CAST: The layout of `struct task_struct` and `CurrentTask` is identical.
197 task: Task::current_raw().cast(),
198 }
199 }
200
201 /// Returns a raw pointer to the task.
202 #[inline]
203 pub fn as_ptr(&self) -> *mut bindings::task_struct {
204 self.0.get()
205 }
206
207 /// Returns the group leader of the given task.
208 pub fn group_leader(&self) -> &Task {
209 // SAFETY: The group leader of a task never changes after initialization, so reading this
210 // field is not a data race.
211 let ptr = unsafe { *ptr::addr_of!((*self.as_ptr()).group_leader) };
212
213 // SAFETY: The lifetime of the returned task reference is tied to the lifetime of `self`,
214 // and given that a task has a reference to its group leader, we know it must be valid for
215 // the lifetime of the returned task reference.
216 unsafe { &*ptr.cast() }
217 }
218
219 /// Returns the PID of the given task.
220 pub fn pid(&self) -> Pid {
221 // SAFETY: The pid of a task never changes after initialization, so reading this field is
222 // not a data race.
223 unsafe { *ptr::addr_of!((*self.as_ptr()).pid) }
224 }
225
226 /// Returns the UID of the given task.
227 #[inline]
228 pub fn uid(&self) -> Kuid {
229 // SAFETY: It's always safe to call `task_uid` on a valid task.
230 Kuid::from_raw(unsafe { bindings::task_uid(self.as_ptr()) })
231 }
232
233 /// Returns the effective UID of the given task.
234 #[inline]
235 pub fn euid(&self) -> Kuid {
236 // SAFETY: It's always safe to call `task_euid` on a valid task.
237 Kuid::from_raw(unsafe { bindings::task_euid(self.as_ptr()) })
238 }
239
240 /// Determines whether the given task has pending signals.
241 #[inline]
242 pub fn signal_pending(&self) -> bool {
243 // SAFETY: It's always safe to call `signal_pending` on a valid task.
244 unsafe { bindings::signal_pending(self.as_ptr()) != 0 }
245 }
246
247 /// Returns task's pid namespace with elevated reference count
248 #[inline]
249 pub fn get_pid_ns(&self) -> Option<ARef<PidNamespace>> {
250 // SAFETY: By the type invariant, we know that `self.0` is valid.
251 let ptr = unsafe { bindings::task_get_pid_ns(self.as_ptr()) };
252 if ptr.is_null() {
253 None
254 } else {
255 // SAFETY: `ptr` is valid by the safety requirements of this function. And we own a
256 // reference count via `task_get_pid_ns()`.
257 // CAST: `Self` is a `repr(transparent)` wrapper around `bindings::pid_namespace`.
258 Some(unsafe { ARef::from_raw(ptr::NonNull::new_unchecked(ptr.cast::<PidNamespace>())) })
259 }
260 }
261
262 /// Returns the given task's pid in the provided pid namespace.
263 #[doc(alias = "task_tgid_nr_ns")]
264 #[inline]
265 pub fn tgid_nr_ns(&self, pidns: Option<&PidNamespace>) -> Pid {
266 let pidns = match pidns {
267 Some(pidns) => pidns.as_ptr(),
268 None => core::ptr::null_mut(),
269 };
270 // SAFETY: By the type invariant, we know that `self.0` is valid. We received a valid
271 // PidNamespace that we can use as a pointer or we received an empty PidNamespace and
272 // thus pass a null pointer. The underlying C function is safe to be used with NULL
273 // pointers.
274 unsafe { bindings::task_tgid_nr_ns(self.as_ptr(), pidns) }
275 }
276
277 /// Wakes up the task.
278 #[inline]
279 pub fn wake_up(&self) {
280 // SAFETY: It's always safe to call `wake_up_process` on a valid task, even if the task
281 // running.
282 unsafe { bindings::wake_up_process(self.as_ptr()) };
283 }
284}
285
286impl CurrentTask {
287 /// Access the address space of the current task.
288 ///
289 /// This function does not touch the refcount of the mm.
290 #[inline]
291 pub fn mm(&self) -> Option<&MmWithUser> {
292 // SAFETY: The `mm` field of `current` is not modified from other threads, so reading it is
293 // not a data race.
294 let mm = unsafe { (*self.as_ptr()).mm };
295
296 if mm.is_null() {
297 return None;
298 }
299
300 // SAFETY: If `current->mm` is non-null, then it references a valid mm with a non-zero
301 // value of `mm_users`. Furthermore, the returned `&MmWithUser` borrows from this
302 // `CurrentTask`, so it cannot escape the scope in which the current pointer was obtained.
303 //
304 // This is safe even if `kthread_use_mm()`/`kthread_unuse_mm()` are used. There are two
305 // relevant cases:
306 // * If the `&CurrentTask` was created before `kthread_use_mm()`, then it cannot be
307 // accessed during the `kthread_use_mm()`/`kthread_unuse_mm()` scope due to the
308 // `NotThreadSafe` field of `CurrentTask`.
309 // * If the `&CurrentTask` was created within a `kthread_use_mm()`/`kthread_unuse_mm()`
310 // scope, then the `&CurrentTask` cannot escape that scope, so the returned `&MmWithUser`
311 // also cannot escape that scope.
312 // In either case, it's not possible to read `current->mm` and keep using it after the
313 // scope is ended with `kthread_unuse_mm()`.
314 Some(unsafe { MmWithUser::from_raw(mm) })
315 }
316
317 /// Access the pid namespace of the current task.
318 ///
319 /// This function does not touch the refcount of the namespace or use RCU protection.
320 ///
321 /// To access the pid namespace of another task, see [`Task::get_pid_ns`].
322 #[doc(alias = "task_active_pid_ns")]
323 #[inline]
324 pub fn active_pid_ns(&self) -> Option<&PidNamespace> {
325 // SAFETY: It is safe to call `task_active_pid_ns` without RCU protection when calling it
326 // on the current task.
327 let active_ns = unsafe { bindings::task_active_pid_ns(self.as_ptr()) };
328
329 if active_ns.is_null() {
330 return None;
331 }
332
333 // The lifetime of `PidNamespace` is bound to `Task` and `struct pid`.
334 //
335 // The `PidNamespace` of a `Task` doesn't ever change once the `Task` is alive.
336 //
337 // From system call context retrieving the `PidNamespace` for the current task is always
338 // safe and requires neither RCU locking nor a reference count to be held. Retrieving the
339 // `PidNamespace` after `release_task()` for current will return `NULL` but no codepath
340 // like that is exposed to Rust.
341 //
342 // SAFETY: If `current`'s pid ns is non-null, then it references a valid pid ns.
343 // Furthermore, the returned `&PidNamespace` borrows from this `CurrentTask`, so it cannot
344 // escape the scope in which the current pointer was obtained, e.g. it cannot live past a
345 // `release_task()` call.
346 Some(unsafe { PidNamespace::from_ptr(active_ns) })
347 }
348}
349
350// SAFETY: The type invariants guarantee that `Task` is always refcounted.
351unsafe impl crate::sync::aref::AlwaysRefCounted for Task {
352 #[inline]
353 fn inc_ref(&self) {
354 // SAFETY: The existence of a shared reference means that the refcount is nonzero.
355 unsafe { bindings::get_task_struct(self.as_ptr()) };
356 }
357
358 #[inline]
359 unsafe fn dec_ref(obj: ptr::NonNull<Self>) {
360 // SAFETY: The safety requirements guarantee that the refcount is nonzero.
361 unsafe { bindings::put_task_struct(obj.cast().as_ptr()) }
362 }
363}
364
365impl Kuid {
366 /// Get the current euid.
367 #[inline]
368 pub fn current_euid() -> Kuid {
369 // SAFETY: Just an FFI call.
370 Self::from_raw(unsafe { bindings::current_euid() })
371 }
372
373 /// Create a `Kuid` given the raw C type.
374 #[inline]
375 pub fn from_raw(kuid: bindings::kuid_t) -> Self {
376 Self { kuid }
377 }
378
379 /// Turn this kuid into the raw C type.
380 #[inline]
381 pub fn into_raw(self) -> bindings::kuid_t {
382 self.kuid
383 }
384
385 /// Converts this kernel UID into a userspace UID.
386 ///
387 /// Uses the namespace of the current task.
388 #[inline]
389 pub fn into_uid_in_current_ns(self) -> bindings::uid_t {
390 // SAFETY: Just an FFI call.
391 unsafe { bindings::from_kuid(bindings::current_user_ns(), self.kuid) }
392 }
393}
394
395impl PartialEq for Kuid {
396 #[inline]
397 fn eq(&self, other: &Kuid) -> bool {
398 // SAFETY: Just an FFI call.
399 unsafe { bindings::uid_eq(self.kuid, other.kuid) }
400 }
401}
402
403impl Eq for Kuid {}
404
405/// Annotation for functions that can sleep.
406///
407/// Equivalent to the C side [`might_sleep()`], this function serves as
408/// a debugging aid and a potential scheduling point.
409///
410/// This function can only be used in a nonatomic context.
411///
412/// [`might_sleep()`]: https://docs.kernel.org/driver-api/basics.html#c.might_sleep
413#[track_caller]
414#[inline]
415pub fn might_sleep() {
416 #[cfg(CONFIG_DEBUG_ATOMIC_SLEEP)]
417 {
418 let loc = core::panic::Location::caller();
419 let file = kernel::file_from_location(loc);
420
421 // SAFETY: `file.as_ptr()` is valid for reading and guaranteed to be nul-terminated.
422 unsafe { crate::bindings::__might_sleep(file.as_ptr().cast(), loc.line() as i32) }
423 }
424
425 // SAFETY: Always safe to call.
426 unsafe { crate::bindings::might_resched() }
427}