1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624
extern crate errno; extern crate interprocess_traits; extern crate libc; extern crate memfd; extern crate thiserror; #[cfg(test)] extern crate sendfd; use std::{ io, mem, ops::Deref, os::unix::io::{AsRawFd, IntoRawFd, RawFd}, ptr, }; use errno::errno; use interprocess_traits::ProcSync; use libc::c_void; use memfd::MemfdOptions; // TODO: Drop the contained `T` when the last `Shared<T>` handle to it is dropped. // This will require: // * To add an atomic reference counter // * To handle properly ZST `T` used with a `size` // TODO: Add a safer interface than AsRawFd/IntoRawFd // This could be done by implementing a trait like `SendShared` / `RecvShared` similar to what is // in `sendfd`, that would use the data-passing functionality of the socket to implement a protocol // that ensures the `T` is the same on both sides. // Note that: // * This will still be unsafe from a rust point of view, as the other side could respect the // protocol but be sending a different fd -- knowing the call is safe requires knowing the other // side is going to respect the protocol // * This will not deprecate the *RawFd series of functions, that will still be required for // fd-passing over interfaces that are not supported by `SendShared` / `RecvShared` // TODO: Consider making this crate work on non-unix systems, just with the fd-passing behavior // suppressed. // It is not obvious that we want to actually do this, as in such cases `Shared` amounts to a less // good `Arc`, and the user should then not be incentivized to use it. /// The error type for errors in this crate. #[derive(Debug, thiserror::Error)] pub enum Error { /// Could not create an in-memory file for shared memory #[error("Could not create an in-memory file for shared memory")] CreateFileError(#[source] memfd::Error), /// Failed to set the length of the shared memory file #[error("Failed to set the length of the shared memory file")] TruncateError(#[source] io::Error), /// Failed to retrieve shared memory file metadata #[error("Failed to retrieve shared memory file metadata")] GetMetadataError(#[source] io::Error), /// File length after truncation does not match requested size #[error( "Failed to truncate the in-memory file to {} bytes, the file is {}B long", expected, actual )] LengthError { expected: usize, actual: usize }, /// Failed to map shared memory #[error("Failed to map shared memory")] MmapError(#[source] io::Error), /// Failed to duplicate the file descriptor #[error("Failed to duplicate the file descriptor")] DupError(#[source] io::Error), } /// A memory-mapped region pointing to an element of type T. /// /// The memory region is owned, and will be unmapped when this is dropped. Note that the element of /// type T will *not* be dropped. struct MmapRegion<T> { /// Start of the memory-mapped region ptr: *mut T, /// Size of the memory-mapped region (usually `mem::size_of::<T>()`, but it can be greater than /// that for eg. `c_void`, which can be used for `!Sized` types) size: usize, } impl<T> MmapRegion<T> { /// Memory-maps `fd` with size `size` and returns a pointer to the mapped memory region. /// /// # Unsafety /// /// This assumes that `fd` points to a file of at least size `size`. unsafe fn new(size: usize, fd: RawFd) -> Result<MmapRegion<T>, Error> { // TODO: investigate the potential impact of MAP_HUGETLB | MAP_HUGE_2MB let ptr = libc::mmap( ptr::null_mut(), size, libc::PROT_READ | libc::PROT_WRITE, // Read-write mapping // MAP_SHARED: Share mapping with `fork`'d processes, validate that all flags are known // MAP_POPULATE: immediately reserve the pages, do not lazily allocate them libc::MAP_SHARED_VALIDATE | libc::MAP_POPULATE, fd, 0, // offset ); // Check the memory map succeeded if ptr == libc::MAP_FAILED { return Err(Error::MmapError(errno().into())); } Ok(MmapRegion { ptr: ptr as *mut T, size, }) } } impl<T> Drop for MmapRegion<T> { fn drop(&mut self) { // Do not drop contents here, see type-level documentation unsafe { // This is safe thanks to the memory region being owned. libc::munmap(self.ptr as *mut c_void, self.size); } // For now, if `munmap` failed, we're ignoring it. // However it means that something happened, so we might some day want to change this, for // instance by logging something. // See also https://github.com/rust-lang/rfcs/pull/2677 } } /// A wrapper for data that can be shared across processes. /// /// The data is owned, but the element of type `T` will *not* be dropped when the `Shared<T>` is /// dropped. Note that an element that has a meaningful `Drop` is likely not `ProcSync` anyway. pub struct Shared<T> { fd: RawFd, region: MmapRegion<T>, } // These implementations are safe thanks to `Shared` giving only a reference-based access to the // underlying `T` unsafe impl<T: Sync> Send for Shared<T> {} unsafe impl<T: Sync> Sync for Shared<T> {} /// Creates an uninitialized `Shared<T>` that points to a memory region of size `size`. /// /// # Unsafety /// /// Returned `Shared` will point to uninitialized memory. /// /// # Panics /// /// Panics if `T` requires an alignment greater than the page size. unsafe fn create_shared<T>(size: usize) -> Result<Shared<T>, Error> { // Check that alignment of T is at most one page let page_size = libc::sysconf(libc::_SC_PAGE_SIZE) as usize; let requested_align = mem::align_of::<T>(); if requested_align > page_size { // Note: This is not implemented as an error, as: // 1. This error condition is *particularly* unlikely to ever happen, requiring both a // developer with appetite for custom-align structs and a user with a somehow reduced // page size. // 2. This makes us forwards-compatible for the day we'll decide we want to actually // support that use case (by allocating lots of memory). panic!("Page size {}B is too low for requested alignment {}", page_size, requested_align); } // Create the file let memfd = MemfdOptions::new() .close_on_exec(true) .create("caring") .map_err(|e| Error::CreateFileError(e))?; let file = memfd.into_file(); // Truncate file.set_len(size as u64).map_err(Error::TruncateError)?; // Check truncation succeeded let actual_size = file.metadata().map_err(Error::GetMetadataError)?.len() as usize; if actual_size != size { return Err(Error::LengthError { expected: size, actual: actual_size, }); } // Retrieve the file descriptor let fd = file.into_raw_fd(); // Memory map the file // The unsafety requirement here is ensured by the “Check truncation succeeded” section above let region = MmapRegion::new(size, fd)?; Ok(Shared { fd, region }) } impl<T> Shared<T> { /// Creates and initializes a `Shared<T>` to value `val`. pub fn new(val: T) -> Result<Shared<T>, Error> { unsafe { // This is safe thanks to `T` being of size `size_of::<T>()` let res = create_shared::<T>(mem::size_of::<T>())?; // This is safe thanks to `res` not yet being shared as it has just been created ptr::write_volatile(res.region.ptr, val); Ok(res) } } } impl Shared<c_void> { // TODO: Make this able to return DSTs (when Rust will have a proper DST story) // This will also require implementing CoerceUnsized for proper !Sized handling /// Creates a `Shared<c_void>` that points to a memory region of size `size`, aligned to page /// boundary. /// /// Proper handling of the returned value is left up to the user. pub fn new_sized(size: usize) -> Result<Shared<c_void>, Error> { unsafe { create_shared(size) // `c_void` is a ZST so this is safe, and usage is up to the caller's judgement } } } impl<T> Shared<T> { /// Creates a `Shared<T>` from a pre-existing `fd`. /// /// # Unsafety /// /// This assumes that `fd` is a file descriptor that has been created by another instance of /// `Shared<T>`, and that it will never be used by anything else than `Shared<T>`. Note that /// not respecting this will at best do as bad as `std::mem::transmute`, and at worst end in /// tears. /// /// This also assumes that `fd` is not shared with another `Shared<T>`, as the output of /// `as_raw_fd()` is only a borrow. Do not forget to call `dup` before passing the file /// descriptor to `from_raw_fd` if you are not using `into_raw_fd` and have not passed the /// `RawFd` over a socket. /// /// Finally, this assumes that `fd` is not coming from another process, as otherwise for safety /// we would need a `T: ProcSync` bound. unsafe fn from_raw_fd_impl(fd: RawFd) -> Result<Shared<T>, Error> { // Retrieve the length of the file let mut statbuf = mem::zeroed::<libc::stat>(); if libc::fstat(fd, &mut statbuf) != 0 { return Err(Error::GetMetadataError(io::Error::last_os_error())); } assert_eq!(statbuf.st_mode & libc::S_IFMT, libc::S_IFREG); let size = statbuf.st_size as usize; // Memory map the file let region = MmapRegion::new(size, fd)?; Ok(Shared { fd, region }) } /// Attempts to clone `data`. /// /// This will remap `data` at another location in memory, in addition to keeping `data` alive. pub fn try_clone(data: &Shared<T>) -> Result<Shared<T>, Error> { unsafe { let fd = libc::dup(data.as_raw_fd()); if fd == -1 { return Err(Error::DupError(errno().into())); } // The unsafety requirements here are satisfied by the successful `dup` and the fact // that types are checked by the signature of this function. Self::from_raw_fd_impl(fd) } } /// Returns a mutable pointer to the data contained by `data`. /// /// Note that using this pointer needs the caller to handle synchronization themselves. pub fn as_mut_ptr(data: &Shared<T>) -> *mut T { data.region.ptr } } impl<T: ProcSync> Shared<T> { /// Creates a `Shared<T>` from pre-existing `fd`. /// /// # Unsafety /// /// This assumes that `fd` is a file descriptor that has been created by another instance of /// `Shared<T>`, and that it will never be used by anything else than `Shared<T>`. Note that /// not respecting this will at best do as bad as `std::mem::transmute`, and at worst end in /// tears. /// /// This also assumes that `fd` is not shared with another `Shared<T>`, as the output of /// `as_raw_fd()` is only a borrow. Do not forget to call `dup` before passing the file /// descriptor to `from_raw_fd` if you are not using `into_raw_fd` and have not passed the /// `RawFd` over a socket. pub unsafe fn from_raw_fd(fd: RawFd) -> Result<Shared<T>, Error> { Self::from_raw_fd_impl(fd) } } impl<T> Drop for Shared<T> { fn drop(&mut self) { unsafe { libc::close(self.fd); } // For now, if `close` failed, we're ignoring it. // However it means that something happened, so we might some day want to change this, for // instance by logging something. // See also https://github.com/rust-lang/rfcs/pull/2677 } } impl<T> Deref for Shared<T> { type Target = T; fn deref(&self) -> &T { // This is safe thanks to the only way of sharing memory being through `from_raw_fd`, which // itself is available only when `T: ProcSync`. There is also no way to safely obtain an // &mut to the memory region. The only way to do it being by unsafely dereferencing a // pointer retrieved from the returned reference, which would be wildly unsafe and most // likely UB. // Note that on the other hand DerefMut would *not* be safe. Long live interior mutability. unsafe { &*self.region.ptr } } } impl<T> AsRawFd for Shared<T> { fn as_raw_fd(&self) -> RawFd { self.fd } } impl<T> IntoRawFd for Shared<T> { fn into_raw_fd(mut self) -> RawFd { let res = self.fd; // Drop self.region without dropping self. // This is safe thanks to `mem::forget`'ing the partially-moved-out-of `self` unsafe { ptr::drop_in_place(&mut self.region); mem::forget(self); } res } } #[cfg(test)] mod tests { use crate::*; use std::{ os::unix::net::UnixDatagram, process, sync::{ atomic::{AtomicBool, AtomicUsize, Ordering}, Arc, }, thread, }; use sendfd::{RecvWithFd, SendWithFd}; macro_rules! test_write_and_read { ($zone:expr, $size:expr) => {{ let zone = $zone; let size = $size; let ptr_mut = &*zone as *const _ as *mut u8; for i in 0..size { ptr::write_volatile(ptr_mut.offset(i as isize), i as u8); } let ptr = &*zone as *const _ as *const u8; for i in 0..size { assert_eq!(ptr::read_volatile(ptr.offset(i as isize)), i as u8); } }}; } #[test] fn new_sized_allocates_properly() { const SIZE: usize = 10 * 4 * 1024 + 1; // 10 pages (maybe) plus one byte let zone = Shared::new_sized(SIZE).unwrap(); unsafe { test_write_and_read!(zone, SIZE) }; } #[test] fn new_allocates_properly() { const SIZE: usize = 4 * 1024 - 1; // 1 page (maybe) minus one byte let zone = Shared::new([0u8; SIZE]).unwrap(); unsafe { test_write_and_read!(zone, SIZE) }; } macro_rules! test_sync_across_threads { ($zone_name:ident, $base_name:ident; $build_zone:stmt, $clone_zone:expr; $v:ident, $incr:expr, $decr:expr) => {{ const $base_name: usize = 42; const INCR: usize = 987650; const DECR: usize = 901230; $build_zone let zone1 = $clone_zone; let zone2 = $clone_zone; let incr = thread::spawn(move || { let $v = &*zone1; for _ in 0..INCR { $incr; } }); let decr = thread::spawn(move || { let $v = &*zone2; for _ in 0..DECR { $decr; } }); incr.join().unwrap(); decr.join().unwrap(); assert_eq!($zone_name.load(Ordering::SeqCst), BASE + INCR - DECR); }}; } macro_rules! test_sync_across_threads_arc { ($v:ident, $incr:expr, $decr:expr) => { test_sync_across_threads!( zone, BASE; let zone = Arc::new(Shared::new(AtomicUsize::new(BASE)).unwrap()), zone.clone(); $v, $incr, $decr ) } } #[test] #[should_panic] fn syncs_across_threads_test_can_fail() { test_sync_across_threads_arc!( v, v.store( v.load(Ordering::SeqCst).overflowing_add(1).0, Ordering::SeqCst ), v.store( v.load(Ordering::SeqCst).overflowing_sub(1).0, Ordering::SeqCst ) ); } #[test] fn syncs_across_threads() { test_sync_across_threads_arc!( v, v.fetch_add(1, Ordering::SeqCst), v.fetch_sub(1, Ordering::SeqCst) ); } macro_rules! test_sync_across_threads_different_shared { ($v:ident, $incr:expr, $decr:expr) => {{ test_sync_across_threads!( zone, BASE; let zone = Shared::new(AtomicUsize::new(BASE)).unwrap(), Shared::try_clone(&zone).unwrap(); $v, $incr, $decr ) }}; } #[test] #[should_panic] fn syncs_across_threads_different_shared_can_fail() { test_sync_across_threads_different_shared!( v, v.store( v.load(Ordering::SeqCst).overflowing_add(1).0, Ordering::SeqCst ), v.store( v.load(Ordering::SeqCst).overflowing_sub(1).0, Ordering::SeqCst ) ); } #[test] fn syncs_across_threads_different_shared() { test_sync_across_threads_different_shared!( v, v.fetch_add(1, Ordering::SeqCst), v.fetch_sub(1, Ordering::SeqCst) ); } macro_rules! test_sync_across_processes_with_fork { ($v:ident, $incr:expr, $decr:expr) => {{ const BASE: usize = 1337; const INCR: usize = 890120; const DECR: usize = 876540; let zone = Shared::new((AtomicUsize::new(BASE), AtomicBool::new(false))).unwrap(); let ($v, child_complete) = &*zone; let child = || { // In the child for _ in 0..INCR { $incr; } child_complete.store(true, Ordering::SeqCst); }; let parent = || { // In the parent for _ in 0..DECR { $decr; } while !child_complete.load(Ordering::SeqCst) { thread::yield_now(); } assert_eq!(zone.0.load(Ordering::SeqCst), BASE + INCR - DECR); }; unsafe { let pid = libc::fork(); if pid == 0 { child(); process::exit(0); } else { parent(); libc::waitpid(pid, ptr::null_mut(), 0); // Reap child } } }}; } #[test] #[should_panic] fn syncs_across_processes_with_fork_test_can_fail() { test_sync_across_processes_with_fork!( v, v.store( v.load(Ordering::SeqCst).overflowing_add(1).0, Ordering::SeqCst ), v.store( v.load(Ordering::SeqCst).overflowing_sub(1).0, Ordering::SeqCst ) ); } #[test] fn syncs_across_processes_with_fork() { test_sync_across_processes_with_fork!( v, v.fetch_add(1, Ordering::SeqCst), v.fetch_sub(1, Ordering::SeqCst) ); } macro_rules! test_sync_across_processes_after_socket_send { ($v:ident, $incr:expr, $decr:expr) => {{ const BASE: usize = 10; const INCR: usize = 900000; const DECR: usize = 800000; let (send, receive) = UnixDatagram::pair().unwrap(); let child = || { // In the child let zone = Shared::new((AtomicUsize::new(BASE), AtomicBool::new(false))).unwrap(); send.send_with_fd(&[], &[zone.as_raw_fd()]) .expect("send should succeed"); let ($v, child_complete) = &*zone; for _ in 0..INCR { $incr; } child_complete.store(true, Ordering::SeqCst); }; let parent = || { // In the parent let mut fd = [0; 1]; receive .recv_with_fd(&mut [], &mut fd) .expect("recv should succeed"); let zone: Shared<(AtomicUsize, AtomicBool)> = unsafe { Shared::from_raw_fd(fd[0]).unwrap() }; let ($v, child_complete) = &*zone; for _ in 0..DECR { $decr; } while !child_complete.load(Ordering::SeqCst) { thread::yield_now(); } assert_eq!($v.load(Ordering::SeqCst), BASE + INCR - DECR); }; unsafe { let pid = libc::fork(); if pid == 0 { child(); process::exit(0); } else { parent(); libc::waitpid(pid, ptr::null_mut(), 0); // Reap child } } }}; } #[test] #[should_panic] fn syncs_across_processes_after_socket_send_test_can_fail() { test_sync_across_processes_after_socket_send!( v, v.store( v.load(Ordering::SeqCst).overflowing_add(1).0, Ordering::SeqCst ), v.store( v.load(Ordering::SeqCst).overflowing_sub(1).0, Ordering::SeqCst ) ); } #[test] fn syncs_across_processes_after_socket_send() { test_sync_across_processes_after_socket_send!( v, v.fetch_add(1, Ordering::SeqCst), v.fetch_sub(1, Ordering::SeqCst) ); } }