diff options
Diffstat (limited to 'third_party/rust/memmap2/src')
| -rw-r--r-- | third_party/rust/memmap2/src/advice.rs | 384 | ||||
| -rw-r--r-- | third_party/rust/memmap2/src/lib.rs | 2179 | ||||
| -rw-r--r-- | third_party/rust/memmap2/src/stub.rs | 81 | ||||
| -rw-r--r-- | third_party/rust/memmap2/src/unix.rs | 458 | ||||
| -rw-r--r-- | third_party/rust/memmap2/src/windows.rs | 524 |
5 files changed, 3626 insertions, 0 deletions
diff --git a/third_party/rust/memmap2/src/advice.rs b/third_party/rust/memmap2/src/advice.rs new file mode 100644 index 0000000000..4316058fc8 --- /dev/null +++ b/third_party/rust/memmap2/src/advice.rs @@ -0,0 +1,384 @@ +/// Values supported by [`Mmap::advise`][crate::Mmap::advise] and [`MmapMut::advise`][crate::MmapMut::advise] functions. +/// +/// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. +#[repr(i32)] +#[derive(Clone, Copy, Debug, Eq, PartialEq, Hash)] +pub enum Advice { + /// **MADV_NORMAL** + /// + /// No special treatment. This is the default. + Normal = libc::MADV_NORMAL, + + /// **MADV_RANDOM** + /// + /// Expect page references in random order. (Hence, read + /// ahead may be less useful than normally.) + Random = libc::MADV_RANDOM, + + /// **MADV_SEQUENTIAL** + /// + /// Expect page references in sequential order. (Hence, pages + /// in the given range can be aggressively read ahead, and may + /// be freed soon after they are accessed.) + Sequential = libc::MADV_SEQUENTIAL, + + /// **MADV_WILLNEED** + /// + /// Expect access in the near future. (Hence, it might be a + /// good idea to read some pages ahead.) + WillNeed = libc::MADV_WILLNEED, + + /// **MADV_DONTFORK** - Linux only (since Linux 2.6.16) + /// + /// Do not make the pages in this range available to the child + /// after a fork(2). This is useful to prevent copy-on-write + /// semantics from changing the physical location of a page if + /// the parent writes to it after a fork(2). (Such page + /// relocations cause problems for hardware that DMAs into the + /// page.) + #[cfg(target_os = "linux")] + DontFork = libc::MADV_DONTFORK, + + /// **MADV_DOFORK** - Linux only (since Linux 2.6.16) + /// + /// Undo the effect of MADV_DONTFORK, restoring the default + /// behavior, whereby a mapping is inherited across fork(2). + #[cfg(target_os = "linux")] + DoFork = libc::MADV_DOFORK, + + /// **MADV_MERGEABLE** - Linux only (since Linux 2.6.32) + /// + /// Enable Kernel Samepage Merging (KSM) for the pages in the + /// range specified by addr and length. The kernel regularly + /// scans those areas of user memory that have been marked as + /// mergeable, looking for pages with identical content. + /// These are replaced by a single write-protected page (which + /// is automatically copied if a process later wants to update + /// the content of the page). KSM merges only private + /// anonymous pages (see mmap(2)). + /// + /// The KSM feature is intended for applications that generate + /// many instances of the same data (e.g., virtualization + /// systems such as KVM). It can consume a lot of processing + /// power; use with care. See the Linux kernel source file + /// Documentation/admin-guide/mm/ksm.rst for more details. + /// + /// The MADV_MERGEABLE and MADV_UNMERGEABLE operations are + /// available only if the kernel was configured with + /// CONFIG_KSM. + #[cfg(target_os = "linux")] + Mergeable = libc::MADV_MERGEABLE, + + /// **MADV_UNMERGEABLE** - Linux only (since Linux 2.6.32) + /// + /// Undo the effect of an earlier MADV_MERGEABLE operation on + /// the specified address range; KSM unmerges whatever pages + /// it had merged in the address range specified by addr and + /// length. + #[cfg(target_os = "linux")] + Unmergeable = libc::MADV_UNMERGEABLE, + + /// **MADV_HUGEPAGE** - Linux only (since Linux 2.6.38) + /// + /// Enable Transparent Huge Pages (THP) for pages in the range + /// specified by addr and length. Currently, Transparent Huge + /// Pages work only with private anonymous pages (see + /// mmap(2)). The kernel will regularly scan the areas marked + /// as huge page candidates to replace them with huge pages. + /// The kernel will also allocate huge pages directly when the + /// region is naturally aligned to the huge page size (see + /// posix_memalign(2)). + /// + /// This feature is primarily aimed at applications that use + /// large mappings of data and access large regions of that + /// memory at a time (e.g., virtualization systems such as + /// QEMU). It can very easily waste memory (e.g., a 2 MB + /// mapping that only ever accesses 1 byte will result in 2 MB + /// of wired memory instead of one 4 KB page). See the Linux + /// kernel source file + /// Documentation/admin-guide/mm/transhuge.rst for more + /// details. + /// + /// Most common kernels configurations provide MADV_HUGEPAGE- + /// style behavior by default, and thus MADV_HUGEPAGE is + /// normally not necessary. It is mostly intended for + /// embedded systems, where MADV_HUGEPAGE-style behavior may + /// not be enabled by default in the kernel. On such systems, + /// this flag can be used in order to selectively enable THP. + /// Whenever MADV_HUGEPAGE is used, it should always be in + /// regions of memory with an access pattern that the + /// developer knows in advance won't risk to increase the + /// memory footprint of the application when transparent + /// hugepages are enabled. + /// + /// The MADV_HUGEPAGE and MADV_NOHUGEPAGE operations are + /// available only if the kernel was configured with + /// CONFIG_TRANSPARENT_HUGEPAGE. + #[cfg(target_os = "linux")] + HugePage = libc::MADV_HUGEPAGE, + + /// **MADV_NOHUGEPAGE** - Linux only (since Linux 2.6.38) + /// + /// Ensures that memory in the address range specified by addr + /// and length will not be backed by transparent hugepages. + #[cfg(target_os = "linux")] + NoHugePage = libc::MADV_NOHUGEPAGE, + + /// **MADV_DONTDUMP** - Linux only (since Linux 3.4) + /// + /// Exclude from a core dump those pages in the range + /// specified by addr and length. This is useful in + /// applications that have large areas of memory that are + /// known not to be useful in a core dump. The effect of + /// **MADV_DONTDUMP** takes precedence over the bit mask that is + /// set via the `/proc/[pid]/coredump_filter` file (see + /// core(5)). + #[cfg(target_os = "linux")] + DontDump = libc::MADV_DONTDUMP, + + /// **MADV_DODUMP** - Linux only (since Linux 3.4) + /// + /// Undo the effect of an earlier MADV_DONTDUMP. + #[cfg(target_os = "linux")] + DoDump = libc::MADV_DODUMP, + + /// **MADV_HWPOISON** - Linux only (since Linux 2.6.32) + /// + /// Poison the pages in the range specified by addr and length + /// and handle subsequent references to those pages like a + /// hardware memory corruption. This operation is available + /// only for privileged (CAP_SYS_ADMIN) processes. This + /// operation may result in the calling process receiving a + /// SIGBUS and the page being unmapped. + /// + /// This feature is intended for testing of memory error- + /// handling code; it is available only if the kernel was + /// configured with CONFIG_MEMORY_FAILURE. + #[cfg(target_os = "linux")] + HwPoison = libc::MADV_HWPOISON, + + /// **MADV_POPULATE_READ** - Linux only (since Linux 5.14) + /// + /// Populate (prefault) page tables readable, faulting in all + /// pages in the range just as if manually reading from each + /// page; however, avoid the actual memory access that would have + /// been performed after handling the fault. + /// + /// In contrast to MAP_POPULATE, MADV_POPULATE_READ does not hide + /// errors, can be applied to (parts of) existing mappings and + /// will always populate (prefault) page tables readable. One + /// example use case is prefaulting a file mapping, reading all + /// file content from disk; however, pages won't be dirtied and + /// consequently won't have to be written back to disk when + /// evicting the pages from memory. + /// + /// Depending on the underlying mapping, map the shared zeropage, + /// preallocate memory or read the underlying file; files with + /// holes might or might not preallocate blocks. If populating + /// fails, a SIGBUS signal is not generated; instead, an error is + /// returned. + /// + /// If MADV_POPULATE_READ succeeds, all page tables have been + /// populated (prefaulted) readable once. If MADV_POPULATE_READ + /// fails, some page tables might have been populated. + /// + /// MADV_POPULATE_READ cannot be applied to mappings without read + /// permissions and special mappings, for example, mappings + /// marked with kernel-internal flags such as VM_PFNMAP or VM_IO, + /// or secret memory regions created using memfd_secret(2). + /// + /// Note that with MADV_POPULATE_READ, the process can be killed + /// at any moment when the system runs out of memory. + #[cfg(target_os = "linux")] + PopulateRead = libc::MADV_POPULATE_READ, + + /// **MADV_POPULATE_WRITE** - Linux only (since Linux 5.14) + /// + /// Populate (prefault) page tables writable, faulting in all + /// pages in the range just as if manually writing to each each + /// page; however, avoid the actual memory access that would have + /// been performed after handling the fault. + /// + /// In contrast to MAP_POPULATE, MADV_POPULATE_WRITE does not + /// hide errors, can be applied to (parts of) existing mappings + /// and will always populate (prefault) page tables writable. + /// One example use case is preallocating memory, breaking any + /// CoW (Copy on Write). + /// + /// Depending on the underlying mapping, preallocate memory or + /// read the underlying file; files with holes will preallocate + /// blocks. If populating fails, a SIGBUS signal is not gener‐ + /// ated; instead, an error is returned. + /// + /// If MADV_POPULATE_WRITE succeeds, all page tables have been + /// populated (prefaulted) writable once. If MADV_POPULATE_WRITE + /// fails, some page tables might have been populated. + /// + /// MADV_POPULATE_WRITE cannot be applied to mappings without + /// write permissions and special mappings, for example, mappings + /// marked with kernel-internal flags such as VM_PFNMAP or VM_IO, + /// or secret memory regions created using memfd_secret(2). + /// + /// Note that with MADV_POPULATE_WRITE, the process can be killed + /// at any moment when the system runs out of memory. + #[cfg(target_os = "linux")] + PopulateWrite = libc::MADV_POPULATE_WRITE, + + /// **MADV_ZERO_WIRED_PAGES** - Darwin only + /// + /// Indicates that the application would like the wired pages in this address range to be + /// zeroed out if the address range is deallocated without first unwiring the pages (i.e. + /// a munmap(2) without a preceding munlock(2) or the application quits). This is used + /// with madvise() system call. + #[cfg(any(target_os = "macos", target_os = "ios"))] + ZeroWiredPages = libc::MADV_ZERO_WIRED_PAGES, +} + +/// Values supported by [`Mmap::unsafe_advise`][crate::Mmap::unsafe_advise] and [`MmapMut::unsafe_advise`][crate::MmapMut::unsafe_advise] functions. +/// +/// These flags can be passed to the [madvise (2)][man_page] system call +/// and effects on the mapped pages which are conceptually writes, +/// i.e. the change the observable contents of these pages which +/// implies undefined behaviour if the mapping is still borrowed. +/// +/// Hence, these potentially unsafe flags must be used with the unsafe +/// methods and the programmer has to justify that the code +/// does not keep any borrows of the mapping active while the mapped pages +/// are updated by the kernel's memory management subsystem. +/// +/// [man_page]: https://man7.org/linux/man-pages/man2/madvise.2.html +#[repr(i32)] +#[derive(Clone, Copy, Debug, Eq, PartialEq, Hash)] +pub enum UncheckedAdvice { + /// **MADV_DONTNEED** + /// + /// Do not expect access in the near future. (For the time + /// being, the application is finished with the given range, + /// so the kernel can free resources associated with it.) + /// + /// After a successful MADV_DONTNEED operation, the semantics + /// of memory access in the specified region are changed: + /// subsequent accesses of pages in the range will succeed, + /// but will result in either repopulating the memory contents + /// from the up-to-date contents of the underlying mapped file + /// (for shared file mappings, shared anonymous mappings, and + /// shmem-based techniques such as System V shared memory + /// segments) or zero-fill-on-demand pages for anonymous + /// private mappings. + /// + /// Note that, when applied to shared mappings, MADV_DONTNEED + /// might not lead to immediate freeing of the pages in the + /// range. The kernel is free to delay freeing the pages + /// until an appropriate moment. The resident set size (RSS) + /// of the calling process will be immediately reduced + /// however. + /// + /// **MADV_DONTNEED** cannot be applied to locked pages, Huge TLB + /// pages, or VM_PFNMAP pages. (Pages marked with the kernel- + /// internal VM_PFNMAP flag are special memory areas that are + /// not managed by the virtual memory subsystem. Such pages + /// are typically created by device drivers that map the pages + /// into user space.) + /// + /// # Safety + /// + /// Using the returned value with conceptually write to the + /// mapped pages, i.e. borrowing the mapping when the pages + /// are freed results in undefined behaviour. + DontNeed = libc::MADV_DONTNEED, + + // + // The rest are Linux-specific + // + /// **MADV_FREE** - Linux (since Linux 4.5) and Darwin + /// + /// The application no longer requires the pages in the range + /// specified by addr and len. The kernel can thus free these + /// pages, but the freeing could be delayed until memory + /// pressure occurs. For each of the pages that has been + /// marked to be freed but has not yet been freed, the free + /// operation will be canceled if the caller writes into the + /// page. After a successful MADV_FREE operation, any stale + /// data (i.e., dirty, unwritten pages) will be lost when the + /// kernel frees the pages. However, subsequent writes to + /// pages in the range will succeed and then kernel cannot + /// free those dirtied pages, so that the caller can always + /// see just written data. If there is no subsequent write, + /// the kernel can free the pages at any time. Once pages in + /// the range have been freed, the caller will see zero-fill- + /// on-demand pages upon subsequent page references. + /// + /// The MADV_FREE operation can be applied only to private + /// anonymous pages (see mmap(2)). In Linux before version + /// 4.12, when freeing pages on a swapless system, the pages + /// in the given range are freed instantly, regardless of + /// memory pressure. + /// + /// # Safety + /// + /// Using the returned value with conceptually write to the + /// mapped pages, i.e. borrowing the mapping while the pages + /// are still being freed results in undefined behaviour. + #[cfg(any(target_os = "linux", target_os = "macos", target_os = "ios"))] + Free = libc::MADV_FREE, + + /// **MADV_REMOVE** - Linux only (since Linux 2.6.16) + /// + /// Free up a given range of pages and its associated backing + /// store. This is equivalent to punching a hole in the + /// corresponding byte range of the backing store (see + /// fallocate(2)). Subsequent accesses in the specified + /// address range will see bytes containing zero. + /// + /// The specified address range must be mapped shared and + /// writable. This flag cannot be applied to locked pages, + /// Huge TLB pages, or VM_PFNMAP pages. + /// + /// In the initial implementation, only tmpfs(5) was supported + /// **MADV_REMOVE**; but since Linux 3.5, any filesystem which + /// supports the fallocate(2) FALLOC_FL_PUNCH_HOLE mode also + /// supports MADV_REMOVE. Hugetlbfs fails with the error + /// EINVAL and other filesystems fail with the error + /// EOPNOTSUPP. + /// + /// # Safety + /// + /// Using the returned value with conceptually write to the + /// mapped pages, i.e. borrowing the mapping when the pages + /// are freed results in undefined behaviour. + #[cfg(target_os = "linux")] + Remove = libc::MADV_REMOVE, + + /// **MADV_FREE_REUSABLE** - Darwin only + /// + /// Behaves like **MADV_FREE**, but the freed pages are accounted for in the RSS of the process. + /// + /// # Safety + /// + /// Using the returned value with conceptually write to the + /// mapped pages, i.e. borrowing the mapping while the pages + /// are still being freed results in undefined behaviour. + #[cfg(any(target_os = "macos", target_os = "ios"))] + FreeReusable = libc::MADV_FREE_REUSABLE, + + /// **MADV_FREE_REUSE** - Darwin only + /// + /// Marks a memory region previously freed by **MADV_FREE_REUSABLE** as non-reusable, accounts + /// for the pages in the RSS of the process. Pages that have been freed will be replaced by + /// zero-filled pages on demand, other pages will be left as is. + /// + /// # Safety + /// + /// Using the returned value with conceptually write to the + /// mapped pages, i.e. borrowing the mapping while the pages + /// are still being freed results in undefined behaviour. + #[cfg(any(target_os = "macos", target_os = "ios"))] + FreeReuse = libc::MADV_FREE_REUSE, +} + +// Future expansion: +// MADV_SOFT_OFFLINE (since Linux 2.6.33) +// MADV_WIPEONFORK (since Linux 4.14) +// MADV_KEEPONFORK (since Linux 4.14) +// MADV_COLD (since Linux 5.4) +// MADV_PAGEOUT (since Linux 5.4) diff --git a/third_party/rust/memmap2/src/lib.rs b/third_party/rust/memmap2/src/lib.rs new file mode 100644 index 0000000000..fb912e6d03 --- /dev/null +++ b/third_party/rust/memmap2/src/lib.rs @@ -0,0 +1,2179 @@ +//! A cross-platform Rust API for memory mapped buffers. +//! +//! The core functionality is provided by either [`Mmap`] or [`MmapMut`], +//! which correspond to mapping a [`File`] to a [`&[u8]`](https://doc.rust-lang.org/std/primitive.slice.html) +//! or [`&mut [u8]`](https://doc.rust-lang.org/std/primitive.slice.html) +//! respectively. Both function by dereferencing to a slice, allowing the +//! [`Mmap`]/[`MmapMut`] to be used in the same way you would the equivalent slice +//! types. +//! +//! [`File`]: std::fs::File +//! +//! # Examples +//! +//! For simple cases [`Mmap`] can be used directly: +//! +//! ``` +//! use std::fs::File; +//! use std::io::Read; +//! +//! use memmap2::Mmap; +//! +//! # fn main() -> std::io::Result<()> { +//! let mut file = File::open("LICENSE-APACHE")?; +//! +//! let mut contents = Vec::new(); +//! file.read_to_end(&mut contents)?; +//! +//! let mmap = unsafe { Mmap::map(&file)? }; +//! +//! assert_eq!(&contents[..], &mmap[..]); +//! # Ok(()) +//! # } +//! ``` +//! +//! However for cases which require configuration of the mapping, then +//! you can use [`MmapOptions`] in order to further configure a mapping +//! before you create it. + +#![allow(clippy::len_without_is_empty, clippy::missing_safety_doc)] + +#[cfg_attr(unix, path = "unix.rs")] +#[cfg_attr(windows, path = "windows.rs")] +#[cfg_attr(not(any(unix, windows)), path = "stub.rs")] +mod os; +use crate::os::{file_len, MmapInner}; + +#[cfg(unix)] +mod advice; +#[cfg(unix)] +pub use crate::advice::{Advice, UncheckedAdvice}; + +use std::fmt; +#[cfg(not(any(unix, windows)))] +use std::fs::File; +use std::io::{Error, ErrorKind, Result}; +use std::isize; +use std::mem; +use std::ops::{Deref, DerefMut}; +#[cfg(unix)] +use std::os::unix::io::{AsRawFd, RawFd}; +#[cfg(windows)] +use std::os::windows::io::{AsRawHandle, RawHandle}; +use std::slice; + +#[cfg(not(any(unix, windows)))] +pub struct MmapRawDescriptor<'a>(&'a File); + +#[cfg(unix)] +pub struct MmapRawDescriptor(RawFd); + +#[cfg(windows)] +pub struct MmapRawDescriptor(RawHandle); + +pub trait MmapAsRawDesc { + fn as_raw_desc(&self) -> MmapRawDescriptor; +} + +#[cfg(not(any(unix, windows)))] +impl MmapAsRawDesc for &File { + fn as_raw_desc(&self) -> MmapRawDescriptor { + MmapRawDescriptor(self) + } +} + +#[cfg(unix)] +impl MmapAsRawDesc for RawFd { + fn as_raw_desc(&self) -> MmapRawDescriptor { + MmapRawDescriptor(*self) + } +} + +#[cfg(unix)] +impl<'a, T> MmapAsRawDesc for &'a T +where + T: AsRawFd, +{ + fn as_raw_desc(&self) -> MmapRawDescriptor { + MmapRawDescriptor(self.as_raw_fd()) + } +} + +#[cfg(windows)] +impl MmapAsRawDesc for RawHandle { + fn as_raw_desc(&self) -> MmapRawDescriptor { + MmapRawDescriptor(*self) + } +} + +#[cfg(windows)] +impl<'a, T> MmapAsRawDesc for &'a T +where + T: AsRawHandle, +{ + fn as_raw_desc(&self) -> MmapRawDescriptor { + MmapRawDescriptor(self.as_raw_handle()) + } +} + +/// A memory map builder, providing advanced options and flags for specifying memory map behavior. +/// +/// `MmapOptions` can be used to create an anonymous memory map using [`map_anon()`], or a +/// file-backed memory map using one of [`map()`], [`map_mut()`], [`map_exec()`], +/// [`map_copy()`], or [`map_copy_read_only()`]. +/// +/// ## Safety +/// +/// All file-backed memory map constructors are marked `unsafe` because of the potential for +/// *Undefined Behavior* (UB) using the map if the underlying file is subsequently modified, in or +/// out of process. Applications must consider the risk and take appropriate precautions when +/// using file-backed maps. Solutions such as file permissions, locks or process-private (e.g. +/// unlinked) files exist but are platform specific and limited. +/// +/// [`map_anon()`]: MmapOptions::map_anon() +/// [`map()`]: MmapOptions::map() +/// [`map_mut()`]: MmapOptions::map_mut() +/// [`map_exec()`]: MmapOptions::map_exec() +/// [`map_copy()`]: MmapOptions::map_copy() +/// [`map_copy_read_only()`]: MmapOptions::map_copy_read_only() +#[derive(Clone, Debug, Default)] +pub struct MmapOptions { + offset: u64, + len: Option<usize>, + huge: Option<u8>, + stack: bool, + populate: bool, +} + +impl MmapOptions { + /// Creates a new set of options for configuring and creating a memory map. + /// + /// # Example + /// + /// ``` + /// use memmap2::{MmapMut, MmapOptions}; + /// # use std::io::Result; + /// + /// # fn main() -> Result<()> { + /// // Create a new memory map builder. + /// let mut mmap_options = MmapOptions::new(); + /// + /// // Configure the memory map builder using option setters, then create + /// // a memory map using one of `mmap_options.map_anon`, `mmap_options.map`, + /// // `mmap_options.map_mut`, `mmap_options.map_exec`, or `mmap_options.map_copy`: + /// let mut mmap: MmapMut = mmap_options.len(36).map_anon()?; + /// + /// // Use the memory map: + /// mmap.copy_from_slice(b"...data to copy to the memory map..."); + /// # Ok(()) + /// # } + /// ``` + pub fn new() -> MmapOptions { + MmapOptions::default() + } + + /// Configures the memory map to start at byte `offset` from the beginning of the file. + /// + /// This option has no effect on anonymous memory maps. + /// + /// By default, the offset is 0. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// use std::fs::File; + /// + /// # fn main() -> std::io::Result<()> { + /// let mmap = unsafe { + /// MmapOptions::new() + /// .offset(30) + /// .map(&File::open("LICENSE-APACHE")?)? + /// }; + /// assert_eq!(&b"Apache License"[..], + /// &mmap[..14]); + /// # Ok(()) + /// # } + /// ``` + pub fn offset(&mut self, offset: u64) -> &mut Self { + self.offset = offset; + self + } + + /// Configures the created memory mapped buffer to be `len` bytes long. + /// + /// This option is mandatory for anonymous memory maps. + /// + /// For file-backed memory maps, the length will default to the file length. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// use std::fs::File; + /// + /// # fn main() -> std::io::Result<()> { + /// let mmap = unsafe { + /// MmapOptions::new() + /// .len(9) + /// .map(&File::open("README.md")?)? + /// }; + /// assert_eq!(&b"# memmap2"[..], &mmap[..]); + /// # Ok(()) + /// # } + /// ``` + pub fn len(&mut self, len: usize) -> &mut Self { + self.len = Some(len); + self + } + + /// Returns the configured length, or the length of the provided file. + fn get_len<T: MmapAsRawDesc>(&self, file: &T) -> Result<usize> { + self.len.map(Ok).unwrap_or_else(|| { + let desc = file.as_raw_desc(); + let file_len = file_len(desc.0)?; + + if file_len < self.offset { + return Err(Error::new( + ErrorKind::InvalidData, + "memory map offset is larger than length", + )); + } + let len = file_len - self.offset; + + // Rust's slice cannot be larger than isize::MAX. + // See https://doc.rust-lang.org/std/slice/fn.from_raw_parts.html + // + // This is not a problem on 64-bit targets, but on 32-bit one + // having a file or an anonymous mapping larger than 2GB is quite normal + // and we have to prevent it. + // + // The code below is essentially the same as in Rust's std: + // https://github.com/rust-lang/rust/blob/db78ab70a88a0a5e89031d7ee4eccec835dcdbde/library/alloc/src/raw_vec.rs#L495 + if mem::size_of::<usize>() < 8 && len > isize::MAX as u64 { + return Err(Error::new( + ErrorKind::InvalidData, + "memory map length overflows isize", + )); + } + + Ok(len as usize) + }) + } + + /// Configures the anonymous memory map to be suitable for a process or thread stack. + /// + /// This option corresponds to the `MAP_STACK` flag on Linux. It has no effect on Windows. + /// + /// This option has no effect on file-backed memory maps. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// + /// # fn main() -> std::io::Result<()> { + /// let stack = MmapOptions::new().stack().len(4096).map_anon(); + /// # Ok(()) + /// # } + /// ``` + pub fn stack(&mut self) -> &mut Self { + self.stack = true; + self + } + + /// Configures the anonymous memory map to be allocated using huge pages. + /// + /// This option corresponds to the `MAP_HUGETLB` flag on Linux. It has no effect on Windows. + /// + /// The size of the requested page can be specified in page bits. If not provided, the system + /// default is requested. The requested length should be a multiple of this, or the mapping + /// will fail. + /// + /// This option has no effect on file-backed memory maps. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// + /// # fn main() -> std::io::Result<()> { + /// let stack = MmapOptions::new().huge(Some(21)).len(2*1024*1024).map_anon(); + /// # Ok(()) + /// # } + /// ``` + pub fn huge(&mut self, page_bits: Option<u8>) -> &mut Self { + self.huge = Some(page_bits.unwrap_or(0)); + self + } + /// Populate (prefault) page tables for a mapping. + /// + /// For a file mapping, this causes read-ahead on the file. This will help to reduce blocking on page faults later. + /// + /// This option corresponds to the `MAP_POPULATE` flag on Linux. It has no effect on Windows. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// use std::fs::File; + /// + /// # fn main() -> std::io::Result<()> { + /// let file = File::open("LICENSE-MIT")?; + /// + /// let mmap = unsafe { + /// MmapOptions::new().populate().map(&file)? + /// }; + /// + /// assert_eq!(&b"Copyright"[..], &mmap[..9]); + /// # Ok(()) + /// # } + /// ``` + pub fn populate(&mut self) -> &mut Self { + self.populate = true; + self + } + + /// Creates a read-only memory map backed by a file. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read permissions. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// use std::fs::File; + /// use std::io::Read; + /// + /// # fn main() -> std::io::Result<()> { + /// let mut file = File::open("LICENSE-APACHE")?; + /// + /// let mut contents = Vec::new(); + /// file.read_to_end(&mut contents)?; + /// + /// let mmap = unsafe { + /// MmapOptions::new().map(&file)? + /// }; + /// + /// assert_eq!(&contents[..], &mmap[..]); + /// # Ok(()) + /// # } + /// ``` + pub unsafe fn map<T: MmapAsRawDesc>(&self, file: T) -> Result<Mmap> { + let desc = file.as_raw_desc(); + + MmapInner::map(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| Mmap { inner }) + } + + /// Creates a readable and executable memory map backed by a file. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read permissions. + pub unsafe fn map_exec<T: MmapAsRawDesc>(&self, file: T) -> Result<Mmap> { + let desc = file.as_raw_desc(); + + MmapInner::map_exec(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| Mmap { inner }) + } + + /// Creates a writeable memory map backed by a file. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read and write permissions. + /// + /// # Example + /// + /// ``` + /// # extern crate memmap2; + /// # extern crate tempfile; + /// # + /// use std::fs::OpenOptions; + /// use std::path::PathBuf; + /// + /// use memmap2::MmapOptions; + /// # + /// # fn main() -> std::io::Result<()> { + /// # let tempdir = tempfile::tempdir()?; + /// let path: PathBuf = /* path to file */ + /// # tempdir.path().join("map_mut"); + /// let file = OpenOptions::new().read(true).write(true).create(true).open(&path)?; + /// file.set_len(13)?; + /// + /// let mut mmap = unsafe { + /// MmapOptions::new().map_mut(&file)? + /// }; + /// + /// mmap.copy_from_slice(b"Hello, world!"); + /// # Ok(()) + /// # } + /// ``` + pub unsafe fn map_mut<T: MmapAsRawDesc>(&self, file: T) -> Result<MmapMut> { + let desc = file.as_raw_desc(); + + MmapInner::map_mut(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| MmapMut { inner }) + } + + /// Creates a copy-on-write memory map backed by a file. + /// + /// Data written to the memory map will not be visible by other processes, + /// and will not be carried through to the underlying file. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with writable permissions. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// use std::fs::File; + /// use std::io::Write; + /// + /// # fn main() -> std::io::Result<()> { + /// let file = File::open("LICENSE-APACHE")?; + /// let mut mmap = unsafe { MmapOptions::new().map_copy(&file)? }; + /// (&mut mmap[..]).write_all(b"Hello, world!")?; + /// # Ok(()) + /// # } + /// ``` + pub unsafe fn map_copy<T: MmapAsRawDesc>(&self, file: T) -> Result<MmapMut> { + let desc = file.as_raw_desc(); + + MmapInner::map_copy(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| MmapMut { inner }) + } + + /// Creates a copy-on-write read-only memory map backed by a file. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read permissions. + /// + /// # Example + /// + /// ``` + /// use memmap2::MmapOptions; + /// use std::fs::File; + /// use std::io::Read; + /// + /// # fn main() -> std::io::Result<()> { + /// let mut file = File::open("README.md")?; + /// + /// let mut contents = Vec::new(); + /// file.read_to_end(&mut contents)?; + /// + /// let mmap = unsafe { + /// MmapOptions::new().map_copy_read_only(&file)? + /// }; + /// + /// assert_eq!(&contents[..], &mmap[..]); + /// # Ok(()) + /// # } + /// ``` + pub unsafe fn map_copy_read_only<T: MmapAsRawDesc>(&self, file: T) -> Result<Mmap> { + let desc = file.as_raw_desc(); + + MmapInner::map_copy_read_only(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| Mmap { inner }) + } + + /// Creates an anonymous memory map. + /// + /// The memory map length should be configured using [`MmapOptions::len()`] + /// before creating an anonymous memory map, otherwise a zero-length mapping + /// will be crated. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails or + /// when `len > isize::MAX`. + pub fn map_anon(&self) -> Result<MmapMut> { + let len = self.len.unwrap_or(0); + + // See get_len() for details. + if mem::size_of::<usize>() < 8 && len > isize::MAX as usize { + return Err(Error::new( + ErrorKind::InvalidData, + "memory map length overflows isize", + )); + } + + MmapInner::map_anon(len, self.stack, self.populate, self.huge) + .map(|inner| MmapMut { inner }) + } + + /// Creates a raw memory map. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read and write permissions. + pub fn map_raw<T: MmapAsRawDesc>(&self, file: T) -> Result<MmapRaw> { + let desc = file.as_raw_desc(); + + MmapInner::map_mut(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| MmapRaw { inner }) + } + + /// Creates a read-only raw memory map + /// + /// This is primarily useful to avoid intermediate `Mmap` instances when + /// read-only access to files modified elsewhere are required. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails + pub fn map_raw_read_only<T: MmapAsRawDesc>(&self, file: T) -> Result<MmapRaw> { + let desc = file.as_raw_desc(); + + MmapInner::map(self.get_len(&file)?, desc.0, self.offset, self.populate) + .map(|inner| MmapRaw { inner }) + } +} + +/// A handle to an immutable memory mapped buffer. +/// +/// A `Mmap` may be backed by a file, or it can be anonymous map, backed by volatile memory. Use +/// [`MmapOptions`] or [`map()`] to create a file-backed memory map. To create an immutable +/// anonymous memory map, first create a mutable anonymous memory map, and then make it immutable +/// with [`MmapMut::make_read_only()`]. +/// +/// A file backed `Mmap` is created by `&File` reference, and will remain valid even after the +/// `File` is dropped. In other words, the `Mmap` handle is completely independent of the `File` +/// used to create it. For consistency, on some platforms this is achieved by duplicating the +/// underlying file handle. The memory will be unmapped when the `Mmap` handle is dropped. +/// +/// Dereferencing and accessing the bytes of the buffer may result in page faults (e.g. swapping +/// the mapped pages into physical memory) though the details of this are platform specific. +/// +/// `Mmap` is [`Sync`] and [`Send`]. +/// +/// ## Safety +/// +/// All file-backed memory map constructors are marked `unsafe` because of the potential for +/// *Undefined Behavior* (UB) using the map if the underlying file is subsequently modified, in or +/// out of process. Applications must consider the risk and take appropriate precautions when using +/// file-backed maps. Solutions such as file permissions, locks or process-private (e.g. unlinked) +/// files exist but are platform specific and limited. +/// +/// ## Example +/// +/// ``` +/// use memmap2::MmapOptions; +/// use std::io::Write; +/// use std::fs::File; +/// +/// # fn main() -> std::io::Result<()> { +/// let file = File::open("README.md")?; +/// let mmap = unsafe { MmapOptions::new().map(&file)? }; +/// assert_eq!(b"# memmap2", &mmap[0..9]); +/// # Ok(()) +/// # } +/// ``` +/// +/// See [`MmapMut`] for the mutable version. +/// +/// [`map()`]: Mmap::map() +pub struct Mmap { + inner: MmapInner, +} + +impl Mmap { + /// Creates a read-only memory map backed by a file. + /// + /// This is equivalent to calling `MmapOptions::new().map(file)`. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read permissions. + /// + /// # Example + /// + /// ``` + /// use std::fs::File; + /// use std::io::Read; + /// + /// use memmap2::Mmap; + /// + /// # fn main() -> std::io::Result<()> { + /// let mut file = File::open("LICENSE-APACHE")?; + /// + /// let mut contents = Vec::new(); + /// file.read_to_end(&mut contents)?; + /// + /// let mmap = unsafe { Mmap::map(&file)? }; + /// + /// assert_eq!(&contents[..], &mmap[..]); + /// # Ok(()) + /// # } + /// ``` + pub unsafe fn map<T: MmapAsRawDesc>(file: T) -> Result<Mmap> { + MmapOptions::new().map(file) + } + + /// Transition the memory map to be writable. + /// + /// If the memory map is file-backed, the file must have been opened with write permissions. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with writable permissions. + /// + /// # Example + /// + /// ``` + /// # extern crate memmap2; + /// # extern crate tempfile; + /// # + /// use memmap2::Mmap; + /// use std::ops::DerefMut; + /// use std::io::Write; + /// # use std::fs::OpenOptions; + /// + /// # fn main() -> std::io::Result<()> { + /// # let tempdir = tempfile::tempdir()?; + /// let file = /* file opened with write permissions */ + /// # OpenOptions::new() + /// # .read(true) + /// # .write(true) + /// # .create(true) + /// # .open(tempdir.path() + /// # .join("make_mut"))?; + /// # file.set_len(128)?; + /// let mmap = unsafe { Mmap::map(&file)? }; + /// // ... use the read-only memory map ... + /// let mut mut_mmap = mmap.make_mut()?; + /// mut_mmap.deref_mut().write_all(b"hello, world!")?; + /// # Ok(()) + /// # } + /// ``` + pub fn make_mut(mut self) -> Result<MmapMut> { + self.inner.make_mut()?; + Ok(MmapMut { inner: self.inner }) + } + + /// Advise OS how this memory map will be accessed. + /// + /// Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn advise(&self, advice: Advice) -> Result<()> { + self.inner + .advise(advice as libc::c_int, 0, self.inner.len()) + } + + /// Advise OS how this memory map will be accessed. + /// + /// Used with the [unchecked flags][UncheckedAdvice]. Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub unsafe fn unchecked_advise(&self, advice: UncheckedAdvice) -> Result<()> { + self.inner + .advise(advice as libc::c_int, 0, self.inner.len()) + } + + /// Advise OS how this range of memory map will be accessed. + /// + /// Only supported on Unix. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn advise_range(&self, advice: Advice, offset: usize, len: usize) -> Result<()> { + self.inner.advise(advice as libc::c_int, offset, len) + } + + /// Advise OS how this range of memory map will be accessed. + /// + /// Used with the [unchecked flags][UncheckedAdvice]. Only supported on Unix. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub unsafe fn unchecked_advise_range( + &self, + advice: UncheckedAdvice, + offset: usize, + len: usize, + ) -> Result<()> { + self.inner.advise(advice as libc::c_int, offset, len) + } + + /// Lock the whole memory map into RAM. Only supported on Unix. + /// + /// See [mlock()](https://man7.org/linux/man-pages/man2/mlock.2.html) map page. + #[cfg(unix)] + pub fn lock(&self) -> Result<()> { + self.inner.lock() + } + + /// Unlock the whole memory map. Only supported on Unix. + /// + /// See [munlock()](https://man7.org/linux/man-pages/man2/munlock.2.html) map page. + #[cfg(unix)] + pub fn unlock(&self) -> Result<()> { + self.inner.unlock() + } + + /// Adjust the size of the memory mapping. + /// + /// This will try to resize the memory mapping in place. If + /// [`RemapOptions::may_move`] is specified it will move the mapping if it + /// could not resize in place, otherwise it will error. + /// + /// Only supported on Linux. + /// + /// See the [`mremap(2)`] man page. + /// + /// # Safety + /// + /// Resizing the memory mapping beyond the end of the mapped file will + /// result in UB should you happen to access memory beyond the end of the + /// file. + /// + /// [`mremap(2)`]: https://man7.org/linux/man-pages/man2/mremap.2.html + #[cfg(target_os = "linux")] + pub unsafe fn remap(&mut self, new_len: usize, options: RemapOptions) -> Result<()> { + self.inner.remap(new_len, options) + } +} + +#[cfg(feature = "stable_deref_trait")] +unsafe impl stable_deref_trait::StableDeref for Mmap {} + +impl Deref for Mmap { + type Target = [u8]; + + #[inline] + fn deref(&self) -> &[u8] { + unsafe { slice::from_raw_parts(self.inner.ptr(), self.inner.len()) } + } +} + +impl AsRef<[u8]> for Mmap { + #[inline] + fn as_ref(&self) -> &[u8] { + self.deref() + } +} + +impl fmt::Debug for Mmap { + fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result { + fmt.debug_struct("Mmap") + .field("ptr", &self.as_ptr()) + .field("len", &self.len()) + .finish() + } +} + +/// A handle to a raw memory mapped buffer. +/// +/// This struct never hands out references to its interior, only raw pointers. +/// This can be helpful when creating shared memory maps between untrusted processes. +pub struct MmapRaw { + inner: MmapInner, +} + +impl MmapRaw { + /// Creates a writeable memory map backed by a file. + /// + /// This is equivalent to calling `MmapOptions::new().map_raw(file)`. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read and write permissions. + pub fn map_raw<T: MmapAsRawDesc>(file: T) -> Result<MmapRaw> { + MmapOptions::new().map_raw(file) + } + + /// Returns a raw pointer to the memory mapped file. + /// + /// Before dereferencing this pointer, you have to make sure that the file has not been + /// truncated since the memory map was created. + /// Avoiding this will not introduce memory safety issues in Rust terms, + /// but will cause SIGBUS (or equivalent) signal. + #[inline] + pub fn as_ptr(&self) -> *const u8 { + self.inner.ptr() + } + + /// Returns an unsafe mutable pointer to the memory mapped file. + /// + /// Before dereferencing this pointer, you have to make sure that the file has not been + /// truncated since the memory map was created. + /// Avoiding this will not introduce memory safety issues in Rust terms, + /// but will cause SIGBUS (or equivalent) signal. + #[inline] + pub fn as_mut_ptr(&self) -> *mut u8 { + self.inner.ptr() as _ + } + + /// Returns the length in bytes of the memory map. + /// + /// Note that truncating the file can cause the length to change (and render this value unusable). + #[inline] + pub fn len(&self) -> usize { + self.inner.len() + } + + /// Flushes outstanding memory map modifications to disk. + /// + /// When this method returns with a non-error result, all outstanding changes to a file-backed + /// memory map are guaranteed to be durably stored. The file's metadata (including last + /// modification timestamp) may not be updated. + /// + /// # Example + /// + /// ``` + /// # extern crate memmap2; + /// # extern crate tempfile; + /// # + /// use std::fs::OpenOptions; + /// use std::io::Write; + /// use std::path::PathBuf; + /// use std::slice; + /// + /// use memmap2::MmapRaw; + /// + /// # fn main() -> std::io::Result<()> { + /// let tempdir = tempfile::tempdir()?; + /// let path: PathBuf = /* path to file */ + /// # tempdir.path().join("flush"); + /// let file = OpenOptions::new().read(true).write(true).create(true).open(&path)?; + /// file.set_len(128)?; + /// + /// let mut mmap = unsafe { MmapRaw::map_raw(&file)? }; + /// + /// let mut memory = unsafe { slice::from_raw_parts_mut(mmap.as_mut_ptr(), 128) }; + /// memory.write_all(b"Hello, world!")?; + /// mmap.flush()?; + /// # Ok(()) + /// # } + /// ``` + pub fn flush(&self) -> Result<()> { + let len = self.len(); + self.inner.flush(0, len) + } + + /// Asynchronously flushes outstanding memory map modifications to disk. + /// + /// This method initiates flushing modified pages to durable storage, but it will not wait for + /// the operation to complete before returning. The file's metadata (including last + /// modification timestamp) may not be updated. + pub fn flush_async(&self) -> Result<()> { + let len = self.len(); + self.inner.flush_async(0, len) + } + + /// Flushes outstanding memory map modifications in the range to disk. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// When this method returns with a non-error result, all outstanding changes to a file-backed + /// memory in the range are guaranteed to be durable stored. The file's metadata (including + /// last modification timestamp) may not be updated. It is not guaranteed the only the changes + /// in the specified range are flushed; other outstanding changes to the memory map may be + /// flushed as well. + pub fn flush_range(&self, offset: usize, len: usize) -> Result<()> { + self.inner.flush(offset, len) + } + + /// Asynchronously flushes outstanding memory map modifications in the range to disk. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// This method initiates flushing modified pages to durable storage, but it will not wait for + /// the operation to complete before returning. The file's metadata (including last + /// modification timestamp) may not be updated. It is not guaranteed that the only changes + /// flushed are those in the specified range; other outstanding changes to the memory map may + /// be flushed as well. + pub fn flush_async_range(&self, offset: usize, len: usize) -> Result<()> { + self.inner.flush_async(offset, len) + } + + /// Advise OS how this memory map will be accessed. + /// + /// Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn advise(&self, advice: Advice) -> Result<()> { + self.inner + .advise(advice as libc::c_int, 0, self.inner.len()) + } + + /// Advise OS how this memory map will be accessed. + /// + /// Used with the [unchecked flags][UncheckedAdvice]. Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub unsafe fn unchecked_advise(&self, advice: UncheckedAdvice) -> Result<()> { + self.inner + .advise(advice as libc::c_int, 0, self.inner.len()) + } + + /// Advise OS how this range of memory map will be accessed. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn advise_range(&self, advice: Advice, offset: usize, len: usize) -> Result<()> { + self.inner.advise(advice as libc::c_int, offset, len) + } + + /// Advise OS how this range of memory map will be accessed. + /// + /// Used with the [unchecked flags][UncheckedAdvice]. Only supported on Unix. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub unsafe fn unchecked_advise_range( + &self, + advice: UncheckedAdvice, + offset: usize, + len: usize, + ) -> Result<()> { + self.inner.advise(advice as libc::c_int, offset, len) + } + + /// Lock the whole memory map into RAM. Only supported on Unix. + /// + /// See [mlock()](https://man7.org/linux/man-pages/man2/mlock.2.html) map page. + #[cfg(unix)] + pub fn lock(&self) -> Result<()> { + self.inner.lock() + } + + /// Unlock the whole memory map. Only supported on Unix. + /// + /// See [munlock()](https://man7.org/linux/man-pages/man2/munlock.2.html) map page. + #[cfg(unix)] + pub fn unlock(&self) -> Result<()> { + self.inner.unlock() + } + + /// Adjust the size of the memory mapping. + /// + /// This will try to resize the memory mapping in place. If + /// [`RemapOptions::may_move`] is specified it will move the mapping if it + /// could not resize in place, otherwise it will error. + /// + /// Only supported on Linux. + /// + /// See the [`mremap(2)`] man page. + /// + /// # Safety + /// + /// Resizing the memory mapping beyond the end of the mapped file will + /// result in UB should you happen to access memory beyond the end of the + /// file. + /// + /// [`mremap(2)`]: https://man7.org/linux/man-pages/man2/mremap.2.html + #[cfg(target_os = "linux")] + pub unsafe fn remap(&mut self, new_len: usize, options: RemapOptions) -> Result<()> { + self.inner.remap(new_len, options) + } +} + +impl fmt::Debug for MmapRaw { + fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result { + fmt.debug_struct("MmapRaw") + .field("ptr", &self.as_ptr()) + .field("len", &self.len()) + .finish() + } +} + +impl From<Mmap> for MmapRaw { + fn from(value: Mmap) -> Self { + Self { inner: value.inner } + } +} + +impl From<MmapMut> for MmapRaw { + fn from(value: MmapMut) -> Self { + Self { inner: value.inner } + } +} + +/// A handle to a mutable memory mapped buffer. +/// +/// A file-backed `MmapMut` buffer may be used to read from or write to a file. An anonymous +/// `MmapMut` buffer may be used any place that an in-memory byte buffer is needed. Use +/// [`MmapMut::map_mut()`] and [`MmapMut::map_anon()`] to create a mutable memory map of the +/// respective types, or [`MmapOptions::map_mut()`] and [`MmapOptions::map_anon()`] if non-default +/// options are required. +/// +/// A file backed `MmapMut` is created by `&File` reference, and will remain valid even after the +/// `File` is dropped. In other words, the `MmapMut` handle is completely independent of the `File` +/// used to create it. For consistency, on some platforms this is achieved by duplicating the +/// underlying file handle. The memory will be unmapped when the `MmapMut` handle is dropped. +/// +/// Dereferencing and accessing the bytes of the buffer may result in page faults (e.g. swapping +/// the mapped pages into physical memory) though the details of this are platform specific. +/// +/// `Mmap` is [`Sync`] and [`Send`]. +/// +/// See [`Mmap`] for the immutable version. +/// +/// ## Safety +/// +/// All file-backed memory map constructors are marked `unsafe` because of the potential for +/// *Undefined Behavior* (UB) using the map if the underlying file is subsequently modified, in or +/// out of process. Applications must consider the risk and take appropriate precautions when using +/// file-backed maps. Solutions such as file permissions, locks or process-private (e.g. unlinked) +/// files exist but are platform specific and limited. +pub struct MmapMut { + inner: MmapInner, +} + +impl MmapMut { + /// Creates a writeable memory map backed by a file. + /// + /// This is equivalent to calling `MmapOptions::new().map_mut(file)`. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file is not open with read and write permissions. + /// + /// # Example + /// + /// ``` + /// # extern crate memmap2; + /// # extern crate tempfile; + /// # + /// use std::fs::OpenOptions; + /// use std::path::PathBuf; + /// + /// use memmap2::MmapMut; + /// # + /// # fn main() -> std::io::Result<()> { + /// # let tempdir = tempfile::tempdir()?; + /// let path: PathBuf = /* path to file */ + /// # tempdir.path().join("map_mut"); + /// let file = OpenOptions::new() + /// .read(true) + /// .write(true) + /// .create(true) + /// .open(&path)?; + /// file.set_len(13)?; + /// + /// let mut mmap = unsafe { MmapMut::map_mut(&file)? }; + /// + /// mmap.copy_from_slice(b"Hello, world!"); + /// # Ok(()) + /// # } + /// ``` + pub unsafe fn map_mut<T: MmapAsRawDesc>(file: T) -> Result<MmapMut> { + MmapOptions::new().map_mut(file) + } + + /// Creates an anonymous memory map. + /// + /// This is equivalent to calling `MmapOptions::new().len(length).map_anon()`. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails or + /// when `len > isize::MAX`. + pub fn map_anon(length: usize) -> Result<MmapMut> { + MmapOptions::new().len(length).map_anon() + } + + /// Flushes outstanding memory map modifications to disk. + /// + /// When this method returns with a non-error result, all outstanding changes to a file-backed + /// memory map are guaranteed to be durably stored. The file's metadata (including last + /// modification timestamp) may not be updated. + /// + /// # Example + /// + /// ``` + /// # extern crate memmap2; + /// # extern crate tempfile; + /// # + /// use std::fs::OpenOptions; + /// use std::io::Write; + /// use std::path::PathBuf; + /// + /// use memmap2::MmapMut; + /// + /// # fn main() -> std::io::Result<()> { + /// # let tempdir = tempfile::tempdir()?; + /// let path: PathBuf = /* path to file */ + /// # tempdir.path().join("flush"); + /// let file = OpenOptions::new().read(true).write(true).create(true).open(&path)?; + /// file.set_len(128)?; + /// + /// let mut mmap = unsafe { MmapMut::map_mut(&file)? }; + /// + /// (&mut mmap[..]).write_all(b"Hello, world!")?; + /// mmap.flush()?; + /// # Ok(()) + /// # } + /// ``` + pub fn flush(&self) -> Result<()> { + let len = self.len(); + self.inner.flush(0, len) + } + + /// Asynchronously flushes outstanding memory map modifications to disk. + /// + /// This method initiates flushing modified pages to durable storage, but it will not wait for + /// the operation to complete before returning. The file's metadata (including last + /// modification timestamp) may not be updated. + pub fn flush_async(&self) -> Result<()> { + let len = self.len(); + self.inner.flush_async(0, len) + } + + /// Flushes outstanding memory map modifications in the range to disk. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// When this method returns with a non-error result, all outstanding changes to a file-backed + /// memory in the range are guaranteed to be durable stored. The file's metadata (including + /// last modification timestamp) may not be updated. It is not guaranteed the only the changes + /// in the specified range are flushed; other outstanding changes to the memory map may be + /// flushed as well. + pub fn flush_range(&self, offset: usize, len: usize) -> Result<()> { + self.inner.flush(offset, len) + } + + /// Asynchronously flushes outstanding memory map modifications in the range to disk. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// This method initiates flushing modified pages to durable storage, but it will not wait for + /// the operation to complete before returning. The file's metadata (including last + /// modification timestamp) may not be updated. It is not guaranteed that the only changes + /// flushed are those in the specified range; other outstanding changes to the memory map may + /// be flushed as well. + pub fn flush_async_range(&self, offset: usize, len: usize) -> Result<()> { + self.inner.flush_async(offset, len) + } + + /// Returns an immutable version of this memory mapped buffer. + /// + /// If the memory map is file-backed, the file must have been opened with read permissions. + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file has not been opened with read permissions. + /// + /// # Example + /// + /// ``` + /// # extern crate memmap2; + /// # + /// use std::io::Write; + /// use std::path::PathBuf; + /// + /// use memmap2::{Mmap, MmapMut}; + /// + /// # fn main() -> std::io::Result<()> { + /// let mut mmap = MmapMut::map_anon(128)?; + /// + /// (&mut mmap[..]).write(b"Hello, world!")?; + /// + /// let mmap: Mmap = mmap.make_read_only()?; + /// # Ok(()) + /// # } + /// ``` + pub fn make_read_only(mut self) -> Result<Mmap> { + self.inner.make_read_only()?; + Ok(Mmap { inner: self.inner }) + } + + /// Transition the memory map to be readable and executable. + /// + /// If the memory map is file-backed, the file must have been opened with execute permissions. + /// + /// On systems with separate instructions and data caches (a category that includes many ARM + /// chips), a platform-specific call may be needed to ensure that the changes are visible to the + /// execution unit (e.g. when using this function to implement a JIT compiler). For more + /// details, see [this ARM write-up](https://community.arm.com/arm-community-blogs/b/architectures-and-processors-blog/posts/caches-and-self-modifying-code) + /// or the `man` page for [`sys_icache_invalidate`](https://developer.apple.com/library/archive/documentation/System/Conceptual/ManPages_iPhoneOS/man3/sys_icache_invalidate.3.html). + /// + /// # Errors + /// + /// This method returns an error when the underlying system call fails, which can happen for a + /// variety of reasons, such as when the file has not been opened with execute permissions. + pub fn make_exec(mut self) -> Result<Mmap> { + self.inner.make_exec()?; + Ok(Mmap { inner: self.inner }) + } + + /// Advise OS how this memory map will be accessed. + /// + /// Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn advise(&self, advice: Advice) -> Result<()> { + self.inner + .advise(advice as libc::c_int, 0, self.inner.len()) + } + + /// Advise OS how this memory map will be accessed. + /// + /// Used with the [unchecked flags][UncheckedAdvice]. Only supported on Unix. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub unsafe fn unchecked_advise(&self, advice: UncheckedAdvice) -> Result<()> { + self.inner + .advise(advice as libc::c_int, 0, self.inner.len()) + } + + /// Advise OS how this range of memory map will be accessed. + /// + /// Only supported on Unix. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn advise_range(&self, advice: Advice, offset: usize, len: usize) -> Result<()> { + self.inner.advise(advice as libc::c_int, offset, len) + } + + /// Advise OS how this range of memory map will be accessed. + /// + /// Used with the [unchecked flags][UncheckedAdvice]. Only supported on Unix. + /// + /// The offset and length must be in the bounds of the memory map. + /// + /// See [madvise()](https://man7.org/linux/man-pages/man2/madvise.2.html) map page. + #[cfg(unix)] + pub fn unchecked_advise_range( + &self, + advice: UncheckedAdvice, + offset: usize, + len: usize, + ) -> Result<()> { + self.inner.advise(advice as libc::c_int, offset, len) + } + + /// Lock the whole memory map into RAM. Only supported on Unix. + /// + /// See [mlock()](https://man7.org/linux/man-pages/man2/mlock.2.html) map page. + #[cfg(unix)] + pub fn lock(&self) -> Result<()> { + self.inner.lock() + } + + /// Unlock the whole memory map. Only supported on Unix. + /// + /// See [munlock()](https://man7.org/linux/man-pages/man2/munlock.2.html) map page. + #[cfg(unix)] + pub fn unlock(&self) -> Result<()> { + self.inner.unlock() + } + + /// Adjust the size of the memory mapping. + /// + /// This will try to resize the memory mapping in place. If + /// [`RemapOptions::may_move`] is specified it will move the mapping if it + /// could not resize in place, otherwise it will error. + /// + /// Only supported on Linux. + /// + /// See the [`mremap(2)`] man page. + /// + /// # Safety + /// + /// Resizing the memory mapping beyond the end of the mapped file will + /// result in UB should you happen to access memory beyond the end of the + /// file. + /// + /// [`mremap(2)`]: https://man7.org/linux/man-pages/man2/mremap.2.html + #[cfg(target_os = "linux")] + pub unsafe fn remap(&mut self, new_len: usize, options: RemapOptions) -> Result<()> { + self.inner.remap(new_len, options) + } +} + +#[cfg(feature = "stable_deref_trait")] +unsafe impl stable_deref_trait::StableDeref for MmapMut {} + +impl Deref for MmapMut { + type Target = [u8]; + + #[inline] + fn deref(&self) -> &[u8] { + unsafe { slice::from_raw_parts(self.inner.ptr(), self.inner.len()) } + } +} + +impl DerefMut for MmapMut { + #[inline] + fn deref_mut(&mut self) -> &mut [u8] { + unsafe { slice::from_raw_parts_mut(self.inner.mut_ptr(), self.inner.len()) } + } +} + +impl AsRef<[u8]> for MmapMut { + #[inline] + fn as_ref(&self) -> &[u8] { + self.deref() + } +} + +impl AsMut<[u8]> for MmapMut { + #[inline] + fn as_mut(&mut self) -> &mut [u8] { + self.deref_mut() + } +} + +impl fmt::Debug for MmapMut { + fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result { + fmt.debug_struct("MmapMut") + .field("ptr", &self.as_ptr()) + .field("len", &self.len()) + .finish() + } +} + +/// Options for [`Mmap::remap`] and [`MmapMut::remap`]. +#[derive(Copy, Clone, Default, Debug)] +#[cfg(target_os = "linux")] +pub struct RemapOptions { + may_move: bool, +} + +#[cfg(target_os = "linux")] +impl RemapOptions { + /// Creates a mew set of options for resizing a memory map. + pub fn new() -> Self { + Self::default() + } + + /// Controls whether the memory map can be moved if it is not possible to + /// resize it in place. + /// + /// If false then the memory map is guaranteed to remain at the same + /// address when being resized but attempting to resize will return an + /// error if the new memory map would overlap with something else in the + /// current process' memory. + /// + /// By default this is false. + /// + /// # `may_move` and `StableDeref` + /// If the `stable_deref_trait` feature is enabled then [`Mmap`] and + /// [`MmapMut`] implement `StableDeref`. `StableDeref` promises that the + /// memory map dereferences to a fixed address, however, calling `remap` + /// with `may_move` set may result in the backing memory of the mapping + /// being moved to a new address. This may cause UB in other code + /// depending on the `StableDeref` guarantees. + pub fn may_move(mut self, may_move: bool) -> Self { + self.may_move = may_move; + self + } + + pub(crate) fn into_flags(self) -> libc::c_int { + if self.may_move { + libc::MREMAP_MAYMOVE + } else { + 0 + } + } +} + +#[cfg(test)] +mod test { + extern crate tempfile; + + #[cfg(unix)] + use crate::advice::Advice; + use std::fs::{File, OpenOptions}; + use std::io::{Read, Write}; + use std::mem; + #[cfg(unix)] + use std::os::unix::io::AsRawFd; + #[cfg(windows)] + use std::os::windows::fs::OpenOptionsExt; + + #[cfg(windows)] + const GENERIC_ALL: u32 = 0x10000000; + + use super::{Mmap, MmapMut, MmapOptions}; + + #[test] + fn map_file() { + let expected_len = 128; + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + + file.set_len(expected_len as u64).unwrap(); + + let mut mmap = unsafe { MmapMut::map_mut(&file).unwrap() }; + let len = mmap.len(); + assert_eq!(expected_len, len); + + let zeros = vec![0; len]; + let incr: Vec<u8> = (0..len as u8).collect(); + + // check that the mmap is empty + assert_eq!(&zeros[..], &mmap[..]); + + // write values into the mmap + (&mut mmap[..]).write_all(&incr[..]).unwrap(); + + // read values back + assert_eq!(&incr[..], &mmap[..]); + } + + #[test] + #[cfg(unix)] + fn map_fd() { + let expected_len = 128; + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + + file.set_len(expected_len as u64).unwrap(); + + let mut mmap = unsafe { MmapMut::map_mut(file.as_raw_fd()).unwrap() }; + let len = mmap.len(); + assert_eq!(expected_len, len); + + let zeros = vec![0; len]; + let incr: Vec<u8> = (0..len as u8).collect(); + + // check that the mmap is empty + assert_eq!(&zeros[..], &mmap[..]); + + // write values into the mmap + (&mut mmap[..]).write_all(&incr[..]).unwrap(); + + // read values back + assert_eq!(&incr[..], &mmap[..]); + } + + /// Checks that "mapping" a 0-length file derefs to an empty slice. + #[test] + fn map_empty_file() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + let mmap = unsafe { Mmap::map(&file).unwrap() }; + assert!(mmap.is_empty()); + assert_eq!(mmap.as_ptr().align_offset(mem::size_of::<usize>()), 0); + let mmap = unsafe { MmapMut::map_mut(&file).unwrap() }; + assert!(mmap.is_empty()); + assert_eq!(mmap.as_ptr().align_offset(mem::size_of::<usize>()), 0); + } + + #[test] + fn map_anon() { + let expected_len = 128; + let mut mmap = MmapMut::map_anon(expected_len).unwrap(); + let len = mmap.len(); + assert_eq!(expected_len, len); + + let zeros = vec![0; len]; + let incr: Vec<u8> = (0..len as u8).collect(); + + // check that the mmap is empty + assert_eq!(&zeros[..], &mmap[..]); + + // write values into the mmap + (&mut mmap[..]).write_all(&incr[..]).unwrap(); + + // read values back + assert_eq!(&incr[..], &mmap[..]); + } + + #[test] + fn map_anon_zero_len() { + assert!(MmapOptions::new().map_anon().unwrap().is_empty()) + } + + #[test] + #[cfg(target_pointer_width = "32")] + fn map_anon_len_overflow() { + let res = MmapMut::map_anon(0x80000000); + + assert_eq!( + res.unwrap_err().to_string(), + "memory map length overflows isize" + ); + } + + #[test] + fn file_write() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let mut file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + file.set_len(128).unwrap(); + + let write = b"abc123"; + let mut read = [0u8; 6]; + + let mut mmap = unsafe { MmapMut::map_mut(&file).unwrap() }; + (&mut mmap[..]).write_all(write).unwrap(); + mmap.flush().unwrap(); + + file.read_exact(&mut read).unwrap(); + assert_eq!(write, &read); + } + + #[test] + fn flush_range() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + file.set_len(128).unwrap(); + let write = b"abc123"; + + let mut mmap = unsafe { + MmapOptions::new() + .offset(2) + .len(write.len()) + .map_mut(&file) + .unwrap() + }; + (&mut mmap[..]).write_all(write).unwrap(); + mmap.flush_async_range(0, write.len()).unwrap(); + mmap.flush_range(0, write.len()).unwrap(); + } + + #[test] + fn map_copy() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let mut file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + file.set_len(128).unwrap(); + + let nulls = b"\0\0\0\0\0\0"; + let write = b"abc123"; + let mut read = [0u8; 6]; + + let mut mmap = unsafe { MmapOptions::new().map_copy(&file).unwrap() }; + + (&mut mmap[..]).write_all(write).unwrap(); + mmap.flush().unwrap(); + + // The mmap contains the write + (&mmap[..]).read_exact(&mut read).unwrap(); + assert_eq!(write, &read); + + // The file does not contain the write + file.read_exact(&mut read).unwrap(); + assert_eq!(nulls, &read); + + // another mmap does not contain the write + let mmap2 = unsafe { MmapOptions::new().map(&file).unwrap() }; + (&mmap2[..]).read_exact(&mut read).unwrap(); + assert_eq!(nulls, &read); + } + + #[test] + fn map_copy_read_only() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + file.set_len(128).unwrap(); + + let nulls = b"\0\0\0\0\0\0"; + let mut read = [0u8; 6]; + + let mmap = unsafe { MmapOptions::new().map_copy_read_only(&file).unwrap() }; + (&mmap[..]).read_exact(&mut read).unwrap(); + assert_eq!(nulls, &read); + + let mmap2 = unsafe { MmapOptions::new().map(&file).unwrap() }; + (&mmap2[..]).read_exact(&mut read).unwrap(); + assert_eq!(nulls, &read); + } + + #[test] + fn map_offset() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + + let offset = u32::MAX as u64 + 2; + let len = 5432; + file.set_len(offset + len as u64).unwrap(); + + // Check inferred length mmap. + let mmap = unsafe { MmapOptions::new().offset(offset).map_mut(&file).unwrap() }; + assert_eq!(len, mmap.len()); + + // Check explicit length mmap. + let mut mmap = unsafe { + MmapOptions::new() + .offset(offset) + .len(len) + .map_mut(&file) + .unwrap() + }; + assert_eq!(len, mmap.len()); + + let zeros = vec![0; len]; + let incr: Vec<_> = (0..len).map(|i| i as u8).collect(); + + // check that the mmap is empty + assert_eq!(&zeros[..], &mmap[..]); + + // write values into the mmap + (&mut mmap[..]).write_all(&incr[..]).unwrap(); + + // read values back + assert_eq!(&incr[..], &mmap[..]); + } + + #[test] + fn index() { + let mut mmap = MmapMut::map_anon(128).unwrap(); + mmap[0] = 42; + assert_eq!(42, mmap[0]); + } + + #[test] + fn sync_send() { + let mmap = MmapMut::map_anon(129).unwrap(); + + fn is_sync_send<T>(_val: T) + where + T: Sync + Send, + { + } + + is_sync_send(mmap); + } + + #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] + fn jit_x86(mut mmap: MmapMut) { + mmap[0] = 0xB8; // mov eax, 0xAB + mmap[1] = 0xAB; + mmap[2] = 0x00; + mmap[3] = 0x00; + mmap[4] = 0x00; + mmap[5] = 0xC3; // ret + + let mmap = mmap.make_exec().expect("make_exec"); + + let jitfn: extern "C" fn() -> u8 = unsafe { mem::transmute(mmap.as_ptr()) }; + assert_eq!(jitfn(), 0xab); + } + + #[test] + #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] + fn jit_x86_anon() { + jit_x86(MmapMut::map_anon(4096).unwrap()); + } + + #[test] + #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] + fn jit_x86_file() { + let tempdir = tempfile::tempdir().unwrap(); + let mut options = OpenOptions::new(); + #[cfg(windows)] + options.access_mode(GENERIC_ALL); + + let file = options + .read(true) + .write(true) + .create(true) + .open(tempdir.path().join("jit_x86")) + .expect("open"); + + file.set_len(4096).expect("set_len"); + jit_x86(unsafe { MmapMut::map_mut(&file).expect("map_mut") }); + } + + #[test] + fn mprotect_file() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let mut options = OpenOptions::new(); + #[cfg(windows)] + options.access_mode(GENERIC_ALL); + + let mut file = options + .read(true) + .write(true) + .create(true) + .open(path) + .expect("open"); + file.set_len(256_u64).expect("set_len"); + + let mmap = unsafe { MmapMut::map_mut(&file).expect("map_mut") }; + + let mmap = mmap.make_read_only().expect("make_read_only"); + let mut mmap = mmap.make_mut().expect("make_mut"); + + let write = b"abc123"; + let mut read = [0u8; 6]; + + (&mut mmap[..]).write_all(write).unwrap(); + mmap.flush().unwrap(); + + // The mmap contains the write + (&mmap[..]).read_exact(&mut read).unwrap(); + assert_eq!(write, &read); + + // The file should contain the write + file.read_exact(&mut read).unwrap(); + assert_eq!(write, &read); + + // another mmap should contain the write + let mmap2 = unsafe { MmapOptions::new().map(&file).unwrap() }; + (&mmap2[..]).read_exact(&mut read).unwrap(); + assert_eq!(write, &read); + + let mmap = mmap.make_exec().expect("make_exec"); + + drop(mmap); + } + + #[test] + fn mprotect_copy() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap"); + + let mut options = OpenOptions::new(); + #[cfg(windows)] + options.access_mode(GENERIC_ALL); + + let mut file = options + .read(true) + .write(true) + .create(true) + .open(path) + .expect("open"); + file.set_len(256_u64).expect("set_len"); + + let mmap = unsafe { MmapOptions::new().map_copy(&file).expect("map_mut") }; + + let mmap = mmap.make_read_only().expect("make_read_only"); + let mut mmap = mmap.make_mut().expect("make_mut"); + + let nulls = b"\0\0\0\0\0\0"; + let write = b"abc123"; + let mut read = [0u8; 6]; + + (&mut mmap[..]).write_all(write).unwrap(); + mmap.flush().unwrap(); + + // The mmap contains the write + (&mmap[..]).read_exact(&mut read).unwrap(); + assert_eq!(write, &read); + + // The file does not contain the write + file.read_exact(&mut read).unwrap(); + assert_eq!(nulls, &read); + + // another mmap does not contain the write + let mmap2 = unsafe { MmapOptions::new().map(&file).unwrap() }; + (&mmap2[..]).read_exact(&mut read).unwrap(); + assert_eq!(nulls, &read); + + let mmap = mmap.make_exec().expect("make_exec"); + + drop(mmap); + } + + #[test] + fn mprotect_anon() { + let mmap = MmapMut::map_anon(256).expect("map_mut"); + + let mmap = mmap.make_read_only().expect("make_read_only"); + let mmap = mmap.make_mut().expect("make_mut"); + let mmap = mmap.make_exec().expect("make_exec"); + drop(mmap); + } + + #[test] + fn raw() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmapraw"); + + let mut options = OpenOptions::new(); + let mut file = options + .read(true) + .write(true) + .create(true) + .open(path) + .expect("open"); + file.write_all(b"abc123").unwrap(); + let mmap = MmapOptions::new().map_raw(&file).unwrap(); + assert_eq!(mmap.len(), 6); + assert!(!mmap.as_ptr().is_null()); + assert_eq!(unsafe { std::ptr::read(mmap.as_ptr()) }, b'a'); + } + + #[test] + fn raw_read_only() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmaprawro"); + + File::create(&path).unwrap().write_all(b"abc123").unwrap(); + + let mmap = MmapOptions::new() + .map_raw_read_only(&File::open(&path).unwrap()) + .unwrap(); + + assert_eq!(mmap.len(), 6); + assert!(!mmap.as_ptr().is_null()); + assert_eq!(unsafe { std::ptr::read(mmap.as_ptr()) }, b'a'); + } + + /// Something that relies on StableDeref + #[test] + #[cfg(feature = "stable_deref_trait")] + fn owning_ref() { + extern crate owning_ref; + + let mut map = MmapMut::map_anon(128).unwrap(); + map[10] = 42; + let owning = owning_ref::OwningRef::new(map); + let sliced = owning.map(|map| &map[10..20]); + assert_eq!(42, sliced[0]); + + let map = sliced.into_owner().make_read_only().unwrap(); + let owning = owning_ref::OwningRef::new(map); + let sliced = owning.map(|map| &map[10..20]); + assert_eq!(42, sliced[0]); + } + + #[test] + #[cfg(unix)] + fn advise() { + let expected_len = 128; + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap_advise"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + + file.set_len(expected_len as u64).unwrap(); + + // Test MmapMut::advise + let mut mmap = unsafe { MmapMut::map_mut(&file).unwrap() }; + mmap.advise(Advice::Random) + .expect("mmap advising should be supported on unix"); + + let len = mmap.len(); + assert_eq!(expected_len, len); + + let zeros = vec![0; len]; + let incr: Vec<u8> = (0..len as u8).collect(); + + // check that the mmap is empty + assert_eq!(&zeros[..], &mmap[..]); + + mmap.advise_range(Advice::Sequential, 0, mmap.len()) + .expect("mmap advising should be supported on unix"); + + // write values into the mmap + (&mut mmap[..]).write_all(&incr[..]).unwrap(); + + // read values back + assert_eq!(&incr[..], &mmap[..]); + + // Set advice and Read from the read-only map + let mmap = unsafe { Mmap::map(&file).unwrap() }; + + mmap.advise(Advice::Random) + .expect("mmap advising should be supported on unix"); + + // read values back + assert_eq!(&incr[..], &mmap[..]); + } + + #[test] + #[cfg(target_os = "linux")] + fn advise_writes_unsafely() { + let page_size = unsafe { libc::sysconf(libc::_SC_PAGESIZE) as usize }; + + let mut mmap = MmapMut::map_anon(page_size).unwrap(); + mmap.as_mut().fill(255); + let mmap = mmap.make_read_only().unwrap(); + + let a = mmap.as_ref()[0]; + unsafe { + mmap.unchecked_advise(crate::UncheckedAdvice::DontNeed) + .unwrap(); + } + let b = mmap.as_ref()[0]; + + assert_eq!(a, 255); + assert_eq!(b, 0); + } + + #[test] + #[cfg(target_os = "linux")] + fn advise_writes_unsafely_to_part_of_map() { + let page_size = unsafe { libc::sysconf(libc::_SC_PAGESIZE) as usize }; + + let mut mmap = MmapMut::map_anon(2 * page_size).unwrap(); + mmap.as_mut().fill(255); + let mmap = mmap.make_read_only().unwrap(); + + let a = mmap.as_ref()[0]; + let b = mmap.as_ref()[page_size]; + unsafe { + mmap.unchecked_advise_range(crate::UncheckedAdvice::DontNeed, page_size, page_size) + .unwrap(); + } + let c = mmap.as_ref()[0]; + let d = mmap.as_ref()[page_size]; + + assert_eq!(a, 255); + assert_eq!(b, 255); + assert_eq!(c, 255); + assert_eq!(d, 0); + } + + /// Returns true if a non-zero amount of memory is locked. + #[cfg(target_os = "linux")] + fn is_locked() -> bool { + let status = &std::fs::read_to_string("/proc/self/status") + .expect("/proc/self/status should be available"); + for line in status.lines() { + if line.starts_with("VmLck:") { + let numbers = line.replace(|c: char| !c.is_ascii_digit(), ""); + return numbers != "0"; + } + } + panic!("cannot get VmLck information") + } + + #[test] + #[cfg(unix)] + fn lock() { + let tempdir = tempfile::tempdir().unwrap(); + let path = tempdir.path().join("mmap_lock"); + + let file = OpenOptions::new() + .read(true) + .write(true) + .create(true) + .open(path) + .unwrap(); + file.set_len(128).unwrap(); + + let mmap = unsafe { Mmap::map(&file).unwrap() }; + #[cfg(target_os = "linux")] + assert!(!is_locked()); + + mmap.lock().expect("mmap lock should be supported on unix"); + #[cfg(target_os = "linux")] + assert!(is_locked()); + + mmap.lock() + .expect("mmap lock again should not cause problems"); + #[cfg(target_os = "linux")] + assert!(is_locked()); + + mmap.unlock() + .expect("mmap unlock should be supported on unix"); + #[cfg(target_os = "linux")] + assert!(!is_locked()); + + mmap.unlock() + .expect("mmap unlock again should not cause problems"); + #[cfg(target_os = "linux")] + assert!(!is_locked()); + } + + #[test] + #[cfg(target_os = "linux")] + fn remap_grow() { + use crate::RemapOptions; + + let initial_len = 128; + let final_len = 2000; + + let zeros = vec![0u8; final_len]; + let incr: Vec<u8> = (0..final_len).map(|v| v as u8).collect(); + + let file = tempfile::tempfile().unwrap(); + file.set_len(final_len as u64).unwrap(); + + let mut mmap = unsafe { MmapOptions::new().len(initial_len).map_mut(&file).unwrap() }; + assert_eq!(mmap.len(), initial_len); + assert_eq!(&mmap[..], &zeros[..initial_len]); + + unsafe { + mmap.remap(final_len, RemapOptions::new().may_move(true)) + .unwrap() + }; + + // The size should have been updated + assert_eq!(mmap.len(), final_len); + + // Should still be all zeros + assert_eq!(&mmap[..], &zeros); + + // Write out to the whole expanded slice. + mmap.copy_from_slice(&incr); + } + + #[test] + #[cfg(target_os = "linux")] + fn remap_shrink() { + use crate::RemapOptions; + + let initial_len = 20000; + let final_len = 400; + + let incr: Vec<u8> = (0..final_len).map(|v| v as u8).collect(); + + let file = tempfile::tempfile().unwrap(); + file.set_len(initial_len as u64).unwrap(); + + let mut mmap = unsafe { MmapMut::map_mut(&file).unwrap() }; + assert_eq!(mmap.len(), initial_len); + + unsafe { mmap.remap(final_len, RemapOptions::new()).unwrap() }; + assert_eq!(mmap.len(), final_len); + + // Check that the mmap is still writable along the slice length + mmap.copy_from_slice(&incr); + } + + #[test] + #[cfg(target_os = "linux")] + #[cfg(target_pointer_width = "32")] + fn remap_len_overflow() { + use crate::RemapOptions; + + let file = tempfile::tempfile().unwrap(); + file.set_len(1024).unwrap(); + let mut mmap = unsafe { MmapOptions::new().len(1024).map(&file).unwrap() }; + + let res = unsafe { mmap.remap(0x80000000, RemapOptions::new().may_move(true)) }; + assert_eq!( + res.unwrap_err().to_string(), + "memory map length overflows isize" + ); + + assert_eq!(mmap.len(), 1024); + } + + #[test] + #[cfg(target_os = "linux")] + fn remap_with_offset() { + use crate::RemapOptions; + + let offset = 77; + let initial_len = 128; + let final_len = 2000; + + let zeros = vec![0u8; final_len]; + let incr: Vec<u8> = (0..final_len).map(|v| v as u8).collect(); + + let file = tempfile::tempfile().unwrap(); + file.set_len(final_len as u64 + offset).unwrap(); + + let mut mmap = unsafe { + MmapOptions::new() + .len(initial_len) + .offset(offset) + .map_mut(&file) + .unwrap() + }; + assert_eq!(mmap.len(), initial_len); + assert_eq!(&mmap[..], &zeros[..initial_len]); + + unsafe { + mmap.remap(final_len, RemapOptions::new().may_move(true)) + .unwrap() + }; + + // The size should have been updated + assert_eq!(mmap.len(), final_len); + + // Should still be all zeros + assert_eq!(&mmap[..], &zeros); + + // Write out to the whole expanded slice. + mmap.copy_from_slice(&incr); + } +} diff --git a/third_party/rust/memmap2/src/stub.rs b/third_party/rust/memmap2/src/stub.rs new file mode 100644 index 0000000000..881983eb18 --- /dev/null +++ b/third_party/rust/memmap2/src/stub.rs @@ -0,0 +1,81 @@ +use std::fs::File; +use std::io; + +// A stable alternative to https://doc.rust-lang.org/stable/std/primitive.never.html +enum Never {} + +pub struct MmapInner { + never: Never, +} + +impl MmapInner { + fn new() -> io::Result<MmapInner> { + Err(io::Error::new( + io::ErrorKind::Other, + "platform not supported", + )) + } + + pub fn map(_: usize, _: &File, _: u64, _: bool) -> io::Result<MmapInner> { + MmapInner::new() + } + + pub fn map_exec(_: usize, _: &File, _: u64, _: bool) -> io::Result<MmapInner> { + MmapInner::new() + } + + pub fn map_mut(_: usize, _: &File, _: u64, _: bool) -> io::Result<MmapInner> { + MmapInner::new() + } + + pub fn map_copy(_: usize, _: &File, _: u64, _: bool) -> io::Result<MmapInner> { + MmapInner::new() + } + + pub fn map_copy_read_only(_: usize, _: &File, _: u64, _: bool) -> io::Result<MmapInner> { + MmapInner::new() + } + + pub fn map_anon(_: usize, _: bool, _: bool, _: Option<u8>) -> io::Result<MmapInner> { + MmapInner::new() + } + + pub fn flush(&self, _: usize, _: usize) -> io::Result<()> { + match self.never {} + } + + pub fn flush_async(&self, _: usize, _: usize) -> io::Result<()> { + match self.never {} + } + + pub fn make_read_only(&mut self) -> io::Result<()> { + match self.never {} + } + + pub fn make_exec(&mut self) -> io::Result<()> { + match self.never {} + } + + pub fn make_mut(&mut self) -> io::Result<()> { + match self.never {} + } + + #[inline] + pub fn ptr(&self) -> *const u8 { + match self.never {} + } + + #[inline] + pub fn mut_ptr(&mut self) -> *mut u8 { + match self.never {} + } + + #[inline] + pub fn len(&self) -> usize { + match self.never {} + } +} + +pub fn file_len(file: &File) -> io::Result<u64> { + Ok(file.metadata()?.len()) +} diff --git a/third_party/rust/memmap2/src/unix.rs b/third_party/rust/memmap2/src/unix.rs new file mode 100644 index 0000000000..bd8fcf32a8 --- /dev/null +++ b/third_party/rust/memmap2/src/unix.rs @@ -0,0 +1,458 @@ +extern crate libc; + +use std::fs::File; +use std::mem::ManuallyDrop; +use std::os::unix::io::{FromRawFd, RawFd}; +use std::sync::atomic::{AtomicUsize, Ordering}; +use std::{io, ptr}; + +#[cfg(any( + all(target_os = "linux", not(target_arch = "mips")), + target_os = "freebsd", + target_os = "android" +))] +const MAP_STACK: libc::c_int = libc::MAP_STACK; + +#[cfg(not(any( + all(target_os = "linux", not(target_arch = "mips")), + target_os = "freebsd", + target_os = "android" +)))] +const MAP_STACK: libc::c_int = 0; + +#[cfg(any(target_os = "linux", target_os = "android"))] +const MAP_POPULATE: libc::c_int = libc::MAP_POPULATE; + +#[cfg(not(any(target_os = "linux", target_os = "android")))] +const MAP_POPULATE: libc::c_int = 0; + +#[cfg(any(target_os = "linux", target_os = "android"))] +const MAP_HUGETLB: libc::c_int = libc::MAP_HUGETLB; + +#[cfg(target_os = "linux")] +const MAP_HUGE_MASK: libc::c_int = libc::MAP_HUGE_MASK; + +#[cfg(any(target_os = "linux", target_os = "android"))] +const MAP_HUGE_SHIFT: libc::c_int = libc::MAP_HUGE_SHIFT; + +#[cfg(not(any(target_os = "linux", target_os = "android")))] +const MAP_HUGETLB: libc::c_int = 0; + +#[cfg(not(target_os = "linux"))] +const MAP_HUGE_MASK: libc::c_int = 0; + +#[cfg(not(any(target_os = "linux", target_os = "android")))] +const MAP_HUGE_SHIFT: libc::c_int = 0; + +#[cfg(any( + target_os = "android", + all(target_os = "linux", not(target_env = "musl")) +))] +use libc::{mmap64 as mmap, off64_t as off_t}; + +#[cfg(not(any( + target_os = "android", + all(target_os = "linux", not(target_env = "musl")) +)))] +use libc::{mmap, off_t}; + +pub struct MmapInner { + ptr: *mut libc::c_void, + len: usize, +} + +impl MmapInner { + /// Creates a new `MmapInner`. + /// + /// This is a thin wrapper around the `mmap` system call. + fn new( + len: usize, + prot: libc::c_int, + flags: libc::c_int, + file: RawFd, + offset: u64, + ) -> io::Result<MmapInner> { + let alignment = offset % page_size() as u64; + let aligned_offset = offset - alignment; + + let (map_len, map_offset) = Self::adjust_mmap_params(len, alignment as usize)?; + + unsafe { + let ptr = mmap( + ptr::null_mut(), + map_len as libc::size_t, + prot, + flags, + file, + aligned_offset as off_t, + ); + + if ptr == libc::MAP_FAILED { + Err(io::Error::last_os_error()) + } else { + Ok(Self::from_raw_parts(ptr, len, map_offset)) + } + } + } + + fn adjust_mmap_params(len: usize, alignment: usize) -> io::Result<(usize, usize)> { + use std::isize; + + // Rust's slice cannot be larger than isize::MAX. + // See https://doc.rust-lang.org/std/slice/fn.from_raw_parts.html + // + // This is not a problem on 64-bit targets, but on 32-bit one + // having a file or an anonymous mapping larger than 2GB is quite normal + // and we have to prevent it. + // + // The code below is essentially the same as in Rust's std: + // https://github.com/rust-lang/rust/blob/db78ab70a88a0a5e89031d7ee4eccec835dcdbde/library/alloc/src/raw_vec.rs#L495 + if std::mem::size_of::<usize>() < 8 && len > isize::MAX as usize { + return Err(io::Error::new( + io::ErrorKind::InvalidData, + "memory map length overflows isize", + )); + } + + let map_len = len + alignment; + let map_offset = alignment; + + // `libc::mmap` does not support zero-size mappings. POSIX defines: + // + // https://pubs.opengroup.org/onlinepubs/9699919799/functions/mmap.html + // > If `len` is zero, `mmap()` shall fail and no mapping shall be established. + // + // So if we would create such a mapping, crate a one-byte mapping instead: + let map_len = map_len.max(1); + + // Note that in that case `MmapInner::len` is still set to zero, + // and `Mmap` will still dereferences to an empty slice. + // + // If this mapping is backed by an empty file, we create a mapping larger than the file. + // This is unusual but well-defined. On the same man page, POSIX further defines: + // + // > The `mmap()` function can be used to map a region of memory that is larger + // > than the current size of the object. + // + // (The object here is the file.) + // + // > Memory access within the mapping but beyond the current end of the underlying + // > objects may result in SIGBUS signals being sent to the process. The reason for this + // > is that the size of the object can be manipulated by other processes and can change + // > at any moment. The implementation should tell the application that a memory reference + // > is outside the object where this can be detected; otherwise, written data may be lost + // > and read data may not reflect actual data in the object. + // + // Because `MmapInner::len` is not incremented, this increment of `aligned_len` + // will not allow accesses past the end of the file and will not cause SIGBUS. + // + // (SIGBUS is still possible by mapping a non-empty file and then truncating it + // to a shorter size, but that is unrelated to this handling of empty files.) + Ok((map_len, map_offset)) + } + + /// Get the current memory mapping as a `(ptr, map_len, offset)` tuple. + /// + /// Note that `map_len` is the length of the memory mapping itself and + /// _not_ the one that would be passed to `from_raw_parts`. + fn as_mmap_params(&self) -> (*mut libc::c_void, usize, usize) { + let offset = self.ptr as usize % page_size(); + let len = self.len + offset; + + // There are two possible memory layouts we could have, depending on + // the length and offset passed when constructing this instance: + // + // 1. The "normal" memory layout looks like this: + // + // |<------------------>|<---------------------->| + // mmap ptr offset ptr public slice + // + // That is, we have + // - The start of the page-aligned memory mapping returned by mmap, + // followed by, + // - Some number of bytes that are memory mapped but ignored since + // they are before the byte offset requested by the user, followed + // by, + // - The actual memory mapped slice requested by the user. + // + // This maps cleanly to a (ptr, len, offset) tuple. + // + // 2. Then, we have the case where the user requested a zero-length + // memory mapping. mmap(2) does not support zero-length mappings so + // this crate works around that by actually making a mapping of + // length one. This means that we have + // - A length zero slice, followed by, + // - A single memory mapped byte + // + // Note that this only happens if the offset within the page is also + // zero. Otherwise, we have a memory map of offset bytes and not a + // zero-length memory map. + // + // This doesn't fit cleanly into a (ptr, len, offset) tuple. Instead, + // we fudge it slightly: a zero-length memory map turns into a + // mapping of length one and can't be told apart outside of this + // method without knowing the original length. + if len == 0 { + (self.ptr, 1, 0) + } else { + (unsafe { self.ptr.offset(-(offset as isize)) }, len, offset) + } + } + + /// Construct this `MmapInner` from its raw components + /// + /// # Safety + /// + /// - `ptr` must point to the start of memory mapping that can be freed + /// using `munmap(2)` (i.e. returned by `mmap(2)` or `mremap(2)`) + /// - The memory mapping at `ptr` must have a length of `len + offset`. + /// - If `len + offset == 0` then the memory mapping must be of length 1. + /// - `offset` must be less than the current page size. + unsafe fn from_raw_parts(ptr: *mut libc::c_void, len: usize, offset: usize) -> Self { + debug_assert_eq!(ptr as usize % page_size(), 0, "ptr not page-aligned"); + debug_assert!(offset < page_size(), "offset larger than page size"); + + Self { + ptr: ptr.add(offset), + len, + } + } + + pub fn map(len: usize, file: RawFd, offset: u64, populate: bool) -> io::Result<MmapInner> { + let populate = if populate { MAP_POPULATE } else { 0 }; + MmapInner::new( + len, + libc::PROT_READ, + libc::MAP_SHARED | populate, + file, + offset, + ) + } + + pub fn map_exec(len: usize, file: RawFd, offset: u64, populate: bool) -> io::Result<MmapInner> { + let populate = if populate { MAP_POPULATE } else { 0 }; + MmapInner::new( + len, + libc::PROT_READ | libc::PROT_EXEC, + libc::MAP_SHARED | populate, + file, + offset, + ) + } + + pub fn map_mut(len: usize, file: RawFd, offset: u64, populate: bool) -> io::Result<MmapInner> { + let populate = if populate { MAP_POPULATE } else { 0 }; + MmapInner::new( + len, + libc::PROT_READ | libc::PROT_WRITE, + libc::MAP_SHARED | populate, + file, + offset, + ) + } + + pub fn map_copy(len: usize, file: RawFd, offset: u64, populate: bool) -> io::Result<MmapInner> { + let populate = if populate { MAP_POPULATE } else { 0 }; + MmapInner::new( + len, + libc::PROT_READ | libc::PROT_WRITE, + libc::MAP_PRIVATE | populate, + file, + offset, + ) + } + + pub fn map_copy_read_only( + len: usize, + file: RawFd, + offset: u64, + populate: bool, + ) -> io::Result<MmapInner> { + let populate = if populate { MAP_POPULATE } else { 0 }; + MmapInner::new( + len, + libc::PROT_READ, + libc::MAP_PRIVATE | populate, + file, + offset, + ) + } + + /// Open an anonymous memory map. + pub fn map_anon( + len: usize, + stack: bool, + populate: bool, + huge: Option<u8>, + ) -> io::Result<MmapInner> { + let stack = if stack { MAP_STACK } else { 0 }; + let populate = if populate { MAP_POPULATE } else { 0 }; + let hugetlb = if huge.is_some() { MAP_HUGETLB } else { 0 }; + let offset = huge + .map(|mask| ((mask as u64) & (MAP_HUGE_MASK as u64)) << MAP_HUGE_SHIFT) + .unwrap_or(0); + MmapInner::new( + len, + libc::PROT_READ | libc::PROT_WRITE, + libc::MAP_PRIVATE | libc::MAP_ANON | stack | populate | hugetlb, + -1, + offset, + ) + } + + pub fn flush(&self, offset: usize, len: usize) -> io::Result<()> { + let alignment = (self.ptr as usize + offset) % page_size(); + let offset = offset as isize - alignment as isize; + let len = len + alignment; + let result = + unsafe { libc::msync(self.ptr.offset(offset), len as libc::size_t, libc::MS_SYNC) }; + if result == 0 { + Ok(()) + } else { + Err(io::Error::last_os_error()) + } + } + + pub fn flush_async(&self, offset: usize, len: usize) -> io::Result<()> { + let alignment = (self.ptr as usize + offset) % page_size(); + let offset = offset as isize - alignment as isize; + let len = len + alignment; + let result = + unsafe { libc::msync(self.ptr.offset(offset), len as libc::size_t, libc::MS_ASYNC) }; + if result == 0 { + Ok(()) + } else { + Err(io::Error::last_os_error()) + } + } + + fn mprotect(&mut self, prot: libc::c_int) -> io::Result<()> { + unsafe { + let alignment = self.ptr as usize % page_size(); + let ptr = self.ptr.offset(-(alignment as isize)); + let len = self.len + alignment; + let len = len.max(1); + if libc::mprotect(ptr, len, prot) == 0 { + Ok(()) + } else { + Err(io::Error::last_os_error()) + } + } + } + + pub fn make_read_only(&mut self) -> io::Result<()> { + self.mprotect(libc::PROT_READ) + } + + pub fn make_exec(&mut self) -> io::Result<()> { + self.mprotect(libc::PROT_READ | libc::PROT_EXEC) + } + + pub fn make_mut(&mut self) -> io::Result<()> { + self.mprotect(libc::PROT_READ | libc::PROT_WRITE) + } + + #[inline] + pub fn ptr(&self) -> *const u8 { + self.ptr as *const u8 + } + + #[inline] + pub fn mut_ptr(&mut self) -> *mut u8 { + self.ptr as *mut u8 + } + + #[inline] + pub fn len(&self) -> usize { + self.len + } + + pub fn advise(&self, advice: libc::c_int, offset: usize, len: usize) -> io::Result<()> { + let alignment = (self.ptr as usize + offset) % page_size(); + let offset = offset as isize - alignment as isize; + let len = len + alignment; + unsafe { + if libc::madvise(self.ptr.offset(offset), len, advice) != 0 { + Err(io::Error::last_os_error()) + } else { + Ok(()) + } + } + } + + #[cfg(target_os = "linux")] + pub fn remap(&mut self, new_len: usize, options: crate::RemapOptions) -> io::Result<()> { + let (old_ptr, old_len, offset) = self.as_mmap_params(); + let (map_len, offset) = Self::adjust_mmap_params(new_len, offset)?; + + unsafe { + let new_ptr = libc::mremap(old_ptr, old_len, map_len, options.into_flags()); + + if new_ptr == libc::MAP_FAILED { + Err(io::Error::last_os_error()) + } else { + // We explicitly don't drop self since the pointer within is no longer valid. + ptr::write(self, Self::from_raw_parts(new_ptr, new_len, offset)); + Ok(()) + } + } + } + + pub fn lock(&self) -> io::Result<()> { + unsafe { + if libc::mlock(self.ptr, self.len) != 0 { + Err(io::Error::last_os_error()) + } else { + Ok(()) + } + } + } + + pub fn unlock(&self) -> io::Result<()> { + unsafe { + if libc::munlock(self.ptr, self.len) != 0 { + Err(io::Error::last_os_error()) + } else { + Ok(()) + } + } + } +} + +impl Drop for MmapInner { + fn drop(&mut self) { + let (ptr, len, _) = self.as_mmap_params(); + + // Any errors during unmapping/closing are ignored as the only way + // to report them would be through panicking which is highly discouraged + // in Drop impls, c.f. https://github.com/rust-lang/lang-team/issues/97 + unsafe { libc::munmap(ptr, len as libc::size_t) }; + } +} + +unsafe impl Sync for MmapInner {} +unsafe impl Send for MmapInner {} + +fn page_size() -> usize { + static PAGE_SIZE: AtomicUsize = AtomicUsize::new(0); + + match PAGE_SIZE.load(Ordering::Relaxed) { + 0 => { + let page_size = unsafe { libc::sysconf(libc::_SC_PAGESIZE) as usize }; + + PAGE_SIZE.store(page_size, Ordering::Relaxed); + + page_size + } + page_size => page_size, + } +} + +pub fn file_len(file: RawFd) -> io::Result<u64> { + // SAFETY: We must not close the passed-in fd by dropping the File we create, + // we ensure this by immediately wrapping it in a ManuallyDrop. + unsafe { + let file = ManuallyDrop::new(File::from_raw_fd(file)); + Ok(file.metadata()?.len()) + } +} diff --git a/third_party/rust/memmap2/src/windows.rs b/third_party/rust/memmap2/src/windows.rs new file mode 100644 index 0000000000..8a870955d2 --- /dev/null +++ b/third_party/rust/memmap2/src/windows.rs @@ -0,0 +1,524 @@ +#![allow(non_camel_case_types)] +#![allow(non_snake_case)] + +use std::fs::File; +use std::mem::ManuallyDrop; +use std::os::raw::c_void; +use std::os::windows::io::{FromRawHandle, RawHandle}; +use std::{io, mem, ptr}; + +type BOOL = i32; +type WORD = u16; +type DWORD = u32; +type WCHAR = u16; +type HANDLE = *mut c_void; +type LPHANDLE = *mut HANDLE; +type LPVOID = *mut c_void; +type LPCVOID = *const c_void; +type ULONG_PTR = usize; +type SIZE_T = ULONG_PTR; +type LPCWSTR = *const WCHAR; +type PDWORD = *mut DWORD; +type DWORD_PTR = ULONG_PTR; +type LPSECURITY_ATTRIBUTES = *mut SECURITY_ATTRIBUTES; +type LPSYSTEM_INFO = *mut SYSTEM_INFO; + +const INVALID_HANDLE_VALUE: HANDLE = -1isize as HANDLE; + +const DUPLICATE_SAME_ACCESS: DWORD = 0x00000002; + +const STANDARD_RIGHTS_REQUIRED: DWORD = 0x000F0000; + +const SECTION_QUERY: DWORD = 0x0001; +const SECTION_MAP_WRITE: DWORD = 0x0002; +const SECTION_MAP_READ: DWORD = 0x0004; +const SECTION_MAP_EXECUTE: DWORD = 0x0008; +const SECTION_EXTEND_SIZE: DWORD = 0x0010; +const SECTION_MAP_EXECUTE_EXPLICIT: DWORD = 0x0020; +const SECTION_ALL_ACCESS: DWORD = STANDARD_RIGHTS_REQUIRED + | SECTION_QUERY + | SECTION_MAP_WRITE + | SECTION_MAP_READ + | SECTION_MAP_EXECUTE + | SECTION_EXTEND_SIZE; + +const PAGE_READONLY: DWORD = 0x02; +const PAGE_READWRITE: DWORD = 0x04; +const PAGE_WRITECOPY: DWORD = 0x08; +const PAGE_EXECUTE_READ: DWORD = 0x20; +const PAGE_EXECUTE_READWRITE: DWORD = 0x40; +const PAGE_EXECUTE_WRITECOPY: DWORD = 0x80; + +const FILE_MAP_WRITE: DWORD = SECTION_MAP_WRITE; +const FILE_MAP_READ: DWORD = SECTION_MAP_READ; +const FILE_MAP_ALL_ACCESS: DWORD = SECTION_ALL_ACCESS; +const FILE_MAP_EXECUTE: DWORD = SECTION_MAP_EXECUTE_EXPLICIT; +const FILE_MAP_COPY: DWORD = 0x00000001; + +#[repr(C)] +struct SECURITY_ATTRIBUTES { + nLength: DWORD, + lpSecurityDescriptor: LPVOID, + bInheritHandle: BOOL, +} + +#[repr(C)] +struct SYSTEM_INFO { + wProcessorArchitecture: WORD, + wReserved: WORD, + dwPageSize: DWORD, + lpMinimumApplicationAddress: LPVOID, + lpMaximumApplicationAddress: LPVOID, + dwActiveProcessorMask: DWORD_PTR, + dwNumberOfProcessors: DWORD, + dwProcessorType: DWORD, + dwAllocationGranularity: DWORD, + wProcessorLevel: WORD, + wProcessorRevision: WORD, +} + +#[repr(C)] +#[derive(Copy, Clone)] +pub struct FILETIME { + pub dwLowDateTime: DWORD, + pub dwHighDateTime: DWORD, +} + +extern "system" { + fn GetCurrentProcess() -> HANDLE; + + fn CloseHandle(hObject: HANDLE) -> BOOL; + + fn DuplicateHandle( + hSourceProcessHandle: HANDLE, + hSourceHandle: HANDLE, + hTargetProcessHandle: HANDLE, + lpTargetHandle: LPHANDLE, + dwDesiredAccess: DWORD, + bInheritHandle: BOOL, + dwOptions: DWORD, + ) -> BOOL; + + fn CreateFileMappingW( + hFile: HANDLE, + lpFileMappingAttributes: LPSECURITY_ATTRIBUTES, + flProtect: DWORD, + dwMaximumSizeHigh: DWORD, + dwMaximumSizeLow: DWORD, + lpName: LPCWSTR, + ) -> HANDLE; + + fn FlushFileBuffers(hFile: HANDLE) -> BOOL; + + fn FlushViewOfFile(lpBaseAddress: LPCVOID, dwNumberOfBytesToFlush: SIZE_T) -> BOOL; + + fn UnmapViewOfFile(lpBaseAddress: LPCVOID) -> BOOL; + + fn MapViewOfFile( + hFileMappingObject: HANDLE, + dwDesiredAccess: DWORD, + dwFileOffsetHigh: DWORD, + dwFileOffsetLow: DWORD, + dwNumberOfBytesToMap: SIZE_T, + ) -> LPVOID; + + fn VirtualProtect( + lpAddress: LPVOID, + dwSize: SIZE_T, + flNewProtect: DWORD, + lpflOldProtect: PDWORD, + ) -> BOOL; + + fn GetSystemInfo(lpSystemInfo: LPSYSTEM_INFO); +} + +/// Returns a fixed aligned pointer that is valid for `slice::from_raw_parts::<u8>` with `len == 0`. +/// +/// This aligns the pointer to `allocation_granularity()` or 1 if unknown. +fn empty_slice_ptr() -> *mut c_void { + let align = allocation_granularity().max(1); + unsafe { mem::transmute(align) } +} + +pub struct MmapInner { + handle: Option<RawHandle>, + ptr: *mut c_void, + len: usize, + copy: bool, +} + +impl MmapInner { + /// Creates a new `MmapInner`. + /// + /// This is a thin wrapper around the `CreateFileMappingW` and `MapViewOfFile` system calls. + pub fn new( + handle: RawHandle, + protect: DWORD, + access: DWORD, + offset: u64, + len: usize, + copy: bool, + ) -> io::Result<MmapInner> { + let alignment = offset % allocation_granularity() as u64; + let aligned_offset = offset - alignment as u64; + let aligned_len = len + alignment as usize; + if aligned_len == 0 { + // `CreateFileMappingW` documents: + // + // https://docs.microsoft.com/en-us/windows/win32/api/memoryapi/nf-memoryapi-createfilemappingw + // > An attempt to map a file with a length of 0 (zero) fails with an error code + // > of ERROR_FILE_INVALID. Applications should test for files with a length of 0 + // > (zero) and reject those files. + // + // For such files, don’t create a mapping at all and use a marker pointer instead. + return Ok(MmapInner { + handle: None, + ptr: empty_slice_ptr(), + len: 0, + copy, + }); + } + + unsafe { + let mapping = CreateFileMappingW(handle, ptr::null_mut(), protect, 0, 0, ptr::null()); + if mapping.is_null() { + return Err(io::Error::last_os_error()); + } + + let ptr = MapViewOfFile( + mapping, + access, + (aligned_offset >> 16 >> 16) as DWORD, + (aligned_offset & 0xffffffff) as DWORD, + aligned_len as SIZE_T, + ); + CloseHandle(mapping); + if ptr.is_null() { + return Err(io::Error::last_os_error()); + } + + let mut new_handle = 0 as RawHandle; + let cur_proc = GetCurrentProcess(); + let ok = DuplicateHandle( + cur_proc, + handle, + cur_proc, + &mut new_handle, + 0, + 0, + DUPLICATE_SAME_ACCESS, + ); + if ok == 0 { + UnmapViewOfFile(ptr); + return Err(io::Error::last_os_error()); + } + + Ok(MmapInner { + handle: Some(new_handle), + ptr: ptr.offset(alignment as isize), + len: len as usize, + copy, + }) + } + } + + pub fn map( + len: usize, + handle: RawHandle, + offset: u64, + _populate: bool, + ) -> io::Result<MmapInner> { + let write = protection_supported(handle, PAGE_READWRITE); + let exec = protection_supported(handle, PAGE_EXECUTE_READ); + let mut access = FILE_MAP_READ; + let protection = match (write, exec) { + (true, true) => { + access |= FILE_MAP_WRITE | FILE_MAP_EXECUTE; + PAGE_EXECUTE_READWRITE + } + (true, false) => { + access |= FILE_MAP_WRITE; + PAGE_READWRITE + } + (false, true) => { + access |= FILE_MAP_EXECUTE; + PAGE_EXECUTE_READ + } + (false, false) => PAGE_READONLY, + }; + + let mut inner = MmapInner::new(handle, protection, access, offset, len, false)?; + if write || exec { + inner.make_read_only()?; + } + Ok(inner) + } + + pub fn map_exec( + len: usize, + handle: RawHandle, + offset: u64, + _populate: bool, + ) -> io::Result<MmapInner> { + let write = protection_supported(handle, PAGE_READWRITE); + let mut access = FILE_MAP_READ | FILE_MAP_EXECUTE; + let protection = if write { + access |= FILE_MAP_WRITE; + PAGE_EXECUTE_READWRITE + } else { + PAGE_EXECUTE_READ + }; + + let mut inner = MmapInner::new(handle, protection, access, offset, len, false)?; + if write { + inner.make_exec()?; + } + Ok(inner) + } + + pub fn map_mut( + len: usize, + handle: RawHandle, + offset: u64, + _populate: bool, + ) -> io::Result<MmapInner> { + let exec = protection_supported(handle, PAGE_EXECUTE_READ); + let mut access = FILE_MAP_READ | FILE_MAP_WRITE; + let protection = if exec { + access |= FILE_MAP_EXECUTE; + PAGE_EXECUTE_READWRITE + } else { + PAGE_READWRITE + }; + + let mut inner = MmapInner::new(handle, protection, access, offset, len, false)?; + if exec { + inner.make_mut()?; + } + Ok(inner) + } + + pub fn map_copy( + len: usize, + handle: RawHandle, + offset: u64, + _populate: bool, + ) -> io::Result<MmapInner> { + let exec = protection_supported(handle, PAGE_EXECUTE_READWRITE); + let mut access = FILE_MAP_COPY; + let protection = if exec { + access |= FILE_MAP_EXECUTE; + PAGE_EXECUTE_WRITECOPY + } else { + PAGE_WRITECOPY + }; + + let mut inner = MmapInner::new(handle, protection, access, offset, len, true)?; + if exec { + inner.make_mut()?; + } + Ok(inner) + } + + pub fn map_copy_read_only( + len: usize, + handle: RawHandle, + offset: u64, + _populate: bool, + ) -> io::Result<MmapInner> { + let write = protection_supported(handle, PAGE_READWRITE); + let exec = protection_supported(handle, PAGE_EXECUTE_READ); + let mut access = FILE_MAP_COPY; + let protection = if exec { + access |= FILE_MAP_EXECUTE; + PAGE_EXECUTE_WRITECOPY + } else { + PAGE_WRITECOPY + }; + + let mut inner = MmapInner::new(handle, protection, access, offset, len, true)?; + if write || exec { + inner.make_read_only()?; + } + Ok(inner) + } + + pub fn map_anon( + len: usize, + _stack: bool, + _populate: bool, + _huge: Option<u8>, + ) -> io::Result<MmapInner> { + // Ensure a non-zero length for the underlying mapping + let mapped_len = len.max(1); + unsafe { + // Create a mapping and view with maximum access permissions, then use `VirtualProtect` + // to set the actual `Protection`. This way, we can set more permissive protection later + // on. + // Also see https://msdn.microsoft.com/en-us/library/windows/desktop/aa366537.aspx + + let mapping = CreateFileMappingW( + INVALID_HANDLE_VALUE, + ptr::null_mut(), + PAGE_EXECUTE_READWRITE, + (mapped_len >> 16 >> 16) as DWORD, + (mapped_len & 0xffffffff) as DWORD, + ptr::null(), + ); + if mapping.is_null() { + return Err(io::Error::last_os_error()); + } + let access = FILE_MAP_ALL_ACCESS | FILE_MAP_EXECUTE; + let ptr = MapViewOfFile(mapping, access, 0, 0, mapped_len as SIZE_T); + CloseHandle(mapping); + + if ptr.is_null() { + return Err(io::Error::last_os_error()); + } + + let mut old = 0; + let result = VirtualProtect(ptr, mapped_len as SIZE_T, PAGE_READWRITE, &mut old); + if result != 0 { + Ok(MmapInner { + handle: None, + ptr, + len: len as usize, + copy: false, + }) + } else { + Err(io::Error::last_os_error()) + } + } + } + + pub fn flush(&self, offset: usize, len: usize) -> io::Result<()> { + self.flush_async(offset, len)?; + + if let Some(handle) = self.handle { + let ok = unsafe { FlushFileBuffers(handle) }; + if ok == 0 { + return Err(io::Error::last_os_error()); + } + } + + Ok(()) + } + + pub fn flush_async(&self, offset: usize, len: usize) -> io::Result<()> { + if self.ptr == empty_slice_ptr() { + return Ok(()); + } + let result = unsafe { FlushViewOfFile(self.ptr.add(offset), len as SIZE_T) }; + if result != 0 { + Ok(()) + } else { + Err(io::Error::last_os_error()) + } + } + + fn virtual_protect(&mut self, protect: DWORD) -> io::Result<()> { + if self.ptr == empty_slice_ptr() { + return Ok(()); + } + unsafe { + let alignment = self.ptr as usize % allocation_granularity(); + let ptr = self.ptr.offset(-(alignment as isize)); + let aligned_len = self.len as SIZE_T + alignment as SIZE_T; + + let mut old = 0; + let result = VirtualProtect(ptr, aligned_len, protect, &mut old); + + if result != 0 { + Ok(()) + } else { + Err(io::Error::last_os_error()) + } + } + } + + pub fn make_read_only(&mut self) -> io::Result<()> { + self.virtual_protect(PAGE_READONLY) + } + + pub fn make_exec(&mut self) -> io::Result<()> { + if self.copy { + self.virtual_protect(PAGE_EXECUTE_WRITECOPY) + } else { + self.virtual_protect(PAGE_EXECUTE_READ) + } + } + + pub fn make_mut(&mut self) -> io::Result<()> { + if self.copy { + self.virtual_protect(PAGE_WRITECOPY) + } else { + self.virtual_protect(PAGE_READWRITE) + } + } + + #[inline] + pub fn ptr(&self) -> *const u8 { + self.ptr as *const u8 + } + + #[inline] + pub fn mut_ptr(&mut self) -> *mut u8 { + self.ptr as *mut u8 + } + + #[inline] + pub fn len(&self) -> usize { + self.len + } +} + +impl Drop for MmapInner { + fn drop(&mut self) { + if self.ptr == empty_slice_ptr() { + return; + } + let alignment = self.ptr as usize % allocation_granularity(); + // Any errors during unmapping/closing are ignored as the only way + // to report them would be through panicking which is highly discouraged + // in Drop impls, c.f. https://github.com/rust-lang/lang-team/issues/97 + unsafe { + let ptr = self.ptr.offset(-(alignment as isize)); + UnmapViewOfFile(ptr); + + if let Some(handle) = self.handle { + CloseHandle(handle); + } + } + } +} + +unsafe impl Sync for MmapInner {} +unsafe impl Send for MmapInner {} + +fn protection_supported(handle: RawHandle, protection: DWORD) -> bool { + unsafe { + let mapping = CreateFileMappingW(handle, ptr::null_mut(), protection, 0, 0, ptr::null()); + if mapping.is_null() { + return false; + } + CloseHandle(mapping); + true + } +} + +fn allocation_granularity() -> usize { + unsafe { + let mut info = mem::zeroed(); + GetSystemInfo(&mut info); + info.dwAllocationGranularity as usize + } +} + +pub fn file_len(handle: RawHandle) -> io::Result<u64> { + // SAFETY: We must not close the passed-in fd by dropping the File we create, + // we ensure this by immediately wrapping it in a ManuallyDrop. + unsafe { + let file = ManuallyDrop::new(File::from_raw_handle(handle)); + Ok(file.metadata()?.len()) + } +} |