base/sys/linux/

mod.rs

// Copyright 2017 The ChromiumOS Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

//! Small system utility modules for usage by other modules.

#[cfg(target_os = "android")]
mod android;
#[cfg(target_os = "android")]
use android as target_os;
#[cfg(target_os = "linux")]
#[allow(clippy::module_inception)]
mod linux;
#[cfg(target_os = "linux")]
use linux as target_os;
use log::warn;
#[macro_use]
pub mod ioctl;
#[macro_use]
pub mod syslog;
mod capabilities;
mod descriptor;
mod event;
mod file;
mod file_traits;
mod mmap;
mod net;
mod notifiers;
pub mod platform_timer_resolution;
mod poll;
mod priority;
mod sched;
mod shm;
pub mod signal;
mod signalfd;
mod terminal;
mod timer;
pub mod vsock;
mod write_zeroes;

use std::ffi::CString;
use std::fs::remove_file;
use std::fs::File;
use std::fs::OpenOptions;
use std::mem;
use std::mem::MaybeUninit;
use std::ops::Deref;
use std::os::fd::AsRawFd;
use std::os::fd::BorrowedFd;
use std::os::unix::io::FromRawFd;
use std::os::unix::io::RawFd;
use std::os::unix::net::UnixDatagram;
use std::os::unix::net::UnixListener;
use std::os::unix::process::ExitStatusExt;
use std::path::Path;
use std::path::PathBuf;
use std::process::ExitStatus;
use std::ptr;
use std::sync::OnceLock;
use std::time::Duration;

pub use capabilities::drop_capabilities;
pub use event::EventExt;
pub(crate) use event::PlatformEvent;
pub use file::find_next_data;
pub use file::FileDataIterator;
pub(crate) use file_traits::lib::*;
pub use ioctl::*;
use libc::c_int;
use libc::c_long;
use libc::fcntl;
use libc::pipe2;
use libc::prctl;
use libc::syscall;
use libc::waitpid;
use libc::SYS_getpid;
use libc::SYS_getppid;
use libc::SYS_gettid;
use libc::EINVAL;
use libc::O_CLOEXEC;
use libc::PR_SET_NAME;
use libc::SIGKILL;
use libc::WNOHANG;
pub use mmap::*;
pub(in crate::sys) use net::sendmsg_nosignal as sendmsg;
pub(in crate::sys) use net::sockaddr_un;
pub(in crate::sys) use net::sockaddrv4_to_lib_c;
pub(in crate::sys) use net::sockaddrv6_to_lib_c;
pub use poll::EventContext;
pub use priority::*;
pub use sched::*;
pub use shm::MemfdSeals;
pub use shm::SharedMemoryLinux;
pub use signal::*;
pub use signalfd::Error as SignalFdError;
pub use signalfd::*;
pub use terminal::*;
pub(crate) use write_zeroes::file_punch_hole;
pub(crate) use write_zeroes::file_write_zeroes_at;

use crate::descriptor::FromRawDescriptor;
use crate::descriptor::SafeDescriptor;
pub use crate::errno::Error;
pub use crate::errno::Result;
pub use crate::errno::*;
use crate::number_of_logical_cores;
use crate::round_up_to_page_size;
pub use crate::sys::unix::descriptor::*;
use crate::syscall;
use crate::AsRawDescriptor;
use crate::IoBuf;
use crate::IoBufMut;
use crate::Pid;

/// Re-export libc types that are part of the API.
pub type Uid = libc::uid_t;
pub type Gid = libc::gid_t;
pub type Mode = libc::mode_t;

// Directory that holds cpu sysinfo files.
const CPU_DIR: &str = "/sys/devices/system/cpu";

/// Safe wrapper for PR_SET_NAME(2const)
#[inline(always)]
pub fn set_thread_name(name: &str) -> Result<()> {
    let name = CString::new(name).or(Err(Error::new(EINVAL)))?;
    // SAFETY: prctl copies name and doesn't expect it to outlive this function.
    let ret = unsafe { prctl(PR_SET_NAME, name.as_c_str()) };
    if ret == 0 {
        Ok(())
    } else {
        errno_result()
    }
}

/// This bypasses `libc`'s caching `getpid(2)` wrapper which can be invalid if a raw clone was used
/// elsewhere.
#[inline(always)]
pub fn getpid() -> Pid {
    // SAFETY:
    // Safe because this syscall can never fail and we give it a valid syscall number.
    unsafe { syscall(SYS_getpid as c_long) as Pid }
}

/// Safe wrapper for the geppid Linux systemcall.
#[inline(always)]
pub fn getppid() -> Pid {
    // SAFETY:
    // Safe because this syscall can never fail and we give it a valid syscall number.
    unsafe { syscall(SYS_getppid as c_long) as Pid }
}

/// Safe wrapper for the gettid Linux systemcall.
pub fn gettid() -> Pid {
    // SAFETY:
    // Calling the gettid() sycall is always safe.
    unsafe { syscall(SYS_gettid as c_long) as Pid }
}

/// Safe wrapper for `geteuid(2)`.
#[inline(always)]
pub fn geteuid() -> Uid {
    // SAFETY:
    // trivially safe
    unsafe { libc::geteuid() }
}

/// Safe wrapper for `getegid(2)`.
#[inline(always)]
pub fn getegid() -> Gid {
    // SAFETY:
    // trivially safe
    unsafe { libc::getegid() }
}

/// The operation to perform with `flock`.
pub enum FlockOperation {
    LockShared,
    LockExclusive,
    Unlock,
}

/// Safe wrapper for flock(2) with the operation `op` and optionally `nonblocking`. The lock will be
/// dropped automatically when `file` is dropped.
#[inline(always)]
pub fn flock<F: AsRawDescriptor>(file: &F, op: FlockOperation, nonblocking: bool) -> Result<()> {
    let mut operation = match op {
        FlockOperation::LockShared => libc::LOCK_SH,
        FlockOperation::LockExclusive => libc::LOCK_EX,
        FlockOperation::Unlock => libc::LOCK_UN,
    };

    if nonblocking {
        operation |= libc::LOCK_NB;
    }

    // SAFETY:
    // Safe since we pass in a valid fd and flock operation, and check the return value.
    syscall!(unsafe { libc::flock(file.as_raw_descriptor(), operation) }).map(|_| ())
}

/// The operation to perform with `fallocate`.
pub enum FallocateMode {
    PunchHole,
    ZeroRange,
    Allocate,
}

impl From<FallocateMode> for i32 {
    fn from(value: FallocateMode) -> Self {
        match value {
            FallocateMode::Allocate => libc::FALLOC_FL_KEEP_SIZE,
            FallocateMode::PunchHole => libc::FALLOC_FL_PUNCH_HOLE | libc::FALLOC_FL_KEEP_SIZE,
            FallocateMode::ZeroRange => libc::FALLOC_FL_ZERO_RANGE | libc::FALLOC_FL_KEEP_SIZE,
        }
    }
}

impl From<FallocateMode> for u32 {
    fn from(value: FallocateMode) -> Self {
        Into::<i32>::into(value) as u32
    }
}

/// Safe wrapper for `fallocate()`.
pub fn fallocate<F: AsRawDescriptor>(
    file: &F,
    mode: FallocateMode,
    offset: u64,
    len: u64,
) -> Result<()> {
    let offset = if offset > libc::off64_t::MAX as u64 {
        return Err(Error::new(libc::EINVAL));
    } else {
        offset as libc::off64_t
    };

    let len = if len > libc::off64_t::MAX as u64 {
        return Err(Error::new(libc::EINVAL));
    } else {
        len as libc::off64_t
    };

    // SAFETY:
    // Safe since we pass in a valid fd and fallocate mode, validate offset and len,
    // and check the return value.
    syscall!(unsafe { libc::fallocate64(file.as_raw_descriptor(), mode.into(), offset, len) })
        .map(|_| ())
}

/// Arguments for how `openat2(2)` should open the target path.
#[repr(C)]
#[derive(Default, Debug, Copy, Clone)]
pub struct open_how {
    /// Flags for the open operation (e.g. `O_RDONLY`).
    pub flags: u64,
    /// Mode for the created file (if `O_CREAT` is set).
    pub mode: u64,
    /// Resolution flags (e.g. `RESOLVE_IN_ROOT`).
    pub resolve: u64,
}

/// Safe wrapper for `openat2(2)`.
pub fn openat2<D: AsRawDescriptor>(dir: &D, name: &std::ffi::CStr, how: &open_how) -> Result<File> {
    // SAFETY:
    // Safe because the syscall is provided with valid arguments and its return value is checked.
    syscall!(unsafe {
        libc::syscall(
            libc::SYS_openat2,
            dir.as_raw_descriptor(),
            name.as_ptr(),
            how as *const open_how,
            std::mem::size_of::<open_how>() as libc::size_t,
        )
    })
    .map(|fd| unsafe { File::from_raw_descriptor(fd as RawFd) })
}

/// Safe wrapper for `fstat()`.
pub fn fstat<F: AsRawDescriptor>(f: &F) -> Result<libc::stat64> {
    let mut st = MaybeUninit::<libc::stat64>::zeroed();

    // SAFETY:
    // Safe because the kernel will only write data in `st` and we check the return
    // value.
    syscall!(unsafe { libc::fstat64(f.as_raw_descriptor(), st.as_mut_ptr()) })?;

    // SAFETY:
    // Safe because the kernel guarantees that the struct is now fully initialized.
    Ok(unsafe { st.assume_init() })
}

/// Checks whether a file is a block device fie or not.
pub fn is_block_file<F: AsRawDescriptor>(file: &F) -> Result<bool> {
    let stat = fstat(file)?;
    Ok((stat.st_mode & libc::S_IFMT) == libc::S_IFBLK)
}

const BLOCK_IO_TYPE: u32 = 0x12;
ioctl_io_nr!(BLKDISCARD, BLOCK_IO_TYPE, 119);

/// Discards the given range of a block file.
pub fn discard_block<F: AsRawDescriptor>(file: &F, offset: u64, len: u64) -> Result<()> {
    let range: [u64; 2] = [offset, len];
    // SAFETY:
    // Safe because
    // - we check the return value.
    // - ioctl(BLKDISCARD) does not hold the descriptor after the call.
    // - ioctl(BLKDISCARD) does not break the file descriptor.
    // - ioctl(BLKDISCARD) does not modify the given range.
    syscall!(unsafe { libc::ioctl(file.as_raw_descriptor(), BLKDISCARD, &range) }).map(|_| ())
}

/// A trait used to abstract types that provide a process id that can be operated on.
pub trait AsRawPid {
    fn as_raw_pid(&self) -> Pid;
}

impl AsRawPid for Pid {
    fn as_raw_pid(&self) -> Pid {
        *self
    }
}

impl AsRawPid for std::process::Child {
    fn as_raw_pid(&self) -> Pid {
        self.id() as Pid
    }
}

/// A safe wrapper around waitpid.
///
/// On success if a process was reaped, it will be returned as the first value.
/// The second returned value is the ExitStatus from the libc::waitpid() call.
///
/// Note: this can block if libc::WNOHANG is not set and EINTR is not handled internally.
pub fn wait_for_pid<A: AsRawPid>(pid: A, options: c_int) -> Result<(Option<Pid>, ExitStatus)> {
    let pid = pid.as_raw_pid();
    let mut status: c_int = 1;
    // SAFETY:
    // Safe because status is owned and the error is checked.
    let ret = unsafe { libc::waitpid(pid, &mut status, options) };
    if ret < 0 {
        return errno_result();
    }
    Ok((
        if ret == 0 { None } else { Some(ret) },
        ExitStatus::from_raw(status),
    ))
}

/// Reaps a child process that has terminated.
///
/// Returns `Ok(pid)` where `pid` is the process that was reaped or `Ok(0)` if none of the children
/// have terminated. An `Error` is with `errno == ECHILD` if there are no children left to reap.
///
/// # Examples
///
/// Reaps all child processes until there are no terminated children to reap.
///
/// ```
/// fn reap_children() {
///     loop {
///         match base::linux::reap_child() {
///             Ok(0) => println!("no children ready to reap"),
///             Ok(pid) => {
///                 println!("reaped {}", pid);
///                 continue
///             },
///             Err(e) if e.errno() == libc::ECHILD => println!("no children left"),
///             Err(e) => println!("error reaping children: {}", e),
///         }
///         break
///     }
/// }
/// ```
pub fn reap_child() -> Result<Pid> {
    // SAFETY:
    // Safe because we pass in no memory, prevent blocking with WNOHANG, and check for error.
    let ret = unsafe { waitpid(-1, ptr::null_mut(), WNOHANG) };
    if ret == -1 {
        errno_result()
    } else {
        Ok(ret)
    }
}

/// Kill all processes in the current process group.
///
/// On success, this kills all processes in the current process group, including the current
/// process, meaning this will not return. This is equivalent to a call to `kill(0, SIGKILL)`.
pub fn kill_process_group() -> Result<()> {
    // SAFETY: Safe because pid is 'self group' and return value doesn't matter.
    unsafe { kill(0, SIGKILL) }?;
    // Kill succeeded, so this process never reaches here.
    unreachable!();
}

/// Spawns a pipe pair where the first pipe is the read end and the second pipe is the write end.
///
/// The `O_CLOEXEC` flag will be set during pipe creation.
pub fn pipe() -> Result<(File, File)> {
    let mut pipe_fds = [-1; 2];
    // SAFETY:
    // Safe because pipe2 will only write 2 element array of i32 to the given pointer, and we check
    // for error.
    let ret = unsafe { pipe2(&mut pipe_fds[0], O_CLOEXEC) };
    if ret == -1 {
        errno_result()
    } else {
        // SAFETY:
        // Safe because both fds must be valid for pipe2 to have returned sucessfully and we have
        // exclusive ownership of them.
        Ok(unsafe {
            (
                File::from_raw_fd(pipe_fds[0]),
                File::from_raw_fd(pipe_fds[1]),
            )
        })
    }
}

/// Sets the pipe signified with fd to `size`.
///
/// Returns the new size of the pipe or an error if the OS fails to set the pipe size.
pub fn set_pipe_size(fd: RawFd, size: usize) -> Result<usize> {
    // SAFETY:
    // Safe because fcntl with the `F_SETPIPE_SZ` arg doesn't touch memory.
    syscall!(unsafe { fcntl(fd, libc::F_SETPIPE_SZ, size as c_int) }).map(|ret| ret as usize)
}

/// Test-only function used to create a pipe that is full. The pipe is created, has its size set to
/// the minimum and then has that much data written to it. Use `new_pipe_full` to test handling of
/// blocking `write` calls in unit tests.
pub fn new_pipe_full() -> Result<(File, File)> {
    use std::io::Write;

    let (rx, mut tx) = pipe()?;
    // The smallest allowed size of a pipe is the system page size on linux.
    let page_size = set_pipe_size(tx.as_raw_descriptor(), round_up_to_page_size(1))?;

    // Fill the pipe with page_size zeros so the next write call will block.
    let buf = vec![0u8; page_size];
    tx.write_all(&buf)?;

    Ok((rx, tx))
}

/// Used to attempt to clean up a named pipe after it is no longer used.
pub struct UnlinkUnixDatagram(pub UnixDatagram);
impl AsRef<UnixDatagram> for UnlinkUnixDatagram {
    fn as_ref(&self) -> &UnixDatagram {
        &self.0
    }
}
impl Drop for UnlinkUnixDatagram {
    fn drop(&mut self) {
        if let Ok(addr) = self.0.local_addr() {
            if let Some(path) = addr.as_pathname() {
                if let Err(e) = remove_file(path) {
                    warn!("failed to remove control socket file: {}", e);
                }
            }
        }
    }
}

/// Used to attempt to clean up a named pipe after it is no longer used.
pub struct UnlinkUnixListener(pub UnixListener);

impl AsRef<UnixListener> for UnlinkUnixListener {
    fn as_ref(&self) -> &UnixListener {
        &self.0
    }
}

impl Deref for UnlinkUnixListener {
    type Target = UnixListener;

    fn deref(&self) -> &UnixListener {
        &self.0
    }
}

impl Drop for UnlinkUnixListener {
    fn drop(&mut self) {
        if let Ok(addr) = self.0.local_addr() {
            if let Some(path) = addr.as_pathname() {
                if let Err(e) = remove_file(path) {
                    warn!("failed to remove control socket file: {}", e);
                }
            }
        }
    }
}

/// Verifies that |raw_descriptor| is actually owned by this process and duplicates it
/// to ensure that we have a unique handle to it.
pub fn validate_raw_descriptor(raw_descriptor: RawDescriptor) -> Result<RawDescriptor> {
    validate_raw_fd(&raw_descriptor)
}

/// Verifies that |raw_fd| is actually owned by this process and duplicates it to ensure that
/// we have a unique handle to it.
pub fn validate_raw_fd(raw_fd: &RawFd) -> Result<RawFd> {
    // Checking that close-on-exec isn't set helps filter out FDs that were opened by
    // crosvm as all crosvm FDs are close on exec.
    // SAFETY:
    // Safe because this doesn't modify any memory and we check the return value.
    let flags = unsafe { libc::fcntl(*raw_fd, libc::F_GETFD) };
    if flags < 0 || (flags & libc::FD_CLOEXEC) != 0 {
        return Err(Error::new(libc::EBADF));
    }

    // SAFETY:
    // Duplicate the fd to ensure that we don't accidentally close an fd previously
    // opened by another subsystem.  Safe because this doesn't modify any memory and
    // we check the return value.
    let dup_fd = unsafe { libc::fcntl(*raw_fd, libc::F_DUPFD_CLOEXEC, 0) };
    if dup_fd < 0 {
        return Err(Error::last());
    }
    Ok(dup_fd as RawFd)
}

/// Utility function that returns true if the given FD is readable without blocking.
///
/// On an error, such as an invalid or incompatible FD, this will return false, which can not be
/// distinguished from a non-ready to read FD.
pub fn poll_in<F: AsRawDescriptor>(fd: &F) -> bool {
    let mut fds = libc::pollfd {
        fd: fd.as_raw_descriptor(),
        events: libc::POLLIN,
        revents: 0,
    };
    // SAFETY:
    // Safe because we give a valid pointer to a list (of 1) FD and check the return value.
    let ret = unsafe { libc::poll(&mut fds, 1, 0) };
    // An error probably indicates an invalid FD, or an FD that can't be polled. Returning false in
    // that case is probably correct as such an FD is unlikely to be readable, although there are
    // probably corner cases in which that is wrong.
    if ret == -1 {
        return false;
    }
    fds.revents & libc::POLLIN != 0
}

/// Return the maximum Duration that can be used with libc::timespec.
pub fn max_timeout() -> Duration {
    Duration::new(libc::time_t::MAX as u64, 999999999)
}

/// If the given path is of the form /proc/self/fd/N for some N, returns `Ok(Some(N))`. Otherwise
/// returns `Ok(None)`.
pub fn safe_descriptor_from_path<P: AsRef<Path>>(path: P) -> Result<Option<SafeDescriptor>> {
    let path = path.as_ref();
    if path.parent() == Some(Path::new("/proc/self/fd")) {
        let raw_descriptor = path
            .file_name()
            .and_then(|fd_osstr| fd_osstr.to_str())
            .and_then(|fd_str| fd_str.parse::<RawFd>().ok())
            .ok_or_else(|| Error::new(EINVAL))?;
        let validated_fd = validate_raw_fd(&raw_descriptor)?;
        Ok(Some(
            // SAFETY:
            // Safe because nothing else has access to validated_fd after this call.
            unsafe { SafeDescriptor::from_raw_descriptor(validated_fd) },
        ))
    } else {
        Ok(None)
    }
}

/// Check FD is not opened by crosvm and returns a FD that is freshly DUPFD_CLOEXEC's.
/// A SafeDescriptor is created from the duplicated fd. It does not take ownership of
/// fd passed by argument.
pub fn safe_descriptor_from_cmdline_fd(fd: &RawFd) -> Result<SafeDescriptor> {
    let validated_fd = validate_raw_fd(fd)?;
    Ok(
        // SAFETY:
        // Safe because nothing else has access to validated_fd after this call.
        unsafe { SafeDescriptor::from_raw_descriptor(validated_fd) },
    )
}

/// Open the file with the given path, or if it is of the form `/proc/self/fd/N` then just use the
/// file descriptor.
///
/// Note that this will not work properly if the same `/proc/self/fd/N` path is used twice in
/// different places, as the metadata (including the offset) will be shared between both file
/// descriptors.
pub fn open_file_or_duplicate<P: AsRef<Path>>(path: P, options: &OpenOptions) -> Result<File> {
    let path = path.as_ref();
    // Special case '/proc/self/fd/*' paths. The FD is already open, just use it.
    Ok(if let Some(fd) = safe_descriptor_from_path(path)? {
        fd.into()
    } else {
        options.open(path)?
    })
}

/// Get the soft and hard limits of max number of open files allowed by the environment.
pub fn max_open_files() -> Result<libc::rlimit64> {
    let mut buf = mem::MaybeUninit::<libc::rlimit64>::zeroed();

    // SAFETY:
    // Safe because this will only modify `buf` and we check the return value.
    let res = unsafe { libc::prlimit64(0, libc::RLIMIT_NOFILE, ptr::null(), buf.as_mut_ptr()) };
    if res == 0 {
        // SAFETY:
        // Safe because the kernel guarantees that the struct is fully initialized.
        let limit = unsafe { buf.assume_init() };
        Ok(limit)
    } else {
        errno_result()
    }
}

/// Executes the given callback with extended soft limit of max number of open files. After the
/// callback executed, restore the limit.
pub fn call_with_extended_max_files<T, E>(
    callback: impl FnOnce() -> std::result::Result<T, E>,
) -> Result<std::result::Result<T, E>> {
    let cur_limit = max_open_files()?;
    let new_limit = libc::rlimit64 {
        rlim_cur: cur_limit.rlim_max,
        ..cur_limit
    };
    let needs_extension = cur_limit.rlim_cur < new_limit.rlim_cur;
    if needs_extension {
        set_max_open_files(new_limit)?;
    }

    let r = callback();

    // Restore the soft limit.
    if needs_extension {
        set_max_open_files(cur_limit)?;
    }

    Ok(r)
}

/// Set the soft and hard limits of max number of open files to the given value.
fn set_max_open_files(limit: libc::rlimit64) -> Result<()> {
    // SAFETY: RLIMIT_NOFILE is known only to read a buffer of size rlimit64, and we have always
    // rlimit64 allocated.
    let res = unsafe { libc::setrlimit64(libc::RLIMIT_NOFILE, &limit) };
    if res == 0 {
        Ok(())
    } else {
        errno_result()
    }
}

/// Moves the requested PID/TID to a particular cgroup
pub fn move_to_cgroup(cgroup_path: PathBuf, id_to_write: Pid, cgroup_file: &str) -> Result<()> {
    use std::io::Write;

    let gpu_cgroup_file = cgroup_path.join(cgroup_file);
    let mut f = File::create(gpu_cgroup_file)?;
    f.write_all(id_to_write.to_string().as_bytes())?;
    Ok(())
}

pub fn move_task_to_cgroup(cgroup_path: PathBuf, thread_id: Pid) -> Result<()> {
    move_to_cgroup(cgroup_path, thread_id, "tasks")
}

pub fn move_proc_to_cgroup(cgroup_path: PathBuf, process_id: Pid) -> Result<()> {
    move_to_cgroup(cgroup_path, process_id, "cgroup.procs")
}

fn read_sysfs_cpu_info_in_dir(cpu_dir: &str, cpu_id: usize, property: &str) -> Result<String> {
    let path = Path::new(cpu_dir)
        .join(format!("cpu{cpu_id}"))
        .join(property);

    std::fs::read_to_string(path).map_err(|e| e.into())
}

/// Queries the property of a specified CPU sysfs node.
fn parse_sysfs_cpu_info_vec(cpu_id: usize, property: &str) -> Result<Vec<u32>> {
    parse_sysfs_cpu_info_vec_in_dir(CPU_DIR, cpu_id, property)
}

fn parse_sysfs_cpu_info_vec_in_dir(
    cpu_dir: &str,
    cpu_id: usize,
    property: &str,
) -> Result<Vec<u32>> {
    read_sysfs_cpu_info_in_dir(cpu_dir, cpu_id, property)?
        .split_whitespace()
        .map(|x| x.parse().map_err(|_| Error::new(libc::EINVAL)))
        .collect()
}

/// Returns a list of supported frequencies in kHz for a given logical core.
pub fn logical_core_frequencies_khz(cpu_id: usize) -> Result<Vec<u32>> {
    parse_sysfs_cpu_info_vec(cpu_id, "cpufreq/scaling_available_frequencies")
}

/// Queries the property of a specified CPU sysfs node.
fn parse_sysfs_cpu_info(cpu_id: usize, property: &str) -> Result<u32> {
    parse_sysfs_cpu_info_in_dir(CPU_DIR, cpu_id, property)
}

fn parse_sysfs_cpu_info_in_dir(cpu_dir: &str, cpu_id: usize, property: &str) -> Result<u32> {
    read_sysfs_cpu_info_in_dir(cpu_dir, cpu_id, property)?
        .trim()
        .parse()
        .map_err(|_| Error::new(libc::EINVAL))
}

/// Returns the capacity (measure of performance) of a given logical core.
pub fn logical_core_capacity(cpu_id: usize) -> Result<u32> {
    static CPU_MAX_FREQS: OnceLock<Option<Vec<u32>>> = OnceLock::new();

    let cpu_capacity = parse_sysfs_cpu_info(cpu_id, "cpu_capacity")?;

    // Collect and cache the maximum frequencies of all cores. We need to know
    // the largest maximum frequency between all cores to reverse normalization,
    // so collect all the values once on the first call to this function.
    let cpu_max_freqs = CPU_MAX_FREQS.get_or_init(|| {
        (0..number_of_logical_cores().ok()?)
            .map(|cpu_id| logical_core_max_freq_khz(cpu_id).ok())
            .collect()
    });

    if let Some(cpu_max_freqs) = cpu_max_freqs {
        let largest_max_freq = *cpu_max_freqs.iter().max().ok_or(Error::new(EINVAL))?;
        let cpu_max_freq = *cpu_max_freqs.get(cpu_id).ok_or(Error::new(EINVAL))?;
        let normalized_cpu_capacity = (u64::from(cpu_capacity) * u64::from(largest_max_freq))
            .checked_div(u64::from(cpu_max_freq))
            .ok_or(Error::new(EINVAL))?;
        normalized_cpu_capacity
            .try_into()
            .map_err(|_| Error::new(EINVAL))
    } else {
        // cpu-freq is not enabled. Fall back to using the normalized capacity.
        Ok(cpu_capacity)
    }
}

/// Returns the cluster ID of a given logical core.
pub fn logical_core_cluster_id(cpu_id: usize) -> Result<u32> {
    parse_sysfs_cpu_info(cpu_id, "topology/physical_package_id")
}

/// Returns the maximum frequency (in kHz) of a given logical core.
pub fn logical_core_max_freq_khz(cpu_id: usize) -> Result<u32> {
    parse_sysfs_cpu_info(cpu_id, "cpufreq/cpuinfo_max_freq")
}

/// Parses a string of comma separated CPU ranges, e.g. "0-2,4,6-8" into a BTreeSet of CPU IDs.
fn parse_online_cpu_range(content: &str) -> std::collections::BTreeSet<usize> {
    let mut cpus = std::collections::BTreeSet::new();
    for part in content.trim().split(',') {
        let part = part.trim();
        if part.is_empty() {
            continue;
        }
        if let Some((start_str, end_str)) = part.split_once('-') {
            if let (Ok(start), Ok(end)) = (start_str.parse::<usize>(), end_str.parse::<usize>()) {
                for i in start..=end {
                    cpus.insert(i);
                }
            }
        } else if let Ok(cpu) = part.parse::<usize>() {
            cpus.insert(cpu);
        }
    }
    cpus
}

/// Returns a bool if the CPU is online. The online status is cached on the first call.
pub fn is_cpu_online(cpu_id: usize) -> bool {
    static ONLINE_CPUS: OnceLock<std::collections::BTreeSet<usize>> = OnceLock::new();

    let online_cpus = ONLINE_CPUS.get_or_init(|| {
        let mut cpus = std::collections::BTreeSet::new();
        let path = Path::new(CPU_DIR).join("online");
        match std::fs::read_to_string(&path) {
            Ok(content) => {
                cpus = parse_online_cpu_range(&content);
            }
            Err(_) => {
                // If we hit an error trying to access cpuX/online files, assume the CPU is online.
                // This prevents permission/EACCES errors on individual files from crashing crosvm.
                if let Ok(total_cores) = crate::number_of_logical_cores() {
                    for id in 0..total_cores {
                        match parse_sysfs_cpu_info(id, "online") {
                            Ok(1) => {
                                cpus.insert(id);
                            }
                            Ok(_) => {}
                            Err(e) => {
                                // Assume online on error to avoid crashes.
                                warn!(
                                    "Assuming CPU {} is online because we couldn't read the sys file: {}",
                                    id, e
                                );
                                cpus.insert(id);
                            }
                        }
                    }
                }
            }
        }
        cpus
    });

    online_cpus.contains(&cpu_id)
}

/// Wrapper around the `preadv2` syscall for standard I/O slices.
///
/// We invoke the Linux syscall directly using `libc::syscall` instead of `libc::preadv2`
/// because some host build environments (e.g. older glibc sysroots like glibc 2.17 used in
/// Android prebuilts for Cuttlefish) do not export `preadv2` in libc.
pub fn preadv2(
    fd: BorrowedFd<'_>,
    iovs: &mut [IoBufMut],
    offset: libc::off_t,
    flags: libc::c_int,
) -> libc::ssize_t {
    if iovs.is_empty() {
        return 0;
    }
    let pos_l = offset as libc::c_ulong;
    let pos_h = 0;
    // SAFETY:
    // Safe because `IoBufMut` is ABI-compatible with `libc::iovec`, the buffers referenced by
    // `iovs` are valid for writes for the duration of the call, and the file descriptor is valid.
    unsafe {
        libc::syscall(
            libc::SYS_preadv2,
            fd.as_raw_fd(),
            iovs.as_mut_ptr() as *mut libc::iovec,
            iovs.len(),
            pos_l,
            pos_h,
            flags,
        ) as libc::ssize_t
    }
}

/// Wrapper around the `pwritev2` syscall for standard I/O slices.
///
/// We invoke the Linux syscall directly using `libc::syscall` instead of `libc::pwritev2`
/// because some host build environments (e.g. older glibc sysroots like glibc 2.17 used in
/// Android prebuilts for Cuttlefish) do not export `pwritev2` in libc.
pub fn pwritev2(
    fd: BorrowedFd<'_>,
    iovs: &[IoBuf],
    offset: libc::off_t,
    flags: libc::c_int,
) -> libc::ssize_t {
    if iovs.is_empty() {
        return 0;
    }
    let pos_l = offset as libc::c_ulong;
    let pos_h = 0;
    // SAFETY:
    // Safe because `IoBuf` is ABI-compatible with `libc::iovec`, the buffers referenced by
    // `iovs` are valid for reads for the duration of the call, and the file descriptor is valid.
    unsafe {
        libc::syscall(
            libc::SYS_pwritev2,
            fd.as_raw_fd(),
            iovs.as_ptr(),
            iovs.len(),
            pos_l,
            pos_h,
            flags,
        ) as libc::ssize_t
    }
}

#[repr(C)]
pub struct sched_attr {
    pub size: u32,

    pub sched_policy: u32,
    pub sched_flags: u64,
    pub sched_nice: i32,

    pub sched_priority: u32,

    pub sched_runtime: u64,
    pub sched_deadline: u64,
    pub sched_period: u64,

    pub sched_util_min: u32,
    pub sched_util_max: u32,
}

impl Default for sched_attr {
    fn default() -> Self {
        Self {
            size: std::mem::size_of::<sched_attr>() as u32,
            sched_policy: 0,
            sched_flags: 0,
            sched_nice: 0,
            sched_priority: 0,
            sched_runtime: 0,
            sched_deadline: 0,
            sched_period: 0,
            sched_util_min: 0,
            sched_util_max: 0,
        }
    }
}

pub fn sched_setattr(pid: Pid, attr: &mut sched_attr, flags: u32) -> Result<()> {
    // SAFETY: Safe becuase all the args are valid and the return valud is checked.
    let ret = unsafe {
        libc::syscall(
            libc::SYS_sched_setattr,
            pid as usize,
            attr as *mut sched_attr as usize,
            flags as usize,
        )
    };

    if ret < 0 {
        return Err(Error::last());
    }
    Ok(())
}

#[cfg(test)]
mod tests {
    use std::fs::create_dir_all;
    use std::fs::File;
    use std::io::Write;
    use std::os::fd::AsRawFd;

    use tempfile::TempDir;

    use super::*;
    use crate::unix::add_fd_flags;

    fn create_temp_file(path: &Path, content: &str) {
        if let Some(parent) = path.parent() {
            create_dir_all(parent).unwrap();
        }
        let mut file = File::create(path).unwrap();
        file.write_all(content.as_bytes()).unwrap();
    }

    #[test]
    fn test_parse_online_cpu_range() {
        let set = parse_online_cpu_range("0-3,5-7");
        assert_eq!(set.len(), 7);
        assert!(set.contains(&0));
        assert!(set.contains(&1));
        assert!(set.contains(&2));
        assert!(set.contains(&3));
        assert!(!set.contains(&4));
        assert!(set.contains(&5));
        assert!(set.contains(&6));
        assert!(set.contains(&7));

        let set = parse_online_cpu_range("0");
        assert_eq!(set.len(), 1);
        assert!(set.contains(&0));

        let set = parse_online_cpu_range("0,2,4");
        assert_eq!(set.len(), 3);
        assert!(set.contains(&0));
        assert!(set.contains(&2));
        assert!(set.contains(&4));

        let set = parse_online_cpu_range("  0-1,  3  ");
        assert_eq!(set.len(), 3);
        assert!(set.contains(&0));
        assert!(set.contains(&1));
        assert!(set.contains(&3));

        let set = parse_online_cpu_range("");
        assert!(set.is_empty());
    }

    #[test]
    fn pipe_size_and_fill() {
        let (_rx, mut tx) = new_pipe_full().expect("Failed to pipe");

        // To  check that setting the size worked, set the descriptor to non blocking and check that
        // write returns an error.
        add_fd_flags(tx.as_raw_fd(), libc::O_NONBLOCK).expect("Failed to set tx non blocking");
        tx.write(&[0u8; 8])
            .expect_err("Write after fill didn't fail");
    }

    #[test]
    fn test_parse_sysfs_cpu_info() {
        let temp_dir = TempDir::new().unwrap();
        let root = temp_dir.path();
        let cpu_dir = root.join("sys/devices/system/cpu");
        let cpu = 0;
        let property = "cpufreq/cpuinfo_max_freq";
        create_temp_file(
            &root.join("sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq"),
            "1000",
        );

        assert_eq!(
            parse_sysfs_cpu_info_in_dir(cpu_dir.to_str().unwrap(), cpu, property).unwrap(),
            1000
        );
    }

    #[test]
    fn test_parse_sysfs_cpu_info_error() {
        let temp_dir = TempDir::new().unwrap();
        let root = temp_dir.path();
        let cpu_dir = root.join("sys/devices/system/cpu");
        let cpu = 0;
        let property = "cpufreq/cpuinfo_max_freq";
        // Not creating the sysinfo file should result in an error trying to read from it.

        let err =
            parse_sysfs_cpu_info_in_dir(cpu_dir.to_str().unwrap(), cpu, property).unwrap_err();
        assert_eq!(err, Error::new(libc::ENOENT));
    }

    #[test]
    fn test_parse_sysfs_cpu_info_vec() {
        let temp_dir = TempDir::new().unwrap();
        let root = temp_dir.path();
        let cpu_dir = root.join("sys/devices/system/cpu");
        let cpu = 0;
        let property = "cpufreq/scaling_available_frequencies";
        create_temp_file(
            &root.join("sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies"),
            "1000 2000",
        );

        assert_eq!(
            parse_sysfs_cpu_info_vec_in_dir(cpu_dir.to_str().unwrap(), cpu, property).unwrap(),
            vec![1000, 2000]
        );
    }

    #[test]
    fn test_parse_sysfs_cpu_info_vec_error() {
        let temp_dir = TempDir::new().unwrap();
        let root = temp_dir.path();
        let cpu_dir = root.join("sys/devices/system/cpu");
        let cpu = 0;
        let property = "cpufreq/scaling_available_frequencies";
        // Not creating the sysinfo file should result in an error trying to read from it.

        let err =
            parse_sysfs_cpu_info_vec_in_dir(cpu_dir.to_str().unwrap(), cpu, property).unwrap_err();
        assert_eq!(err, Error::new(libc::ENOENT));
    }
}