starnix_core/task/

syscalls.rs

// Copyright 2021 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

use crate::execution::execute_task;
use crate::mm::{DumpPolicy, MemoryAccessor, MemoryAccessorExt, PAGE_SIZE};
use crate::ptrace::{
    PR_SET_PTRACER_ANY, PR_SET_PTRACER_ANY_ARCH32, PtraceAllowedPtracers, PtraceAttachType,
    PtraceOptions, ptrace_attach, ptrace_dispatch, ptrace_traceme,
};
use crate::security;
use crate::signals::syscalls::RUsagePtr;
use crate::task::{
    CurrentTask, ExitStatus, NormalPriority, SchedulingPolicy, SeccompAction, SeccompStateValue,
    SyslogAccess, Task, ThreadGroup, max_priority_for_sched_policy, min_priority_for_sched_policy,
};
use crate::vfs::{
    FdNumber, FileHandle, MountNamespaceFile, PidFdFileObject, UserBuffersOutputBuffer,
    VecOutputBuffer,
};
use starnix_logging::{log_error, log_info, log_trace, track_stub};
use starnix_syscalls::SyscallResult;
use starnix_task_command::TaskCommand;
use starnix_types::time::timeval_from_duration;
use starnix_uapi::auth::{
    CAP_SETGID, CAP_SETPCAP, CAP_SETUID, CAP_SYS_ADMIN, CAP_SYS_NICE, CAP_SYS_RESOURCE,
    CAP_SYS_TTY_CONFIG, Capabilities, Credentials, PTRACE_MODE_READ_REALCREDS, SecureBits,
};
use starnix_uapi::errors::{ENAMETOOLONG, Errno};
use starnix_uapi::kcmp::KcmpResource;
use starnix_uapi::open_flags::OpenFlags;
use starnix_uapi::resource_limits::Resource;
use starnix_uapi::signals::{Signal, UncheckedSignal};
use starnix_uapi::syslog::SyslogAction;
use starnix_uapi::user_address::{
    ArchSpecific, MappingMultiArchUserRef, MultiArchUserRef, UserAddress, UserCString,
    UserCStringPtr, UserRef,
};
use starnix_uapi::{
    __user_cap_data_struct, __user_cap_header_struct, _LINUX_CAPABILITY_VERSION_1,
    _LINUX_CAPABILITY_VERSION_2, _LINUX_CAPABILITY_VERSION_3, AT_EMPTY_PATH, AT_SYMLINK_NOFOLLOW,
    BPF_MAXINSNS, CLONE_ARGS_SIZE_VER0, CLONE_ARGS_SIZE_VER1, CLONE_ARGS_SIZE_VER2, CLONE_FILES,
    CLONE_FS, CLONE_NEWNS, CLONE_NEWUTS, CLONE_SETTLS, CLONE_VFORK, NGROUPS_MAX, PR_CAP_AMBIENT,
    PR_CAP_AMBIENT_CLEAR_ALL, PR_CAP_AMBIENT_IS_SET, PR_CAP_AMBIENT_LOWER, PR_CAP_AMBIENT_RAISE,
    PR_CAPBSET_DROP, PR_CAPBSET_READ, PR_GET_CHILD_SUBREAPER, PR_GET_DUMPABLE, PR_GET_KEEPCAPS,
    PR_GET_NAME, PR_GET_NO_NEW_PRIVS, PR_GET_PDEATHSIG, PR_GET_SECCOMP, PR_GET_SECUREBITS,
    PR_GET_TIMERSLACK, PR_SET_CHILD_SUBREAPER, PR_SET_DUMPABLE, PR_SET_KEEPCAPS, PR_SET_NAME,
    PR_SET_NO_NEW_PRIVS, PR_SET_PDEATHSIG, PR_SET_PTRACER, PR_SET_SECCOMP, PR_SET_SECUREBITS,
    PR_SET_TIMERSLACK, PR_SET_VMA, PR_SET_VMA_ANON_NAME, PRIO_PROCESS, PTRACE_ATTACH, PTRACE_SEIZE,
    PTRACE_TRACEME, RUSAGE_CHILDREN, SCHED_RESET_ON_FORK, SECCOMP_FILTER_FLAG_LOG,
    SECCOMP_FILTER_FLAG_NEW_LISTENER, SECCOMP_FILTER_FLAG_SPEC_ALLOW, SECCOMP_FILTER_FLAG_TSYNC,
    SECCOMP_FILTER_FLAG_TSYNC_ESRCH, SECCOMP_GET_ACTION_AVAIL, SECCOMP_GET_NOTIF_SIZES,
    SECCOMP_MODE_FILTER, SECCOMP_MODE_STRICT, SECCOMP_SET_MODE_FILTER, SECCOMP_SET_MODE_STRICT,
    c_char, c_int, clone_args, errno, error, gid_t, pid_t, rlimit, rusage, sched_param,
    sock_filter, uapi, uid_t,
};
use static_assertions::const_assert;
use std::cmp;
use std::ffi::CString;
use std::sync::{Arc, LazyLock};
use zerocopy::{FromBytes, Immutable, IntoBytes, KnownLayout};

#[cfg(target_arch = "aarch64")]
use starnix_uapi::{PR_GET_TAGGED_ADDR_CTRL, PR_SET_TAGGED_ADDR_CTRL, PR_TAGGED_ADDR_ENABLE};

pub type SockFProgPtr =
    MappingMultiArchUserRef<SockFProg, uapi::sock_fprog, uapi::arch32::sock_fprog>;
pub type SockFilterPtr = MultiArchUserRef<uapi::sock_filter, uapi::arch32::sock_filter>;

pub struct SockFProg {
    pub len: u32,
    pub filter: SockFilterPtr,
}

uapi::arch_map_data! {
    BidiTryFrom<SockFProg, sock_fprog> {
        len = len;
        filter = filter;
    }
}

uapi::check_arch_independent_layout! {
    sched_param {
        sched_priority,
    }
}

pub fn do_clone(current_task: &mut CurrentTask, args: &clone_args) -> Result<pid_t, Errno> {
    security::check_task_create_access(current_task)?;

    let child_exit_signal = if args.exit_signal == 0 {
        None
    } else {
        Some(Signal::try_from(UncheckedSignal::new(args.exit_signal))?)
    };

    let mut new_task = current_task.clone_task(
        args.flags,
        child_exit_signal,
        UserRef::<pid_t>::new(UserAddress::from(args.parent_tid)),
        UserRef::<pid_t>::new(UserAddress::from(args.child_tid)),
        UserRef::<FdNumber>::new(UserAddress::from(args.pidfd)),
    )?;

    // Set the result register to 0 for the return value from clone in the
    // cloned process.
    new_task.thread_state.registers.set_return_register(0);
    let (trace_kind, ptrace_state) = current_task.get_ptrace_core_state_for_clone(args);

    if args.stack != 0 {
        // In clone() the `stack` argument points to the top of the stack, while in clone3()
        // `stack` points to the bottom of the stack. Therefore, in clone3() we need to add
        // `stack_size` to calculate the stack pointer. Note that in clone() `stack_size` is 0.
        new_task
            .thread_state
            .registers
            .set_stack_pointer_register(args.stack.wrapping_add(args.stack_size));
    }

    if args.flags & (CLONE_SETTLS as u64) != 0 {
        new_task.thread_state.registers.set_thread_pointer_register(args.tls);
    }

    let tid = new_task.task.tid.id;
    let task_ref = Arc::downgrade(&new_task.task);
    execute_task(new_task, |_| Ok(()), |_| {}, ptrace_state)?;

    current_task.ptrace_event(trace_kind, tid as u64);

    if args.flags & (CLONE_VFORK as u64) != 0 {
        current_task.wait_for_execve(task_ref)?;
        current_task.ptrace_event(PtraceOptions::TRACEVFORKDONE, tid as u64);
    }

    Ok(tid)
}

pub fn sys_clone3(
    current_task: &mut CurrentTask,
    user_clone_args: UserRef<clone_args>,
    user_clone_args_size: usize,
) -> Result<pid_t, Errno> {
    // Only these specific sized versions are supported.
    if !(user_clone_args_size == CLONE_ARGS_SIZE_VER0 as usize
        || user_clone_args_size == CLONE_ARGS_SIZE_VER1 as usize
        || user_clone_args_size == CLONE_ARGS_SIZE_VER2 as usize)
    {
        return error!(EINVAL);
    }

    // The most recent version of the struct size should match our definition.
    const_assert!(std::mem::size_of::<clone_args>() == CLONE_ARGS_SIZE_VER2 as usize);

    let clone_args = current_task.read_object_partial(user_clone_args, user_clone_args_size)?;
    do_clone(current_task, &clone_args)
}

fn read_c_string_vector(
    mm: &CurrentTask,
    user_vector: UserCStringPtr,
    elem_limit: usize,
    vec_limit: usize,
) -> Result<(Vec<CString>, usize), Errno> {
    let mut user_current = user_vector;
    let mut vector: Vec<CString> = vec![];
    let mut vec_size: usize = 0;
    loop {
        let user_string = mm.read_multi_arch_ptr(user_current)?;
        if user_string.is_null() {
            break;
        }
        let string = mm
            .read_c_string_to_vec(user_string, elem_limit)
            .map_err(|e| if e.code == ENAMETOOLONG { errno!(E2BIG) } else { e })?;
        let cstring = CString::new(string).map_err(|_| errno!(EINVAL))?;
        vec_size =
            vec_size.checked_add(cstring.as_bytes_with_nul().len()).ok_or_else(|| errno!(E2BIG))?;
        if vec_size > vec_limit {
            return error!(E2BIG);
        }
        vector.push(cstring);
        user_current = user_current.next()?;
    }
    Ok((vector, vec_size))
}

pub fn sys_execve(
    current_task: &mut CurrentTask,
    user_path: UserCString,
    user_argv: UserCStringPtr,
    user_environ: UserCStringPtr,
) -> Result<(), Errno> {
    sys_execveat(current_task, FdNumber::AT_FDCWD, user_path, user_argv, user_environ, 0)
}

pub fn sys_execveat(
    current_task: &mut CurrentTask,
    dir_fd: FdNumber,
    user_path: UserCString,
    user_argv: UserCStringPtr,
    user_environ: UserCStringPtr,
    flags: u32,
) -> Result<(), Errno> {
    if flags & !(AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW) != 0 {
        return error!(EINVAL);
    }

    // Calculate the limit for argv and environ size as 1/4 of the stack size, floored at 32 pages.
    // See the Limits sections in https://man7.org/linux/man-pages/man2/execve.2.html
    const PAGE_LIMIT: usize = 32;
    let page_limit_size: usize = PAGE_LIMIT * *PAGE_SIZE as usize;
    let rlimit = current_task.thread_group().get_rlimit(Resource::STACK);
    let stack_limit = rlimit / 4;
    let argv_env_limit = cmp::max(page_limit_size, stack_limit as usize);

    // The limit per argument or environment variable is 32 pages.
    // See the Limits sections in https://man7.org/linux/man-pages/man2/execve.2.html
    let (argv, argv_size) = if user_argv.is_null() {
        (Vec::new(), 0)
    } else {
        read_c_string_vector(current_task, user_argv, page_limit_size, argv_env_limit)?
    };

    let (environ, _) = if user_environ.is_null() {
        (Vec::new(), 0)
    } else {
        read_c_string_vector(
            current_task,
            user_environ,
            page_limit_size,
            argv_env_limit - argv_size,
        )?
    };

    let path = &current_task.read_path(user_path)?;

    log_trace!(argv:?, environ:?, flags:?; "execveat({dir_fd}, {path})");

    let mut open_flags = OpenFlags::empty();

    if flags & AT_SYMLINK_NOFOLLOW != 0 {
        open_flags |= OpenFlags::NOFOLLOW;
    }

    if path.is_empty() && flags & AT_EMPTY_PATH == 0 {
        // If AT_EMPTY_PATH is not set, an empty path is an error.
        return error!(ENOENT);
    }

    let executable = current_task.open_file_for_exec(dir_fd, path.as_ref(), open_flags)?;

    // This path can affect script resolution (the path is appended to the script args)
    // and the auxiliary value `AT_EXECFN` from the syscall `getauxval()`
    let path = if dir_fd == FdNumber::AT_FDCWD {
        // The file descriptor is CWD, so the path is exactly
        // what the user specified.
        path.to_vec()
    } else {
        // The path is `/dev/fd/N/P` where N is the file descriptor
        // number and P is the user-provided path (if relative and non-empty).
        //
        // See https://man7.org/linux/man-pages/man2/execveat.2.html#NOTES
        match path.first() {
            Some(b'/') => {
                // The user-provided path is absolute, so dir_fd is ignored.
                path.to_vec()
            }
            Some(_) => {
                // User-provided path is relative, append it.
                let mut new_path = format!("/dev/fd/{}/", dir_fd.raw()).into_bytes();
                new_path.append(&mut path.to_vec());
                new_path
            }
            // User-provided path is empty
            None => format!("/dev/fd/{}", dir_fd.raw()).into_bytes(),
        }
    };

    let path = CString::new(path).map_err(|_| errno!(EINVAL))?;

    current_task.exec(executable, path, argv, environ)?;
    Ok(())
}

pub fn sys_getcpu(
    current_task: &CurrentTask,
    cpu_out: UserRef<u32>,
    node_out: UserRef<u32>,
) -> Result<(), Errno> {
    // "When either cpu or node is NULL nothing is written to the respective pointer."
    // from https://man7.org/linux/man-pages/man2/getcpu.2.html
    if !cpu_out.is_null() {
        let thread_stats = current_task
            .running_state()
            .thread
            .get()
            .expect("current thread is never None when executing")
            .thread
            .stats()
            .map_err(|e| errno!(EINVAL, format!("getting thread stats failed {e:?}")))?;
        current_task.write_object(cpu_out, &thread_stats.last_scheduled_cpu)?;
    }
    if !node_out.is_null() {
        // Zircon does not yet have a concept of NUMA task scheduling, always tell userspace that
        // it's on the "first" node which should be true for non-NUMA systems.
        track_stub!(TODO("https://fxbug.dev/325643815"), "getcpu() numa node");
        current_task.write_object(node_out, &0)?;
    }
    Ok(())
}

pub fn sys_getpid(current_task: &CurrentTask) -> Result<pid_t, Errno> {
    Ok(current_task.get_pid())
}

pub fn sys_gettid(current_task: &CurrentTask) -> Result<pid_t, Errno> {
    Ok(current_task.get_tid())
}

pub fn sys_getppid(current_task: &CurrentTask) -> Result<pid_t, Errno> {
    Ok(current_task.thread_group().read().get_ppid())
}

fn get_task_or_current(current_task: &CurrentTask, pid: pid_t) -> Result<Arc<Task>, Errno> {
    if pid == 0 { Ok(current_task.task.clone()) } else { current_task.get_task(pid) }
}

pub fn sys_getsid(current_task: &CurrentTask, pid: pid_t) -> Result<pid_t, Errno> {
    let target_task = get_task_or_current(current_task, pid)?;
    if pid != 0 {
        security::check_task_getsid(current_task, &target_task)?;
    }
    let sid = target_task.thread_group().read().session.id;
    Ok(sid)
}

pub fn sys_getpgid(current_task: &CurrentTask, pid: pid_t) -> Result<pid_t, Errno> {
    let task = get_task_or_current(current_task, pid)?;

    security::check_getpgid_access(current_task, &task)?;
    let pgid = task.thread_group().read().process_group.id;
    Ok(pgid)
}

pub fn sys_setpgid(current_task: &CurrentTask, pid: pid_t, pgid: pid_t) -> Result<(), Errno> {
    if pgid < 0 {
        return error!(EINVAL);
    }
    let task = get_task_or_current(current_task, pid)?;
    let pgid = if pgid == 0 {
        task.pid.clone()
    } else {
        current_task.kernel().pids.get(pgid).map_err(|_| errno!(EPERM))?
    };

    current_task.thread_group().setpgid(current_task, &task, &pgid)
}

impl CurrentTask {
    /// Returns true if the `current_task`'s effective user ID (EUID) is the same as the
    /// EUID or UID of the `target_task`. We describe this as the current task being
    /// "EUID-friendly" to the target and it enables actions to be performed that would
    /// otherwise require additional privileges.
    ///
    /// See "The caller needs an effective user ID equal to the real user ID or effective
    /// user ID of the [target]" at sched_setaffinity(2), comparable language at
    /// setpriority(2), more ambiguous language at sched_setscheduler(2), and no
    /// particular specification at sched_setparam(2).
    fn is_euid_friendly_with(&self, target_task: &Task) -> bool {
        let self_creds = self.current_creds();
        let target_creds = target_task.real_creds();
        self_creds.euid == target_creds.uid || self_creds.euid == target_creds.euid
    }
}

// A non-root process is allowed to set any of its three uids to the value of any other. The
// CAP_SETUID capability bypasses these checks and allows setting any uid to any integer. Likewise
// for gids.
fn new_uid_allowed(current_task: &CurrentTask, uid: uid_t) -> bool {
    let current_creds = current_task.current_creds();
    uid == current_creds.uid
        || uid == current_creds.euid
        || uid == current_creds.saved_uid
        || security::is_task_capable_noaudit(current_task, CAP_SETUID)
}

fn new_gid_allowed(current_task: &CurrentTask, gid: gid_t) -> bool {
    let current_creds = current_task.current_creds();
    gid == current_creds.gid
        || gid == current_creds.egid
        || gid == current_creds.saved_gid
        || security::is_task_capable_noaudit(current_task, CAP_SETGID)
}

pub fn sys_getuid(current_task: &CurrentTask) -> Result<uid_t, Errno> {
    Ok(current_task.current_creds().uid)
}

pub fn sys_getgid(current_task: &CurrentTask) -> Result<gid_t, Errno> {
    Ok(current_task.current_creds().gid)
}

pub fn sys_setuid(current_task: &CurrentTask, uid: uid_t) -> Result<(), Errno> {
    if uid == uid_t::MAX {
        return error!(EINVAL);
    }
    if !new_uid_allowed(&current_task, uid) {
        return error!(EPERM);
    }

    let prev = current_task.current_creds();
    let mut creds = Credentials::clone(&prev);
    creds.euid = uid;
    creds.fsuid = uid;
    if security::is_task_capable_noaudit(current_task, CAP_SETUID) {
        creds.uid = uid;
        creds.saved_uid = uid;
    }

    creds.update_capabilities(&prev);
    std::mem::drop(prev);
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_setgid(current_task: &CurrentTask, gid: gid_t) -> Result<(), Errno> {
    if gid == gid_t::MAX {
        return error!(EINVAL);
    }
    if !new_gid_allowed(&current_task, gid) {
        return error!(EPERM);
    }

    let mut creds = Credentials::clone(&current_task.current_creds());
    creds.egid = gid;
    creds.fsgid = gid;
    if security::is_task_capable_noaudit(current_task, CAP_SETGID) {
        creds.gid = gid;
        creds.saved_gid = gid;
    }
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_geteuid(current_task: &CurrentTask) -> Result<uid_t, Errno> {
    Ok(current_task.current_creds().euid)
}

pub fn sys_getegid(current_task: &CurrentTask) -> Result<gid_t, Errno> {
    Ok(current_task.current_creds().egid)
}

pub fn sys_setfsuid(current_task: &CurrentTask, fsuid: uid_t) -> Result<uid_t, Errno> {
    let prev = current_task.current_creds();
    let prev_fsuid = prev.fsuid;
    if fsuid != u32::MAX && new_uid_allowed(&current_task, fsuid) {
        let mut creds = Credentials::clone(&prev);
        creds.fsuid = fsuid;
        creds.update_capabilities(&prev);
        std::mem::drop(prev);
        current_task.set_creds(creds);
    }

    Ok(prev_fsuid)
}

pub fn sys_setfsgid(current_task: &CurrentTask, fsgid: gid_t) -> Result<gid_t, Errno> {
    let prev = current_task.current_creds();
    let prev_fsgid = prev.fsgid;

    if fsgid != u32::MAX && new_gid_allowed(&current_task, fsgid) {
        let mut creds = Credentials::clone(&prev);
        creds.fsgid = fsgid;
        creds.update_capabilities(&prev);
        std::mem::drop(prev);
        current_task.set_creds(creds);
    }

    Ok(prev_fsgid)
}

pub fn sys_getresuid(
    current_task: &CurrentTask,
    ruid_addr: UserRef<uid_t>,
    euid_addr: UserRef<uid_t>,
    suid_addr: UserRef<uid_t>,
) -> Result<(), Errno> {
    let creds = current_task.current_creds();
    current_task.write_object(ruid_addr, &creds.uid)?;
    current_task.write_object(euid_addr, &creds.euid)?;
    current_task.write_object(suid_addr, &creds.saved_uid)?;
    Ok(())
}

pub fn sys_getresgid(
    current_task: &CurrentTask,
    rgid_addr: UserRef<gid_t>,
    egid_addr: UserRef<gid_t>,
    sgid_addr: UserRef<gid_t>,
) -> Result<(), Errno> {
    let creds = current_task.current_creds();
    current_task.write_object(rgid_addr, &creds.gid)?;
    current_task.write_object(egid_addr, &creds.egid)?;
    current_task.write_object(sgid_addr, &creds.saved_gid)?;
    Ok(())
}

pub fn sys_setreuid(current_task: &CurrentTask, ruid: uid_t, euid: uid_t) -> Result<(), Errno> {
    // Linux __sys_setreuid() uses asymmetric checks: ruid cannot be set
    // to saved_uid, while euid can. This prevents regaining root via
    // setreuid after a privilege drop when setresuid would be required.
    let validate_ruid = |uid: uid_t| {
        let creds = current_task.current_creds();
        uid == u32::MAX
            || uid == creds.uid
            || uid == creds.euid
            || security::is_task_capable_noaudit(current_task, CAP_SETUID)
    };
    let validate_euid = |uid: uid_t| {
        let creds = current_task.current_creds();
        uid == u32::MAX
            || uid == creds.uid
            || uid == creds.euid
            || uid == creds.saved_uid
            || security::is_task_capable_noaudit(current_task, CAP_SETUID)
    };
    if !validate_ruid(ruid) || !validate_euid(euid) {
        return error!(EPERM);
    }

    let prev = current_task.current_creds();
    let mut creds = Credentials::clone(&prev);
    let is_ruid_set = ruid != u32::MAX;
    if is_ruid_set {
        creds.uid = ruid;
    }
    let is_euid_set = euid != u32::MAX;
    if is_euid_set {
        creds.euid = euid;
        creds.fsuid = euid;
    }

    // If the real user ID is set (i.e., ruid is not -1) or the effective
    // user ID is set to a value not equal to the previous real user ID,
    // the saved set-user-ID will be set to the new effective user ID.
    if is_ruid_set || (is_euid_set && euid != prev.uid) {
        creds.saved_uid = creds.euid;
    }

    creds.update_capabilities(&prev);
    std::mem::drop(prev);
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_setregid(current_task: &CurrentTask, rgid: gid_t, egid: gid_t) -> Result<(), Errno> {
    // Same asymmetric permission model as setreuid - see above.
    let validate_rgid = |gid: gid_t| {
        let creds = current_task.current_creds();
        gid == u32::MAX
            || gid == creds.gid
            || gid == creds.egid
            || security::is_task_capable_noaudit(current_task, CAP_SETGID)
    };
    let validate_egid = |gid: gid_t| {
        let creds = current_task.current_creds();
        gid == u32::MAX
            || gid == creds.gid
            || gid == creds.egid
            || gid == creds.saved_gid
            || security::is_task_capable_noaudit(current_task, CAP_SETGID)
    };
    if !validate_rgid(rgid) || !validate_egid(egid) {
        return error!(EPERM);
    }

    let mut creds = Credentials::clone(&current_task.current_creds());
    let previous_rgid = creds.gid;
    let is_rgid_set = rgid != u32::MAX;
    if is_rgid_set {
        creds.gid = rgid;
    }
    let is_egid_set = egid != u32::MAX;
    if is_egid_set {
        creds.egid = egid;
        creds.fsgid = egid;
    }

    // If the real group ID is set (i.e., rgid is not -1) or the effective
    // group ID is set to a value not equal to the previous real group ID,
    // the saved set-group-ID will be set to the new effective group ID.
    if is_rgid_set || (is_egid_set && egid != previous_rgid) {
        creds.saved_gid = creds.egid;
    }

    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_setresuid(
    current_task: &CurrentTask,
    ruid: uid_t,
    euid: uid_t,
    suid: uid_t,
) -> Result<(), Errno> {
    let allowed = |uid| uid == u32::MAX || new_uid_allowed(&current_task, uid);
    if !allowed(ruid) || !allowed(euid) || !allowed(suid) {
        return error!(EPERM);
    }

    let prev = current_task.current_creds();
    let mut creds = Credentials::clone(&prev);
    if ruid != u32::MAX {
        creds.uid = ruid;
    }
    if euid != u32::MAX {
        creds.euid = euid;
        creds.fsuid = euid;
    }
    if suid != u32::MAX {
        creds.saved_uid = suid;
    }
    creds.update_capabilities(&prev);
    std::mem::drop(prev);
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_setresgid(
    current_task: &CurrentTask,
    rgid: gid_t,
    egid: gid_t,
    sgid: gid_t,
) -> Result<(), Errno> {
    let allowed = |gid| gid == u32::MAX || new_gid_allowed(&current_task, gid);
    if !allowed(rgid) || !allowed(egid) || !allowed(sgid) {
        return error!(EPERM);
    }

    let mut creds = Credentials::clone(&current_task.current_creds());
    if rgid != u32::MAX {
        creds.gid = rgid;
    }
    if egid != u32::MAX {
        creds.egid = egid;
        creds.fsgid = egid;
    }
    if sgid != u32::MAX {
        creds.saved_gid = sgid;
    }
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_exit(current_task: &CurrentTask, code: i32) -> Result<(), Errno> {
    // Only change the current exit status if this has not been already set by exit_group, as
    // otherwise it has priority.
    current_task.write().set_exit_status_if_not_already(ExitStatus::Exit(code as u8));
    Ok(())
}

pub fn sys_exit_group(current_task: &mut CurrentTask, code: i32) -> Result<(), Errno> {
    current_task.kill_thread_group(ExitStatus::Exit(code as u8));
    Ok(())
}

pub fn sys_sched_getscheduler(current_task: &CurrentTask, pid: pid_t) -> Result<u32, Errno> {
    if pid < 0 {
        return error!(EINVAL);
    }

    let target_task = get_task_or_current(current_task, pid)?;
    security::check_task_getscheduler_access(current_task, target_task.as_ref())?;
    Ok(target_task.read().scheduler_state.policy_for_sched_getscheduler())
}

pub fn sys_sched_setscheduler(
    current_task: &CurrentTask,
    pid: pid_t,
    policy: u32,
    param: UserRef<sched_param>,
) -> Result<(), Errno> {
    // Parse & validate the arguments.
    if pid < 0 || param.is_null() {
        return error!(EINVAL);
    }

    let target_task = get_task_or_current(current_task, pid)?;

    let reset_on_fork = policy & SCHED_RESET_ON_FORK != 0;

    let policy = SchedulingPolicy::try_from(policy & !SCHED_RESET_ON_FORK)?;
    let realtime_priority =
        policy.realtime_priority_from(current_task.read_object(param)?.sched_priority)?;

    // TODO: https://fxbug.dev/425143440 - we probably want to improve the locking here.
    let current_state = target_task.read().scheduler_state.clone();

    // Check capabilities and permissions, if required, for the operation.
    let euid_friendly = current_task.is_euid_friendly_with(&target_task);
    let strengthening = current_state.realtime_priority < realtime_priority;
    let rlimited = strengthening
        && realtime_priority.exceeds(target_task.thread_group().get_rlimit(Resource::RTPRIO));
    let clearing_reset_on_fork = current_state.reset_on_fork && !reset_on_fork;
    let caught_in_idle_trap = current_state.policy == SchedulingPolicy::Idle
        && policy != SchedulingPolicy::Idle
        && current_state
            .normal_priority
            .exceeds(target_task.thread_group().get_rlimit(Resource::NICE));
    if !euid_friendly || rlimited || clearing_reset_on_fork || caught_in_idle_trap {
        security::check_task_capable(current_task, CAP_SYS_NICE)?;
    }

    security::check_task_setscheduler_access(current_task, &target_task)?;

    // Apply the new scheduler configuration to the task.
    target_task.set_scheduler_policy_priority_and_reset_on_fork(
        policy,
        realtime_priority,
        reset_on_fork,
    )?;

    Ok(())
}

const CPU_SET_SIZE: usize = 128;

#[repr(C)]
#[derive(Debug, Copy, Clone, IntoBytes, FromBytes, KnownLayout, Immutable)]
pub struct CpuSet {
    bits: [u8; CPU_SET_SIZE],
}

impl Default for CpuSet {
    fn default() -> Self {
        Self { bits: [0; CPU_SET_SIZE] }
    }
}

fn check_cpu_set_alignment(current_task: &CurrentTask, cpusetsize: u32) -> Result<(), Errno> {
    let alignment = if current_task.is_arch32() { 4 } else { 8 };
    if cpusetsize < alignment || cpusetsize % alignment != 0 {
        return error!(EINVAL);
    }
    Ok(())
}

fn get_default_cpu_set() -> CpuSet {
    let mut result = CpuSet::default();
    let mut cpus_count = zx::system_get_num_cpus();
    let cpus_count_max = (CPU_SET_SIZE * 8) as u32;
    if cpus_count > cpus_count_max {
        log_error!("cpus_count={cpus_count}, greater than the {cpus_count_max} max supported.");
        cpus_count = cpus_count_max;
    }
    let mut index = 0;
    while cpus_count > 0 {
        let count = std::cmp::min(cpus_count, 8);
        let (shl, overflow) = 1_u8.overflowing_shl(count);
        let mask = if overflow { u8::MAX } else { shl - 1 };
        result.bits[index] = mask;
        index += 1;
        cpus_count -= count;
    }
    result
}

pub fn sys_sched_getaffinity(
    current_task: &CurrentTask,
    pid: pid_t,
    cpusetsize: u32,
    user_mask: UserAddress,
) -> Result<usize, Errno> {
    if pid < 0 {
        return error!(EINVAL);
    }

    check_cpu_set_alignment(current_task, cpusetsize)?;

    let target_task = get_task_or_current(current_task, pid)?;
    security::check_task_getscheduler_access(current_task, &target_task)?;

    // sched_setaffinity() is not implemented. Fake affinity mask based on the number of CPUs.
    let mask = get_default_cpu_set();
    let mask_size = std::cmp::min(cpusetsize as usize, CPU_SET_SIZE);
    current_task.write_memory(user_mask, &mask.bits[..mask_size])?;
    track_stub!(TODO("https://fxbug.dev/322874659"), "sched_getaffinity");
    Ok(mask_size)
}

pub fn sys_sched_setaffinity(
    current_task: &CurrentTask,
    pid: pid_t,
    cpusetsize: u32,
    user_mask: UserAddress,
) -> Result<(), Errno> {
    if pid < 0 {
        return error!(EINVAL);
    }
    let target_task = get_task_or_current(current_task, pid)?;

    check_cpu_set_alignment(current_task, cpusetsize)?;

    let mask_size = std::cmp::min(cpusetsize as usize, CPU_SET_SIZE);
    let mut mask = CpuSet::default();
    current_task.read_memory_to_slice(user_mask, &mut mask.bits[..mask_size])?;

    // Specified mask must include at least one valid CPU.
    let max_mask = get_default_cpu_set();
    let mut has_valid_cpu_in_mask = false;
    for (l1, l2) in std::iter::zip(max_mask.bits, mask.bits) {
        has_valid_cpu_in_mask = has_valid_cpu_in_mask || (l1 & l2 > 0);
    }
    if !has_valid_cpu_in_mask {
        return error!(EINVAL);
    }

    if !current_task.is_euid_friendly_with(&target_task) {
        security::check_task_capable(current_task, CAP_SYS_NICE)?;
    }

    security::check_task_setscheduler_access(current_task, &target_task)?;

    // Currently, we ignore the mask and act as if the system reset the mask
    // immediately to allowing all CPUs.
    track_stub!(TODO("https://fxbug.dev/322874889"), "sched_setaffinity");
    Ok(())
}

pub fn sys_sched_getparam(
    current_task: &CurrentTask,
    pid: pid_t,
    param: UserRef<sched_param>,
) -> Result<(), Errno> {
    if pid < 0 || param.is_null() {
        return error!(EINVAL);
    }

    let target_task = get_task_or_current(current_task, pid)?;
    let param_value = target_task.read().scheduler_state.get_sched_param();
    current_task.write_object(param, &param_value)?;
    Ok(())
}

pub fn sys_sched_setparam(
    current_task: &CurrentTask,
    pid: pid_t,
    param: UserRef<sched_param>,
) -> Result<(), Errno> {
    // Parse & validate the arguments.
    if pid < 0 || param.is_null() {
        return error!(EINVAL);
    }
    let target_task = get_task_or_current(current_task, pid)?;

    // TODO: https://fxbug.dev/425143440 - we probably want to improve the locking here.
    let current_state = target_task.read().scheduler_state.clone();

    let realtime_priority = current_state
        .policy
        .realtime_priority_from(current_task.read_object(param)?.sched_priority)?;

    // Check capabilities and permissions, if required, for the operation.
    let euid_friendly = current_task.is_euid_friendly_with(&target_task);
    let strengthening = current_state.realtime_priority < realtime_priority;
    let rlimited = strengthening
        && realtime_priority.exceeds(target_task.thread_group().get_rlimit(Resource::RTPRIO));
    if !euid_friendly || rlimited {
        security::check_task_capable(current_task, CAP_SYS_NICE)?;
    }

    security::check_task_setscheduler_access(current_task, &target_task)?;

    // Apply the new scheduler configuration to the task.
    target_task.set_scheduler_priority(realtime_priority)?;

    Ok(())
}

pub fn sys_sched_get_priority_min(_ctx: &CurrentTask, policy: u32) -> Result<u8, Errno> {
    min_priority_for_sched_policy(policy)
}

pub fn sys_sched_get_priority_max(_ctx: &CurrentTask, policy: u32) -> Result<u8, Errno> {
    max_priority_for_sched_policy(policy)
}

pub fn sys_ioprio_set(
    _current_task: &mut CurrentTask,
    _which: i32,
    _who: i32,
    _ioprio: i32,
) -> Result<(), Errno> {
    track_stub!(TODO("https://fxbug.dev/297591758"), "ioprio_set()");
    error!(ENOSYS)
}

pub fn sys_prctl(
    current_task: &mut CurrentTask,
    option: u32,
    arg2: u64,
    arg3: u64,
    arg4: u64,
    arg5: u64,
) -> Result<SyscallResult, Errno> {
    match option {
        PR_SET_VMA => {
            if arg2 != PR_SET_VMA_ANON_NAME as u64 {
                track_stub!(TODO("https://fxbug.dev/322874826"), "prctl PR_SET_VMA", arg2);
                return error!(ENOSYS);
            }
            let addr = UserAddress::from(arg3);
            let length = arg4 as usize;
            let name_addr = UserAddress::from(arg5);
            let name = if name_addr.is_null() {
                None
            } else {
                let name = UserCString::new(current_task, UserAddress::from(arg5));
                let name = current_task.read_c_string_to_vec(name, 256).map_err(|e| {
                    // An overly long name produces EINVAL and not ENAMETOOLONG in Linux 5.15.
                    if e.code == ENAMETOOLONG { errno!(EINVAL) } else { e }
                })?;
                // Some characters are forbidden in VMA names.
                if name.iter().any(|b| {
                    matches!(b,
                        0..=0x1f |
                        0x7f..=0xff |
                        b'\\' | b'`' | b'$' | b'[' | b']'
                    )
                }) {
                    return error!(EINVAL);
                }
                Some(name)
            };
            current_task.mm()?.set_mapping_name(addr, length, name)?;
            Ok(().into())
        }
        PR_SET_DUMPABLE => {
            let mm = current_task.mm()?;
            mm.dumpable.store(match arg2 {
                0 => DumpPolicy::Disable,
                1 => DumpPolicy::User,
                _ => return error!(EINVAL),
            });
            Ok(().into())
        }
        PR_GET_DUMPABLE => {
            let mm = current_task.mm()?;
            Ok(match mm.dumpable.load() {
                DumpPolicy::Disable => 0.into(),
                DumpPolicy::User => 1.into(),
            })
        }
        PR_SET_PDEATHSIG => {
            // From <https://man7.org/linux/man-pages/man2/PR_SET_PDEATHSIG.2const.html>:
            //
            //   Set the parent-death signal of the calling process to sig (either a
            //   signal value in the range [1, NSIG - 1], or 0 to clear).  This is the
            //   signal that the calling process will get when its parent dies.
            let signal =
                if arg2 == 0 { None } else { Some(Signal::try_from(UncheckedSignal::new(arg2))?) };
            current_task.thread_group().write().set_parent_death_signal(signal);
            Ok(().into())
        }
        PR_GET_PDEATHSIG => {
            // From <https://man7.org/linux/man-pages/man2/PR_GET_PDEATHSIG.2const.html>:
            //
            //   Return the parent-death signal number of the calling process, in the
            //   location pointed to by sig.
            let signal_ptr = UserRef::<i32>::new(UserAddress::from(arg2));
            let parent_death_signal = current_task.thread_group().read().parent_death_signal();
            let signal_number = parent_death_signal.map_or(0, |signal| signal.number() as i32);
            current_task.write_object(signal_ptr, &signal_number)?;
            Ok(().into())
        }
        PR_SET_NAME => {
            let addr = UserAddress::from(arg2);
            let name = TaskCommand::new(&current_task.read_memory_to_array::<16>(addr)?);
            current_task.set_command_name(name);
            if current_task.tid == current_task.pid {
                current_task.thread_group.sync_syscall_log_level();
            }
            Ok(0.into())
        }
        PR_GET_NAME => {
            let addr = UserAddress::from(arg2);
            let name = current_task.command().prctl_name();
            current_task.write_memory(addr, &name[..])?;
            Ok(().into())
        }
        PR_SET_PTRACER => {
            let allowed_ptracers = if arg2 == PR_SET_PTRACER_ANY
                || (current_task.is_arch32() && arg2 == PR_SET_PTRACER_ANY_ARCH32)
            {
                PtraceAllowedPtracers::Any
            } else if arg2 == 0 {
                PtraceAllowedPtracers::None
            } else {
                let task = current_task.get_task(arg2 as i32).map_err(|_| errno!(EINVAL))?;
                PtraceAllowedPtracers::Some(task.pid.clone())
            };
            current_task.thread_group().write().allowed_ptracers = allowed_ptracers;
            Ok(().into())
        }
        PR_GET_KEEPCAPS => {
            Ok(current_task.current_creds().securebits.contains(SecureBits::KEEP_CAPS).into())
        }
        PR_SET_KEEPCAPS => {
            if arg2 != 0 && arg2 != 1 {
                return error!(EINVAL);
            }
            // Per PR_SET_KEEPCAPS(2const):
            //   EPERM: The caller's SECBIT_KEEP_CAPS_LOCKED flag is set.
            if current_task.current_creds().securebits.contains(SecureBits::KEEP_CAPS_LOCKED) {
                return error!(EPERM);
            }
            let mut creds = Credentials::clone(&current_task.current_creds());
            creds.securebits.set(SecureBits::KEEP_CAPS, arg2 != 0);
            current_task.set_creds(creds);
            Ok(().into())
        }
        PR_SET_NO_NEW_PRIVS => {
            // If any args are set other than arg2 to 1, this should return einval
            if arg2 != 1 || arg3 != 0 || arg4 != 0 || arg5 != 0 {
                return error!(EINVAL);
            }
            current_task.write().enable_no_new_privs();
            Ok(().into())
        }
        PR_GET_NO_NEW_PRIVS => {
            // If any args are set, this should return einval
            if arg2 != 0 || arg3 != 0 || arg4 != 0 {
                return error!(EINVAL);
            }
            Ok(current_task.read().no_new_privs().into())
        }
        PR_GET_SECCOMP => {
            if current_task.seccomp_filter_state.get() == SeccompStateValue::None {
                Ok(0.into())
            } else {
                Ok(2.into())
            }
        }
        PR_SET_SECCOMP => {
            if arg2 == SECCOMP_MODE_STRICT as u64 {
                return sys_seccomp(current_task, SECCOMP_SET_MODE_STRICT, 0, UserAddress::NULL);
            } else if arg2 == SECCOMP_MODE_FILTER as u64 {
                return sys_seccomp(current_task, SECCOMP_SET_MODE_FILTER, 0, arg3.into());
            }
            Ok(().into())
        }
        PR_GET_CHILD_SUBREAPER => {
            let addr = UserAddress::from(arg2);
            #[allow(clippy::bool_to_int_with_if)]
            let value: i32 =
                if current_task.thread_group().read().is_child_subreaper { 1 } else { 0 };
            current_task.write_object(addr.into(), &value)?;
            Ok(().into())
        }
        PR_SET_CHILD_SUBREAPER => {
            current_task.thread_group().write().is_child_subreaper = arg2 != 0;
            Ok(().into())
        }
        PR_GET_SECUREBITS => Ok(current_task.current_creds().securebits.bits().into()),
        PR_SET_SECUREBITS => {
            security::check_task_capable(current_task, CAP_SETPCAP)?;

            // Observed: Linux returns EPERM (not EINVAL) when unrecognized bits are set in flags.
            let flags = u32::try_from(arg2).map_err(|_| errno!(EPERM))?;
            let securebits = SecureBits::from_bits(flags).ok_or_else(|| errno!(EPERM))?;

            let mut creds = Credentials::clone(&current_task.current_creds());
            creds.set_securebits(securebits)?;
            current_task.set_creds(creds);
            Ok(().into())
        }
        PR_CAPBSET_READ => {
            let cap = Capabilities::try_from(arg2)?;
            Ok(current_task.current_creds().cap_bounding.contains(cap).into())
        }
        PR_CAPBSET_DROP => {
            let mut creds = Credentials::clone(&current_task.current_creds());
            security::check_task_capable(current_task, CAP_SETPCAP)?;

            creds.cap_bounding.remove(Capabilities::try_from(arg2)?);
            current_task.set_creds(creds);
            Ok(().into())
        }
        PR_CAP_AMBIENT => {
            let operation = arg2 as u32;
            let capability_arg = Capabilities::try_from(arg3)?;
            if arg4 != 0 || arg5 != 0 {
                return error!(EINVAL);
            }

            // TODO(security): We don't currently validate capabilities, but this should return an
            // error if the capability_arg is invalid.
            match operation {
                PR_CAP_AMBIENT_RAISE => {
                    let mut creds = Credentials::clone(&current_task.current_creds());
                    if !(creds.cap_permitted.contains(capability_arg)
                        && creds.cap_inheritable.contains(capability_arg))
                    {
                        return error!(EPERM);
                    }
                    if creds.securebits.contains(SecureBits::NO_CAP_AMBIENT_RAISE) {
                        return error!(EPERM);
                    }

                    creds.cap_ambient.insert(capability_arg);
                    current_task.set_creds(creds);
                    Ok(().into())
                }
                PR_CAP_AMBIENT_LOWER => {
                    let mut creds = Credentials::clone(&current_task.current_creds());
                    creds.cap_ambient.remove(capability_arg);
                    current_task.set_creds(creds);
                    Ok(().into())
                }
                PR_CAP_AMBIENT_IS_SET => {
                    Ok(current_task.current_creds().cap_ambient.contains(capability_arg).into())
                }
                PR_CAP_AMBIENT_CLEAR_ALL => {
                    if arg3 != 0 {
                        return error!(EINVAL);
                    }

                    let mut creds = Credentials::clone(&current_task.current_creds());
                    creds.cap_ambient = Capabilities::empty();
                    current_task.set_creds(creds);
                    Ok(().into())
                }
                _ => error!(EINVAL),
            }
        }
        PR_SET_TIMERSLACK => {
            current_task.write().set_timerslack_ns(arg2);
            Ok(().into())
        }
        PR_GET_TIMERSLACK => Ok(current_task.read().timerslack_ns.into()),
        #[cfg(target_arch = "aarch64")]
        PR_GET_TAGGED_ADDR_CTRL => {
            track_stub!(TODO("https://fxbug.dev/408554469"), "PR_GET_TAGGED_ADDR_CTRL");
            Ok(0.into())
        }
        #[cfg(target_arch = "aarch64")]
        PR_SET_TAGGED_ADDR_CTRL => match u32::try_from(arg2).map_err(|_| errno!(EINVAL))? {
            // Only untagged pointers are allowed, the default.
            0 => Ok(().into()),
            PR_TAGGED_ADDR_ENABLE => {
                track_stub!(TODO("https://fxbug.dev/408554469"), "PR_TAGGED_ADDR_ENABLE");
                error!(EINVAL)
            }
            unknown_mode => {
                track_stub!(
                    TODO("https://fxbug.dev/408554469"),
                    "PR_SET_TAGGED_ADDR_CTRL unknown mode",
                    unknown_mode,
                );
                error!(EINVAL)
            }
        },
        _ => {
            track_stub!(TODO("https://fxbug.dev/322874733"), "prctl fallthrough", option);
            error!(ENOSYS)
        }
    }
}

pub fn sys_ptrace(
    current_task: &mut CurrentTask,
    request: u32,
    pid: pid_t,
    addr: UserAddress,
    data: UserAddress,
) -> Result<SyscallResult, Errno> {
    if request == PTRACE_TRACEME {
        return ptrace_traceme(current_task);
    }
    let pid = current_task.kernel().pids.get(pid)?;
    match request {
        PTRACE_ATTACH => ptrace_attach(current_task, &pid, PtraceAttachType::Attach, data),
        PTRACE_SEIZE => ptrace_attach(current_task, &pid, PtraceAttachType::Seize, data),
        _ => ptrace_dispatch(current_task, request, &pid, addr, data),
    }
}

pub fn sys_set_tid_address(
    current_task: &CurrentTask,
    user_tid: UserRef<pid_t>,
) -> Result<pid_t, Errno> {
    current_task.write().clear_child_tid = user_tid;
    Ok(current_task.get_tid())
}

pub fn sys_getrusage(
    current_task: &CurrentTask,
    who: i32,
    user_usage: RUsagePtr,
) -> Result<(), Errno> {
    const RUSAGE_SELF: i32 = starnix_uapi::uapi::RUSAGE_SELF as i32;
    const RUSAGE_THREAD: i32 = starnix_uapi::uapi::RUSAGE_THREAD as i32;
    track_stub!(TODO("https://fxbug.dev/297370242"), "real rusage");
    let time_stats = match who {
        RUSAGE_CHILDREN => current_task.task.thread_group().read().children_time_stats,
        RUSAGE_SELF => current_task.task.thread_group().time_stats(),
        RUSAGE_THREAD => current_task.task.time_stats(),
        _ => return error!(EINVAL),
    };

    let usage = rusage {
        ru_utime: timeval_from_duration(time_stats.user_time),
        ru_stime: timeval_from_duration(time_stats.system_time),
        ..rusage::default()
    };
    current_task.write_multi_arch_object(user_usage, usage)?;

    Ok(())
}

type PrLimitRef = MultiArchUserRef<uapi::rlimit, uapi::arch32::rlimit>;

pub fn sys_getrlimit(
    current_task: &CurrentTask,
    resource: u32,
    user_rlimit: PrLimitRef,
) -> Result<(), Errno> {
    do_prlimit64(current_task, 0, resource, PrLimitRef::null(current_task), user_rlimit)
}

pub fn sys_setrlimit(
    current_task: &CurrentTask,
    resource: u32,
    user_rlimit: PrLimitRef,
) -> Result<(), Errno> {
    do_prlimit64(current_task, 0, resource, user_rlimit, PrLimitRef::null(current_task))
}

pub fn sys_prlimit64(
    current_task: &CurrentTask,
    pid: pid_t,
    user_resource: u32,
    new_limit_ref: UserRef<uapi::rlimit>,
    old_limit_ref: UserRef<uapi::rlimit>,
) -> Result<(), Errno> {
    do_prlimit64::<uapi::rlimit>(
        current_task,
        pid,
        user_resource,
        new_limit_ref.into(),
        old_limit_ref.into(),
    )
}

pub fn do_prlimit64<T>(
    current_task: &CurrentTask,
    pid: pid_t,
    user_resource: u32,
    new_limit_ref: MultiArchUserRef<uapi::rlimit, T>,
    old_limit_ref: MultiArchUserRef<uapi::rlimit, T>,
) -> Result<(), Errno>
where
    T: FromBytes + IntoBytes + Immutable + From<uapi::rlimit> + Into<uapi::rlimit>,
{
    let target_task = get_task_or_current(current_task, pid)?;

    // To get or set the resource of a process other than itself, the caller must have either:
    // * the same `uid`, `euid`, `saved_uid`, `gid`, `egid`, `saved_gid` as the target.
    // * the CAP_SYS_RESOURCE
    if current_task.get_pid() != target_task.get_pid() {
        let self_creds = current_task.current_creds();
        let target_creds = target_task.real_creds();
        if self_creds.uid != target_creds.uid
            || self_creds.euid != target_creds.euid
            || self_creds.saved_uid != target_creds.saved_uid
            || self_creds.gid != target_creds.gid
            || self_creds.egid != target_creds.egid
            || self_creds.saved_gid != target_creds.saved_gid
        {
            security::check_task_capable(current_task, CAP_SYS_RESOURCE)?;
        }
        security::task_prlimit(
            current_task,
            &target_task,
            !old_limit_ref.is_null(),
            !new_limit_ref.is_null(),
        )?;
    }

    let resource = Resource::from_raw(user_resource)?;

    let old_limit = match resource {
        // TODO: Integrate Resource::STACK with generic ResourceLimits machinery.
        Resource::STACK => {
            if !new_limit_ref.is_null() {
                track_stub!(
                    TODO("https://fxbug.dev/322874791"),
                    "prlimit64 cannot set RLIMIT_STACK"
                );
            }
            // The stack size is fixed at the moment, but
            // if MAP_GROWSDOWN is implemented this should
            // report the limit that it can be grown.
            let mm = target_task.mm()?;
            let mm_state = mm.state.read();
            let stack_size = mm_state.stack_size as u64;
            rlimit { rlim_cur: stack_size, rlim_max: stack_size }
        }
        _ => {
            let new_limit = if new_limit_ref.is_null() {
                None
            } else {
                let new_limit = current_task.read_multi_arch_object(new_limit_ref)?;
                if new_limit.rlim_cur > new_limit.rlim_max {
                    return error!(EINVAL);
                }
                Some(new_limit)
            };
            ThreadGroup::adjust_rlimits(current_task, &target_task, resource, new_limit)?
        }
    };
    if !old_limit_ref.is_null() {
        current_task.write_multi_arch_object(old_limit_ref, old_limit)?;
    }
    Ok(())
}

pub fn sys_quotactl(
    _current_task: &CurrentTask,
    _cmd: i32,
    _special: UserRef<c_char>,
    _id: i32,
    _addr: UserRef<c_char>,
) -> Result<SyscallResult, Errno> {
    track_stub!(TODO("https://fxbug.dev/297302197"), "quotacl()");
    error!(ENOSYS)
}

pub fn sys_capget(
    current_task: &CurrentTask,
    user_header: UserRef<__user_cap_header_struct>,
    user_data: UserRef<__user_cap_data_struct>,
) -> Result<(), Errno> {
    let mut header = current_task.read_object(user_header)?;
    let is_version_valid =
        [_LINUX_CAPABILITY_VERSION_1, _LINUX_CAPABILITY_VERSION_2, _LINUX_CAPABILITY_VERSION_3]
            .contains(&header.version);
    if !is_version_valid {
        header.version = _LINUX_CAPABILITY_VERSION_3;
        current_task.write_object(user_header, &header)?;
    }
    if user_data.is_null() {
        return Ok(());
    }
    if !is_version_valid || header.pid < 0 {
        return error!(EINVAL);
    }

    let target_task = get_task_or_current(current_task, header.pid)?;

    security::check_getcap_access(current_task, &target_task)?;

    let (permitted, effective, inheritable) = {
        let creds = &target_task.real_creds();
        (creds.cap_permitted, creds.cap_effective, creds.cap_inheritable)
    };

    match header.version {
        _LINUX_CAPABILITY_VERSION_1 => {
            let data: [__user_cap_data_struct; 1] = [__user_cap_data_struct {
                effective: effective.as_abi_v1(),
                inheritable: inheritable.as_abi_v1(),
                permitted: permitted.as_abi_v1(),
            }];
            current_task.write_objects(user_data, &data)?;
        }
        _LINUX_CAPABILITY_VERSION_2 | _LINUX_CAPABILITY_VERSION_3 => {
            // Return 64 bit capabilities as two sets of 32 bit capabilities, little endian
            let (permitted, effective, inheritable) =
                (permitted.as_abi_v3(), effective.as_abi_v3(), inheritable.as_abi_v3());
            let data: [__user_cap_data_struct; 2] = [
                __user_cap_data_struct {
                    effective: effective.0,
                    inheritable: inheritable.0,
                    permitted: permitted.0,
                },
                __user_cap_data_struct {
                    effective: effective.1,
                    inheritable: inheritable.1,
                    permitted: permitted.1,
                },
            ];
            current_task.write_objects(user_data, &data)?;
        }
        _ => {
            unreachable!("already returned if Linux capability version is not valid")
        }
    }
    Ok(())
}

pub fn sys_capset(
    current_task: &CurrentTask,
    user_header: UserRef<__user_cap_header_struct>,
    user_data: UserRef<__user_cap_data_struct>,
) -> Result<(), Errno> {
    let mut header = current_task.read_object(user_header)?;
    let is_version_valid =
        [_LINUX_CAPABILITY_VERSION_1, _LINUX_CAPABILITY_VERSION_2, _LINUX_CAPABILITY_VERSION_3]
            .contains(&header.version);
    if !is_version_valid {
        header.version = _LINUX_CAPABILITY_VERSION_3;
        current_task.write_object(user_header, &header)?;
        return error!(EINVAL);
    }
    if header.pid != 0 && header.pid != current_task.tid.id {
        return error!(EPERM);
    }

    let (new_permitted, new_effective, new_inheritable) = match header.version {
        _LINUX_CAPABILITY_VERSION_1 => {
            let data = current_task.read_object(user_data)?;
            (
                Capabilities::from_abi_v1(data.permitted),
                Capabilities::from_abi_v1(data.effective),
                Capabilities::from_abi_v1(data.inheritable),
            )
        }
        _LINUX_CAPABILITY_VERSION_2 | _LINUX_CAPABILITY_VERSION_3 => {
            let data =
                current_task.read_objects_to_array::<__user_cap_data_struct, 2>(user_data)?;
            (
                Capabilities::from_abi_v3((data[0].permitted, data[1].permitted)),
                Capabilities::from_abi_v3((data[0].effective, data[1].effective)),
                Capabilities::from_abi_v3((data[0].inheritable, data[1].inheritable)),
            )
        }
        _ => {
            unreachable!("already returned if Linux capability version is not valid")
        }
    };

    // Permission checks. Copied out of TLPI section 39.7.
    let mut creds = Credentials::clone(&current_task.current_creds());
    {
        log_trace!(
            "Capabilities({{permitted={:?} from {:?}, effective={:?} from {:?}, inheritable={:?} from {:?}}}, bounding={:?})",
            new_permitted,
            creds.cap_permitted,
            new_effective,
            creds.cap_effective,
            new_inheritable,
            creds.cap_inheritable,
            creds.cap_bounding
        );
        if !creds.cap_inheritable.union(creds.cap_permitted).contains(new_inheritable) {
            security::check_task_capable(current_task, CAP_SETPCAP)?;
        }

        if !creds.cap_inheritable.union(creds.cap_bounding).contains(new_inheritable) {
            return error!(EPERM);
        }
        if !creds.cap_permitted.contains(new_permitted) {
            return error!(EPERM);
        }
        if !new_permitted.contains(new_effective) {
            return error!(EPERM);
        }
    }
    let target_task = get_task_or_current(current_task, header.pid)?;

    security::check_setcap_access(current_task, &target_task)?;

    creds.cap_permitted = new_permitted;
    creds.cap_effective = new_effective;
    creds.cap_inheritable = new_inheritable;
    creds.cap_ambient = new_permitted & new_inheritable & creds.cap_ambient;
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_seccomp(
    current_task: &mut CurrentTask,
    operation: u32,
    flags: u32,
    args: UserAddress,
) -> Result<SyscallResult, Errno> {
    match operation {
        SECCOMP_SET_MODE_STRICT => {
            if flags != 0 || args != UserAddress::NULL {
                return error!(EINVAL);
            }
            current_task.set_seccomp_state(SeccompStateValue::Strict)?;
            Ok(().into())
        }
        SECCOMP_SET_MODE_FILTER => {
            if flags
                & (SECCOMP_FILTER_FLAG_LOG
                    | SECCOMP_FILTER_FLAG_NEW_LISTENER
                    | SECCOMP_FILTER_FLAG_SPEC_ALLOW
                    | SECCOMP_FILTER_FLAG_TSYNC
                    | SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
                != flags
            {
                return error!(EINVAL);
            }
            if (flags & SECCOMP_FILTER_FLAG_TSYNC == 0)
                && (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH != 0)
            {
                return error!(EINVAL);
            }
            if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER != 0)
                && (flags & SECCOMP_FILTER_FLAG_TSYNC != 0)
                && (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH == 0)
            {
                return error!(EINVAL);
            }
            let fprog =
                current_task.read_multi_arch_object(SockFProgPtr::new(current_task, args))?;
            if fprog.len > BPF_MAXINSNS || fprog.len == 0 {
                return error!(EINVAL);
            }
            let code: Vec<sock_filter> =
                current_task.read_multi_arch_objects_to_vec(fprog.filter, fprog.len as usize)?;

            if !current_task.read().no_new_privs() {
                security::check_task_capable(current_task, CAP_SYS_ADMIN)
                    .map_err(|_| errno!(EACCES))?;
            }
            current_task.add_seccomp_filter(code, flags)
        }
        SECCOMP_GET_ACTION_AVAIL => {
            if flags != 0 || args.is_null() {
                return error!(EINVAL);
            }
            let action: u32 = current_task.read_object(UserRef::new(args))?;
            SeccompAction::is_action_available(action)
        }
        SECCOMP_GET_NOTIF_SIZES => {
            if flags != 0 {
                return error!(EINVAL);
            }
            track_stub!(TODO("https://fxbug.dev/322874791"), "SECCOMP_GET_NOTIF_SIZES");
            error!(ENOSYS)
        }
        _ => {
            track_stub!(TODO("https://fxbug.dev/322874916"), "seccomp fallthrough", operation);
            error!(EINVAL)
        }
    }
}

pub fn sys_setgroups(
    current_task: &CurrentTask,
    size: usize,
    groups_addr: UserAddress,
) -> Result<(), Errno> {
    if size > NGROUPS_MAX as usize {
        return error!(EINVAL);
    }
    let groups = current_task.read_objects_to_vec::<gid_t>(groups_addr.into(), size)?;
    security::check_task_capable(current_task, CAP_SETGID)?;
    let mut creds = Credentials::clone(&current_task.current_creds());
    creds.groups = groups;
    current_task.set_creds(creds);
    Ok(())
}

pub fn sys_getgroups(
    current_task: &CurrentTask,
    size: usize,
    groups_addr: UserAddress,
) -> Result<usize, Errno> {
    if size > NGROUPS_MAX as usize {
        return error!(EINVAL);
    }
    let creds = current_task.current_creds();
    if size != 0 {
        if size < creds.groups.len() {
            return error!(EINVAL);
        }
        current_task.write_memory(groups_addr, creds.groups.as_slice().as_bytes())?;
    }
    Ok(creds.groups.len())
}

pub fn sys_setsid(current_task: &CurrentTask) -> Result<pid_t, Errno> {
    current_task.thread_group().setsid()?;
    Ok(current_task.get_pid())
}

// Note the asymmetry with sys_setpriority: this returns "kernel nice" which ranges
// from 1 (weakest) to 40 (strongest). (It is part of Linux history that this syscall
// deals with niceness but has "priority" in its name.)
pub fn sys_getpriority(current_task: &CurrentTask, which: u32, who: i32) -> Result<u8, Errno> {
    match which {
        PRIO_PROCESS => {}
        // TODO: https://fxbug.dev/287121196 - support PRIO_PGRP and PRIO_USER?
        _ => return error!(EINVAL),
    }
    track_stub!(TODO("https://fxbug.dev/322893809"), "getpriority permissions");
    let target_task = get_task_or_current(current_task, who)?;
    let state = target_task.read();
    Ok(state.scheduler_state.normal_priority.raw_priority())
}

// Note the asymmetry with sys_getpriority: this call's `priority` parameter is a
// "user nice" which ranges from -20 (strongest) to 19 (weakest) (other values can be
// passed and are clamped to that range and interpretation). (It is part of Linux
// history that this syscall deals with niceness but has "priority" in its name.)
pub fn sys_setpriority(
    current_task: &CurrentTask,
    which: u32,
    who: i32,
    priority: i32,
) -> Result<(), Errno> {
    // Parse & validate the arguments.
    match which {
        PRIO_PROCESS => {}
        // TODO: https://fxbug.dev/287121196 - support PRIO_PGRP and PRIO_USER?
        _ => return error!(EINVAL),
    }

    let target_task = get_task_or_current(current_task, who)?;

    let normal_priority = NormalPriority::from_setpriority_syscall(priority);

    // TODO: https://fxbug.dev/425143440 - we probably want to improve the locking here.
    let current_state = target_task.read().scheduler_state.clone();

    // Check capabilities and permissions, if required, for the operation.
    let euid_friendly = current_task.is_euid_friendly_with(&target_task);
    let strengthening = current_state.normal_priority < normal_priority;
    let rlimited = strengthening
        && normal_priority.exceeds(target_task.thread_group().get_rlimit(Resource::NICE));
    if !euid_friendly {
        security::check_task_capable(current_task, CAP_SYS_NICE)?;
    } else if rlimited {
        security::check_task_capable(current_task, CAP_SYS_NICE).map_err(|_| errno!(EACCES))?;
    }

    security::check_task_setnice_access(current_task, &target_task)?;

    // Apply the new scheduler configuration to the task.
    target_task.set_scheduler_nice(normal_priority)?;

    Ok(())
}

pub fn sys_setns(current_task: &CurrentTask, ns_fd: FdNumber, ns_type: c_int) -> Result<(), Errno> {
    let file_handle = current_task.files().get(ns_fd)?;

    // From man pages this is not quite right because some namespace types require more capabilities
    // or require this capability in multiple namespaces, but it should cover our current test
    // cases and we can make this more nuanced once more namespace types are supported.
    security::check_task_capable(current_task, CAP_SYS_ADMIN)?;

    if let Some(mount_ns) = file_handle.downcast_file::<MountNamespaceFile>() {
        if !(ns_type == 0 || ns_type == CLONE_NEWNS as i32) {
            log_trace!("invalid type");
            return error!(EINVAL);
        }

        track_stub!(TODO("https://fxbug.dev/297312091"), "setns CLONE_FS limitations");
        current_task.fs().set_namespace(mount_ns.0.clone())?;
        return Ok(());
    }

    if let Some(_pidfd) = file_handle.downcast_file::<PidFdFileObject>() {
        track_stub!(TODO("https://fxbug.dev/297312844"), "setns w/ pidfd");
        return error!(ENOSYS);
    }

    track_stub!(TODO("https://fxbug.dev/322893829"), "unknown ns file for setns, see logs");
    log_info!("ns_fd was not a supported namespace file: {}", file_handle.ops_type_name());
    error!(EINVAL)
}

pub fn sys_unshare(current_task: &CurrentTask, flags: u32) -> Result<(), Errno> {
    const IMPLEMENTED_FLAGS: u32 = CLONE_FILES | CLONE_FS | CLONE_NEWNS | CLONE_NEWUTS;
    if flags & !IMPLEMENTED_FLAGS != 0 {
        track_stub!(TODO("https://fxbug.dev/322893372"), "unshare", flags & !IMPLEMENTED_FLAGS);
        return error!(EINVAL);
    }

    if (flags & CLONE_FILES) != 0 {
        current_task.running_state().unshare_files(current_task);
    }

    if (flags & CLONE_FS) != 0 {
        current_task.unshare_fs();
    }

    if (flags & CLONE_NEWNS) != 0 {
        security::check_task_capable(current_task, CAP_SYS_ADMIN)?;
        current_task.fs().unshare_namespace();
    }

    if (flags & CLONE_NEWUTS) != 0 {
        security::check_task_capable(current_task, CAP_SYS_ADMIN)?;
        // Fork the UTS namespace.
        let mut task_state = current_task.write();
        let new_uts_ns = task_state.uts_ns.read().clone();
        task_state.uts_ns = Arc::new(new_uts_ns.into());
    }

    Ok(())
}

pub fn sys_swapon(
    current_task: &CurrentTask,
    user_path: UserCString,
    flags: u32,
) -> Result<(), Errno> {
    const MAX_SWAPFILES: usize = 32; // See https://man7.org/linux/man-pages/man2/swapon.2.html
    const SWAP_FLAGS_VALID: u32 = 0x7ffff; // Defined in include/linux/swap.h

    if (flags & !SWAP_FLAGS_VALID) != 0 {
        return error!(EINVAL);
    }

    security::check_task_capable(current_task, CAP_SYS_ADMIN)?;

    if flags != 0 {
        track_stub!(TODO("https://fxbug.dev/322893905"), "swapon flags", flags);
    }

    let path = current_task.read_path(user_path)?;
    let file = current_task.open_file(path.as_ref(), OpenFlags::RDWR)?;

    let node = file.node();
    let mode = node.info().mode;
    if !mode.is_reg() && !mode.is_blk() {
        return error!(EINVAL);
    }

    // We determined this magic number by using the mkswap tool and the file tool. The mkswap tool
    // populates a few bytes in the file, including a UUID, which can be replaced with zeros while
    // still being recognized by the file tool. This string appears at a fixed offset
    // (MAGIC_OFFSET) in the file, which looks quite like a magic number.
    const MAGIC_OFFSET: usize = 0xff6;
    let swap_magic = b"SWAPSPACE2";
    let mut buffer = VecOutputBuffer::new(swap_magic.len());
    if file.read_at(current_task, MAGIC_OFFSET, &mut buffer)? != swap_magic.len()
        || buffer.data() != swap_magic
    {
        return error!(EINVAL);
    }

    let mut swap_files = current_task.kernel().swap_files.lock();
    for swap_node in swap_files.iter() {
        if Arc::ptr_eq(swap_node, node) {
            return error!(EBUSY);
        }
    }
    if swap_files.len() >= MAX_SWAPFILES {
        return error!(EPERM);
    }
    swap_files.push(node.clone());
    Ok(())
}

pub fn sys_swapoff(current_task: &CurrentTask, user_path: UserCString) -> Result<(), Errno> {
    security::check_task_capable(current_task, CAP_SYS_ADMIN)?;

    let path = current_task.read_path(user_path)?;
    let file = current_task.open_file(path.as_ref(), OpenFlags::RDWR)?;
    let node = file.node();

    let mut swap_files = current_task.kernel().swap_files.lock();
    let original_length = swap_files.len();
    swap_files.retain(|swap_node| !Arc::ptr_eq(swap_node, node));
    if swap_files.len() == original_length {
        return error!(EINVAL);
    }
    Ok(())
}

#[derive(Default, Debug, IntoBytes, KnownLayout, FromBytes, Immutable)]
#[repr(C)]
struct KcmpParams {
    mask: usize,
    shuffle: usize,
}

static KCMP_PARAMS: LazyLock<KcmpParams> = LazyLock::new(|| {
    let mut params = KcmpParams::default();
    starnix_crypto::cprng_draw(params.as_mut_bytes());
    // Ensure the shuffle is odd so that multiplying a usize by this value is a permutation.
    params.shuffle |= 1;
    params
});

fn obfuscate_value(value: usize) -> usize {
    let KcmpParams { mask, shuffle } = *KCMP_PARAMS;
    (value ^ mask).wrapping_mul(shuffle)
}

fn obfuscate_ptr<T>(ptr: *const T) -> usize {
    obfuscate_value(ptr as usize)
}

fn obfuscate_arc<T>(arc: &Arc<T>) -> usize {
    obfuscate_ptr(Arc::as_ptr(arc))
}

pub fn sys_kcmp(
    current_task: &CurrentTask,
    pid1: pid_t,
    pid2: pid_t,
    resource_type: u32,
    index1: u64,
    index2: u64,
) -> Result<u32, Errno> {
    let task1 = current_task.get_task(pid1)?;
    let task2 = current_task.get_task(pid2)?;

    current_task.check_ptrace_access_mode(PTRACE_MODE_READ_REALCREDS, &task1)?;
    current_task.check_ptrace_access_mode(PTRACE_MODE_READ_REALCREDS, &task2)?;

    let resource_type = KcmpResource::from_raw(resource_type)?;

    // Output encoding (see <https://man7.org/linux/man-pages/man2/kcmp.2.html>):
    //
    //   0  v1 is equal to v2; in other words, the two processes share the resource.
    //   1  v1 is less than v2.
    //   2  v1 is greater than v2.
    //   3  v1 is not equal to v2, but ordering information is unavailable.
    //
    fn encode_ordering(value: cmp::Ordering) -> u32 {
        match value {
            cmp::Ordering::Equal => 0,
            cmp::Ordering::Less => 1,
            cmp::Ordering::Greater => 2,
        }
    }

    match resource_type {
        KcmpResource::FILE => {
            fn get_file(task: &Task, index: u64) -> Result<FileHandle, Errno> {
                // TODO: Test whether O_PATH is allowed here. Conceptually, seems like
                //       O_PATH should be allowed, but we haven't tested it yet.
                task.files()?.get_allowing_opath(FdNumber::from_raw(
                    index.try_into().map_err(|_| errno!(EBADF))?,
                ))
            }
            let file1 = get_file(&task1, index1)?;
            let file2 = get_file(&task2, index2)?;
            Ok(encode_ordering(obfuscate_arc(&file1).cmp(&obfuscate_arc(&file2))))
        }
        KcmpResource::FILES => {
            let files1 = task1.files()?.id();
            let files2 = task2.files()?.id();
            Ok(encode_ordering(obfuscate_value(files1.raw()).cmp(&obfuscate_value(files2.raw()))))
        }
        KcmpResource::FS => {
            let fs1 = task1.fs()?;
            let fs2 = task2.fs()?;
            Ok(encode_ordering(obfuscate_arc(&fs1).cmp(&obfuscate_arc(&fs2))))
        }
        KcmpResource::SIGHAND => Ok(encode_ordering(
            obfuscate_arc(&task1.thread_group().signal_actions)
                .cmp(&obfuscate_arc(&task2.thread_group().signal_actions)),
        )),
        KcmpResource::VM => {
            Ok(encode_ordering(obfuscate_arc(&task1.mm()?).cmp(&obfuscate_arc(&task2.mm()?))))
        }
        KcmpResource::SYSVSEM => error!(EOPNOTSUPP),
        _ => error!(EINVAL),
    }
}

pub fn sys_syslog(
    current_task: &CurrentTask,
    action_type: i32,
    address: UserAddress,
    length: i32,
) -> Result<i32, Errno> {
    let action = SyslogAction::try_from(action_type)?;
    let syslog =
        current_task.kernel().syslog.access(&current_task, SyslogAccess::Syscall(action))?;
    match action {
        SyslogAction::Read => {
            if address.is_null() || length < 0 {
                return error!(EINVAL);
            }
            let mut output_buffer =
                UserBuffersOutputBuffer::unified_new_at(current_task, address, length as usize)?;
            syslog.blocking_read(current_task, &mut output_buffer)
        }
        SyslogAction::ReadAll => {
            if address.is_null() || length < 0 {
                return error!(EINVAL);
            }
            let mut output_buffer =
                UserBuffersOutputBuffer::unified_new_at(current_task, address, length as usize)?;
            syslog.read_all(current_task.kernel(), &mut output_buffer)
        }
        SyslogAction::SizeUnread => syslog.size_unread(),
        SyslogAction::SizeBuffer => syslog.size_buffer(),
        SyslogAction::Close | SyslogAction::Open => Ok(0),
        SyslogAction::ReadClear => {
            track_stub!(TODO("https://fxbug.dev/322894145"), "syslog: read clear");
            Ok(0)
        }
        SyslogAction::Clear => {
            track_stub!(TODO("https://fxbug.dev/322893673"), "syslog: clear");
            Ok(0)
        }
        SyslogAction::ConsoleOff => {
            track_stub!(TODO("https://fxbug.dev/322894399"), "syslog: console off");
            Ok(0)
        }
        SyslogAction::ConsoleOn => {
            track_stub!(TODO("https://fxbug.dev/322894106"), "syslog: console on");
            Ok(0)
        }
        SyslogAction::ConsoleLevel => {
            if length <= 0 || length >= 8 {
                return error!(EINVAL);
            }
            track_stub!(TODO("https://fxbug.dev/322894199"), "syslog: console level");
            Ok(0)
        }
    }
}

pub fn sys_vhangup(current_task: &CurrentTask) -> Result<(), Errno> {
    security::check_task_capable(current_task, CAP_SYS_TTY_CONFIG)?;
    track_stub!(TODO("https://fxbug.dev/324079257"), "vhangup");
    Ok(())
}

// Syscalls for arch32 usage
#[cfg(target_arch = "aarch64")]
mod arch32 {
    pub use super::{
        sys_execve as sys_arch32_execve, sys_execveat as sys_arch32_execveat,
        sys_getegid as sys_arch32_getegid32, sys_geteuid as sys_arch32_geteuid32,
        sys_getgid as sys_arch32_getgid32, sys_getgroups as sys_arch32_getgroups32,
        sys_getpgid as sys_arch32_getpgid, sys_getppid as sys_arch32_getppid,
        sys_getpriority as sys_arch32_getpriority, sys_getresgid as sys_arch32_getresgid32,
        sys_getresuid as sys_arch32_getresuid32, sys_getrlimit as sys_arch32_ugetrlimit,
        sys_getrusage as sys_arch32_getrusage, sys_getuid as sys_arch32_getuid32,
        sys_ioprio_set as sys_arch32_ioprio_set, sys_kcmp as sys_arch32_kcmp,
        sys_ptrace as sys_arch32_ptrace, sys_quotactl as sys_arch32_quotactl,
        sys_sched_get_priority_max as sys_arch32_sched_get_priority_max,
        sys_sched_get_priority_min as sys_arch32_sched_get_priority_min,
        sys_sched_getaffinity as sys_arch32_sched_getaffinity,
        sys_sched_getparam as sys_arch32_sched_getparam,
        sys_sched_setaffinity as sys_arch32_sched_setaffinity,
        sys_sched_setparam as sys_arch32_sched_setparam,
        sys_sched_setscheduler as sys_arch32_sched_setscheduler, sys_seccomp as sys_arch32_seccomp,
        sys_setfsgid as sys_arch32_setfsgid, sys_setfsgid as sys_arch32_setfsgid32,
        sys_setfsuid as sys_arch32_setfsuid, sys_setfsuid as sys_arch32_setfsuid32,
        sys_setgid as sys_arch32_setgid32, sys_setgroups as sys_arch32_setgroups32,
        sys_setns as sys_arch32_setns, sys_setpgid as sys_arch32_setpgid,
        sys_setpriority as sys_arch32_setpriority, sys_setregid as sys_arch32_setregid32,
        sys_setresgid as sys_arch32_setresgid32, sys_setresuid as sys_arch32_setresuid32,
        sys_setreuid as sys_arch32_setreuid32, sys_setreuid as sys_arch32_setreuid,
        sys_setrlimit as sys_arch32_setrlimit, sys_setsid as sys_arch32_setsid,
        sys_syslog as sys_arch32_syslog, sys_unshare as sys_arch32_unshare,
    };
}

#[cfg(target_arch = "aarch64")]
pub use arch32::*;

#[cfg(test)]
mod tests {
    use super::*;
    use crate::mm::syscalls::sys_munmap;
    use crate::testing::{AutoReleasableTask, map_memory, spawn_kernel_and_run};
    use starnix_syscalls::SUCCESS;
    use starnix_task_command::TaskCommand;
    use starnix_uapi::auth::Credentials;
    use starnix_uapi::signals::SIGCHLD;
    use starnix_uapi::uapi::{kcmp_type_KCMP_SYSVSEM, kcmp_type_KCMP_TYPES, kcmp_type_KCMP_VM};
    use starnix_uapi::{SCHED_FIFO, SCHED_NORMAL};
    use std::ffi::CString;

    #[::fuchsia::test]
    async fn test_prctl_set_vma_anon_name() {
        spawn_kernel_and_run(async |current_task| {
            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            let name_addr = (mapped_address + 128u64).unwrap();
            let name = "test-name\0";
            current_task.write_memory(name_addr, name.as_bytes()).expect("failed to write name");
            sys_prctl(
                current_task,
                PR_SET_VMA,
                PR_SET_VMA_ANON_NAME as u64,
                mapped_address.ptr() as u64,
                32,
                name_addr.ptr() as u64,
            )
            .expect("failed to set name");
            assert_eq!(
                "test-name",
                current_task
                    .mm()
                    .unwrap()
                    .get_mapping_name((mapped_address + 24u64).unwrap())
                    .expect("failed to get address")
                    .unwrap()
                    .to_string(),
            );

            sys_munmap(&current_task, mapped_address, *PAGE_SIZE as usize)
                .expect("failed to unmap memory");
            assert_eq!(
                error!(EFAULT),
                current_task.mm().unwrap().get_mapping_name((mapped_address + 24u64).unwrap())
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_set_vma_name_special_chars() {
        spawn_kernel_and_run(async |current_task| {
            let name_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);

            let mapping_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);

            for c in 1..255 {
                let vma_name = CString::new([c]).unwrap();
                current_task.write_memory(name_addr, vma_name.as_bytes_with_nul()).unwrap();

                let result = sys_prctl(
                    current_task,
                    PR_SET_VMA,
                    PR_SET_VMA_ANON_NAME as u64,
                    mapping_addr.ptr() as u64,
                    *PAGE_SIZE,
                    name_addr.ptr() as u64,
                );

                if c > 0x1f
                    && c < 0x7f
                    && c != b'\\'
                    && c != b'`'
                    && c != b'$'
                    && c != b'['
                    && c != b']'
                {
                    assert_eq!(result, Ok(SUCCESS));
                } else {
                    assert_eq!(result, error!(EINVAL));
                }
            }
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_set_vma_name_long() {
        spawn_kernel_and_run(async |current_task| {
            let name_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);

            let mapping_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);

            let name_too_long = CString::new(vec![b'a'; 256]).unwrap();

            current_task.write_memory(name_addr, name_too_long.as_bytes_with_nul()).unwrap();

            assert_eq!(
                sys_prctl(
                    current_task,
                    PR_SET_VMA,
                    PR_SET_VMA_ANON_NAME as u64,
                    mapping_addr.ptr() as u64,
                    *PAGE_SIZE,
                    name_addr.ptr() as u64,
                ),
                error!(EINVAL)
            );

            let name_just_long_enough = CString::new(vec![b'a'; 255]).unwrap();

            current_task
                .write_memory(name_addr, name_just_long_enough.as_bytes_with_nul())
                .unwrap();

            assert_eq!(
                sys_prctl(
                    current_task,
                    PR_SET_VMA,
                    PR_SET_VMA_ANON_NAME as u64,
                    mapping_addr.ptr() as u64,
                    *PAGE_SIZE,
                    name_addr.ptr() as u64,
                ),
                Ok(SUCCESS)
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_set_vma_name_misaligned() {
        spawn_kernel_and_run(async |current_task| {
            let name_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);

            let mapping_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);

            let name = CString::new("name").unwrap();
            current_task.write_memory(name_addr, name.as_bytes_with_nul()).unwrap();

            // Passing a misaligned pointer to the start of the named region fails.
            assert_eq!(
                sys_prctl(
                    current_task,
                    PR_SET_VMA,
                    PR_SET_VMA_ANON_NAME as u64,
                    1 + mapping_addr.ptr() as u64,
                    *PAGE_SIZE - 1,
                    name_addr.ptr() as u64,
                ),
                error!(EINVAL)
            );

            // Passing an unaligned length does work, however.
            assert_eq!(
                sys_prctl(
                    current_task,
                    PR_SET_VMA,
                    PR_SET_VMA_ANON_NAME as u64,
                    mapping_addr.ptr() as u64,
                    *PAGE_SIZE - 1,
                    name_addr.ptr() as u64,
                ),
                Ok(SUCCESS)
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_prctl_get_set_dumpable() {
        spawn_kernel_and_run(async |current_task| {
            assert_eq!(sys_prctl(current_task, PR_GET_DUMPABLE, 0, 0, 0, 0), Ok(1.into()));

            sys_prctl(current_task, PR_SET_DUMPABLE, 0, 0, 0, 0).expect("failed to set dumpable");
            assert_eq!(sys_prctl(current_task, PR_GET_DUMPABLE, 0, 0, 0, 0), Ok(0.into()));

            sys_prctl(current_task, PR_SET_DUMPABLE, 1, 0, 0, 0).expect("failed to set dumpable");
            assert_eq!(sys_prctl(current_task, PR_GET_DUMPABLE, 0, 0, 0, 0), Ok(1.into()));

            // SUID_DUMP_ROOT not supported.
            assert_eq!(sys_prctl(current_task, PR_SET_DUMPABLE, 2, 0, 0, 0), error!(EINVAL));
            assert_eq!(sys_prctl(current_task, PR_GET_DUMPABLE, 0, 0, 0, 0), Ok(1.into()));
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_sys_getsid() {
        spawn_kernel_and_run(async |current_task| {
            let kernel = current_task.kernel();
            assert_eq!(
                current_task.get_tid(),
                sys_getsid(&current_task, 0).expect("failed to get sid")
            );

            let second_task = crate::execution::create_init_child_process(
                &kernel.weak_self.upgrade().unwrap(),
                TaskCommand::new(b"second task"),
                Credentials::with_ids(0, 0),
                None,
            )
            .expect("failed to create second task");
            let second_current = AutoReleasableTask::from(second_task);

            assert_eq!(
                second_current.get_tid(),
                sys_getsid(&current_task, second_current.get_tid()).expect("failed to get sid")
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_get_affinity_size() {
        spawn_kernel_and_run(async |current_task| {
            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            let pid = current_task.get_pid();
            assert_eq!(sys_sched_getaffinity(&current_task, pid, 16, mapped_address), Ok(16));
            assert_eq!(
                sys_sched_getaffinity(&current_task, pid, 1024, mapped_address),
                Ok(std::mem::size_of::<CpuSet>())
            );
            assert_eq!(
                sys_sched_getaffinity(&current_task, pid, 1, mapped_address),
                error!(EINVAL)
            );
            assert_eq!(
                sys_sched_getaffinity(&current_task, pid, 9, mapped_address),
                error!(EINVAL)
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_set_affinity_size() {
        spawn_kernel_and_run(async |current_task| {
            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            current_task.write_memory(mapped_address, &[0xffu8]).expect("failed to cpumask");
            let pid = current_task.get_pid();
            assert_eq!(
                sys_sched_setaffinity(&current_task, pid, *PAGE_SIZE as u32, mapped_address),
                Ok(())
            );
            assert_eq!(
                sys_sched_setaffinity(&current_task, pid, 1, mapped_address),
                error!(EINVAL)
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_task_name() {
        spawn_kernel_and_run(async |current_task| {
            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            let name = "my-task-name\0";
            current_task
                .write_memory(mapped_address, name.as_bytes())
                .expect("failed to write name");

            let result =
                sys_prctl(current_task, PR_SET_NAME, mapped_address.ptr() as u64, 0, 0, 0).unwrap();
            assert_eq!(SUCCESS, result);

            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            let result =
                sys_prctl(current_task, PR_GET_NAME, mapped_address.ptr() as u64, 0, 0, 0).unwrap();
            assert_eq!(SUCCESS, result);

            let name_length = name.len();

            let out_name = current_task.read_memory_to_vec(mapped_address, name_length).unwrap();
            assert_eq!(name.as_bytes(), &out_name);
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_sched_get_priority_min_max() {
        spawn_kernel_and_run(async |current_task| {
            let non_rt_min = sys_sched_get_priority_min(&current_task, SCHED_NORMAL).unwrap();
            assert_eq!(non_rt_min, 0);
            let non_rt_max = sys_sched_get_priority_max(&current_task, SCHED_NORMAL).unwrap();
            assert_eq!(non_rt_max, 0);

            let rt_min = sys_sched_get_priority_min(&current_task, SCHED_FIFO).unwrap();
            assert_eq!(rt_min, 1);
            let rt_max = sys_sched_get_priority_max(&current_task, SCHED_FIFO).unwrap();
            assert_eq!(rt_max, 99);

            let min_bad_policy_error =
                sys_sched_get_priority_min(&current_task, u32::MAX).unwrap_err();
            assert_eq!(min_bad_policy_error, errno!(EINVAL));

            let max_bad_policy_error =
                sys_sched_get_priority_max(&current_task, u32::MAX).unwrap_err();
            assert_eq!(max_bad_policy_error, errno!(EINVAL));
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_sched_setscheduler() {
        spawn_kernel_and_run(async |current_task| {
            current_task
                .thread_group()
                .limits
                .lock()
                .set(Resource::RTPRIO, rlimit { rlim_cur: 255, rlim_max: 255 });

            let scheduler = sys_sched_getscheduler(&current_task, 0).unwrap();
            assert_eq!(scheduler, SCHED_NORMAL, "tasks should have normal scheduler by default");

            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            let requested_params = sched_param { sched_priority: 15 };
            current_task.write_object(mapped_address.into(), &requested_params).unwrap();

            sys_sched_setscheduler(&current_task, 0, SCHED_FIFO, mapped_address.into()).unwrap();

            let new_scheduler = sys_sched_getscheduler(&current_task, 0).unwrap();
            assert_eq!(new_scheduler, SCHED_FIFO, "task should have been assigned fifo scheduler");

            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            sys_sched_getparam(&current_task, 0, mapped_address.into()).expect("sched_getparam");
            let param_value: sched_param =
                current_task.read_object(mapped_address.into()).expect("read_object");
            assert_eq!(param_value.sched_priority, 15);
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_sched_getparam() {
        spawn_kernel_and_run(async |current_task| {
            let mapped_address = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            sys_sched_getparam(&current_task, 0, mapped_address.into()).expect("sched_getparam");
            let param_value: sched_param =
                current_task.read_object(mapped_address.into()).expect("read_object");
            assert_eq!(param_value.sched_priority, 0);
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_setuid() {
        spawn_kernel_and_run(async |current_task| {
            // Test for root.
            current_task.set_creds(Credentials::with_ids(0, 0));
            sys_setuid(&current_task, 42).expect("setuid");
            let mut creds = Credentials::clone(&current_task.current_creds());
            assert_eq!(creds.euid, 42);
            assert_eq!(creds.uid, 42);
            assert_eq!(creds.saved_uid, 42);

            // Remove the CAP_SETUID capability to avoid overwriting permission checks.
            creds.cap_effective.remove(CAP_SETUID);
            current_task.set_creds(creds);

            // Test for non root, which task now is.
            assert_eq!(sys_setuid(&current_task, 0), error!(EPERM));
            assert_eq!(sys_setuid(&current_task, 43), error!(EPERM));

            sys_setuid(&current_task, 42).expect("setuid");
            assert_eq!(current_task.current_creds().euid, 42);
            assert_eq!(current_task.current_creds().uid, 42);
            assert_eq!(current_task.current_creds().saved_uid, 42);

            // Change uid and saved_uid, and check that one can set the euid to these.
            let mut creds = Credentials::clone(&current_task.current_creds());
            creds.uid = 41;
            creds.euid = 42;
            creds.saved_uid = 43;
            current_task.set_creds(creds);

            sys_setuid(&current_task, 41).expect("setuid");
            assert_eq!(current_task.current_creds().euid, 41);
            assert_eq!(current_task.current_creds().uid, 41);
            assert_eq!(current_task.current_creds().saved_uid, 43);

            let mut creds = Credentials::clone(&current_task.current_creds());
            creds.uid = 41;
            creds.euid = 42;
            creds.saved_uid = 43;
            current_task.set_creds(creds);

            sys_setuid(&current_task, 43).expect("setuid");
            assert_eq!(current_task.current_creds().euid, 43);
            assert_eq!(current_task.current_creds().uid, 41);
            assert_eq!(current_task.current_creds().saved_uid, 43);
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_read_c_string_vector() {
        spawn_kernel_and_run(async |current_task| {
            let arg_addr = map_memory(&current_task, UserAddress::default(), *PAGE_SIZE);
            let arg = b"test-arg\0";
            current_task.write_memory(arg_addr, arg).expect("failed to write test arg");
            let arg_usercstr = UserCString::new(current_task, arg_addr);
            let null_usercstr = UserCString::null(current_task);

            let argv_addr = UserCStringPtr::new(
                current_task,
                map_memory(&current_task, UserAddress::default(), *PAGE_SIZE),
            );
            current_task
                .write_multi_arch_ptr(argv_addr.addr(), arg_usercstr)
                .expect("failed to write UserCString");
            current_task
                .write_multi_arch_ptr(argv_addr.next().unwrap().addr(), null_usercstr)
                .expect("failed to write UserCString");

            // The arguments size limit should include the null terminator.
            assert!(read_c_string_vector(&current_task, argv_addr, 100, arg.len()).is_ok());
            assert_eq!(
                read_c_string_vector(
                    &current_task,
                    argv_addr,
                    100,
                    std::str::from_utf8(arg).unwrap().trim_matches('\0').len()
                ),
                error!(E2BIG)
            );
        })
        .await;
    }

    #[::fuchsia::test]
    async fn test_kcmp() {
        spawn_kernel_and_run(async |current_task| {
            let pid = current_task.get_pid();
            assert_eq!(sys_kcmp(&current_task, pid, pid, kcmp_type_KCMP_VM, 0, 0), Ok(0));
            assert_eq!(
                sys_kcmp(&current_task, pid, pid, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(EOPNOTSUPP)
            );
            assert_eq!(
                sys_kcmp(&current_task, pid, pid, kcmp_type_KCMP_SYSVSEM, 1, 2),
                error!(EOPNOTSUPP)
            );
            assert_eq!(
                sys_kcmp(&current_task, pid, pid, kcmp_type_KCMP_SYSVSEM, u64::MAX, u64::MAX),
                error!(EOPNOTSUPP)
            );
            assert_eq!(
                sys_kcmp(&current_task, pid, pid, kcmp_type_KCMP_TYPES, 0, 0),
                error!(EINVAL)
            );
            assert_eq!(sys_kcmp(&current_task, pid, pid, u32::MAX, 0, 0), error!(EINVAL));
            assert_eq!(
                sys_kcmp(&current_task, 0, pid, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(ESRCH)
            );
            assert_eq!(
                sys_kcmp(&current_task, pid, 0, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(ESRCH)
            );
            assert_eq!(
                sys_kcmp(&current_task, -1, pid, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(ESRCH)
            );
            assert_eq!(
                sys_kcmp(&current_task, pid, -1, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(ESRCH)
            );

            // Verify ptrace access checks take precedence over resource type errors for both tasks.
            let target_task = current_task.clone_task_for_test(0, Some(SIGCHLD));
            let target_pid = target_task.get_pid();
            current_task.set_creds(Credentials::with_ids(1, 1));
            assert_eq!(
                sys_kcmp(&current_task, target_pid, target_pid, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(EPERM)
            );
            assert_eq!(
                sys_kcmp(&current_task, pid, target_pid, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(EPERM)
            );
            assert_eq!(
                sys_kcmp(&current_task, target_pid, pid, kcmp_type_KCMP_SYSVSEM, 0, 0),
                error!(EPERM)
            );
        })
        .await;
    }
}