Add syscall eventfd and eventfd2

docs/src/kernel/linux-compatibility.md

@@ -304,13 +304,13 @@ provided by Linux on x86-64 architecture.
 | 281     | epoll_pwait      | ✅              |
 | 282     | signalfd         | ❌              |
 | 283     | timerfd_create   | ❌              |
-| 284     | eventfd          | ❌              |
+| 284     | eventfd          | ✅              |
 | 285     | fallocate        | ❌              |
 | 286     | timerfd_settime  | ❌              |
 | 287     | timerfd_gettime  | ❌              |
 | 288     | accept4          | ❌              |
 | 289     | signalfd4        | ❌              |
-| 290     | eventfd2         | ❌              |
+| 290     | eventfd2         | ✅              |
 | 291     | epoll_create1    | ✅              |
 | 292     | dup3             | ❌              |
 | 293     | pipe2            | ✅              |

kernel/aster-nix/src/fs/file_handle.rs

@@ -80,7 +80,7 @@ pub trait FileLike: Send + Sync + Any {
     }
 
     fn seek(&self, seek_from: SeekFrom) -> Result<usize> {
-        return_errno_with_message!(Errno::EINVAL, "seek is not supported");
+        return_errno_with_message!(Errno::ESPIPE, "seek is not supported");
     }
 
     fn clean_for_close(&self) -> Result<()> {

kernel/aster-nix/src/syscall/eventfd.rs

@@ -0,0 +1,259 @@
+// SPDX-License-Identifier: MPL-2.0
+
+//! `eventfd()` creates an "eventfd object" (we name it as `EventFile`)
+//! which serves as a mechanism for event wait/notify.
+//!
+//! `EventFile` holds a u64 integer counter.
+//! Writing to `EventFile` increments the counter by the written value.
+//! Reading from `EventFile` returns the current counter value and resets it
+//! (It is also possible to only read 1,
+//! depending on whether the `EFD_SEMAPHORE` flag is set).
+//! The read/write operations may be blocked based on file flags.
+//!
+//! For more detailed information about this syscall,
+//! refer to the man 2 eventfd documentation.
+//!
+
+use super::{SyscallReturn, SYS_EVENTFD, SYS_EVENTFD2};
+use crate::{
+    events::{IoEvents, Observer},
+    fs::{
+        file_handle::FileLike,
+        file_table::{FdFlags, FileDesc},
+        utils::{CreationFlags, StatusFlags},
+    },
+    log_syscall_entry,
+    prelude::*,
+    process::signal::{Pauser, Pollee, Poller},
+};
+
+pub fn sys_eventfd(init_val: u64) -> Result<SyscallReturn> {
+    log_syscall_entry!(SYS_EVENTFD);
+    debug!("init_val = 0x{:x}", init_val);
+
+    let fd = do_sys_eventfd2(init_val, Flags::empty());
+
+    Ok(SyscallReturn::Return(fd as _))
+}
+
+pub fn sys_eventfd2(init_val: u64, flags: u32) -> Result<SyscallReturn> {
+    log_syscall_entry!(SYS_EVENTFD2);
+    trace!("raw flags = {}", flags);
+    let flags = Flags::from_bits(flags)
+        .ok_or_else(|| Error::with_message(Errno::EINVAL, "unknown flags"))?;
+    debug!("init_val = 0x{:x}, flags = {:?}", init_val, flags);
+
+    let fd = do_sys_eventfd2(init_val, flags);
+
+    Ok(SyscallReturn::Return(fd as _))
+}
+
+fn do_sys_eventfd2(init_val: u64, flags: Flags) -> FileDesc {
+    let event_file = EventFile::new(init_val, flags);
+    let fd = {
+        let current = current!();
+        let mut file_table = current.file_table().lock();
+        let fd_flags = if flags.contains(Flags::EFD_CLOEXEC) {
+            FdFlags::CLOEXEC
+        } else {
+            FdFlags::empty()
+        };
+        file_table.insert(Arc::new(event_file), fd_flags)
+    };
+    fd
+}
+
+bitflags! {
+    struct Flags: u32 {
+        const EFD_SEMAPHORE = 1;
+        const EFD_CLOEXEC = CreationFlags::O_CLOEXEC.bits();
+        const EFD_NONBLOCK = StatusFlags::O_NONBLOCK.bits();
+    }
+}
+
+struct EventFile {
+    counter: Mutex<u64>,
+    pollee: Pollee,
+    flags: Mutex<Flags>,
+    write_pauser: Arc<Pauser>,
+}
+
+impl EventFile {
+    const MAX_COUNTER_VALUE: u64 = u64::MAX - 1;
+
+    fn new(init_val: u64, flags: Flags) -> Self {
+        let counter = Mutex::new(init_val);
+        let pollee = Pollee::new(IoEvents::OUT);
+        let write_pauser = Pauser::new();
+        Self {
+            counter,
+            pollee,
+            flags: Mutex::new(flags),
+            write_pauser,
+        }
+    }
+
+    fn is_nonblocking(&self) -> bool {
+        self.flags.lock().contains(Flags::EFD_NONBLOCK)
+    }
+
+    fn update_io_state(&self, counter: &MutexGuard<u64>) {
+        let is_readable = **counter != 0;
+
+        // if it is possible to write a value of at least "1"
+        // without blocking, the file is writable
+        let is_writable = **counter < Self::MAX_COUNTER_VALUE;
+
+        if is_writable {
+            if is_readable {
+                self.pollee.add_events(IoEvents::IN | IoEvents::OUT);
+            } else {
+                self.pollee.add_events(IoEvents::OUT);
+                self.pollee.del_events(IoEvents::IN);
+            }
+
+            self.write_pauser.resume_all();
+
+            return;
+        }
+
+        if is_readable {
+            self.pollee.add_events(IoEvents::IN);
+            self.pollee.del_events(IoEvents::OUT);
+            return;
+        }
+
+        self.pollee.del_events(IoEvents::IN | IoEvents::OUT);
+
+        // TODO: deal with overflow logic
+    }
+
+    /// Adds val to the counter.
+    ///
+    /// If the new_value is overflowed or exceeds MAX_COUNTER_VALUE, the counter value
+    /// will not be modified, and this method returns `Err(EINVAL)`.
+    fn add_counter_val(&self, val: u64) -> Result<()> {
+        let mut counter = self.counter.lock();
+
+        let new_value = (*counter)
+            .checked_add(val)
+            .ok_or_else(|| Error::with_message(Errno::EINVAL, "arithmetic overflow"))?;
+
+        if new_value <= Self::MAX_COUNTER_VALUE {
+            *counter = new_value;
+            self.update_io_state(&counter);
+            return Ok(());
+        }
+
+        return_errno_with_message!(Errno::EINVAL, "new value exceeds MAX_COUNTER_VALUE");
+    }
+}
+
+impl FileLike for EventFile {
+    fn read(&self, buf: &mut [u8]) -> Result<usize> {
+        let read_len = core::mem::size_of::<u64>();
+        if buf.len() < read_len {
+            return_errno_with_message!(Errno::EINVAL, "buf len is less len u64 size");
+        }
+
+        loop {
+            let mut counter = self.counter.lock();
+
+            // Wait until the counter becomes non-zero
+            if *counter == 0 {
+                if self.is_nonblocking() {
+                    return_errno_with_message!(Errno::EAGAIN, "try reading event file again");
+                }
+
+                self.update_io_state(&counter);
+                drop(counter);
+
+                let poller = Poller::new();
+                if self.pollee.poll(IoEvents::IN, Some(&poller)).is_empty() {
+                    poller.wait()?;
+                }
+                continue;
+            }
+
+            // Copy value from counter, and set the new counter value
+            if self.flags.lock().contains(Flags::EFD_SEMAPHORE) {
+                buf[..read_len].copy_from_slice(1u64.as_bytes());
+                *counter -= 1;
+            } else {
+                buf[..read_len].copy_from_slice((*counter).as_bytes());
+                *counter = 0;
+            }
+
+            self.update_io_state(&counter);
+            break;
+        }
+
+        Ok(read_len)
+    }
+
+    fn write(&self, buf: &[u8]) -> Result<usize> {
+        let write_len = core::mem::size_of::<u64>();
+        if buf.len() < write_len {
+            return_errno_with_message!(Errno::EINVAL, "buf len is less than the size of u64");
+        }
+
+        let supplied_value = u64::from_bytes(buf);
+
+        // Try to add counter val at first
+        if self.add_counter_val(supplied_value).is_ok() {
+            return Ok(write_len);
+        }
+
+        if self.is_nonblocking() {
+            return_errno_with_message!(Errno::EAGAIN, "try writing to event file again");
+        }
+
+        // Wait until counter can be added val to
+        self.write_pauser
+            .pause_until(|| self.add_counter_val(supplied_value).ok())?;
+
+        Ok(write_len)
+    }
+
+    fn poll(&self, mask: IoEvents, poller: Option<&Poller>) -> IoEvents {
+        self.pollee.poll(mask, poller)
+    }
+
+    fn status_flags(&self) -> StatusFlags {
+        if self.is_nonblocking() {
+            StatusFlags::O_NONBLOCK
+        } else {
+            StatusFlags::empty()
+        }
+    }
+
+    fn set_status_flags(&self, new_flags: StatusFlags) -> Result<()> {
+        let mut flags = self.flags.lock();
+
+        if new_flags.contains(StatusFlags::O_NONBLOCK) {
+            *flags |= Flags::EFD_NONBLOCK;
+        } else {
+            *flags &= !Flags::EFD_NONBLOCK;
+        }
+
+        // TODO: deal with other flags
+
+        Ok(())
+    }
+
+    fn register_observer(
+        &self,
+        observer: Weak<dyn crate::events::Observer<IoEvents>>,
+        mask: IoEvents,
+    ) -> Result<()> {
+        self.pollee.register_observer(observer, mask);
+        Ok(())
+    }
+
+    fn unregister_observer(
+        &self,
+        observer: &Weak<dyn Observer<IoEvents>>,
+    ) -> Option<Weak<dyn Observer<IoEvents>>> {
+        self.pollee.unregister_observer(observer)
+    }
+}

kernel/aster-nix/src/syscall/mod.rs

@@ -5,15 +5,38 @@
 use aster_frame::cpu::UserContext;
 
 use self::{
-    accept::sys_accept, alarm::sys_alarm, bind::sys_bind, connect::sys_connect,
-    execve::sys_execveat, getgroups::sys_getgroups, getpeername::sys_getpeername,
-    getrandom::sys_getrandom, getresgid::sys_getresgid, getresuid::sys_getresuid,
-    getsid::sys_getsid, getsockname::sys_getsockname, getsockopt::sys_getsockopt,
-    listen::sys_listen, pread64::sys_pread64, recvfrom::sys_recvfrom, sendto::sys_sendto,
-    setfsgid::sys_setfsgid, setfsuid::sys_setfsuid, setgid::sys_setgid, setgroups::sys_setgroups,
-    setregid::sys_setregid, setresgid::sys_setresgid, setresuid::sys_setresuid,
-    setreuid::sys_setreuid, setsid::sys_setsid, setsockopt::sys_setsockopt, setuid::sys_setuid,
-    shutdown::sys_shutdown, sigaltstack::sys_sigaltstack, socket::sys_socket,
+    accept::sys_accept,
+    alarm::sys_alarm,
+    bind::sys_bind,
+    connect::sys_connect,
+    eventfd::{sys_eventfd, sys_eventfd2},
+    execve::sys_execveat,
+    getgroups::sys_getgroups,
+    getpeername::sys_getpeername,
+    getrandom::sys_getrandom,
+    getresgid::sys_getresgid,
+    getresuid::sys_getresuid,
+    getsid::sys_getsid,
+    getsockname::sys_getsockname,
+    getsockopt::sys_getsockopt,
+    listen::sys_listen,
+    pread64::sys_pread64,
+    recvfrom::sys_recvfrom,
+    sendto::sys_sendto,
+    setfsgid::sys_setfsgid,
+    setfsuid::sys_setfsuid,
+    setgid::sys_setgid,
+    setgroups::sys_setgroups,
+    setregid::sys_setregid,
+    setresgid::sys_setresgid,
+    setresuid::sys_setresuid,
+    setreuid::sys_setreuid,
+    setsid::sys_setsid,
+    setsockopt::sys_setsockopt,
+    setuid::sys_setuid,
+    shutdown::sys_shutdown,
+    sigaltstack::sys_sigaltstack,
+    socket::sys_socket,
     socketpair::sys_socketpair,
 };
 use crate::{
@@ -115,6 +138,7 @@ mod connect;
 mod constants;
 mod dup;
 mod epoll;
+mod eventfd;
 mod execve;
 mod exit;
 mod exit_group;
@@ -363,6 +387,8 @@ define_syscall_nums!(
     SYS_SET_ROBUST_LIST = 273,
     SYS_UTIMENSAT = 280,
     SYS_EPOLL_PWAIT = 281,
+    SYS_EVENTFD = 284,
+    SYS_EVENTFD2 = 290,
     SYS_EPOLL_CREATE1 = 291,
     SYS_PIPE2 = 293,
     SYS_PRLIMIT64 = 302,
@@ -553,6 +579,8 @@ pub fn syscall_dispatch(
         SYS_SET_ROBUST_LIST => syscall_handler!(2, sys_set_robust_list, args),
         SYS_UTIMENSAT => syscall_handler!(4, sys_utimensat, args),
         SYS_EPOLL_PWAIT => syscall_handler!(5, sys_epoll_pwait, args),
+        SYS_EVENTFD => syscall_handler!(1, sys_eventfd, args),
+        SYS_EVENTFD2 => syscall_handler!(2, sys_eventfd2, args),
         SYS_EPOLL_CREATE1 => syscall_handler!(1, sys_epoll_create1, args),
         SYS_PIPE2 => syscall_handler!(2, sys_pipe2, args),
         SYS_PRLIMIT64 => syscall_handler!(4, sys_prlimit64, args),

kernel/aster-nix/src/syscall/read.rs

@@ -9,9 +9,13 @@ pub fn sys_read(fd: FileDesc, user_buf_addr: Vaddr, buf_len: usize) -> Result<Sy
         "fd = {}, user_buf_ptr = 0x{:x}, buf_len = 0x{:x}",
         fd, user_buf_addr, buf_len
     );
+
+    let file = {
         let current = current!();
         let file_table = current.file_table().lock();
-    let file = file_table.get_file(fd)?;
+        file_table.get_file(fd)?.clone()
+    };
+
     let mut read_buf = vec![0u8; buf_len];
     let read_len = file.read(&mut read_buf)?;
     write_bytes_to_user(user_buf_addr, &read_buf)?;

kernel/aster-nix/src/syscall/write.rs

@@ -16,9 +16,12 @@ pub fn sys_write(fd: FileDesc, user_buf_ptr: Vaddr, user_buf_len: usize) -> Resu
         fd, user_buf_ptr, user_buf_len
     );
 
+    let file = {
         let current = current!();
         let file_table = current.file_table().lock();
-    let file = file_table.get_file(fd)?;
+        file_table.get_file(fd)?.clone()
+    };
+
     if user_buf_len == 0 {
         return Ok(SyscallReturn::Return(0));
     }

regression/apps/Makefile

@@ -11,6 +11,7 @@ REGRESSION_BUILD_DIR ?= $(INITRAMFS)/regression
 # These test apps are sorted by name
 TEST_APPS := \
 	execve \
+	eventfd2 \
 	fork \
 	fork_c \
 	getpid \

regression/apps/eventfd2/Makefile

@@ -0,0 +1,5 @@
+# SPDX-License-Identifier: MPL-2.0
+
+include ../test_common.mk
+
+EXTRA_C_FLAGS :=

regression/apps/eventfd2/eventfd2.c

@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: MPL-2.0
+
+#include <err.h>
+#include <inttypes.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/eventfd.h>
+#include <unistd.h>
+
+int main()
+{
+	int efd;
+	uint64_t u;
+	ssize_t s;
+
+	uint64_t values[] = { 11, 222, 3333 };
+	size_t length = sizeof(values) / sizeof(values[0]);
+
+	efd = eventfd(0, 0);
+	if (efd == -1)
+		err(EXIT_FAILURE, "eventfd");
+
+	switch (fork()) {
+	case 0:
+		for (size_t j = 0; j < length; j++) {
+			printf("Child writing %ld to efd\n", values[j]);
+			u = values[j]; /* strtoull() allows various bases */
+			s = write(efd, &u, sizeof(uint64_t));
+			if (s != sizeof(uint64_t))
+				err(EXIT_FAILURE, "write");
+		}
+
+		printf("Child completed write loop\n");
+
+		exit(EXIT_SUCCESS);
+
+	default:
+		sleep(2);
+
+		printf("Parent about to read\n");
+		s = read(efd, &u, sizeof(uint64_t));
+		if (s != sizeof(uint64_t))
+			err(EXIT_FAILURE, "read");
+		printf("Parent read %" PRIu64 " (%#" PRIx64 ") from efd\n", u,
+		       u);
+		if (u != 11 + 222 + 3333) {
+			err(EXIT_FAILURE, "read eventfd");
+			exit(EXIT_FAILURE);
+		}
+		exit(EXIT_SUCCESS);
+
+	case -1:
+		err(EXIT_FAILURE, "fork");
+	}
+}

regression/apps/scripts/process.sh

@@ -8,7 +8,7 @@ SCRIPT_DIR=/regression
 cd ${SCRIPT_DIR}/..
 
 echo "Start process test......"
-tests="hello_world/hello_world fork/fork execve/execve fork_c/fork signal_c/signal_test pthread/pthread_test hello_pie/hello pty/open_pty getpid/getpid"
+tests="hello_world/hello_world fork/fork execve/execve fork_c/fork signal_c/signal_test pthread/pthread_test hello_pie/hello pty/open_pty getpid/getpid eventfd2/eventfd2"
 for testcase in ${tests}
 do 
     echo "Running test ${testcase}......"

regression/syscall_test/Makefile

@@ -11,6 +11,7 @@ TESTS ?= \
 	chown_test \
 	chroot_test \
 	epoll_test \
+	eventfd_test \
 	fsync_test \
 	getdents_test \
 	link_test \

regression/syscall_test/blocklists/eventfd_test

@@ -0,0 +1,3 @@
+EventfdTest.IllegalPwrite
+EventfdTest.SpliceFromPipePartialSucceeds
+EventfdTest.NotifyNonZero_NoRandomSave