Add eventfd file type and system call

src/Enclave.edl

@@ -112,5 +112,10 @@ enclave {
             [out] int* msg_flags_recv,
             int flags
         ) propagate_errno;
+
+        int occlum_ocall_eventfd(
+            unsigned int initval,
+            int flags
+        ) propagate_errno;
     };
 };

src/libos/src/fs/event_file.rs

@@ -0,0 +1,108 @@
+use super::*;
+
+/// Native Linux eventfd
+// TODO: move the implementaion of eventfd into libos to defend against Iago attacks from OCalls
+#[derive(Debug)]
+pub struct EventFile {
+    host_fd: c_int,
+}
+
+impl EventFile {
+    pub fn new(init_val: u32, flags: EventCreationFlags) -> Result<Self> {
+        let host_fd = try_libc!({
+            let mut ret: i32 = 0;
+            let status = occlum_ocall_eventfd(&mut ret, init_val, flags.bits());
+            assert!(status == sgx_status_t::SGX_SUCCESS);
+            ret
+        });
+        Ok(Self { host_fd })
+    }
+
+    pub fn get_host_fd(&self) -> c_int {
+        self.host_fd
+    }
+}
+
+bitflags! {
+    pub struct EventCreationFlags: i32 {
+        /// Provides semaphore-like semantics for reads from the new file descriptor
+        const EFD_SEMAPHORE = 1 << 0;
+        /// Non-blocking
+        const EFD_NONBLOCK  = 1 << 11;
+        /// Close on exec
+        const EFD_CLOEXEC   = 1 << 19;
+    }
+}
+
+extern "C" {
+    fn occlum_ocall_eventfd(ret: *mut i32, init_val: u32, flags: i32) -> sgx_status_t;
+}
+
+impl Drop for EventFile {
+    fn drop(&mut self) {
+        let ret = unsafe { libc::ocall::close(self.host_fd) };
+        assert!(ret == 0);
+    }
+}
+
+impl File for EventFile {
+    fn read(&self, buf: &mut [u8]) -> Result<usize> {
+        let ret = try_libc!(libc::ocall::read(
+            self.host_fd,
+            buf.as_mut_ptr() as *mut c_void,
+            buf.len()
+        )) as usize;
+        assert!(ret <= buf.len());
+        Ok(ret)
+    }
+
+    fn write(&self, buf: &[u8]) -> Result<usize> {
+        let ret = try_libc!(libc::ocall::write(
+            self.host_fd,
+            buf.as_ptr() as *const c_void,
+            buf.len()
+        )) as usize;
+        assert!(ret <= buf.len());
+        Ok(ret)
+    }
+
+    fn get_access_mode(&self) -> Result<AccessMode> {
+        Ok(AccessMode::O_RDWR)
+    }
+
+    fn get_status_flags(&self) -> Result<StatusFlags> {
+        let ret = try_libc!(libc::ocall::fcntl_arg0(self.get_host_fd(), libc::F_GETFL));
+        Ok(StatusFlags::from_bits_truncate(ret as u32))
+    }
+
+    fn set_status_flags(&self, new_status_flags: StatusFlags) -> Result<()> {
+        let valid_flags_mask = StatusFlags::O_APPEND
+            | StatusFlags::O_ASYNC
+            | StatusFlags::O_DIRECT
+            | StatusFlags::O_NOATIME
+            | StatusFlags::O_NONBLOCK;
+        let raw_status_flags = (new_status_flags & valid_flags_mask).bits();
+        try_libc!(libc::ocall::fcntl_arg1(
+            self.get_host_fd(),
+            libc::F_SETFL,
+            raw_status_flags as c_int
+        ));
+        Ok(())
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+}
+
+pub trait AsEvent {
+    fn as_event(&self) -> Result<&EventFile>;
+}
+
+impl AsEvent for FileRef {
+    fn as_event(&self) -> Result<&EventFile> {
+        self.as_any()
+            .downcast_ref::<EventFile>()
+            .ok_or_else(|| errno!(EBADF, "not an event file"))
+    }
+}

src/libos/src/fs/mod.rs

@@ -10,6 +10,7 @@ use std::mem::MaybeUninit;
 use std::path::Path;
 
 pub use self::dev_fs::AsDevRandom;
+pub use self::event_file::{AsEvent, EventFile};
 pub use self::file::{File, FileRef};
 pub use self::file_ops::{AccessMode, CreationFlags, Stat, StatusFlags};
 pub use self::file_ops::{Flock, FlockType};
@@ -22,6 +23,7 @@ pub use self::stdio::{StdinFile, StdoutFile};
 pub use self::syscalls::*;
 
 mod dev_fs;
+mod event_file;
 mod file;
 mod file_ops;
 mod file_table;

src/libos/src/fs/syscalls.rs

@@ -1,3 +1,4 @@
+use super::event_file::EventCreationFlags;
 use super::file_ops;
 use super::file_ops::{
     AccessibilityCheckFlags, AccessibilityCheckMode, DirFd, FcntlCmd, StatFlags,
@@ -12,6 +13,26 @@ pub struct iovec_t {
     len: size_t,
 }
 
+pub fn do_eventfd2(init_val: u32, flags: i32) -> Result<isize> {
+    info!("eventfd: initval {}, flags {} ", init_val, flags);
+
+    let inner_flags =
+        EventCreationFlags::from_bits(flags).ok_or_else(|| errno!(EINVAL, "invalid flags"))?;
+    let file_ref: Arc<Box<dyn File>> = {
+        let event = EventFile::new(init_val, inner_flags)?;
+        Arc::new(Box::new(event))
+    };
+
+    let current_ref = process::get_current();
+    let mut proc = current_ref.lock().unwrap();
+
+    let fd = proc.get_files().lock().unwrap().put(
+        file_ref,
+        inner_flags.contains(EventCreationFlags::EFD_CLOEXEC),
+    );
+    Ok(fd as isize)
+}
+
 pub fn do_open(path: *const i8, flags: u32, mode: u32) -> Result<isize> {
     let path = from_user::clone_cstring_safely(path)?
         .to_string_lossy()

src/libos/src/net/io_multiplexing.rs

@@ -1,5 +1,5 @@
 use super::*;
-use fs::{AsDevRandom, File, FileDesc, FileRef};
+use fs::{AsDevRandom, AsEvent, File, FileDesc, FileRef};
 use std::any::Any;
 use std::collections::btree_map::BTreeMap;
 use std::fmt;
@@ -25,6 +25,7 @@ pub fn do_select(
     let file_table_ref = proc.get_files().lock().unwrap();
 
     for fd in 0..nfds {
+        let fd_ref = file_table_ref.get(fd as FileDesc)?;
         let (r, w, e) = (
             readfds.is_set(fd),
             writefds.is_set(fd),
@@ -33,7 +34,7 @@ pub fn do_select(
         if !(r || w || e) {
             continue;
         }
-        if let Ok(socket) = file_table_ref.get(fd as FileDesc)?.as_unix_socket() {
+        if let Ok(socket) = fd_ref.as_unix_socket() {
             warn!("select unix socket is unimplemented, spin for read");
             readfds.clear();
             writefds.clear();
@@ -56,7 +57,13 @@ pub fn do_select(
             }
             return Ok(1);
         }
-        let host_fd = file_table_ref.get(fd as FileDesc)?.as_socket()?.fd();
+        let host_fd = if let Ok(socket) = fd_ref.as_socket() {
+            socket.fd()
+        } else if let Ok(eventfd) = fd_ref.as_event() {
+            eventfd.get_host_fd()
+        } else {
+            return_errno!(EBADF, "unsupported file type");
+        };
 
         host_to_libos_fd[host_fd as usize] = fd;
         let mut events = 0;
@@ -132,6 +139,9 @@ pub fn do_poll(pollfds: &mut [libc::pollfd], timeout: c_int) -> Result<usize> {
             // convert libos fd to host fd in the copy to keep pollfds unchanged
             u_pollfds[i].fd = socket.fd();
             u_pollfds[i].revents = 0;
+        } else if let Ok(eventfd) = file_ref.as_event() {
+            u_pollfds[i].fd = eventfd.get_host_fd();
+            u_pollfds[i].revents = 0;
         } else if let Ok(socket) = file_ref.as_unix_socket() {
             // FIXME: spin poll until can read (hack for php)
             while (pollfd.events & libc::POLLIN) != 0 && socket.poll()?.0 == false {
@@ -210,13 +220,16 @@ pub fn do_epoll_ctl(
     let mut epoll = file_ref.as_epoll()?.inner.lock().unwrap();
 
     let fd_ref = file_table_ref.get(fd)?;
-    let sock_result = fd_ref.as_socket();
-    if sock_result.is_err() {
-        //FIXME: workaround for grpc, other fd types including pipe should be supported
-        return Ok(());
-    }
 
-    let host_fd = sock_result.unwrap().fd() as FileDesc;
+    let host_fd = if let Ok(socket) = fd_ref.as_socket() {
+        socket.fd() as FileDesc
+    } else if let Ok(eventfd) = fd_ref.as_event() {
+        eventfd.get_host_fd() as FileDesc
+    } else {
+        warn!("unsupported file type");
+        return Ok(());
+    };
+
     epoll.ctl(op, host_fd, event)?;
 
     Ok(())

src/libos/src/net/syscalls.rs

@@ -232,7 +232,11 @@ pub fn do_epoll_pwait(
     timeout: c_int,
     sigmask: *const usize, //TODO:add sigset_t
 ) -> Result<isize> {
-    info!("epoll_pwait");
-    //TODO:add signal support
-    do_epoll_wait(epfd, events, maxevents, 0)
+    if sigmask.is_null() {
+        info!("epoll_wait");
+    } else {
+        info!("epoll_pwait")
+    }
+    //TODO:add signal support
+    do_epoll_wait(epfd, events, maxevents, timeout)
 }

src/libos/src/syscall/mod.rs

@@ -221,6 +221,8 @@ pub extern "C" fn dispatch_syscall(
 
         SysPipe => fs::do_pipe2(arg0 as *mut i32, 0),
         SysPipe2 => fs::do_pipe2(arg0 as *mut i32, arg1 as u32),
+        SysEventfd => fs::do_eventfd2(arg0 as u32, 0),
+        SysEventfd2 => fs::do_eventfd2(arg0 as u32, arg1 as i32),
         SysDup => fs::do_dup(arg0 as FileDesc),
         SysDup2 => fs::do_dup2(arg0 as FileDesc, arg1 as FileDesc),
         SysDup3 => fs::do_dup3(arg0 as FileDesc, arg1 as FileDesc, arg2 as u32),

src/pal/src/ocalls/fs.c

@@ -1,6 +1,11 @@
 #include <unistd.h>
 #include "ocalls.h"
+#include <sys/eventfd.h>
 
 void occlum_ocall_sync(void) {
     sync();
 }
+
+int occlum_ocall_eventfd(unsigned int initval, int flags) {
+    return eventfd(initval, flags);
+}

test/Makefile

@@ -14,7 +14,7 @@ TEST_DEPS := client data_sink
 TESTS ?= empty env hello_world malloc mmap file fs_perms getpid spawn sched pipe time \
 	truncate readdir mkdir open stat link symlink tls pthread uname rlimit server \
 	server_epoll unix_socket cout hostfs cpuid rdtsc device sleep exit_group \
-	ioctl fcntl
+	ioctl fcntl eventfd
 # Benchmarks: need to be compiled and run by bench-% target
 BENCHES := spawn_and_exit_latency pipe_throughput unix_socket_throughput
 

test/eventfd/Makefile

@@ -0,0 +1,5 @@
+include ../test_common.mk
+
+EXTRA_C_FLAGS := -Wno-incompatible-pointer-types-discards-qualifiers
+EXTRA_LINK_FLAGS :=
+BIN_ARGS :=

test/eventfd/main.c

@@ -0,0 +1,251 @@
+#include <sys/epoll.h>
+#include <sys/eventfd.h>
+#include <sys/select.h>
+#include <sys/socket.h>
+#include <sys/syscall.h>
+#include <sys/wait.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <poll.h>
+#include <pthread.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <spawn.h>
+#include <string.h>
+
+#include "test.h"
+
+#define MAXEVENTS 1
+
+// ============================================================================
+// Test cases
+// ============================================================================
+
+int test_fcntl_get_flags() {
+    int event_fd = eventfd(0, 0);
+    if (event_fd < 0) {
+        THROW_ERROR("failed to create an eventfd");
+    }
+
+    if ((fcntl(event_fd, F_GETFL, 0) != O_RDWR)) {
+        close(event_fd);
+        THROW_ERROR("fcntl get flags failed");
+    }
+
+    close(event_fd);
+    return 0;
+}
+
+int test_fcntl_set_flags() {
+    int event_fd = eventfd(0, 0);
+    if (event_fd < 0) {
+        THROW_ERROR("failed to create an eventfd");
+    }
+
+    fcntl(event_fd, F_SETFL, O_NONBLOCK);
+    if ((fcntl(event_fd, F_GETFL, 0) != (O_NONBLOCK | O_RDWR))) {
+        close(event_fd);
+        THROW_ERROR("fcntl set flags failed");
+    }
+
+    close(event_fd);
+    return 0;
+}
+
+int test_create_with_flags() {
+    int event_fd = eventfd(0, EFD_NONBLOCK);
+    if (event_fd < 0) {
+        THROW_ERROR("failed to create an eventfd");
+    }
+
+    if ((fcntl(event_fd, F_GETFL, 0) != (O_NONBLOCK | O_RDWR))) {
+        close(event_fd);
+        THROW_ERROR("create flags failed\n");
+    }
+
+    close(event_fd);
+    return 0;
+}
+
+struct thread_arg {
+    pthread_t tid;
+    int fd;
+    uint64_t data;
+};
+
+#define TEST_DATA 678
+#define CHILD_NUM 16
+
+static void *thread_child(void *arg) {
+    struct thread_arg *child_arg = arg;
+    write(child_arg->fd, &(child_arg->data), sizeof(child_arg->data));
+    return NULL;
+}
+
+int create_child(struct thread_arg *arg) {
+    pthread_attr_t attr;
+    if (pthread_attr_init(&attr) != 0) {
+        THROW_ERROR("failed to initialize attribute");
+    }
+
+    if (pthread_create(&(arg->tid), &attr, &thread_child, arg) != 0) {
+        if (pthread_attr_destroy(&attr) != 0) {
+            THROW_ERROR("failed to destroy attr");
+        }
+        THROW_ERROR("failed to create the thread");
+    }
+
+    if (pthread_attr_destroy(&attr) != 0) {
+        THROW_ERROR("failed to destroy attr");
+    }
+
+    return 0;
+}
+
+int test_read_write() {
+    int event_fd = eventfd(0, 0);
+    if (event_fd < 0) {
+        THROW_ERROR("failed to create an eventfd");
+    }
+
+    struct thread_arg child_arg[CHILD_NUM] = {0};
+
+    // Create child threads and send eventfd and data
+    for (int i = 0; i < CHILD_NUM; i++) {
+        child_arg[i].fd = event_fd;
+        child_arg[i].data = TEST_DATA;
+        if (create_child(&child_arg[i]) != 0) {
+            close(event_fd);
+            THROW_ERROR("failed to create children");
+        }
+    }
+
+    // Check the data sent from children
+    uint64_t data_recv = 0;
+
+    do {
+        uint64_t cur_data = 0;
+        ssize_t len_recv = read(event_fd, &cur_data, sizeof(uint64_t));
+        if (len_recv != sizeof(uint64_t)) {
+            close(event_fd);
+            THROW_ERROR("received length is not as expected");
+        }
+        data_recv += cur_data;
+    } while (data_recv != TEST_DATA*CHILD_NUM);
+
+    close(event_fd);
+
+    for (int i = 0; i < CHILD_NUM; i++) {
+        if (pthread_join(child_arg[i].tid, NULL) != 0) {
+            THROW_ERROR("pthread_join");
+        }
+    }
+
+    return 0;
+}
+
+int test_select_with_socket() {
+    fd_set wfds;
+
+    struct timeval tv = { .tv_sec = 60, .tv_usec = 0 };
+
+    int sock = socket(AF_INET, SOCK_STREAM, 0);
+    int event_fd = eventfd(0, 0);
+    if (event_fd < 0 || sock < 0) {
+        THROW_ERROR("failed to create files");
+    }
+
+    FD_ZERO(&wfds);
+    FD_SET(sock, &wfds);
+    FD_SET(event_fd, &wfds);
+
+    if (select(sock > event_fd? sock + 1: event_fd + 1, NULL, &wfds, NULL, &tv) <= 0) {
+        close_files(2, sock, event_fd);
+        THROW_ERROR("select failed");
+    }
+
+    close_files(2, sock, event_fd);
+    return 0;
+}
+
+int test_poll_with_socket() {
+    int sock = socket(AF_INET, SOCK_STREAM, 0);
+    int event_fd = eventfd(0, 0);
+    if (event_fd < 0 || sock < 0) {
+        THROW_ERROR("failed to create files");
+    }
+
+    struct pollfd pollfds[] = {
+        { .fd = sock, .events = POLLIN, .revents = 0, },
+        { .fd = event_fd, .events = POLLOUT, .revents = 0 },
+    };
+
+    int ret = poll(pollfds, 2, -1);
+    if (ret <= 0) {
+        close_files(2, event_fd, sock);
+        THROW_ERROR("poll error");
+    }
+
+    close_files(2, event_fd, sock);
+    return 0;
+}
+
+int test_epoll_with_socket() {
+    int event_fd = eventfd(0, EFD_NONBLOCK);
+    int sock = socket(AF_INET, SOCK_STREAM, 0);
+    int epfd = epoll_create1(0);
+
+    if (event_fd < 0 || sock < 0 || epfd < 0) {
+        THROW_ERROR("failed to create files");
+    }
+
+    struct epoll_event ctl_events[2] = {0};
+    // Add eventfd to the interest list
+	ctl_events[0].data.fd = event_fd;
+	ctl_events[0].events = EPOLLIN | EPOLLET;
+    // Add socket to the interest list
+	ctl_events[1].data.fd = sock;
+	ctl_events[1].events = EPOLLIN | EPOLLET;
+	if (epoll_ctl(epfd, EPOLL_CTL_ADD, event_fd, &ctl_events[0]) == -1 ||
+            epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ctl_events[1]) == -1) {
+        close_files(3, event_fd, sock, epfd);
+		THROW_ERROR("epoll_ctl");
+    }
+
+    struct thread_arg child_arg = { .tid = 0, .fd = event_fd, .data = TEST_DATA };
+    if (create_child(&child_arg) != 0) {
+        close_files(3, event_fd, sock, epfd);
+        THROW_ERROR("failed to create child");
+    }
+
+    struct epoll_event events[MAXEVENTS] = {0};
+    if (epoll_pwait(epfd, events, MAXEVENTS, -1, NULL) <= 0){
+        close_files(3, event_fd, sock, epfd);
+        THROW_ERROR("epoll failed");
+    }
+
+    close_files(3, event_fd, sock, epfd);
+
+    if (pthread_join(child_arg.tid, NULL) != 0) {
+        THROW_ERROR("pthread_join");
+    }
+
+    return 0;
+}
+// ============================================================================
+// Test suite
+// ============================================================================
+
+static test_case_t test_cases[] = {
+    TEST_CASE(test_fcntl_get_flags),
+    TEST_CASE(test_fcntl_set_flags),
+    TEST_CASE(test_create_with_flags),
+    TEST_CASE(test_read_write),
+    TEST_CASE(test_epoll_with_socket),
+    TEST_CASE(test_poll_with_socket),
+    TEST_CASE(test_select_with_socket),
+};
+
+int main(int argc, const char* argv[]) {
+    return test_suite_run(test_cases, ARRAY_SIZE(test_cases));
+}

test/include/test.h

@@ -2,6 +2,7 @@
 #define __TEST_H
 
 #include <stdio.h>
+#include <stdarg.h>
 
 #define _STR(x)             #x
 #define STR(x)              _STR(x)
@@ -37,4 +38,13 @@ int test_suite_run(test_case_t* test_cases, int num_test_cases) {
     return 0;
 }
 
+void close_files(int count, ...) {
+    va_list ap;
+    va_start(ap, count);
+    for (int i = 0; i < count; i++) {
+        close(va_arg(ap, int));
+    }
+    va_end(ap);
+}
+
 #endif /* __TEST_H */