feat(fdtable): add CLOEXEC flag

chore: apply code style fixes and update documentation

api/ruxos_posix_api/src/imp/execve/mod.rs

@@ -135,6 +135,14 @@ pub fn sys_execve(pathname: *const c_char, argv: usize, envp: usize) -> ! {
         stack.stack_size()
     );
 
+    current()
+        .fs
+        .lock()
+        .as_mut()
+        .unwrap()
+        .fd_table
+        .do_close_on_exec();
+
     set_sp_and_jmp(sp, entry);
 }
 

api/ruxos_posix_api/src/imp/fd_ops.rs

@@ -133,44 +133,40 @@ pub fn sys_close(fd: c_int) -> c_int {
     syscall_body!(sys_close, close_file_like(fd).map(|_| 0))
 }
 
-fn dup_fd(old_fd: c_int) -> LinuxResult<c_int> {
-    let f = get_file_like(old_fd)?;
-    let new_fd = add_file_like(f)?;
-    Ok(new_fd as _)
-}
-
 /// Duplicate a file descriptor.
 pub fn sys_dup(old_fd: c_int) -> c_int {
     debug!("sys_dup <= {}", old_fd);
-    syscall_body!(sys_dup, dup_fd(old_fd))
+    syscall_body!(sys_dup, {
+        let binding_task = current();
+        let mut binding_fs = binding_task.fs.lock();
+        let new_fd = binding_fs.as_mut().unwrap().fd_table.dup(old_fd as _)?;
+        Ok(new_fd as c_int)
+    })
 }
 
 /// Duplicate a file descriptor, but it uses the file descriptor number specified in `new_fd`.
 ///
-/// TODO: `dup2` should forcibly close new_fd if it is already opened.
+/// The close-on-exec flag for the duplicate descriptor is off.
 pub fn sys_dup2(old_fd: c_int, new_fd: c_int) -> c_int {
     debug!("sys_dup2 <= old_fd: {}, new_fd: {}", old_fd, new_fd);
     syscall_body!(sys_dup2, {
         if old_fd == new_fd {
-            let r = sys_fcntl(old_fd, ctypes::F_GETFD as _, 0);
-            if r >= 0 {
-                return Ok(old_fd);
-            } else {
-                return Ok(r);
-            }
+            // check if `oldfd` isn't an open file descriptor. If it not, return `EBADF`
+            get_file_like(old_fd as _)?;
+            return Ok(new_fd);
         }
         if new_fd as usize >= RUX_FILE_LIMIT {
             return Err(LinuxError::EBADF);
         }
-        close_file_like(new_fd as _)?;
 
-        let f = get_file_like(old_fd as _)?;
-        let binding_task = current();
-        let mut binding_fs = binding_task.fs.lock();
-        let fd_table = &mut binding_fs.as_mut().unwrap().fd_table;
-        fd_table
-            .add_at(new_fd as usize, f)
-            .ok_or(LinuxError::EMFILE)?;
+        // The steps of closing and reusing the file descriptor newfd are performed atomically.
+        current()
+            .fs
+            .lock()
+            .as_mut()
+            .unwrap()
+            .fd_table
+            .dup3(old_fd as _, new_fd as _, false)?;
 
         Ok(new_fd)
     })
@@ -187,14 +183,15 @@ pub fn sys_dup3(old_fd: c_int, new_fd: c_int, flags: c_int) -> c_int {
         if old_fd == new_fd {
             return Err(LinuxError::EINVAL);
         }
-        sys_dup2(old_fd, new_fd);
-        if flags as u32 & ctypes::O_CLOEXEC != 0 {
-            sys_fcntl(
-                new_fd,
-                ctypes::F_SETFD as c_int,
-                ctypes::FD_CLOEXEC as usize,
-            );
-        }
+        let cloexec = (flags as u32 & ctypes::O_CLOEXEC) != 0;
+        let binding_task = current();
+        let mut binding_fs = binding_task.fs.lock();
+        binding_fs
+            .as_mut()
+            .unwrap()
+            .fd_table
+            .dup3(old_fd as _, new_fd as _, cloexec)?;
+
         Ok(new_fd)
     })
 }
@@ -206,15 +203,39 @@ pub fn sys_fcntl(fd: c_int, cmd: c_int, arg: usize) -> c_int {
     debug!("sys_fcntl <= fd: {} cmd: {} arg: {}", fd, cmd, arg);
     syscall_body!(sys_fcntl, {
         match cmd as u32 {
-            ctypes::F_DUPFD => dup_fd(fd),
+            ctypes::F_DUPFD => {
+                let new_fd = current()
+                    .fs
+                    .lock()
+                    .as_mut()
+                    .unwrap()
+                    .fd_table
+                    .dup_with_low_bound(fd as _, arg as _, false)?;
+                Ok(new_fd as _)
+            }
             ctypes::F_GETFD => {
-                // Return (as the function result) the file descriptor flags; the arg is ignored.
-                // temporary unsupport CLOEXEC flag
-                Ok(0)
+                let binding_task = current();
+                let mut binding_fs = binding_task.fs.lock();
+                let fd_table = &mut binding_fs.as_mut().unwrap().fd_table;
+                if fd_table.get(fd as _).is_none() {
+                    return Err(LinuxError::EBADF);
+                }
+                let cloexec = if fd_table.get_cloexec(fd as _) {
+                    ctypes::FD_CLOEXEC
+                } else {
+                    0
+                };
+                Ok(cloexec as _)
             }
             ctypes::F_DUPFD_CLOEXEC => {
-                // TODO: Change fd flags
-                dup_fd(fd)
+                let new_fd = current()
+                    .fs
+                    .lock()
+                    .as_mut()
+                    .unwrap()
+                    .fd_table
+                    .dup_with_low_bound(fd as _, arg as _, true)?;
+                Ok(new_fd as _)
             }
             ctypes::F_SETFL => {
                 if fd == 0 || fd == 1 || fd == 2 {
@@ -238,16 +259,14 @@ pub fn sys_fcntl(fd: c_int, cmd: c_int, arg: usize) -> c_int {
                 Ok(flags as c_int)
             }
             ctypes::F_SETFD => {
-                if arg == 0 || arg == 1 || arg == 2 {
-                    return Ok(0);
-                }
+                let cloexec = (arg as u32 & ctypes::FD_CLOEXEC) == 1;
                 let binding_task = current();
                 let mut binding_fs = binding_task.fs.lock();
                 let fd_table = &mut binding_fs.as_mut().unwrap().fd_table;
-                fd_table
-                    .add_at(arg, get_file_like(fd)?)
-                    .ok_or(LinuxError::EMFILE)?;
-                let _ = close_file_like(fd);
+                if fd_table.get(fd as _).is_none() {
+                    return Err(LinuxError::EBADF);
+                }
+                fd_table.set_cloexec(fd as _, cloexec);
                 Ok(0)
             }
             _ => {

api/ruxos_posix_api/src/imp/fs.rs

@@ -72,9 +72,10 @@ pub fn sys_open(filename: *const c_char, flags: c_int, mode: ctypes::mode_t) ->
     let filename = char_ptr_to_str(filename);
     debug!("sys_open <= {:?} {:#o} {:#o}", filename, flags, mode);
     syscall_body!(sys_open, {
+        let cloexec = (flags as u32 & ctypes::O_CLOEXEC) != 0;
         let options = flags_to_options(flags, mode);
         let file = ruxfs::fops::File::open(filename?, &options)?;
-        File::new(file).add_to_fd_table()
+        File::new(file).add_to_fd_table(cloexec)
     })
 }
 
@@ -84,6 +85,7 @@ pub fn sys_openat(fd: usize, path: *const c_char, flags: c_int, mode: ctypes::mo
     let fd: c_int = fd as c_int;
     debug!("sys_openat <= {}, {:?}, {:#o} {:#o}", fd, path, flags, mode);
     syscall_body!(sys_openat, {
+        let cloexec = (flags as u32 & ctypes::O_CLOEXEC) != 0;
         let options = flags_to_options(flags, mode);
         if (flags as u32) & ctypes::O_DIRECTORY != 0 {
             let dir = if fd == ctypes::AT_FDCWD {
@@ -94,7 +96,7 @@ pub fn sys_openat(fd: usize, path: *const c_char, flags: c_int, mode: ctypes::mo
                     .write()
                     .open_dir_at(path?, &options)?
             };
-            Directory::new(dir).add_to_fd_table()
+            Directory::new(dir).add_to_fd_table(cloexec)
         } else {
             let file = if fd == ctypes::AT_FDCWD {
                 ruxfs::fops::File::open(path?, &options)?
@@ -104,7 +106,7 @@ pub fn sys_openat(fd: usize, path: *const c_char, flags: c_int, mode: ctypes::mo
                     .write()
                     .open_file_at(path?, &options)?
             };
-            File::new(file).add_to_fd_table()
+            File::new(file).add_to_fd_table(cloexec)
         }
     })
 }

api/ruxos_posix_api/src/imp/io_mpx/epoll.rs

@@ -160,14 +160,15 @@ impl FileLike for EpollInstance {
 /// Creates a new epoll instance.
 ///
 /// It returns a file descriptor referring to the new epoll instance.
-pub fn sys_epoll_create(size: c_int) -> c_int {
-    debug!("sys_epoll_create <= {}", size);
+pub fn sys_epoll_create1(flags: c_int) -> c_int {
+    debug!("sys_epoll_create <= {}", flags);
     syscall_body!(sys_epoll_create, {
-        if size < 0 {
+        if flags < 0 {
             return Err(LinuxError::EINVAL);
         }
         let epoll_instance = EpollInstance::new(0);
-        add_file_like(Arc::new(epoll_instance))
+        let cloexec = (flags as u32 & ctypes::EPOLL_CLOEXEC) != 0;
+        add_file_like(Arc::new(epoll_instance), cloexec)
     })
 }
 

api/ruxos_posix_api/src/imp/io_mpx/mod.rs

@@ -23,7 +23,7 @@ mod poll;
 mod select;
 
 #[cfg(feature = "epoll")]
-pub use self::epoll::{sys_epoll_create, sys_epoll_ctl, sys_epoll_pwait, sys_epoll_wait};
+pub use self::epoll::{sys_epoll_create1, sys_epoll_ctl, sys_epoll_pwait, sys_epoll_wait};
 #[cfg(feature = "poll")]
 pub use self::poll::{sys_poll, sys_ppoll};
 #[cfg(feature = "select")]

api/ruxos_posix_api/src/imp/net.rs

@@ -42,8 +42,8 @@ pub enum Socket {
 }
 
 impl Socket {
-    fn add_to_fd_table(self) -> LinuxResult<c_int> {
-        ruxtask::fs::add_file_like(Arc::new(self))
+    fn add_to_fd_table(self, cloexec: bool) -> LinuxResult<c_int> {
+        ruxtask::fs::add_file_like(Arc::new(self), cloexec)
     }
 
     fn from_fd(fd: c_int) -> LinuxResult<Arc<Self>> {
@@ -338,25 +338,26 @@ fn from_sockaddr(
 pub fn sys_socket(domain: c_int, socktype: c_int, protocol: c_int) -> c_int {
     debug!("sys_socket <= {} {} {}", domain, socktype, protocol);
     let (domain, socktype, protocol) = (domain as u32, socktype as u32, protocol as u32);
-    pub const _SOCK_STREAM_NONBLOCK: u32 = ctypes::SOCK_STREAM | ctypes::SOCK_NONBLOCK;
+    let cloexec = (socktype & ctypes::SOCK_CLOEXEC) != 0;
+    let nonblock = (socktype & ctypes::SOCK_NONBLOCK) != 0;
+    let socktype = socktype & !ctypes::SOCK_CLOEXEC & !ctypes::SOCK_NONBLOCK;
     syscall_body!(sys_socket, {
         match (domain, socktype, protocol) {
             (ctypes::AF_INET, ctypes::SOCK_STREAM, ctypes::IPPROTO_TCP)
             | (ctypes::AF_INET, ctypes::SOCK_STREAM, 0) => {
-                Socket::Tcp(Mutex::new(TcpSocket::new())).add_to_fd_table()
+                let tcp_socket = TcpSocket::new();
+                if nonblock {
+                    tcp_socket.set_nonblocking(true);
+                }
+                Socket::Tcp(Mutex::new(tcp_socket)).add_to_fd_table(cloexec)
             }
             (ctypes::AF_INET, ctypes::SOCK_DGRAM, ctypes::IPPROTO_UDP)
             | (ctypes::AF_INET, ctypes::SOCK_DGRAM, 0) => {
-                Socket::Udp(Mutex::new(UdpSocket::new())).add_to_fd_table()
-            }
-            (ctypes::AF_INET, _SOCK_STREAM_NONBLOCK, ctypes::IPPROTO_TCP) => {
-                let tcp_socket = TcpSocket::new();
-                tcp_socket.set_nonblocking(true);
-                Socket::Tcp(Mutex::new(tcp_socket)).add_to_fd_table()
+                Socket::Udp(Mutex::new(UdpSocket::new())).add_to_fd_table(cloexec)
             }
             (ctypes::AF_UNIX, ctypes::SOCK_STREAM, 0) => {
                 Socket::Unix(Mutex::new(UnixSocket::new(UnixSocketType::SockStream)))
-                    .add_to_fd_table()
+                    .add_to_fd_table(cloexec)
             }
             _ => Err(LinuxError::EINVAL),
         }
@@ -558,7 +559,8 @@ pub unsafe fn sys_accept(
         let socket = Socket::from_fd(socket_fd)?;
         let new_socket = socket.accept()?;
         let addr = new_socket.peer_addr()?;
-        let new_fd = Socket::add_to_fd_table(new_socket)?;
+        let cloexec = false;
+        let new_fd = Socket::add_to_fd_table(new_socket, cloexec)?;
         match addr {
             UnifiedSocketAddress::Net(addr) => unsafe {
                 (*socket_addr, *socket_len) = into_sockaddr(addr);

api/ruxos_posix_api/src/imp/pipe.rs

@@ -233,8 +233,9 @@ pub fn sys_pipe(fds: &mut [c_int]) -> c_int {
         }
 
         let (read_end, write_end) = Pipe::new();
-        let read_fd = add_file_like(Arc::new(read_end))?;
-        let write_fd = add_file_like(Arc::new(write_end)).inspect_err(|_| {
+        let cloexec = false;
+        let read_fd = add_file_like(Arc::new(read_end), cloexec)?;
+        let write_fd = add_file_like(Arc::new(write_end), cloexec).inspect_err(|_| {
             close_file_like(read_fd).ok();
         })?;
 

api/ruxos_posix_api/src/lib.rs

@@ -76,7 +76,7 @@ pub use imp::fs::{
     sys_rmdir, sys_stat, sys_unlink, sys_unlinkat,
 };
 #[cfg(feature = "epoll")]
-pub use imp::io_mpx::{sys_epoll_create, sys_epoll_ctl, sys_epoll_pwait, sys_epoll_wait};
+pub use imp::io_mpx::{sys_epoll_create1, sys_epoll_ctl, sys_epoll_pwait, sys_epoll_wait};
 #[cfg(feature = "poll")]
 pub use imp::io_mpx::{sys_poll, sys_ppoll};
 #[cfg(feature = "select")]

crates/flatten_objects/src/lib.rs

@@ -144,6 +144,25 @@ impl<T, const CAP: usize> FlattenObjects<T, CAP> {
         Some(id)
     }
 
+    /// Add an object and assign it a unique ID, but only search for IDs starting from `low_bound`.
+    ///
+    /// Returns the assigned ID if an available slot is found; otherwise, returns `None`.
+    pub fn add_with_low_bound(&mut self, value: T, low_bound: usize) -> Option<usize> {
+        let id = if low_bound == 0 {
+            self.id_bitmap.first_false_index()
+        } else {
+            self.id_bitmap.next_false_index(low_bound - 1)
+        }?;
+        if id < CAP {
+            self.count += 1;
+            self.id_bitmap.set(id, true);
+            self.objects[id].write(value);
+            Some(id)
+        } else {
+            None
+        }
+    }
+
     /// Removes the object with the given ID.
     ///
     /// Returns the object if there is one assigned to the ID. Otherwise,

modules/ruxtask/Cargo.toml

@@ -2,10 +2,7 @@
 name = "ruxtask"
 version = "0.1.0"
 edition = "2021"
-authors = [
-    "Yuekai Jia <[email protected]>",
-    "AuYang261 <[email protected]>",
-]
+authors = ["Yuekai Jia <[email protected]>", "AuYang261 <[email protected]>"]
 description = "Ruxos task management module"
 license = "GPL-3.0-or-later OR Apache-2.0"
 homepage = "https://github.com/syswonder/ruxos"
@@ -15,8 +12,15 @@ repository = "https://github.com/syswonder/ruxos/tree/main/modules/ruxtask"
 default = ["fs"]
 
 multitask = [
-    "dep:ruxconfig", "dep:percpu", "dep:spinlock", "dep:lazy_init", "dep:memory_addr",
-    "dep:scheduler", "dep:timer_list", "kernel_guard", "dep:crate_interface"
+    "dep:ruxconfig",
+    "dep:percpu",
+    "dep:spinlock",
+    "dep:lazy_init",
+    "dep:memory_addr",
+    "dep:scheduler",
+    "dep:timer_list",
+    "kernel_guard",
+    "dep:crate_interface",
 ]
 irq = []
 tls = ["ruxhal/tls"]
@@ -53,16 +57,16 @@ kernel_guard = { version = "0.1.0", optional = true }
 crate_interface = { version = "0.1.1", optional = true }
 flatten_objects = { path = "../../crates/flatten_objects" }
 spin = "0.9"
-axio ={ path = "../../crates/axio" }
+axio = { path = "../../crates/axio" }
 lazy_static = { version = "1.4", features = ["spin_no_std"] }
 page_table = { path = "../../crates/page_table" }
 page_table_entry = { path = "../../crates/page_table_entry" }
 
+bitmaps = { version = "3.2", default-features = false }
 # for testing
-axfs_vfs = { path = "../../crates/axfs_vfs"}
+axfs_vfs = { path = "../../crates/axfs_vfs" }
 
 [dev-dependencies]
 rand = "0.8"
 ruxhal = { path = "../ruxhal", features = ["fp_simd"] }
 ruxtask = { path = ".", features = ["test"] }
-