net: sockets: move poll implementation to zvfs

Move the implementation of zsock_poll to zvfs_poll. This allows
other types of file descriptors to also make use of poll()
functionality even when the network subsystem is not enabled.

Additionally, it partially removes a dependency cycle between
posix and networking by moving functionality into a mutual
dependency.

Signed-off-by: Chris Friedt <[email protected]>

include/zephyr/net/socket.h

@@ -625,7 +625,10 @@ static inline int zsock_ioctl_wrapper(int sock, unsigned long request, ...)
  * it may conflict with generic POSIX ``poll()`` function).
  * @endrst
  */
-__syscall int zsock_poll(struct zsock_pollfd *fds, int nfds, int timeout);
+static inline int zsock_poll(struct zsock_pollfd *fds, int nfds, int timeout)
+{
+	return zvfs_poll(fds, nfds, timeout);
+}
 
 /**
  * @brief Get various socket options

include/zephyr/net/socket_poll.h

@@ -7,6 +7,8 @@
 #ifndef ZEPHYR_INCLUDE_NET_SOCKET_POLL_H_
 #define ZEPHYR_INCLUDE_NET_SOCKET_POLL_H_
 
+#include <zephyr/sys/fdtable.h>
+
 /* Setting for pollfd to avoid circular inclusion */
 
 /**
@@ -25,11 +27,7 @@ extern "C" {
  *
  * An array of these descriptors is passed as an argument to poll().
  */
-struct zsock_pollfd {
-	int fd;        /**< Socket descriptor */
-	short events;  /**< Requested events */
-	short revents; /**< Returned events */
-};
+#define zsock_pollfd zvfs_pollfd
 
 #ifdef __cplusplus
 }

include/zephyr/sys/fdtable.h

@@ -11,6 +11,7 @@
 /* FIXME: For native_posix ssize_t, off_t. */
 #include <zephyr/fs/fs.h>
 #include <zephyr/sys/mutex.h>
+#include <zephyr/sys/util.h>
 
 /* File mode bits */
 #define ZVFS_MODE_IFMT   0170000
@@ -26,6 +27,13 @@
 #define ZVFS_MODE_IFLNK  0120000
 #define ZVFS_MODE_IFSOCK 0140000
 
+#define ZVFS_POLLIN   BIT(0)
+#define ZVFS_POLLPRI  BIT(1)
+#define ZVFS_POLLOUT  BIT(2)
+#define ZVFS_POLLERR  BIT(3)
+#define ZVFS_POLLHUP  BIT(4)
+#define ZVFS_POLLNVAL BIT(5)
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -191,6 +199,17 @@ static inline int zvfs_fdtable_call_ioctl(const struct fd_op_vtable *vtable, voi
 	return res;
 }
 
+struct zvfs_pollfd {
+	int fd;
+	short events;
+	short revents;
+};
+
+__syscall int zvfs_poll(struct zvfs_pollfd *fds, int nfds, int poll_timeout);
+
+struct zsock_fd_set {
+	uint32_t bitset[(CONFIG_ZVFS_OPEN_MAX + 31) / 32];
+};
 /**
  * Request codes for fd_op_vtable.ioctl().
  *

lib/os/CMakeLists.txt

@@ -12,6 +12,9 @@ zephyr_sources(
   )
 
 zephyr_sources_ifdef(CONFIG_FDTABLE fdtable.c)
+zephyr_syscall_header_ifdef(CONFIG_FDTABLE
+  ${ZEPHYR_BASE}/include/zephyr/sys/fdtable.h
+)
 
 zephyr_sources_ifdef(CONFIG_CBPRINTF_COMPLETE cbprintf_complete.c)
 zephyr_sources_ifdef(CONFIG_CBPRINTF_NANO cbprintf_nano.c)

lib/os/zvfs/CMakeLists.txt

@@ -2,3 +2,4 @@
 
 zephyr_library()
 zephyr_library_sources_ifdef(CONFIG_ZVFS_EVENTFD zvfs_eventfd.c)
+zephyr_library_sources_ifdef(CONFIG_ZVFS_POLL zvfs_poll.c)

lib/os/zvfs/Kconfig

@@ -16,7 +16,7 @@ if ZVFS
 
 config ZVFS_EVENTFD
 	bool "ZVFS event file descriptor support"
-	select POLL
+	imply ZVFS_POLL
 	help
 	  Enable support for ZVFS event file descriptors. An eventfd can
 	  be used as an event wait/notify mechanism together with POSIX calls
@@ -33,4 +33,22 @@ config ZVFS_EVENTFD_MAX
 
 endif # ZVFS_EVENTFD
 
+config ZVFS_POLL
+	bool "ZVFS poll"
+	select POLL
+	help
+	  Enable support for zvfs_poll().
+
+if ZVFS_POLL
+
+config ZVFS_POLL_MAX
+	int "Max number of supported zvfs_poll() entries"
+	default 6 if WIFI_NM_WPA_SUPPLICANT
+	default 4 if SHELL_BACKEND_TELNET
+	default 3
+	help
+	  Maximum number of entries supported for poll() call.
+
+endif # ZVFS_POLL
+
 endif # ZVFS

lib/os/zvfs/zvfs_poll.c

@@ -0,0 +1,212 @@
+/*
+ * Copyright (c) 2017-2018 Linaro Limited
+ * Copyright (c) 2021 Nordic Semiconductor
+ * Copyright (c) 2023 Arm Limited (or its affiliates). All rights reserved.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include <zephyr/kernel.h>
+#include <zephyr/internal/syscall_handler.h>
+#include <zephyr/sys/fdtable.h>
+
+#if defined(CONFIG_NET_SOCKETS_SOCKOPT_TLS)
+bool net_socket_is_tls(void *obj);
+#else
+#define net_socket_is_tls(obj) false
+#endif
+
+int zvfs_poll_internal(struct zvfs_pollfd *fds, int nfds, k_timeout_t timeout)
+{
+	bool retry;
+	int ret = 0;
+	int i;
+	struct zvfs_pollfd *pfd;
+	struct k_poll_event poll_events[CONFIG_ZVFS_POLL_MAX];
+	struct k_poll_event *pev;
+	struct k_poll_event *pev_end = poll_events + ARRAY_SIZE(poll_events);
+	const struct fd_op_vtable *vtable;
+	struct k_mutex *lock;
+	k_timepoint_t end;
+	bool offload = false;
+	const struct fd_op_vtable *offl_vtable = NULL;
+	void *offl_ctx = NULL;
+
+	end = sys_timepoint_calc(timeout);
+
+	pev = poll_events;
+	for (pfd = fds, i = nfds; i--; pfd++) {
+		void *ctx;
+		int result;
+
+		/* Per POSIX, negative fd's are just ignored */
+		if (pfd->fd < 0) {
+			continue;
+		}
+
+		ctx = zvfs_get_fd_obj_and_vtable(pfd->fd, &vtable, &lock);
+		if (ctx == NULL) {
+			/* Will set POLLNVAL in return loop */
+			continue;
+		}
+
+		(void)k_mutex_lock(lock, K_FOREVER);
+
+		result = zvfs_fdtable_call_ioctl(vtable, ctx, ZFD_IOCTL_POLL_PREPARE, pfd, &pev,
+						 pev_end);
+		if (result == -EALREADY) {
+			/* If POLL_PREPARE returned with EALREADY, it means
+			 * it already detected that some socket is ready. In
+			 * this case, we still perform a k_poll to pick up
+			 * as many events as possible, but without any wait.
+			 */
+			timeout = K_NO_WAIT;
+			end = sys_timepoint_calc(timeout);
+			result = 0;
+		} else if (result == -EXDEV) {
+			/* If POLL_PREPARE returned EXDEV, it means
+			 * it detected an offloaded socket.
+			 * If offloaded socket is used with native TLS, the TLS
+			 * wrapper for the offloaded poll will be used.
+			 * In case the fds array contains a mixup of offloaded
+			 * and non-offloaded sockets, the offloaded poll handler
+			 * shall return an error.
+			 */
+			offload = true;
+			if (offl_vtable == NULL || net_socket_is_tls(ctx)) {
+				offl_vtable = vtable;
+				offl_ctx = ctx;
+			}
+
+			result = 0;
+		}
+
+		k_mutex_unlock(lock);
+
+		if (result < 0) {
+			errno = -result;
+			return -1;
+		}
+	}
+
+	if (offload) {
+		int poll_timeout;
+
+		if (K_TIMEOUT_EQ(timeout, K_FOREVER)) {
+			poll_timeout = SYS_FOREVER_MS;
+		} else {
+			poll_timeout = k_ticks_to_ms_floor32(timeout.ticks);
+		}
+
+		return zvfs_fdtable_call_ioctl(offl_vtable, offl_ctx, ZFD_IOCTL_POLL_OFFLOAD, fds,
+					       nfds, poll_timeout);
+	}
+
+	timeout = sys_timepoint_timeout(end);
+
+	do {
+		ret = k_poll(poll_events, pev - poll_events, timeout);
+		/* EAGAIN when timeout expired, EINTR when cancelled (i.e. EOF) */
+		if (ret != 0 && ret != -EAGAIN && ret != -EINTR) {
+			errno = -ret;
+			return -1;
+		}
+
+		retry = false;
+		ret = 0;
+
+		pev = poll_events;
+		for (pfd = fds, i = nfds; i--; pfd++) {
+			void *ctx;
+			int result;
+
+			pfd->revents = 0;
+
+			if (pfd->fd < 0) {
+				continue;
+			}
+
+			ctx = zvfs_get_fd_obj_and_vtable(pfd->fd, &vtable, &lock);
+			if (ctx == NULL) {
+				pfd->revents = ZVFS_POLLNVAL;
+				ret++;
+				continue;
+			}
+
+			(void)k_mutex_lock(lock, K_FOREVER);
+
+			result = zvfs_fdtable_call_ioctl(vtable, ctx, ZFD_IOCTL_POLL_UPDATE, pfd,
+							 &pev);
+			k_mutex_unlock(lock);
+
+			if (result == -EAGAIN) {
+				retry = true;
+				continue;
+			} else if (result != 0) {
+				errno = -result;
+				return -1;
+			}
+
+			if (pfd->revents != 0) {
+				ret++;
+			}
+		}
+
+		if (retry) {
+			if (ret > 0) {
+				break;
+			}
+
+			timeout = sys_timepoint_timeout(end);
+
+			if (K_TIMEOUT_EQ(timeout, K_NO_WAIT)) {
+				break;
+			}
+		}
+	} while (retry);
+
+	return ret;
+}
+
+int z_impl_zvfs_poll(struct zvfs_pollfd *fds, int nfds, int poll_timeout)
+{
+	k_timeout_t timeout;
+
+	if (poll_timeout < 0) {
+		timeout = K_FOREVER;
+	} else {
+		timeout = K_MSEC(poll_timeout);
+	}
+
+	return zvfs_poll_internal(fds, nfds, timeout);
+}
+
+#ifdef CONFIG_USERSPACE
+static inline int z_vrfy_zvfs_poll(struct zvfs_pollfd *fds, int nfds, int timeout)
+{
+	struct zvfs_pollfd *fds_copy;
+	size_t fds_size;
+	int ret;
+
+	/* Copy fds array from user mode */
+	if (size_mul_overflow(nfds, sizeof(struct zvfs_pollfd), &fds_size)) {
+		errno = EFAULT;
+		return -1;
+	}
+	fds_copy = k_usermode_alloc_from_copy((void *)fds, fds_size);
+	if (!fds_copy) {
+		errno = ENOMEM;
+		return -1;
+	}
+
+	ret = z_impl_zvfs_poll(fds_copy, nfds, timeout);
+
+	if (ret >= 0) {
+		k_usermode_to_copy((void *)fds, fds_copy, fds_size);
+	}
+	k_free(fds_copy);
+
+	return ret;
+}
+#include <zephyr/syscalls/zvfs_poll_mrsh.c>
+#endif

lib/posix/options/Kconfig.device_io

@@ -6,9 +6,10 @@ menu "POSIX device I/O"
 
 config POSIX_DEVICE_IO
 	bool "POSIX device I/O [EXPERIMENTAL]"
-	select FDTABLE
 	select EXPERIMENTAL
 	select REQUIRES_FULL_LIBC
+	select ZVFS
+	select ZVFS_POLL
 	help
 	  Select 'y' here and Zephyr will provide an implementation of the POSIX_DEVICE_IO Option
 	  Group such as FD_CLR(), FD_ISSET(), FD_SET(), FD_ZERO(), close(), fdopen(), fileno(), open(),

lib/posix/options/device_io.c

@@ -37,8 +37,7 @@ FUNC_ALIAS(open, _open, int);
 
 int poll(struct pollfd *fds, int nfds, int timeout)
 {
-	/* TODO: create  zvfs_poll() and dispatch to subsystems based on file type */
-	return zsock_poll(fds, nfds, timeout);
+	return zvfs_poll(fds, nfds, timeout);
 }
 
 ssize_t pread(int fd, void *buf, size_t count, off_t offset)

subsys/net/lib/sockets/Kconfig

@@ -5,7 +5,8 @@
 
 menuconfig NET_SOCKETS
 	bool "BSD Sockets compatible API"
-	select FDTABLE
+	select ZVFS
+	select ZVFS_POLL
 	help
 	  Provide BSD Sockets like API on top of native Zephyr networking API.
 
@@ -47,9 +48,7 @@ config NET_SOCKETS_POSIX_NAMES
 
 config NET_SOCKETS_POLL_MAX
 	int "Max number of supported poll() entries"
-	default 6 if WIFI_NM_WPA_SUPPLICANT
-	default 4 if SHELL_BACKEND_TELNET
-	default 3
+	default ZVFS_POLL_MAX
 	help
 	  Maximum number of entries supported for poll() call.