Merge pull request torvalds#202 from opurdila/lkl-virtio-net-fd

lkl tools: virtio net fd: make it posix compatible

Author: Octavian Purdila

arch/lkl/kernel/setup.c

@@ -117,8 +117,6 @@ void machine_restart(char *unused)
 	machine_halt();
 }
 
-extern int lkl_netdevs_remove(void);
-
 long lkl_sys_halt(void)
 {
 	long err;
@@ -141,10 +139,8 @@ long lkl_sys_halt(void)
 	lkl_ops->sem_free(init_sem);
 
 	free_initial_syscall_thread();
-	if (lkl_netdevs_remove() == 0)
-		/* We know that there is nothing else touching our
-		 * memory. */
-		free_mem();
+
+	free_mem();
 
 	return 0;
 }

tools/lkl/include/lkl.h

@@ -270,15 +270,14 @@ struct lkl_netdev_args {
 int lkl_netdev_add(struct lkl_netdev *nd, struct lkl_netdev_args* args);
 
 /**
-* lkl_netdevs_remove - destroy all network devices
+* lkl_netdev_remove - remove a previously added network device
 *
-* Attempts to release all resources held by network devices created
+* Attempts to release all resources held by a network device created
 * via lkl_netdev_add.
 *
-* @returns 0 if all devices are successfully removed, -1 if at least
-* one fails.
+* @id - the network device id, as return by @lkl_netdev_add
 */
-int lkl_netdevs_remove(void);
+void lkl_netdev_remove(int id);
 
 /**
  * lkl_netdev_get_ifindex - retrieve the interface index for a given network

tools/lkl/include/lkl_host.h

@@ -49,22 +49,24 @@ struct lkl_dev_blk_ops {
 
 struct lkl_netdev {
 	struct lkl_dev_net_ops *ops;
-	lkl_thread_t rx_tid, tx_tid;
 	uint8_t has_vnet_hdr: 1;
 };
 
 struct lkl_dev_net_ops {
-	/* Writes a L2 packet into the net device.
+	/*
+	 * Writes a L2 packet into the net device.
 	 *
 	 * The data buffer can only hold 0 or 1 complete packets.
 	 *
 	 * @nd - pointer to the network device
 	 * @iov - pointer to the buffer vector
 	 * @cnt - # of vectors in iov.
 	 * @returns number of bytes transmitted
-	 */ 
+	 */
 	int (*tx)(struct lkl_netdev *nd, struct lkl_dev_buf *iov, int cnt);
-	/* Reads a packet from the net device.
+
+	/*
+	 * Reads a packet from the net device.
 	 *
 	 * It must only read one complete packet if present.
 	 *
@@ -75,43 +77,32 @@ struct lkl_dev_net_ops {
 	 * @iov - pointer to the buffer vector to store the packet
 	 * @cnt - # of vectors in iov.
 	 * @returns number of bytes read for success or < 0 if error
-	 */ 
+	 */
 	int (*rx)(struct lkl_netdev *nd, struct lkl_dev_buf *iov, int cnt);
+
 #define LKL_DEV_NET_POLL_RX		1
 #define LKL_DEV_NET_POLL_TX		2
-	/* Polls a net device.
-	 *
-	 * Supports only one of two events: LKL_DEV_NET_POLL_RX (readable) and
-	 * LKL_DEV_NET_POLL_TX (writable). Blocks until one event is available.
-	 *
-	 * Implementation can assume only one of LKL_DEV_NET_POLL_RX or
-	 * LKL_DEV_NET_POLL_TX is set in @events.
+#define LKL_DEV_NET_POLL_HUP		4
+
+	/*
+	 * Polls a net device.
 	 *
-	 * Both LKL_DEV_NET_POLL_RX and LKL_DEV_NET_POLL_TX can be
-	 * level-triggered or edge-triggered. When it's level-triggered,
-	 * rx/tx thread can become a busy waiting loop which burns out CPU.
-	 * This is more of a problem for tx, because LKL_DEV_NET_POLL_TX event
-	 * is present most of the time.
+	 * Supports the following events: LKL_DEV_NET_POLL_RX (readable),
+	 * LKL_DEV_NET_POLL_TX (writable) or LKL_DEV_NET_POLL_HUP (the close
+	 * operations has been issued and we need to clean up). Blocks until one
+	 * event is available.
 	 *
 	 * @nd - pointer to the network device
-	 * @events - a bit mask specifying the events to poll on. Only one of
-	 * LKL_DEV_NET_POLL_RX or LKL_DEV_NET_POLL_TX is set.
-	 * @returns the events triggered for success. -1 for failure.
 	 */
-	int (*poll)(struct lkl_netdev *nd, int events);
-	/* Closes a net device.
-	 *
-	 * Implementation can choose to release any resources releated to it. In
-	 * particular, the polling threads are to be killed in this function.
-	 *
-	 * Implemenation must guarantee it's safe to call free_mem() after this
-	 * function call.
-	 *
-	 * Not implemented by all netdev types.
+	int (*poll)(struct lkl_netdev *nd);
+
+	/*
+	 * Closes a net device.
 	 *
-	 * @returns 0 for success. -1 for failure.
+	 * Implementation must release its resources and poll must wakeup and
+	 * return LKL_DEV_NET_POLL_HUP.
 	 */
-	int (*close)(struct lkl_netdev *nd);
+	void (*close)(struct lkl_netdev *nd);
 };
 
 #ifdef __cplusplus

tools/lkl/lib/Build

@@ -12,7 +12,7 @@ lkl-y += virtio.o
 lkl-y += dbg.o
 lkl-y += dbg_handler.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net.o
-lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_linux_fdnet.o
+lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_fd.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_tap.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_raw.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_macvtap.o

tools/lkl/lib/hijack/hijack.c

@@ -27,6 +27,7 @@
 #include <lkl_host.h>
 
 #include "xlate.h"
+#include "init.h"
 
 static int is_lklfd(int fd)
 {
@@ -75,6 +76,29 @@ static host_call host_calls[__lkl__NR_syscalls];
 	asm(".global " #name);						\
 	asm(".set " #name "," #name "_hook");				\
 
+#define HOOK_CALL_USE_HOST_BEFORE_START(name)				\
+	static void __attribute__((constructor(101)))			\
+	init_host_##name(void)						\
+	{								\
+		host_calls[__lkl__NR_##name] = resolve_sym(#name);	\
+	}								\
+									\
+	long name##_hook(long p1, long p2, long p3, long p4, long p5,	\
+			 long p6)					\
+	{								\
+		long p[6] = {p1, p2, p3, p4, p5, p6 };			\
+									\
+		if (!host_calls[__lkl__NR_##name])			\
+			host_calls[__lkl__NR_##name] = resolve_sym(#name); \
+		if (!lkl_running)					\
+			return host_calls[__lkl__NR_##name](p1, p2, p3,	\
+							    p4, p5, p6); \
+									\
+		return lkl_set_errno(lkl_syscall(__lkl__NR_##name, p));	\
+	}								\
+	asm(".global " #name);						\
+	asm(".set " #name "," #name "_hook")
+
 #define HOST_CALL(name)							\
 	static long (*host_##name)();					\
 	static void __attribute__((constructor(101)))			\
@@ -131,7 +155,7 @@ HOOK_FD_CALL(read)
 HOOK_FD_CALL(recvfrom)
 HOOK_FD_CALL(recv)
 HOOK_FD_CALL(epoll_wait)
-HOOK_CALL(pipe);
+HOOK_CALL_USE_HOST_BEFORE_START(pipe);
 
 HOST_CALL(setsockopt);
 int setsockopt(int fd, int level, int optname, const void *optval,
@@ -264,7 +288,7 @@ int select(int nfds, fd_set *r, fd_set *w, fd_set *e, struct timeval *t)
 	return lkl_call(__lkl__NR_select, 5, nfds, r, w, e, t);
 }
 
-HOOK_CALL(epoll_create)
+HOOK_CALL_USE_HOST_BEFORE_START(epoll_create);
 
 HOST_CALL(epoll_ctl);
 int epoll_ctl(int epollfd, int op, int fd, struct epoll_event *event)
@@ -282,6 +306,12 @@ int epoll_ctl(int epollfd, int op, int fd, struct epoll_event *event)
 
 int eventfd(unsigned int count, int flags)
 {
+	if (!lkl_running) {
+		int (*f)(unsigned int, int) = resolve_sym("eventfd");
+
+		return f(count, flags);
+	}
+
 	return lkl_sys_eventfd2(count, flags);
 }
 

tools/lkl/lib/hijack/init.c

@@ -25,7 +25,6 @@
 #include <lkl_host.h>
 
 #include "xlate.h"
-#include "../virtio_net_linux_fdnet.h"
 
 #define __USE_GNU
 #include <dlfcn.h>
@@ -155,13 +154,6 @@ static void mount_cmds_exec(char *_cmds, int (*callback)(char*))
 	free(cmds);
 }
 
-void fixup_netdev_linux_fdnet_ops(void)
-{
-	/* It's okay if this is NULL, because then netdev close will
-	 * fall back onto an uncloseable implementation. */
-	lkl_netdev_linux_fdnet_ops.eventfd = dlsym(RTLD_NEXT, "eventfd");
-}
-
 static void PinToCpus(const cpu_set_t* cpus)
 {
 	if (sched_setaffinity(0, sizeof(cpu_set_t), cpus)) {
@@ -184,12 +176,14 @@ static void PinToFirstCpu(const cpu_set_t* cpus)
 	}
 }
 
-int lkl_debug;
+int lkl_debug, lkl_running;
+
+static int nd_id = -1;
 
 void __attribute__((constructor(102)))
 hijack_init(void)
 {
-	int ret, i, dev_null, nd_id = -1, nd_ifindex = -1;
+	int ret, i, dev_null, nd_ifindex = -1;
 	/* OBSOLETE: should use IFTYPE and IFPARAMS */
 	char *tap = getenv("LKL_HIJACK_NET_TAP");
 	char *iftype = getenv("LKL_HIJACK_NET_IFTYPE");
@@ -261,9 +255,6 @@ hijack_init(void)
 	if (single_cpu_mode == 2)
 		PinToFirstCpu(&ori_cpu);
 
-	/* Must be run before lkl_netdev_tap_create */
-	fixup_netdev_linux_fdnet_ops();
-
 	if (tap) {
 		fprintf(stderr,
 			"WARN: variable LKL_HIJACK_NET_TAP is now obsoleted.\n"
@@ -329,6 +320,8 @@ hijack_init(void)
 		return;
 	}
 
+	lkl_running = 1;
+
 	/* restore cpu affinity */
 	if (single_cpu_mode)
 		PinToCpus(&ori_cpu);
@@ -440,6 +433,8 @@ hijack_fini(void)
 	for (i = 0; i < LKL_FD_OFFSET; i++)
 		lkl_sys_close(i);
 
+	if (nd_id >= 0)
+		lkl_netdev_remove(nd_id);
 
 	lkl_sys_halt();
 }

tools/lkl/lib/hijack/init.h

@@ -0,0 +1,6 @@
+#ifndef _LKL_HIJACK_INIT_H
+#define _LKL_HIJACK_INIT_H
+
+extern int lkl_running;
+
+#endif /*_LKL_HIJACK_INIT_H */