【Skynet】Socket源码剖析二

socket_server 层使用简单封装后的 epoll ，向上提供一些列 socket_server_*** 的API。

skynet_socket封装：

为了进一步适用于Skynet框架，又进行一步对socket_server进行了封装，所有常用的接口都封装在skynet_socket.c 。相比 socket_server.c 的1600多行代码，skynet_socket.c 则只有200多行，其主要功能都由 socket_server 提供的API来完成。

先来看看 skynet_socket.h

#ifndef skynet_socket_h#define skynet_socket_hstruct skynet_context;// 可以看做是对 socket_server 里返回值类型的封装#define SKYNET_SOCKET_TYPE_DATA 1#define SKYNET_SOCKET_TYPE_CONNECT 2#define SKYNET_SOCKET_TYPE_CLOSE 3#define SKYNET_SOCKET_TYPE_ACCEPT 4#define SKYNET_SOCKET_TYPE_ERROR 5#define SKYNET_SOCKET_TYPE_UDP 6#define SKYNET_SOCKET_TYPE_WARNING 7struct skynet_socket_message {int type;int id;int ud;char * buffer;};void skynet_socket_init();void skynet_socket_exit();void skynet_socket_free();int skynet_socket_poll();int skynet_socket_send(struct skynet_context *ctx, int id, void *buffer, int sz);void skynet_socket_send_lowpriority(struct skynet_context *ctx, int id, void *buffer, int sz);int skynet_socket_listen(struct skynet_context *ctx, const char *host, int port, int backlog);int skynet_socket_connect(struct skynet_context *ctx, const char *host, int port);int skynet_socket_bind(struct skynet_context *ctx, int fd);void skynet_socket_close(struct skynet_context *ctx, int id);void skynet_socket_shutdown(struct skynet_context *ctx, int id);void skynet_socket_start(struct skynet_context *ctx, int id);void skynet_socket_nodelay(struct skynet_context *ctx, int id);int skynet_socket_udp(struct skynet_context *ctx, const char * addr, int port);int skynet_socket_udp_connect(struct skynet_context *ctx, int id, const char * addr, int port);int skynet_socket_udp_send(struct skynet_context *ctx, int id, const char * address, const void *buffer, int sz);const char * skynet_socket_udp_address(struct skynet_socket_message *, int *addrsz);#endif

对比一下之前 socket_server.h

#ifndef skynet_socket_server_h#define skynet_socket_server_h#include <stdint.h>#define SOCKET_DATA 0#define SOCKET_CLOSE 1#define SOCKET_OPEN 2#define SOCKET_ACCEPT 3#define SOCKET_ERROR 4#define SOCKET_EXIT 5#define SOCKET_UDP 6struct socket_server;struct socket_message {int id;uintptr_t opaque;int ud;// for accept, ud is new connection id ; for data, ud is size of data char * data;};struct socket_server * socket_server_create();void socket_server_release(struct socket_server *);int socket_server_poll(struct socket_server *, struct socket_message *result, int *more);void socket_server_exit(struct socket_server *);void socket_server_close(struct socket_server *, uintptr_t opaque, int id);void socket_server_shutdown(struct socket_server *, uintptr_t opaque, int id);void socket_server_start(struct socket_server *, uintptr_t opaque, int id);// return -1 when errorint64_t socket_server_send(struct socket_server *, int id, const void * buffer, int sz);void socket_server_send_lowpriority(struct socket_server *, int id, const void * buffer, int sz);// ctrl command below returns idint socket_server_listen(struct socket_server *, uintptr_t opaque, const char * addr, int port, int backlog);int socket_server_connect(struct socket_server *, uintptr_t opaque, const char * addr, int port);int socket_server_bind(struct socket_server *, uintptr_t opaque, int fd);// for tcpvoid socket_server_nodelay(struct socket_server *, int id);struct socket_udp_address;int socket_server_udp(struct socket_server *, uintptr_t opaque, const char * addr, int port);int socket_server_udp_connect(struct socket_server *, int id, const char * addr, int port);int64_t socket_server_udp_send(struct socket_server *, int id, const struct socket_udp_address *, const void *buffer, int sz);const struct socket_udp_address * socket_server_udp_address(struct socket_server *, struct socket_message *, int *addrsz);struct socket_object_interface {void * (*buffer)(void *);int (*size)(void *);void (*free)(void *);};void socket_server_userobject(struct socket_server *, struct socket_object_interface *soi);#endif

很是相似吧。

核心是 skynet_socket_poll()，该函数调用了socket_server 层的核心 socket_server_poll()，写法类似 socket-server 中 test.c 的 _poll() 函数，用于处理 socket_server_poll() 的返回值。

int skynet_socket_poll() {struct socket_server *ss = SOCKET_SERVER;assert(ss);struct socket_message result;int more = 1;int type = socket_server_poll(ss, &result, &more);switch (type) {case SOCKET_EXIT:return 0;case SOCKET_DATA:forward_message(SKYNET_SOCKET_TYPE_DATA, false, &result);break;case SOCKET_CLOSE:forward_message(SKYNET_SOCKET_TYPE_CLOSE, false, &result);break;case SOCKET_OPEN:forward_message(SKYNET_SOCKET_TYPE_CONNECT, true, &result);break;case SOCKET_ERROR:forward_message(SKYNET_SOCKET_TYPE_ERROR, true, &result);break;case SOCKET_ACCEPT:forward_message(SKYNET_SOCKET_TYPE_ACCEPT, true, &result);break;case SOCKET_UDP:forward_message(SKYNET_SOCKET_TYPE_UDP, false, &result);break;default:skynet_error(NULL, "Unknown socket message type %d.",type);return -1;}if (more) {return -1;}return 1;}

绝大部分的返回值都交给了 forward_message() 函数处理，并把之前得到的 result 传入。这里重点提出三种用于消息传输的结构体：

// 对应于 socket_server 服务中的消息传输struct socket_message {      int id;      uintptr_t opaque;   // 在skynet中对应一个Actor实体的handle句柄      int ud;         // 对于accept连接来说, ud是新连接的id;对于数据（data）来说, ud是数据的大小      char * data;  // 对于数据收发事件，存放数据；对于socket连接事件，存放地址；其他情况下还可以存放打印日志。 };  // 对应 skynet_socket_server 服务中消息传输struct skynet_socket_message {int type;int id;int ud;char * buffer;};// 在skynet不同服务（Actor）间进行通信struct skynet_message {uint32_t source;int session;void * data;size_t sz;};

再来看看 forward_message()：

// mainloop threadstatic voidforward_message(int type, bool padding, struct socket_message * result) {struct skynet_socket_message *sm;size_t sz = sizeof(*sm);if (padding) {if (result->data) {size_t msg_sz = strlen(result->data);if (msg_sz > 128) {msg_sz = 128;}sz += msg_sz;} else {result->data = "";}}sm = (struct skynet_socket_message *)skynet_malloc(sz);// skynet_malloc() 内部使用了 jemalloc 分配内存sm->type = type;sm->id = result->id;sm->ud = result->ud;if (padding) {sm->buffer = NULL;memcpy(sm+1, result->data, sz - sizeof(*sm));} else {sm->buffer = result->data;}struct skynet_message message;message.source = 0;message.session = 0;message.data = sm;message.sz = sz | ((size_t)PTYPE_SOCKET << MESSAGE_TYPE_SHIFT);// 很有意思的写法，其实就是将类型 PTYPE_SOCKET 值置于 sz 的高8bit，再赋值给 message.sz// sz的值最大不超过sizeof(struct skynet_socket_message) + 128，该值并不大，高八位并没有值，没有数据覆盖问题。if (skynet_context_push((uint32_t)result->opaque, &message)) {// 看到 opaque 的作用了吧，其实就是上层handle的标记，按这个标记将信息向上层传递// don't call skynet_socket_close here (It will block mainloop)skynet_free(sm->buffer);skynet_free(sm);}}

skynet_context_push() 已经涉及到 skynet 的消息调度，关于 skynet 消息调度机制，后面将有博文专门学习。这里我们就不探寻 skynet_socket_push() 之后的步骤了。

lua_socket与lua封装：

lua_socket 这一层将 C 代码封装成 lua 接口，最后在进行一次lua封装完成最终提供给用户的lua API。

intluaopen_socketdriver(lua_State *L) {// socketdriver，对应lua中 require “socketdriver”luaL_checkversion(L);// 参见云风：https://blog.codingnow.com/2012/01/lua_link_bug.htmlluaL_Reg l[] = {{ "buffer", lnewbuffer },// {"(lua调用时使用的函数名)",C定义函数名}{ "push", lpushbuffer },{ "pop", lpopbuffer },{ "drop", ldrop },{ "readall", lreadall },{ "clear", lclearbuffer },{ "readline", lreadline },{ "str2p", lstr2p },{ "header", lheader },{ "unpack", lunpack },{ NULL, NULL },};luaL_newlib(L,l);luaL_Reg l2[] = {{ "connect", lconnect },{ "close", lclose },{ "shutdown", lshutdown },{ "listen", llisten },{ "send", lsend },{ "lsend", lsendlow },{ "bind", lbind },{ "start", lstart },{ "nodelay", lnodelay },{ "udp", ludp },{ "udp_connect", ludp_connect },{ "udp_send", ludp_send },{ "udp_address", ludp_address },{ NULL, NULL },};lua_getfield(L, LUA_REGISTRYINDEX, "skynet_context");// 这两句话大致就是存在lua栈LUA_REGISTRYINDEX处的 skynet_context 变量压入栈顶，同时将其首地址交给 ctx struct skynet_context *ctx = lua_touserdata(L,-1);// 可是 LUA_REGISTRYINDEX 到底存放了什么，什么时候赋的值？if (ctx == NULL) {return luaL_error(L, "Init skynet context first");}luaL_setfuncs(L,l2,1);return 1;}

向lua注册的函数可以参见表 l 和 l2。我们以 llisten() 为例，它调用了 C 接口。

static intllisten(lua_State *L) {const char * host = luaL_checkstring(L,1);// 从栈获取 address 字符串int port = luaL_checkinteger(L,2);// 从栈获取 port 整型int backlog = luaL_optinteger(L,3,BACKLOG);// 从栈获取 backlog 整型，若没有该参数，使用默认值BACKLOGstruct skynet_context * ctx = lua_touserdata(L, lua_upvalueindex(1));int id = skynet_socket_listen(ctx, host,port,backlog);if (id < 0) {return luaL_error(L, "Listen error");}lua_pushinteger(L,id);// 将返回值压栈return 1;// 返回值个数}

接着是最后一次封装：

local driver = require "socketdriver"function socket.listen(host, port, backlog)if port == nil thenhost, port = string.match(host, "([^:]+):(.+)$")port = tonumber(port)endreturn driver.listen(host, port, backlog)end

有的函数可能没做封装，直接赋值，比如 socket.write = assert(driver.send)，socket.lwrite = assert(driver.lsend) 的写法。最后看看云风wiki上向外提供的对应 lua API：

socket.listen(address, port) 监听一个端口，返回一个 id ，供 start 使用。// 官方文档就是这样写的。可能有笔误，没写 backlog 参数。

这里有几个相关的宏，目前还不是很清楚其用法：

// ./3rd/lua/luaconfig.h/*@@ LUAI_MAXSTACK limits the size of the Lua stack.** CHANGE it if you need a different limit. This limit is arbitrary;** its only purpose is to stop Lua from consuming unlimited stack** space (and to reserve some numbers for pseudo-indices).*/#if LUAI_BITSINT >= 32#define LUAI_MAXSTACK1000000#else#define LUAI_MAXSTACK15000#endif// ./3rd/lua/lua.h/*** Pseudo-indices** (-LUAI_MAXSTACK is the minimum valid index; we keep some free empty** space after that to help overflow detection)*/#define LUA_REGISTRYINDEX(-LUAI_MAXSTACK - 1000)#define lua_upvalueindex(i)(LUA_REGISTRYINDEX - (i))

其他接口类似。

接着看看 skynet 向外提供的 socket lua 接口，常见的接口如下，更多详细内容见 ./lualib/socket.lua，以及 skynet Socket WiKi

*   发起一个TCP连接：       socket.open(address, port)*   启动epoll监听：       socket.start(id)*   从socket读数据：    socket.read(id, sz)*   向socket写数据：     socket.write(id, str)*   开启一个TCP监听：          socket.listen(address, port)*   关闭socket：           socket.close(id)