基于NETLINK的内核与用户空间共享内存的实现_VMware, Unix及操作系统讨论区_Weblogic技术|Tuxedo技术|中间件技术|Oracle论坛|JAVA论坛|Linux/Unix技术|hadoop论坛

总帖数

每页帖数

1/1页

返回列表

发起投票

查看: 4347 | 回复: 0

主题： 基于NETLINK的内核与用户空间共享内存的实现

barry

注册用户

等级：中校
经验：1534
发帖：236
精华：2
注册：2012-1-13
状态：离线
发送短消息息给barry

加好友发送短消息息给barry

发消息

发表于：

2012-2-10 10:47:19 | [全部帖] [楼主帖]

楼主

http://www.chinaunix.net 作者:bripengandre  发表于：2009-07-12 10:06:59 【发表评论】【查看原文】【Linux讨论区】【关闭】
[size=5]    ps:刚不小把文章发《内核问题》版去了，在这个版找不到还以为被删了，小气了下~~
还是不会从blog导入文章，只好复制粘贴了，原文在my blog：[url=http://blog.chinaunix.net/u3/94771/showart_1945422.html]http://blog.chinaunix.net/u3/94771/showart_1945422.html
[/color][/size][align=center][size=5][color=#0000ff]基于NETLINK的内核与用户空间共享内存的实现[/color][/size][/align][align=center][size=2][color=#000000]author:bripengandre Email:bripengandre@126.com[/color][/size][/align][size=4][color=#0080ff]一、前言[/color][/size]
    前些日子，开发中用到了netlink来实现内核与用户空间共享内存，写点笔记与大家分享。因为我对这块也不了解，写出来的东西一定存在很多错误，请大家批评指正~
[size=2]    内核与用户空间共享内存的关键是，用户空间必须得知共享内存的起始地址，这就要求内核空间应该有一种通信机制来通知用户空间。已经有[color=#0000ff]Godbach版主等人[/color][color=#000000]用proc文件系统实现了（可以google '共享内存内核用户空间'）,很显然任何内核空间与用户空间的通信方法都可资利用。本文主要讲基于NETLINK机制的实现。[/color][/size]

[size=4][color=#0080ff]二、NETLINK简介[/color][/size]
    netlink在linux的内核与用户空间通信中用得很多（但具体例子我举不出，因为我不清楚~~请google之），其最大优势是接口与网络编程中的socket相似，且内核要主动发信息给用户空间很方便。
   但通过实践，我发现netlink通信机制的最大弊病在于其在各内核版本中接口变化太大，让人难以适从（可从后文列出的源码中的kernel_receive的声明窥一斑）。
  既然涉及到内核与用户空间两个空间，就应该在两个空间各有一套接口。用户空间的接口很简单，与一般的socket接口相似，内核空间则稍先复杂，但简单的应用只需简单地了解即可：首先也是建立描述符，建立描述符时会注册一个回调函数（源码中的kernel_receive即是），然后当用户空间有消息发过来时，我们的函数将被调用，显然在这个函数里我们可做相应的处理；当内核要主动发消息给用户进程时，直接调用一个类send函数即可（netlink_unicast系列函数）。当然这个过程中，有很多结构体变量需要填充。具体用法请google，我差不多忘光了~。

[size=2][size=4][color=#0080ff]三、基于netlink的共享内存[/color][/size]   [/size]
   这里的共享内存是指内核与用户空间，而不是通常的用户进程间的。
   大概流程如下。
  内核：__get_free__pages分配连续的物理内存页（貌似返回的其实是虚拟地址）-->SetPageReserved每一页（每一页都需这个操作，参见源码）-->如果用户空间通过netlink要求获取共享内存的起始物理地址，将__get_free__pages返回的地址__pa下发给用户空间。
   用户空间：open "/dev/shm"（一个读写物理内存的设备，具体请google"linux读写物理内存"）-->发netlink消息给内核，得到共享内存的起始物理地址-->mmap上步得到的物理地址。

[size=4][color=#0080ff]四、源码说明[/color][/size]
   正如二中提到的，netlink接口在各版本中接口变化很大，本人懒惰及时间紧，只实验了比较新的内核2.6.25，源码如需移植到老版本上，需要一些改动，敬请原谅。
   另外，由于本源码是从一个比较大的程序里抠出来的，所以命名什么的可能有点怪异~
  源码包括shm_k.c（内核模块）和用户空间程序（shm_u.c）两部分。shm_k.c模块的工作是：分配8KB内存供共享，并把前几十个字节置为“hello, use share memory withnetlink"字样。shm_u.c工作是：读取共享内存的前几十个字节，将内容输出在stdout上。
[size=2]   [color=#0000ff] [/color][color=#0000ff]特别说明：[/color][color=#ff00ff]该程序只适用于2.6.25左右的新版本！[/color][color=#0000ff]用__get_free_pages分配连续内存时，宜先用get_order获取页数，然后需将各页都SetPageReserved下，同样地，释放内存时，需要对每一页调用ClearPageReserved。[/color][/size]
[size=2][color=#0000ff]    我成功用该程序分配了4MB共享内存，运行还比较稳定。[/color][color=#ff00ff]因为linux内核的默认设置，一般情况下用get_free_pages只能分配到4MB内存左右，如需增大，可能需改相应的参数并重新编译内核[/color][/size][size=2][color=#ff00ff]。[/color][/size]

[size=2][/size]

[size=4][color=deepskyblue]五、运行结果[/color][/size]
[size=4][color=#00bfff] [size=2][color=#000000]我是在mpc8377上运行的，全过程如下（有少许编辑，如dmesg信息只列出新增的）。[/color][/size][/color][/size]

-sh-2.05b# insmod shm_k.ko
-sh-2.05b# dmesg
SHM_TEST: linux version:00020619
SHM_TEST: init_netlink ok.
SHM_TEST: size=00002000, page_cnt=2
SHM_TEST: __get_free_pages ok.
SHM_TEST: init_mem_pool ok.
-sh-2.05b# ./shm_u
SHM_TEST: 24, errno=0.Success, 0defe000, 00002000
the first 30 bytes of shm are: hello, use share memory with netlink
-sh-2.05b# dmesg
SHM_TEST: begin kernel_receive
SHM_TEST: receiv TA_GET_SHM_INFO
SHM_TEST: nlk_get_mem_addr ok
-sh-2.05b# rmmod shm_k
-sh-2.05b# dmesg
SHM_TEST: ta_exit ok.
-sh-2.05b#
[size=2][color=#ff00ff][/color][/size]

[size=4][color=#0080ff]六、内核源码[/color][/size]

[size=2] [color=#ff00ff] 1、common.h（内核与用户空间都用到的头文件）[/color][/size]

#ifndef _COMMON_H_

#define _COMMON_H_

/* protocol type */

#define SHM_NETLINK 30

/* message type */

#define SHM_GET_SHM_INFO 1

/* you can add othe message type here */

#define SHM_WITH_NETLINK "hello, use share memory with netlink"

typedef struct _nlk_msg

{

       union _data

       {

             struct _shm_info

             {

                   uint32_t mem_addr;

                   uint32_t mem_size;

             }shm_info;


             /* you can add other content here */

       }data;

}nlk_msg_t;

#endif /* _COMMON_H_ */

[size=2][color=#ff00ff]2、shm_k.c（内核模块）[/color][/size]

#include <linux/init.h>

#include <linux/module.h>

#include <linux/version.h>

#include <linux/types.h>

#include <linux/skbuff.h>

#include <linux/netlink.h>

#include <net/sock.h>

#include <linux/spinlock.h>

#include "common.h"

#define SHM_TEST_DEBUG

#ifdef SHM_TEST_DEBUG

#define SHM_DBG(args...) printk(KERN_DEBUG "SHM_TEST: " args)

#else

#define SHM_DBG(args...)

#endif

#define SHM_ERR(args...) printk(KERN_ERR "SHM_TEST: " args)

static struct _glb_para

{

       struct _shm_para

       {

             uint32_t mem_addr; /* memory starting address */

             uint32_t mem_size; /* memory size */

             uint32_t page_cnt; /* memory page count*/

             uint16_t order;

             uint8_t mem_init_flag; /* 0, init failed; 1, init successful */

       }shm_para;


       struct sock *nlfd; /* netlink descriptor */

       uint32_t pid; /* user-space process's pid */

       rwlock_t lock;

}glb_para;

static void init_glb_para(void);

static int init_netlink(void);

static void kernel_receive(struct sk_buff* __skb);

static int nlk_get_mem_addr(struct nlmsghdr *pnhdr);

static void clean_netlink(void);

static int init_shm(void);

static void clean_shm(void);

static int __init init_shm_test(void);

static void clean_shm_test(void);

static void init_glb_para(void)

{

       memset(&glb_para, 0, sizeof(glb_para));

}

static int init_netlink(void)

{

       rwlock_init(&glb_para.lock);

       SHM_DBG("linux version:%08x\n", LINUX_VERSION_CODE);

      #if(LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))

       glb_para.nlfd = netlink_kernel_create(SHM_NETLINK, kernel_receive);

      #elif(LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24))

       glb_para.nlfd = netlink_kernel_create(SHM_NETLINK, 0, kernel_receive, THIS_MODULE));

      #else

       glb_para.nlfd = netlink_kernel_create(&init_net, SHM_NETLINK, 0, kernel_receive, NULL, THIS_MODULE);

      #endif


       if(glb_para.nlfd == NULL)

       {

             SHM_ERR("init_netlink::netlink_kernel_create error\n";);

             return (-1);

       }


       return (0);

}

static void kernel_receive(struct sk_buff* __skb)

{

      struct sk_buff *skb;

       struct nlmsghdr *nlh = NULL;

       int invalid;


      SHM_DBG("begin kernel_receive\n";);

      skb = skb_get(__skb);

      invalid = 0;

      if(skb->len >= sizeof(struct nlmsghdr))

       {

             nlh = (struct nlmsghdr *)skb->data;

             if((nlh->nlmsg_len >= sizeof(struct nlmsghdr))

             && (skb->len >= nlh->nlmsg_len))

             {

                   switch(nlh->nlmsg_type)

                   {

                         case SHM_GET_SHM_INFO:

                         SHM_DBG("receiv TA_GET_SHM_INFO\n";);

                         nlk_get_mem_addr(nlh);

                         break;

                         default:

                         break;

                   }

             }

      }

       kfree_skb(skb);

}

static int nlk_get_mem_addr(struct nlmsghdr *pnhdr)

{

       int ret, size;

       unsigned char *old_tail;

       struct sk_buff *skb;

       struct nlmsghdr *nlh;

       struct _nlk_msg *p;



       glb_para.pid = pnhdr->nlmsg_pid; /* get the user-space process's pid */


       size = NLMSG_SPACE(sizeof(struct _nlk_msg)); /* compute the needed memory size */

       if( (skb = alloc_skb(size, GFP_ATOMIC)) == NULL) /* allocate memory */

       {

             SHM_DBG("nlk_hello_test::alloc_skb error.\n";);

             return (-1);

       }


       old_tail = skb->tail;

       nlh = NLMSG_PUT(skb, 0, 0, SHM_GET_SHM_INFO, size-sizeof(struct nlmsghdr)); /* put netlink message structure into memory */


       p = NLMSG_DATA(nlh); /* get netlink message body pointer */

       p->data.shm_info.mem_addr = __pa(glb_para.shm_para.mem_addr); /*__pa:convert virtual address to physical address, which needed by/dev/mem */

       p->data.shm_info.mem_size = glb_para.shm_para.mem_size;


       nlh->nlmsg_len = skb->tail - old_tail;

       NETLINK_CB(skb).pid = 0; /* from kernel */

       NETLINK_CB(skb).dst_group = 0;

       read_lock_bh(&glb_para.lock);

       ret = netlink_unicast(glb_para.nlfd, skb, glb_para.pid, MSG_DONTWAIT); /* send message to user-space process */

       read_unlock_bh(&glb_para.lock);

       SHM_DBG("nlk_get_mem_addr ok.\n";);

       return (ret);


      nlmsg_failure:

       SHM_DBG("nlmsg_failure\n";);

       if(skb)

       {

             kfree_skb(skb);

       }

       return (-1);

}

static void clean_netlink(void)

{

       if(glb_para.nlfd != NULL)

       {

             sock_release(glb_para.nlfd->sk_socket);

       }

}

static int init_shm(void)

{

       int i;

       char *p;

       uint32_t page_addr;


       glb_para.shm_para.order = get_order(1024*8); /* allocate 8kB */

       glb_para.shm_para.mem_addr = __get_free_pages(GFP_KERNEL, glb_para.shm_para.order);

       if(glb_para.shm_para.mem_addr == 0)

       {

             SHM_ERR("init_mem_pool::__get_free_pages error.\n";);

             glb_para.shm_para.mem_init_flag = 0;

             return (-1);

       }

       else

       {

             glb_para.shm_para.page_cnt = (1<<glb_para.shm_para.order);

             glb_para.shm_para.mem_size = glb_para.shm_para.page_cnt*PAGE_SIZE;

             glb_para.shm_para.mem_init_flag = 1;

             page_addr = glb_para.shm_para.mem_addr;

             SHM_DBG("size=%08x, page_cnt=%d\n", glb_para.shm_para.mem_size, glb_para.shm_para.page_cnt);

             for(i = 0; i < glb_para.shm_para.page_cnt; i++)

             {

                   SetPageReserved(virt_to_page(page_addr)); /* reserved for used */

                   page_addr += PAGE_SIZE;

             }


             p = (char *)glb_para.shm_para.mem_addr;

             strcpy(p, SHM_WITH_NETLINK); /* write */

             SHM_DBG("__get_free_pages ok.\n";);

       }


       return (0);

}

static void clean_shm(void)

{

       int i;

       uint32_t page_addr;


       if(glb_para.shm_para.mem_init_flag == 1)

       {

             page_addr = glb_para.shm_para.mem_addr;

             for(i = 0; i < glb_para.shm_para.page_cnt; i++)

             {

                   ClearPageReserved(virt_to_page(page_addr));

                   page_addr += PAGE_SIZE;

             }

             free_pages(glb_para.shm_para.mem_addr, glb_para.shm_para.order);

       }

}

static int __init init_shm_test(void)

{

       init_glb_para();

       if(init_netlink() < 0)

       {

             SHM_ERR("init_shm_test::init_netlink error.\n";);

             return (-1);

       }

       SHM_DBG("init_netlink ok.\n";);


       if(init_shm() < 0)

       {

             SHM_ERR("init_shm_test::init_mem_pool error.\n");

             clean_shm_test();

             return (-1);

       }

       SHM_DBG("init_mem_pool ok.\n");


       return (0);

}

static void clean_shm_test(void)

{

       clean_shm();

       clean_netlink();


       SHM_DBG("ta_exit ok.\n");

}

module_init(init_shm_test);

module_exit(clean_shm_test);

MODULE_LICENSE("GPL");

MODULE_AUTHOR("bripengandre (bripengandre@126.com)");

MODULE_DESCRIPTION("Memory Share between user-space and kernel-space with netlink.");

[size=2][color=#ff00ff]3、shm_u.c(用户进程)[/color][/size]

#include <stdio.h>

#include <sys/socket.h>

#include <arpa/inet.h>

#include <linux/netlink.h>

#include <stdlib.h>

#include <string.h>

#include <errno.h>

#include <sys/types.h>

#include <unistd.h>

#include <fcntl.h>

#include <sys/mman.h>

#include "common.h"

/* netlink */

#define MAX_SEND_BUF_SIZE 2500

#define MAX_RECV_BUF_SIZE 2500

#define SHM_TEST_DEBUG

#ifdef SHM_TEST_DEBUG

#define SHM_DBG(args...) fprintf(stderr, "SHM_TEST: " args)

#else

#define SHM_DBG(args...)

#endif

#define SHM_ERR(args...) fprintf(stderr, "SHM_TEST: " args)

struct _glb_para

{

       struct _shm_para

       {

             uint32_t mem_addr;

             uint32_t mem_size;

       }shm_para;


       int nlk_fd;

       char send_buf[MAX_SEND_BUF_SIZE];

       char recv_buf[MAX_RECV_BUF_SIZE];

}glb_para;

static void init_glb_para(void);

static int create_nlk_connect(void);

static int nlk_get_shm_info(void);

static int init_mem_pool(void);

int main(int argc ,char *argv[])

{

       char *p;


       init_glb_para();

       if(create_nlk_connect() < 0)

       {

             SHM_ERR("main::create_nlk_connect error.\n");

             return (1);

       }


       if(nlk_get_shm_info() < 0)

       {

             SHM_ERR("main::nlk_get_shm_info error.\n");

             return (1);

       }


       init_mem_pool();

       /* printf the first 30 bytes */

       p = (char *)glb_para.shm_para.mem_addr;

       p[strlen(SHM_WITH_NETLINK)] = '\0';

       printf("the first 30 bytes of shm are: %s\n", p);


       return (0);

}

static void init_glb_para(void)

{

       memset(&glb_para, 0, sizeof(glb_para));

}

static int create_nlk_connect(void)

{

       int sockfd;

       struct sockaddr_nl local;


       sockfd = socket(PF_NETLINK, SOCK_RAW, SHM_NETLINK);

       if(sockfd < 0)

       {

             SHM_ERR("create_nlk_connect::socket error:%s\n", strerror(errno));

             return (-1);

       }

       memset(&local, 0, sizeof(local));

       local.nl_family = AF_NETLINK;

       local.nl_pid = getpid();

       local.nl_groups = 0;

       if(bind(sockfd, (struct sockaddr*)&local, sizeof(local)) != 0)

       {

             SHM_ERR("create_nlk_connect::bind error: %s\n", strerror(errno));

             return -1;

       }


       glb_para.nlk_fd = sockfd;


       return (sockfd);

}

static int nlk_get_shm_info(void)

{

       struct nlmsghdr *nlh;

       struct _nlk_msg *p;

       struct sockaddr_nl kpeer;

       int recv_len, kpeerlen;


       memset(&kpeer, 0, sizeof(kpeer));

       kpeer.nl_family = AF_NETLINK;

       kpeer.nl_pid = 0;

       kpeer.nl_groups = 0;


       memset(glb_para.send_buf, 0, sizeof(glb_para.send_buf));

       nlh = (struct nlmsghdr *)glb_para.send_buf;

       nlh->nlmsg_len = NLMSG_SPACE(0);

       nlh->nlmsg_flags = 0;

       nlh->nlmsg_type = SHM_GET_SHM_INFO;

       nlh->nlmsg_pid = getpid();

       sendto(glb_para.nlk_fd, nlh, nlh->nlmsg_len, 0, (struct sockaddr*)&kpeer, sizeof(kpeer));

       memset(glb_para.send_buf, 0, sizeof(glb_para.send_buf));

       kpeerlen = sizeof(struct sockaddr_nl);

      recv_len = recvfrom(glb_para.nlk_fd, glb_para.recv_buf,sizeof(glb_para.recv_buf), 0, (struct sockaddr*)&kpeer,&kpeerlen);

      p = NLMSG_DATA((struct nlmsghdr *) glb_para.recv_buf);

      SHM_DBG("%d, errno=%d.%s, %08x, %08x\n", recv_len, errno,strerror(errno),p->data.shm_info.mem_addr, p->data.shm_info.mem_size);

      glb_para.shm_para.mem_addr = p->data.shm_info.mem_addr;

      glb_para.shm_para.mem_size = p->data.shm_info.mem_size;


      return (0);

}

static int init_mem_pool(void)

{

       int map_fd;

       void *map_addr;


       map_fd = open("/dev/mem", O_RDWR);

       if(map_fd < 0)

       {

             SHM_ERR("init_mem_pool::open %s error: %s\n", "/dev/mem", strerror(errno));

             return (-1);

       }


       map_addr = mmap(0, glb_para.shm_para.mem_size, PROT_READ|PROT_WRITE, MAP_SHARED, map_fd, glb_para.shm_para.mem_addr);

       if(map_addr == NULL)

       {

             SHM_ERR("init_mem_pool::mmap error: %s\n", strerror(errno));

             return (-1);

       }

       glb_para.shm_para.mem_addr = (uint32_t)map_addr;

       return (0);

}

[size=2][color=#ff00ff]4、Makefile[/color][/size]

引用：#PREFIX = powerpc-e300c3-linux-gnu-

CC ?= $(PREFIX)gcc
KERNELDIR ?= /lib/modules/`uname -r`/build
all: modules app
obj-m:= shm_k.o
module-objs := shm_k.c
modules:
make -C $(KERNELDIR) M=`pwd` modules
app: shm_u.o
$(CC) -o shm_u shm_u.c
clean:
rm -rf *.o Module.symvers modules.order shm_u shm_k.ko shm_k.mod.c .tmp_versions .shm_k.*

[color=#000000][font=song,Verdana][size=12px]
[/size][/font]

本版精华
热门帖子

操作引用/回复

总帖数

每页帖数

1/1页

返回列表

用户登录

Weblogic中间件技术论坛

Tuxedo中间件技术论坛

数据库论坛

Java论坛

Linux/unix论坛

网站地图