devtmpfs文件系统创建设备节点

  1. 分类： LINUX

    原文地址：devtmpfs文件系统创建设备节点 作者：wangbaolin719

  2. http://blog.chinaunix.net/uid-27097876-id-4334356.html
  4. 一、devtmpfs概述
  5. 1.devtmpfs 的功用是在 Linux 核心 启动早期建立一个初步的 /dev，令一般启动程序不用等待 udev，缩短 GNU/Linux 的开机时间。

  7. 2.重要解释
  8. Devtmpfs lets the kernel create a tmpfs very early at kernel initialization, before any driver core device is registered. Every device with a major/minor will have a device node created in this tmpfs instance. After the rootfs is mounted by the kernel, the populated tmpfs is mounted at /dev. In initramfs, it can be moved to the manually mounted root filesystem before /sbin/init is executed.

  10. 3.menuconfig 中加入devtmpfs支持
  11. make menuconfig-->Device Drivers-->Generic Driver Options
  12. Maintain a devtmpfs filesystem to mount at /dev
  13. Automount devtmpfs at /dev, after the kernel mounted the rootfs

  15. 4.df -T显示devtmpfs
  16. 文件系统 类型 1K-块 已用 可用 已用% 挂载点
  17. /dev/sda1 ext4 31621016 14985712 15029008 50% /
  18. none devtmpfs 399552 276 399276 1% /dev
  19. none tmpfs 403804 24 403780 1% /dev/shm
  20. none tmpfs 403804 108 403696 1% /var/run
  21. none tmpfs 403804 0 403804 0% /var/lock
  22. none tmpfs 403804 0 403804 0% /lib/init/rw
  23. .host:/ vmhgfs 67151668 54038400 13113268 81% /mnt/hgfs
  24. /dev/loop0 ext2 16119 8528 6772 56% /mnt/loop

  26. 二、devtmpfs文件系统初始化
  27. void __init driver_init(void)
  28. {
  29.     /* These are the core pieces */
  30.     devtmpfs_init();//devtmpfs文件系统初始化
  31.     devices_init();
  32.     buses_init();
  33.     classes_init();
  34.     firmware_init();
  35.     hypervisor_init();
  36.     platform_bus_init();
  37.     system_bus_init();
  38.     cpu_dev_init();
  39.     memory_dev_init();
  40. }

  42. static struct file_system_type dev_fs_type = {
  43.     .name = "devtmpfs",
  44.     .mount = dev_mount,
  45.     .kill_sb = kill_litter_super,
  46. };
  47.    
  48. int __init devtmpfs_init(void)
  49. {
  50.     int err = register_filesystem(&dev_fs_type);//注册dev_fs_type文件系统，即将dev_fs_type添加到内核全局总链表中file_systems
  51.     if (err) {
  52.         printk(KERN_ERR "devtmpfs: unable to register devtmpfs ""type %i\n", err);
  53.         return err;
  54.     }
  55.     
  56.     thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");//创建并启动一个内核线程devtmpfsd
  57.     if (!IS_ERR(thread)) {
  58.         wait_for_completion(&setup_done);//进行一个不可打断的等待,允许一个线程告诉另一个线程工作已经完成
  59.     } else {
  60.         err = PTR_ERR(thread);
  61.         thread = NULL;
  62.     }
  63.     
  64.     if (err) {
  65.         printk(KERN_ERR "devtmpfs: unable to create devtmpfs %i\n", err);
  66.         unregister_filesystem(&dev_fs_type);
  67.         return err;
  68.     }
  69.     
  70.     printk(KERN_INFO "devtmpfs: initialized\n");
  71.     return 0;
  72. }

  74. //请求创建设备节点的请求队列req结构
  75. static struct req {
  76.     struct req *next;
  77.     struct completion done;
  78.     int err;
  79.     const char *name;
  80.     umode_t mode;//0代表删除
  81.     struct device *dev;
  82. } *requests;

  84. //内核线程devtmpfsd
  85. static int devtmpfsd(void *p)
  86. {
  87.     char options[] = "mode=0755";
  88.     int *err = p;
  89.     
  90.     *err = sys_unshare(CLONE_NEWNS);
  91.     if (*err)
  92.         goto out;
  93.         
  94.     //挂载devtmpfs文件系统
  95.     //devtmpfs是待安装设备的路径名，“/”是安装点路径名，”devtmpfs“表示文件系统类型，MS_SILENT=32768，即0x8000
  96.     *err = sys_mount("devtmpfs", "/", "devtmpfs", MS_SILENT, options);
  97.     if (*err)
  98.         goto out;
  99.     sys_chdir("/.."); //将进程的当前工作目录(pwd)设定为devtmpfs文件系统的根目录/* will traverse into overmounted root */
  100.     sys_chroot(".");
  101.     complete(&setup_done);//允许一个线程告诉另一个线程工作已经完成
  102.     while (1) {
  103.         spin_lock(&req_lock);
  104.         while (requests) {//扫描请求链表，每当要创建一个设备节点时，都需要向requests链表中添加请求
  105.             struct req *req = requests;//赋值给临时req
  106.             requests = NULL;//清空
  107.             spin_unlock(&req_lock);
  108.             while (req) {//遍历刚才requests的请求链表
  109.                 struct req *next = req->next;
  110.                 req->err = handle(req->name, req->mode, req->dev);//对链表中的每一个请求调用handle函数
  111.                 complete(&req->done);
  112.                 req = next;
  113.             }
  114.             spin_lock(&req_lock);
  115.         }
  116.         __set_current_state(TASK_INTERRUPTIBLE);//设置为睡眠状态
  117.         spin_unlock(&req_lock);
  118.         schedule();//系统切换
  119.     }
  120.     return 0;
  121. out:
  122.     complete(&setup_done);
  123.     return *err;
  124. }

  126. static int handle(const char *name, umode_t mode, struct device *dev)
  127. {
  128.     if (mode)
  129.         return handle_create(name, mode, dev);
  130.     else
  131.         return handle_remove(name, dev);
  132. }

  134. static int handle_create(const char *nodename, umode_t mode, struct device *dev)
  135. {
  136.     struct dentry *dentry;
  137.     struct path path;
  138.     int err;
  139.     
  140.     //查找节点名称的路径以及返回节点对应的父目录dentry结构，即在此目录下创建一个设备节点，即是/dev目录对应的dentry结构
  141.     dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
  142.     if (dentry == ERR_PTR(-ENOENT)) {
  143.         create_path(nodename);
  144.         dentry = kern_path_create(AT_FDCWD, nodename, &path, 0);
  145.     }
  146.     if (IS_ERR(dentry))
  147.         return PTR_ERR(dentry);
  148.     
  149.     //创建设备节点
  150.     err = vfs_mknod(path.dentry->d_inode,dentry, mode, dev->devt);
  151.     if (!err) {
  152.         struct iattr newattrs;
  153.         newattrs.ia_mode = mode;/* fixup possibly umasked mode */
  154.         newattrs.ia_valid = ATTR_MODE;
  155.         mutex_lock(&dentry->d_inode->i_mutex);
  156.         notify_change(dentry, &newattrs);
  157.         mutex_unlock(&dentry->d_inode->i_mutex);
  158.         dentry->d_inode->i_private = &thread;/* mark as kernel-created inode */
  159.     }
  160.     done_path_create(&path, dentry);//与前边kern_path_create对应，减少path和dentry的计数等
  161.     return err;
  162. }

  164. int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
  165. {
  166.     int error = may_create(dir, dentry);//检查是否可以创建设备文件节点
  167.     
  168.     if (error)
  169.         return error;
  170.     
  171.     //必须是字符设备或者块设备，且具有创建节点的权限
  172.     if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
  173.         return -EPERM;
  174.     
  175.     if (!dir->i_op->mknod)
  176.         return -EPERM;
  177.     
  178.     error = devcgroup_inode_mknod(mode, dev);
  179.     if (error)
  180.         return error;
  181.     
  182.     error = security_inode_mknod(dir, dentry, mode, dev);
  183.     if (error)
  184.         return error;
  185.     
  186.     //调用具体文件系统的mknod()函数
  187.     //mount时调用shmem_fill_super()-->shmem_get_inode()分配inode节点时做出的初始化
  188.     /*那么在shmem_get_inode中
  189.         caseS_IFDIR:
  190.         inc_nlink(inode);
  191.         inode->i_size= 2 * BOGO_DIRENT_SIZE;
  192.         inode->i_op= &shmem_dir_inode_operations;
  193.         inode->i_fop= &simple_dir_operations;
  194.         由于mountpoint是dev这个目录，所以dev对应的inode的i_op就是shmem_dir_inode_operations。
  195.         staticconst struct inode_operations shmem_dir_inode_operations = {
  196.             #ifdefCONFIG_TMPFS
  197.             .create =shmem_create,
  198.             .lookup =simple_lookup,
  199.             .link =shmem_link,
  200.             .unlink =shmem_unlink,
  201.             .symlink =shmem_symlink,
  202.             .mkdir =shmem_mkdir,
  203.             .rmdir =shmem_rmdir,
  204.             .mknod =shmem_mknod,
  205.             .rename =shmem_rename,
  206.             #endif
  207.             #ifdefCONFIG_TMPFS_POSIX_ACL
  208.             .setattr =shmem_notify_change,
  209.             .setxattr =generic_setxattr,
  210.             .getxattr =generic_getxattr,
  211.             .listxattr =generic_listxattr,
  212.             .removexattr =generic_removexattr,
  213.             .check_acl =generic_check_acl,
  214.             #endif
  215.             };
  216.         */
  217.     error = dir->i_op->mknod(dir, dentry, mode, dev);//所以这里调用的就是shmem_mknod
  218.     if (!error)
  219.         fsnotify_create(dir, dentry);
  220.     return error;
  221. }

  223. shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
  224. {
  225.     struct inode *inode;
  226.     int error = -ENOSPC;
  227.     
  228.     inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);//获得一个要创建的设备节点的inode，并初始化
  229.     if (inode) {
  230.         error = security_inode_init_security(inode, dir,&dentry->d_name,shmem_initxattrs, NULL);
  231.         if (error) {
  232.             if (error != -EOPNOTSUPP) {
  233.                 iput(inode);
  234.                 return error;
  235.             }
  236.         }
  237. #ifdef CONFIG_TMPFS_POSIX_ACL
  238.         error = generic_acl_init(inode, dir);
  239.         if (error) {
  240.             iput(inode);
  241.             return error;
  242.         }
  243. #else
  244.         error = 0;
  245. #endif
  246.         dir->i_size += BOGO_DIRENT_SIZE;
  247.         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
  248.         d_instantiate(dentry, inode);//与dentry建立关,此时就可以在/dev下看到这个字符设备节点了
  249.         dget(dentry); //递减dentry的计数
  250.     }
  251.     return error;
  252. }

  254. 三、文件系统的mount
  255. 内核主要是通过kernel_init调用prepare_namespace()函数执行安装实际根文件系统的操作：
  256. void __init prepare_namespace(void)
  257. {
  258.     int is_floppy;
  259.   
  260.     if (root_delay) {
  261.         printk(KERN_INFO "Waiting %dsec before mounting root device...\n",
  262.                root_delay);
  263.         ssleep(root_delay);
  264.     }
  265.     wait_for_device_probe();
  266.   
  267.     md_run_setup();
  268.     /* 把root_device_name变量置为从启动参数“root”中获取的设备文件名。
  269.   * 同样，把ROOT_DEV变量置为同一设备文件的主设备号和次设备号。*/
  270.     if (saved_root_name[0]) {
  271.         root_device_name = saved_root_name;
  272.         if (!strncmp(root_device_name, "mtd", 3) ||
  273.             !strncmp(root_device_name, "ubi", 3)) {
  274.             mount_block_root(root_device_name, root_mountflags);
  275.             goto out;
  276.         }
  277.         ROOT_DEV = name_to_dev_t(root_device_name);//转换为设备号/dev/mtdblock2.
  278.         if (strncmp(root_device_name, "/dev/", 5) == 0)
  279.             root_device_name += 5;
  280.     }
  281.   
  282.     if (initrd_load())
  283.         goto out;
  284.   
  285.     /* wait for any asynchronous scanning to complete */
  286.     if ((ROOT_DEV == 0) && root_wait) {
  287.         printk(KERN_INFO "Waiting for root device %s...\n",
  288.             saved_root_name);
  289.         while (driver_probe_done() != 0 ||
  290.             (ROOT_DEV = name_to_dev_t(saved_root_name)) == 0)
  291.             msleep(100);
  292.         async_synchronize_full();
  293.     }
  294.   
  295.     is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
  296.   
  297.     if (is_floppy && rd_doload && rd_load_disk(0))
  298.         ROOT_DEV = Root_RAM0;
  299.   
  300.     mount_root();
  301. out:
  302.     devtmpfs_mount("dev");//挂载devtmpfs文件系统
  303.     sys_mount(".", "/", NULL, MS_MOVE, NULL); /* 移动rootfs文件系统根目录上的已安装文件系统的安装点。 */
  304.     sys_chroot(".");
  305. }

  307. int devtmpfs_mount(const char *mntdir)
  308. {
  309.     int err;
  310.     
  311.     if (!mount_dev)
  312.         return 0;
  313.     
  314.     if (!thread)
  315.         return 0;
  316.     //将devtmpfs文件系统挂载到/dev目录下
  317.     err = sys_mount("devtmpfs", (char *)mntdir, "devtmpfs", MS_SILENT, NULL);
  318.     if (err)
  319.         printk(KERN_INFO "devtmpfs: error mounting %i\n", err);
  320.     else
  321.         printk(KERN_INFO "devtmpfs: mounted\n");
  322.     return err;
  323. }


  326. 四、devtmpfs创建节点
  327. 系统在启动过程中，扫描到的设备会通过devtmpfs_create_node()函数来添加设备节点。
  328. int devtmpfs_create_node(struct device *dev)
  329. {
  330.     const char *tmp = NULL;
  331.     struct req req;
  332.     
  333.     if (!thread)
  334.     return 0;
  335.     
  336.     req.mode = 0;
  337.     req.name = device_get_devnode(dev, &req.mode, &tmp);//获得设备名
  338.     if (!req.name)
  339.         return -ENOMEM;
  340.     
  341.     if (req.mode == 0)
  342.         req.mode = 0600;
  343.     if (is_blockdev(dev))
  344.         req.mode |= S_IFBLK;//块设备
  345.     else
  346.         req.mode |= S_IFCHR;//字符设备
  347.     
  348.     req.dev = dev;
  349.     
  350.     init_completion(&req.done);
  351.     
  352.     spin_lock(&req_lock);
  353.     req.next = requests;//请求添加到requests链表
  354.     requests = &req;
  355.     spin_unlock(&req_lock);
  356.     
  357.     wake_up_process(thread);//唤醒内核线程devtmpfsd添加设备节点
  358.     wait_for_completion(&req.done);
  359.     
  360.     kfree(tmp);
  361.     
  362.     return req.err;
  363. }

  365. const char *device_get_devnode(struct device *dev,umode_t *mode, const char **tmp)
  366. {
  367.     char *s;
  368.     
  369.     *tmp = NULL;
  370.     
  371.     /* the device type may provide a specific name */
  372.     if (dev->type && dev->type->devnode)
  373.         *tmp = dev->type->devnode(dev, mode);
  374.     if (*tmp)
  375.         return *tmp;
  376.     
  377.     /* the class may provide a specific name */
  378.     if (dev->class && dev->class->devnode)
  379.         *tmp = dev->class->devnode(dev, mode);
  380.     if (*tmp)
  381.         return *tmp;
  382.     
  383.     /* return name without allocation, tmp == NULL */
  384.     if (strchr(dev_name(dev), '!') == NULL)
  385.         return dev_name(dev);
  386.     
  387.     /* replace '!' in the name with '/' */
  388.     *tmp = kstrdup(dev_name(dev), GFP_KERNEL);
  389.     if (!*tmp)
  390.         return NULL;
  391.     while ((s = strchr(*tmp, '!')))
  392.         s[0] = '/';
  393.     return *tmp;
  394. }
posted @ 2015-09-11 16:23  mull  阅读(1431)  评论(0)    收藏  举报