浅析kern_mount加载sysfs的流程

浅析kern_mount加载sysfs的流程

   因为sysfs是一个内存文件系统,所以文件的物理存储关系就需要使用sd来维护,因此sysfs_dirent即sd就类似于硬盘中的磁道.
   sysfs文件系统是一个排它式的文件系统,不论被mount多少次都只产生一个sb超级块,
如果尝试再次mount,即尝试再次调用sysfs_get_sb获取另一个sb超级块,那么将执行atomic_inc(old->s_active);增加
已被mount的引用计数,然后如果s已经执行了alloc_super,那么调用destroy_super将其销毁,然后返回这个已被mount了的
super_block超级块old,这样就实现了sysfs文件系统不论被mount多少次都只产生一个sb超级块的效果,所以取名为get_sb_single[luther.gliethttp]

start_kernel
=>vfs_caches_init
=>mnt_init
=>sysfs_init
sysfs_mount = kern_mount(&sysfs_fs_type);
static struct file_system_type sysfs_fs_type = {
    .name = "sysfs",
    .get_sb = sysfs_get_sb,
    .kill_sb = kill_anon_super,
};
#define kern_mount(type) kern_mount_data(type, NULL)
#define MS_KERNMOUNT (1<<22) /* this is a kern_mount call */
struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
{
    return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
}
struct vfsmount *
vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
{
    struct vfsmount *mnt;
    char *secdata = NULL;
    int error;

    if (!type)
        return ERR_PTR(-ENODEV);

    error = -ENOMEM;
    mnt = alloc_vfsmnt(name);//获取一个mnt结构体,并做一些初始化工作
    if (!mnt)
        goto out;

    if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
        secdata = alloc_secdata();
        if (!secdata)
            goto out_mnt;

        error = security_sb_copy_data(data, secdata);
        if (error)
            goto out_free_secdata;
    }
//调用文件系统控制结构体的get_sb()方法,获取文件系统的超级块,这里就是sysfs_get_sb.
    error = type->get_sb(type, flags, name, data, mnt);
    if (error < 0)
        goto out_free_secdata;
    BUG_ON(!mnt->mnt_sb);

     error = security_sb_kern_mount(mnt->mnt_sb, secdata);
     if (error)
         goto out_sb;

    mnt->mnt_mountpoint = mnt->mnt_root;//mount点就是sysfs的'/'根目录项
    mnt->mnt_parent = mnt;//指向自己
    up_write(&mnt->mnt_sb->s_umount);
    free_secdata(secdata);
    return mnt;//成功完成mnt的创建和sb超级块等信息的填充[luther.gliethttp]
out_sb:
    dput(mnt->mnt_root);
    up_write(&mnt->mnt_sb->s_umount);
    deactivate_super(mnt->mnt_sb);
out_free_secdata:
    free_secdata(secdata);
out_mnt:
    free_vfsmnt(mnt);
out:
    return ERR_PTR(error);
}
struct vfsmount *alloc_vfsmnt(const char *name)
{
    struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);//从cache上摘一个slab空闲对象
    if (mnt) {
//成功获取mnt内存空间,初始化之[luther.gliethttp]
        atomic_set(&mnt->mnt_count, 1);
        INIT_LIST_HEAD(&mnt->mnt_hash);
        INIT_LIST_HEAD(&mnt->mnt_child);
        INIT_LIST_HEAD(&mnt->mnt_mounts);
        INIT_LIST_HEAD(&mnt->mnt_list);
        INIT_LIST_HEAD(&mnt->mnt_expire);
        INIT_LIST_HEAD(&mnt->mnt_share);
        INIT_LIST_HEAD(&mnt->mnt_slave_list);
        INIT_LIST_HEAD(&mnt->mnt_slave);
        if (name) {
            int size = strlen(name) + 1;
            char *newname = kmalloc(size, GFP_KERNEL);
            if (newname) {
                memcpy(newname, name, size);
                mnt->mnt_devname = newname;//比如: "sysfs"
            }
        }
    }
    return mnt;
}
//sysfs文件系统是一个排它式的文件系统,不论被mount多少次都只产生一个sb超级块,
//如果尝试再次mount,即尝试再次调用sysfs_get_sb获取另一个sb超级块,那么将执行atomic_inc(old->s_active);增加
//已被mount的引用计数,然后如果s已经执行了alloc_super,那么调用destroy_super将其销毁,然后返回这个已被mount了的
//super_block超级块old,这样就实现了sysfs文件系统不论被mount多少次都只产生一个sb超级块的效果,所以取名为get_sb_single[luther.gliethttp]
static int sysfs_get_sb(struct file_system_type *fs_type,
    int flags, const char *dev_name, void *data, struct vfsmount *mnt)
{
    return get_sb_single(fs_type, flags, data, sysfs_fill_super, mnt);
}

int get_sb_single(struct file_system_type *fs_type,
    int flags, void *data,
    int (*fill_super)(struct super_block *, void *, int),
    struct vfsmount *mnt)
{
    struct super_block *s;
    int error;

    s = sget(fs_type, compare_single, set_anon_super, NULL);//获取超级块,如果之前已经创建了sb,那么这里将不再创建,将返回上一次创建的sb,所以这表明sb超级块将只被生成1个 [luther.gliethttp]
    if (IS_ERR(s))
        return PTR_ERR(s);
    if (!s->s_root) {
//如果是第一次mount该sysfs文件系统,那么首先填充sb超级块
        s->s_flags = flags;
        error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);//这里对应sysfs_fill_super函数,细化sb超级块
        if (error) {
            up_write(&s->s_umount);
            deactivate_super(s);
            return error;
        }
        s->s_flags |= MS_ACTIVE;
    }
    do_remount_sb(s, flags, data, 0);//asks filesystem to change mount options,在sysfs中sysfs_ops没有实现remount_fs
    return simple_set_mnt(mnt, s);//将s超级块安装到mnt这个mount节点上
}
int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
{
    mnt->mnt_sb = sb;//mnt的sb
    mnt->mnt_root = dget(sb->s_root);//对应的根目录项
    return 0;
}
static int compare_single(struct super_block *s, void *p)
{
    return 1;
}
struct super_block *sget(struct file_system_type *type,
            int (*test)(struct super_block *,void *),
            int (*set)(struct super_block *,void *),
            void *data)
{
    struct super_block *s = NULL;
    struct super_block *old;
    int err;

retry:
    spin_lock(&sb_lock);
    if (test) {
        list_for_each_entry(old, &type->fs_supers, s_instances) {
//sysfs文件系统是一个排它式的文件系统,不论被mount多少次都只产生一个sb超级块,
//如果尝试再次mount,即尝试再次调用sysfs_get_sb获取另一个sb超级块,那么将执行atomic_inc(old->s_active);增加
//已被mount的引用计数,然后如果s已经执行了alloc_super,那么调用destroy_super将其销毁,然后返回这个已被mount了的
//super_block超级块old,这样就实现了sysfs文件系统不论被mount多少次都只产生一个sb超级块的效果,所以取名为get_sb_single[luther.gliethttp]
            if (!test(old, data))//compare_single一直返回1
                continue;
            if (!grab_super(old))
                goto retry;
            if (s)
                destroy_super(s);
            return old;
        }
    }
    if (!s) {
//没有找到
        spin_unlock(&sb_lock);
        s = alloc_super(type);//获取一个sb超级块的控制内存,同时做部分结构初始化
        if (!s)
            return ERR_PTR(-ENOMEM);
        goto retry;//继续尝试一次,看看是否重复了
    }
       
    err = set(s, data);//sysfs对应set_anon_super,获取一个sb超级块的设备号,major为0
    if (err) {
        spin_unlock(&sb_lock);
        destroy_super(s);
        return ERR_PTR(err);
    }
    s->s_type = type;//填充该sb超级块的type
    strlcpy(s->s_id, type->name, sizeof(s->s_id));//该sb超级块的s_id为type->name,比如:"sysfs"
    list_add_tail(&s->s_list, &super_blocks);//将该超级块添加到全局超级块链表super_blocks上
    list_add(&s->s_instances, &type->fs_supers);//sb超级块将自己添加到type->fs_supers管理链表上
    spin_unlock(&sb_lock);
    get_filesystem(type);//
    return s;
}
//type->fs_supers链表上挂上了有效的sb超级块之后,才会执行到这里
static int grab_super(struct super_block *s) __releases(sb_lock)
{
    s->s_count++;//alloc_super时将设置s->s_count = S_BIAS;
    spin_unlock(&sb_lock);
    down_write(&s->s_umount);
    if (s->s_root) {
        spin_lock(&sb_lock);
        if (s->s_count > S_BIAS) {
            atomic_inc(&s->s_active);
            s->s_count--;
            spin_unlock(&sb_lock);
            return 1;//确实有一个有效的sb超级块对应的s_root目录项,那么返回1,然后由destroy_super将其销毁
        }
        spin_unlock(&sb_lock);
    }
    up_write(&s->s_umount);
    put_super(s);
    yield();
    return 0;
}
//获取一个sb超级块的控制内存,同时做部分结构初始化
static struct super_block *alloc_super(struct file_system_type *type)
{
    struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
    static struct super_operations default_op;

    if (s) {
        if (security_sb_alloc(s)) {
            kfree(s);
            s = NULL;
            goto out;
        }
        INIT_LIST_HEAD(&s->s_dirty);
        INIT_LIST_HEAD(&s->s_io);
        INIT_LIST_HEAD(&s->s_more_io);
        INIT_LIST_HEAD(&s->s_files);
        INIT_LIST_HEAD(&s->s_instances);
        INIT_HLIST_HEAD(&s->s_anon);
        INIT_LIST_HEAD(&s->s_inodes);
        init_rwsem(&s->s_umount);
        mutex_init(&s->s_lock);
        lockdep_set_class(&s->s_umount, &type->s_umount_key);
        /*
         * The locking rules for s_lock are up to the
         * filesystem. For example ext3fs has different
         * lock ordering than usbfs:
         */
        lockdep_set_class(&s->s_lock, &type->s_lock_key);
        down_write(&s->s_umount);
        s->s_count = S_BIAS;
        atomic_set(&s->s_active, 1);
        mutex_init(&s->s_vfs_rename_mutex);
        mutex_init(&s->s_dquot.dqio_mutex);
        mutex_init(&s->s_dquot.dqonoff_mutex);
        init_rwsem(&s->s_dquot.dqptr_sem);
        init_waitqueue_head(&s->s_wait_unfrozen);
        s->s_maxbytes = MAX_NON_LFS;
        s->dq_op = sb_dquot_ops;
        s->s_qcop = sb_quotactl_ops;
        s->s_op = &default_op;
        s->s_time_gran = 1000000000;
    }
out:
    return s;
}
int set_anon_super(struct super_block *s, void *data)
{
    int dev;
    int error;

 retry:
    if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
        return -ENOMEM;
    spin_lock(&unnamed_dev_lock);
    error = idr_get_new(&unnamed_dev_idr, NULL, &dev);//从radix树中,递增式的获取一个唯一整数值到&dev
    spin_unlock(&unnamed_dev_lock);
    if (error == -EAGAIN)
        /* We raced and lost with another CPU. */
        goto retry;
    else if (error)
        return -EAGAIN;

    if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
        spin_lock(&unnamed_dev_lock);
        idr_remove(&unnamed_dev_idr, dev);
        spin_unlock(&unnamed_dev_lock);
        return -EMFILE;
    }
    s->s_dev = MKDEV(0, dev & MINORMASK);//生成major为0的超级块设备号
    return 0;
}
//给申请的sb超级块填充细化数据
static int sysfs_fill_super(struct super_block *sb, void *data, int silent)
{
    struct inode *inode;
    struct dentry *root;

    sb->s_blocksize = PAGE_CACHE_SIZE;//块大小4k
    sb->s_blocksize_bits = PAGE_CACHE_SHIFT;//12
    sb->s_magic = SYSFS_MAGIC;//magic代号
    sb->s_op = &sysfs_ops;//超级块操作函数集
    sb->s_time_gran = 1;
    sysfs_sb = sb;//保存

    /* get root inode, initialize and unlock it */
    inode = sysfs_get_inode(&sysfs_root);//生成sysfs_root中所定义要求的inode内存节点
    if (!inode) {
        pr_debug("sysfs: could not get root inode\n");
        return -ENOMEM;
    }

    /* instantiate and link root dentry */
    root = d_alloc_root(inode);//将inode安装到'/'根目录项上
    if (!root) {
        pr_debug("%s: could not get root dentry!\n",__FUNCTION__);
        iput(inode);
        return -ENOMEM;
    }
    root->d_fsdata = &sysfs_root;//根目录项的文件系统数据fsdata指向sysfs_root
    sb->s_root = root;//填充sb超级块对应的根目录项
    return 0;
}
struct sysfs_dirent sysfs_root = {
    .s_name = "",//0空
    .s_count = ATOMIC_INIT(1),
    .s_flags = SYSFS_DIR,//目录
    .s_mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO,
    .s_ino = 1,//节点号搞为1
};
struct inode * sysfs_get_inode(struct sysfs_dirent *sd)
{
    struct inode *inode;

    inode = iget_locked(sysfs_sb, sd->s_ino);
    if (inode && (inode->i_state & I_NEW))
        sysfs_init_inode(sd, inode);//这是一个新创建的inode节点,那么调用sysfs对它进一步做符合sysfs要求的初始化[luther.gliethttp]

    return inode;
}
//对新创建的inode节点,进行sysfs特性格式化[luther.gliethttp]
static void sysfs_init_inode(struct sysfs_dirent *sd, struct inode *inode)
{
    struct bin_attribute *bin_attr;

    inode->i_blocks = 0;
    inode->i_mapping->a_ops = &sysfs_aops;
    inode->i_mapping->backing_dev_info = &sysfs_backing_dev_info;
    inode->i_op = &sysfs_inode_operations;//更改inode的方法集
/*
static const struct inode_operations sysfs_inode_operations ={
    .setattr = sysfs_setattr,
};
*/
    inode->i_ino = sd->s_ino;//inode的节点号要和sd->s_ino目录项节点号一致
//★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
自此
sysfs_fill_super=>sysfs_sb = sb;//保存

sysfs_create_dir
=>parent_sd = &sysfs_root;
sysfs_addrm_start
inode = ilookup5_nowait(sysfs_sb, parent_sd->s_ino, sysfs_ilookup_test, parent_sd);
=>就能够找到sysfs_root对应的inode节点了,所以
start_kernel
=>vfs_caches_init
=>mnt_init
=>sysfs_init执行完之后,driver们也就能够在稍后的时候顺利使用sysfs文件系统进行文件创建了.

start_kernel
=>rest_init
=>kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
kernel_init
=>do_basic_setup
=>driver_init会创建如下：
/sys/devices
/sys/bus
/sys/class
/sys/firmware
等sysfs目录和文件
然后do_basic_setup继续调用
=>do_initcalls
调用所有build in到kernel中的module_init驱动程序,
所以这时的驱动程序就可以自由使用sysfs以及上面创建的那些sysfs文件系统下的目录和文件了[luther.gliethttp]
void __init driver_init(void)
{
    /* These are the core pieces */
    devices_init();
    buses_init();
    classes_init();
    firmware_init();
    hypervisor_init();

    /* These are also core pieces, but must come after the
     * core core pieces.
     */
    platform_bus_init();
    system_bus_init();
    cpu_dev_init();
    memory_dev_init();
}
//★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
    lockdep_set_class(&inode->i_mutex, &sysfs_inode_imutex_key);

    if (sd->s_iattr) {
        /* sysfs_dirent has non-default attributes
         * get them for the new inode from persistent copy
         * in sysfs_dirent
         */
        set_inode_attr(inode, sd->s_iattr);
    } else
        set_default_inode_attr(inode, sd->s_mode);//设备inode默认attr属性们

    /* initialize inode according to type */
    switch (sysfs_type(sd)) {//根据sd目录项信息,来反向订制该inode节点的fops等操作函数集[luther.gliethttp]
    case SYSFS_DIR:
        inode->i_op = &sysfs_dir_inode_operations;
        inode->i_fop = &sysfs_dir_operations;//dir目录的fops操作函数集
        inode->i_nlink = sysfs_count_nlink(sd);
        break;
    case SYSFS_KOBJ_ATTR:
        inode->i_size = PAGE_SIZE;
        inode->i_fop = &sysfs_file_operations;//attr属性文件的fops操作函数集
        break;
    case SYSFS_KOBJ_BIN_ATTR:
        bin_attr = sd->s_bin_attr.bin_attr;//bin_attr属性文件的fops操作函数集
        inode->i_size = bin_attr->size;
        inode->i_fop = &bin_fops;
        break;
    case SYSFS_KOBJ_LINK:
        inode->i_op = &sysfs_symlink_inode_operations;//link文件的fops操作函数集
        break;
    default:
        BUG();
    }

    unlock_new_inode(inode);
}
static inline void set_default_inode_attr(struct inode * inode, mode_t mode)
{
    inode->i_mode = mode;
    inode->i_uid = 0;
    inode->i_gid = 0;
    inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
}
struct inode *iget_locked(struct super_block *sb, unsigned long ino)
{
    struct hlist_head *head = inode_hashtable + hash(sb, ino);
    struct inode *inode;

    inode = ifind_fast(sb, head, ino);
    if (inode)//第一次的话,肯定inode=NULL;
        return inode;
    /*
     * get_new_inode_fast() will do the right thing, re-trying the search
     * in case it had to block at any point.
     */
    return get_new_inode_fast(sb, head, ino);//为root获取一个inode存储结构体,对于sysfs就是获取一个内存空间
    //对于ext3和yaffs2文件系统就是申请flash或者硬盘上的物理空间了[luther.gliethttp].
}
static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
{
    struct inode * inode;

    inode = alloc_inode(sb);//从sb超级块上获取一个新的inode
    if (inode) {
        struct inode * old;

        spin_lock(&inode_lock);//锁住
        /* We released the lock, so.. */
        old = find_inode_fast(sb, head, ino);//可能他人已经创建完成了,所以再查一遍
        if (!old) {
//确实ino节点号仍然没有被创建,那么我们可以安全的将inode作为ino节点号对应的inode了[luther.gliethttp]
            inode->i_ino = ino;//节点对应的节点号
            inodes_stat.nr_inodes++;
            list_add(&inode->i_list, &inode_in_use);//挂到全局量inode_in_use链表上
            list_add(&inode->i_sb_list, &sb->s_inodes);//挂到超级块的s_inodes链表上
            hlist_add_head(&inode->i_hash, head);//将inode添加到所有节点都挂接到的hash数组inode_hashtable中.
            inode->i_state = I_LOCK|I_NEW;//标记该inode为新建节点
            spin_unlock(&inode_lock);//解锁

            /* Return the locked inode with I_NEW set, the
             * caller is responsible for filling in the contents
             */
            return inode;//返回这个新生成的inode节点
        }

        /*
         * Uhhuh, somebody else created the same inode under
         * us. Use the old inode instead of the one we just
         * allocated.
         */
//很不走运,该ino对应的inode节点别人已经在你上边某一步时,提前搞定了,所以我们需要把刚才申请的咚咚释放掉[luther.gliethttp]
        __iget(old);
        spin_unlock(&inode_lock);
        destroy_inode(inode);//销毁前面申请的inode节点,因为别人已经成功添加了.
        inode = old;
        wait_on_inode(inode);
    }
    return inode;//inode=NULL
}
static const struct super_operations sysfs_ops = {
    .statfs = simple_statfs,
    .drop_inode = generic_delete_inode,
};
sb->s_op = &sysfs_ops;//超级块操作函数集
static struct inode *alloc_inode(struct super_block *sb)
{
    static const struct address_space_operations empty_aops;
    static struct inode_operations empty_iops;
    static const struct file_operations empty_fops;
    struct inode *inode;

    if (sb->s_op->alloc_inode)//该sb超级块提供自定义,特殊的alloc_inode函数,对于sysfs没有提供专门的函数[luther.gliethttp]
        inode = sb->s_op->alloc_inode(sb);
    else
        inode = (struct inode *) kmem_cache_alloc(inode_cachep, GFP_KERNEL);//所以从cache中获取一个空闲染色的slab对象

    if (inode) {
//细化inode的部分结单元
        struct address_space * const mapping = &inode->i_data;

        inode->i_sb = sb;
        inode->i_blkbits = sb->s_blocksize_bits;
        inode->i_flags = 0;
        atomic_set(&inode->i_count, 1);
        inode->i_op = &empty_iops;
        inode->i_fop = &empty_fops;
        inode->i_nlink = 1;
        atomic_set(&inode->i_writecount, 0);
        inode->i_size = 0;
        inode->i_blocks = 0;
        inode->i_bytes = 0;
        inode->i_generation = 0;
#ifdef CONFIG_QUOTA
        memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
#endif
        inode->i_pipe = NULL;
        inode->i_bdev = NULL;
        inode->i_cdev = NULL;
        inode->i_rdev = 0;
        inode->dirtied_when = 0;
        if (security_inode_alloc(inode)) {
            if (inode->i_sb->s_op->destroy_inode)
                inode->i_sb->s_op->destroy_inode(inode);
            else
                kmem_cache_free(inode_cachep, (inode));
            return NULL;
        }

        spin_lock_init(&inode->i_lock);
        lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);

        mutex_init(&inode->i_mutex);
        lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);

        init_rwsem(&inode->i_alloc_sem);
        lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);

        mapping->a_ops = &empty_aops;
         mapping->host = inode;
        mapping->flags = 0;
        mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE);
        mapping->assoc_mapping = NULL;
        mapping->backing_dev_info = &default_backing_dev_info;

        /*
         * If the block_device provides a backing_dev_info for client
         * inodes then use that. Otherwise the inode share the bdev's
         * backing_dev_info.
         */
        if (sb->s_bdev) {
            struct backing_dev_info *bdi;

            bdi = sb->s_bdev->bd_inode_backing_dev_info;
            if (!bdi)
                bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
            mapping->backing_dev_info = bdi;
        }
        inode->i_private = NULL;
        inode->i_mapping = mapping;
    }
    return inode;
}

struct dentry * d_alloc_root(struct inode * root_inode)
{
    struct dentry *res = NULL;

    if (root_inode) {
        static const struct qstr name = { .name = "/", .len = 1 };

        res = d_alloc(NULL, &name);//申请一个根目录项,调用dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
        if (res) {
            res->d_sb = root_inode->i_sb;//根目录项的超级块
            res->d_parent = res;//自己指向自己
            d_instantiate(res, root_inode);//将inode安装到根目录项上
        }
    }
    return res;
}
void d_instantiate(struct dentry *entry, struct inode * inode)
{
    BUG_ON(!list_empty(&entry->d_alias));
    spin_lock(&dcache_lock);
    if (inode)
        list_add(&entry->d_alias, &inode->i_dentry);//将目录向链接到inode上,所以一个inode可以有多个目录项指向,
    //这些目录项可以通过扫描inode->i_dentry链表获得
    entry->d_inode = inode;//目录项对应的所管理的inode节点
    fsnotify_d_instantiate(entry, inode);
    spin_unlock(&dcache_lock);
    security_d_instantiate(entry, inode);
}