引言
在前篇中介绍了Linux是如何挂载Overlay文件系统的，重点关注了Overlay的挂载流程的实现和关键数据结构的关系，而在本文中主要介绍Overlay FS的目录相关的关键数据结构和目录的相关接口，其中比较复杂的接口将会在下一篇中展开介绍。
一、Overlay 关键数据结构
1、目录文件
struct ovl_dir_file {
bool is_real;
bool is_upper;
struct ovl_dir_cache *cache;
struct list_head *cursor;
struct file *realfile;
struct file *upperfile;
};2、目录接口
// fs/overlayfs/readdir.c
const struct file_operations ovl_dir_operations = {
.read = generic_read_dir,
.open = ovl_dir_open,
.iterate = ovl_iterate,
.llseek = ovl_dir_llseek,
.fsync = ovl_dir_fsync,
.release = ovl_dir_release,
};3、目录inode操作
// fs/overlayfs/dir.c
const struct inode_operations ovl_dir_inode_operations = {
.lookup = ovl_lookup, // fs/overlayfs/namei.c
.mkdir = ovl_mkdir, // fs/overlayfs/dir.c
.symlink = ovl_symlink,
.unlink = ovl_unlink,
.rmdir = ovl_rmdir,
.rename = ovl_rename,
.link = ovl_link,
.setattr = ovl_setattr,
.create = ovl_create,
.mknod = ovl_mknod,
.permission = ovl_permission,
.getattr = ovl_getattr,
.listxattr = ovl_listxattr,
.get_acl = ovl_get_acl,
.update_time = ovl_update_time,
};二、目录操作
1、创建目录
（1）创建目录系统调用
// ./fs/namei.c
SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
|-> do_mkdirat(AT_FDCWD, pathname, mode);
|-> struct dentry *dentry;
|-> struct path path;
| // (1)创建目录项：创建ovl_dentry，同时将其加入到其父目录的子目录项链表中
|-> dentry = user_path_create(dfd, pathname, &path, lookup_flags); // path保存父目录路径；dentry保存新建目录项内存对象
| |-> filename_create(dfd, getname(pathname), path, lookup_flags);
| |-> name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
| |-> dentry = __lookup_hash(&last, path->dentry, lookup_flags)
| |-> dentry = d_alloc(base, name);
|
| // (2)创建ovl inode：调用对应文件系统目录ops中的mkdir创建inode，并关联inode到dentry
|-> vfs_mkdir(path.dentry->d_inode, dentry, mode);
|-> dir->i_op->mkdir(dir, dentry, mode) // 参数：父目录inode、新目录的目录项、模式（2）Overlay FS中ovl_mkdir创建目录的过程
在Overlay fs联合文件系统下，创建目录有2种情况：
（1）upper层和lower层均无同名目录，则直接在upper层创建。
（2）lower层有同名目录，但是之后被删除了，则需要在已经存在whiteout文件的情况下创建目录。
ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) // fs/overlayfs/dir.c
|-> ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL)
| // （1）创建ovl_inode节点
|-> inode = ovl_new_inode(dentry->d_sb, mode, rdev);
|-> inode->i_state |= I_CREATING;
|
| // （2）联合文件系统调用（实际文件系统）创建节点，并连接（link）虚拟文件系统inode和实际文件系统inode
| // 实际文件系统创建新目录，并将实际文件系统的inode赋值给虚拟文件系统的目录项（dentry），同时虚拟文件系统的目录项作为虚拟文件系统目录项的一个别名
|-> ovl_create_or_link(dentry, inode, &attr, false) // 参数：虚拟文件系统目录项、虚拟文件系统inode节点、属性、非原始节点
| // 父目录进行copyup：将parent的lower目录内容copyup到upper层。若存在upper层，则直接返回。
|-> ovl_copy_up(parent);// copyup将会在后续的文章中详细介绍
|-> if (!ovl_dentry_is_whiteout(dentry))
| // 不存在whiteout文件，则直接在upper层创建新目录
|-> ovl_create_upper(dentry, inode, attr);
|-> else
| // 存在whiteout文件，则在upper层创建新目录并..???..
|-> ovl_create_over_whiteout(dentry, inode, attr);（3）直接在upper层创建新目录
参数inode为ovl_inode，而dentry则是与之对应的overlayfs文件系统目录节点的目录项。
ovl_create_upper(struct dentry *dentry, struct inode *inode, struct ovl_cattr *attr)
| // 将overlayfs目录转换成源文件路径
|-> upperdir = ovl_dentry_upper(dentry->d_parent); // 原始实际文件系统中的父目录
|-> struct inode *udir = upperdir->d_inode;
|-> struct dentry *newdentry;
| // （1）创建父目录的upper层创建新目录
|-> newdentry = ovl_create_real(udir, lookup_one_len(dentry->d_name.name, upperdir, dentry->d_name.len), attr);
| |-> ovl_mkdir_real(dir, &newdentry, attr->mode);
| |-> ovl_do_mkdir(dir, dentry, mode)
| |-> vfs_mkdir()
| |-> dir->i_op->mkdir(dir, dentry, mode);
|
| // 实例化overlayfs目录项和ovl_inode:
| // (1)ovl_inode的i_private和__upperdentry分别指向upper的inode和目录项;(2)ovl目录项的d_inode指向ovl_inode
|-> ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink); // ovl: dentry; ovl: inode; realfs: newdentry, ..
|-> struct ovl_inode_params oip = {
| .upperdentry = newdentry, // 实际文件系统目录项
| .newinode = inode, // ovl_inode
| };
|
| // 返回ovl_inode的指针：ovl_inode的i_private和__upperdentry分别指向upper的inode和目录项
|-> inode = ovl_get_inode(dentry->d_sb, &oip);
| | // 返回ovl_inode的指针,并将realinode作为ovl_inode私有数据
| |-> inode = ovl_iget5(sb, oip->newinode, key);
| | // 设置ovlinode的__upperdentry指向upper的目录项upperdentry
| |-> ovl_inode_init(inode, oip, ino, fsid);
| |-> OVL_I(inode)->__upperdentry = oip->upperdentry;
|
| // 设置ovl目录项的d_inode指向ovl_inode
|-> d_instantiate(dentry, inode); // overlayfs文件系统目录项和inode
|-> __d_instantiate(entry, inode);
|-> __d_set_inode_and_type(dentry, inode, add_flags);
|-> dentry->d_inode = inode;ovl_create_upper()执行完后，新创建的overlayfs目录结构，如下图所示：
overlayfs: struct ovl_inode
struct dentry +---------------------------------+
+--------------------------------+ |redirect: char*; |
|d_parent: struct dentry *; | |mnt_mountpoint: struct dentry*; |
|d_name: struct qstr; | |vfs_inode: struct inode; |
|d_inode: struct inode*; -------+--->|+-struct inode -----------------+|
|d_op: struct dentry_operations*;| ||i_op: struct inode_operations*;||
|d_sb: struct super_block; | ||i_sb: struct super_block *; ||
|d_fsdata: void *; | || ... ||
| ... | ||i_private: void * ------------++---+
+--------------------------------+ |+-------------------------------+| |
|__upperdentry: struct dentry*; --+-+ |
|lower: struct inode*; | | |
|... | | |
+---------------------------------+ | |
| |
upper: +-------------------------------------------------------------------------+ |
| +-------------------------------------+
| |
V V
struct dentry struct inode
+--------------------------------+ +->+-------------------------------+
|d_parent: struct dentry *; | | |i_op: struct inode_operations*;|
|d_name: struct qstr; | | |i_sb: struct super_block *; |
|d_inode: struct inode*; --------+-+ |i_ino: unsigned long; |
|d_op: struct dentry_operations*;| | ... |
|d_sb: struct super_block; | +-------------------------------+
| ... |
+--------------------------------+（4）创建新目录，但lower层存在被删除的同名目录
参数ovl_create_upper()一样：inode为ovl_inode，而dentry则是与之对应的overlayfs文件系统目录节点的目录项。
ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode, struct ovl_cattr *attr)
|-> struct dentry *workdir = ovl_workdir(dentry);
|-> struct inode *wdir = workdir->d_inode;
|-> struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent); // upper父目录项
|-> struct inode *udir = upperdir->d_inode; // upper父目录inode
|-> struct dentry *upper; // ovl目录项
|-> struct dentry *newdentry; // upper目录项
|
| // (1)在upper层父目录中查找目录项（whiteout）
|-> upper = lookup_one_len(dentry->d_name.name, upperdir, dentry->d_name.len);
| // (2)在workdir目录下创建临时目录
|-> newdentry = ovl_create_temp(workdir, cattr);
| |-> ovl_create_real(d_inode(workdir), ovl_lookup_temp(workdir), attr);
| // (3)设置worddir下新建临时目录属性为opaque
|-> ovl_set_opaque(dentry, newdentry);
| // (4)workdir新建的临时目录与upper的whiteout目录做交互rename
|-> ovl_do_rename(wdir, newdentry, udir, upper, RENAME_EXCHANGE);
| // (5)清除rename后workdir下的废弃文件（whiteout）
|-> ovl_cleanup(wdir, upper);
|
| // 实例化overlayfs目录项和ovl_inode：
| // (1)ovl_inode的i_private和__upperdentry分别指向upper的inode和目录项;(2)ovl目录项的d_inode指向ovl_inode.
|-> ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);由上面流程可见，与ovl_create_upper()相比，ovl_create_over_whiteout()先在work目录下创建一个临时目录，并设置目录属性为opaque，以屏蔽lower层同名目录。然后将workdir目录下临时目录与upper层的whiteout文件做重命名互换，并清除互换后的垃圾文件。
2、删除目录ovl_rmdir()
将要被删除的联合文件系统目录主要有3种途径：
（1）目录只来自upper层，则直接删除即可；
（2）目录只来自lower层，则需要建立对lower层同名目录的whiteout文件；
（3）目录是由uppwer层和lower层目录合成，则既要删除upper层目录，又要建立lower层同名目录的whiteout文件；
但是上面的情况，经过简化合并为两种情况：
- 不存在于lower层，则直接删除即可
- 存在于lower层，需要生成whiteout文件屏蔽底层目录。
下面代码为Overlay fs联合文件系统删除目录的过程，留意下列过程中根据lower_positive是否为true，即是否存在lower层，分别调用了函数ovl_remove_upper()或者ovl_remove_and_whiteout()。
ovl_rmdir(struct inode *dir, struct dentry *dentry)
|-> ovl_do_remove(dentry, true);
|-> ovl_check_empty_dir(dentry, &list);
| | // 获取目录下的所有目录和文件列挂到list链表
| |-> ovl_dir_read_merged(dentry, list, &root);
| | |-> for (idx = 0; idx != -1; idx = next) {
| | |-> next = ovl_path_next(idx, dentry, &realpath);
| | |-> ovl_dir_read(&realpath, &rdd);
| | | |-> realfile = ovl_path_open(realpath, O_RDONLY | O_LARGEFILE);
| | | |-> iterate_dir(realfile, &rdd->ctx);
| | |-> }
| |->
|
| // copyup父目录：若存在upper层，则ovl_copy_up()直接返回。
|-> ovl_copy_up(dentry->d_parent);
|-> if (!lower_positive)
|-> err = ovl_remove_upper(dentry, is_dir, &list);
|-> else
|-> err = ovl_remove_and_whiteout(dentry, &list);
|
|-> clear_nlink(dentry->d_inode);
|
|-> upperdentry = ovl_dentry_upper(dentry);
|-> if (upperdentry)
|-> ovl_copyattr(d_inode(upperdentry), d_inode(dentry));（1）删除不存在lower层的目录
ovl_remove_upper(dentry, is_dir, &list)
|-> struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent)
|-> struct inode *dir = upperdir->d_inode
|-> struct dentry *opaquedir = NULL
|-> opaquedir = ovl_clear_empty(dentry, list) // 返回opaque目录
|-> upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,dentry->d_name.len)// 找到需要删除的dengtry
|-> err = ovl_do_rmdir(ofs, dir, upper)
| |-> vfs_rmdir(ovl_upper_mnt_userns(ofs), dir, dentry) // 直接调用vfs_mkdir进行删除
| |...（2）删除存在lower层的目录
ovl_remove_and_whiteout(dentry, &list);
|-> struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
|-> workdir = ovl_workdir(dentry);
| // 返回opaque目录：已经完成了目标目录的删除
|-> opaquedir = ovl_clear_empty(dentry, list);
| |-> opaquedir = ovl_create_temp(workdir, OVL_CATTR(stat.mode));
| |-> ovl_copy_xattr(dentry->d_sb, upper, opaquedir);
| |-> ovl_set_opaque(dentry, opaquedir);
| | // rename后opaquedir与upper的inode内容互换
| |-> ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
| | // 递归删除目标目录下的upper层内容
| |-> ovl_cleanup_whiteouts(upper, list);
| | // 删除目标目录
| |-> ovl_cleanup(wdir, upper);
| |-> ovl_do_rmdir(wdir, wdentry);
| |-> vfs_rmdir(dir, dentry);
| // 此时，upper即opaquedir
|-> upper = lookup_one_len(dentry->d_name.name, upperdir, dentry->d_name.len);
|-> ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
|-> whiteout = ovl_whiteout(ofs);
| |-> whiteout = ovl_lookup_temp(workdir);
| | |-> snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
| | |-> temp = lookup_one_len(name, workdir, strlen(name));
| |-> ovl_do_whiteout(wdir, whiteout);
|-> ovl_do_rename(wdir, whiteout, dir, dentry, flags);
|-> ...如果 要删除的文件是upper层覆盖lower层的文件，要删除的目录是上下层合并的目录，其实是前两个场景的合并，Overlay fs即需要删除upper层对应文件系统中的文件或目录，也需要在对应位置创建同名whiteout文件，让upper层的文件被删除后不至于lower层的文件被暴露出来。
3、ovl_dir_open打开目录
假设目录是merged目录：
ovl_dir_open(struct inode *inode, struct file *file) // fs/overlayfs/readdir.c
|-> struct ovl_dir_file *od;
|-> od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
| // 获取目录路径：若目录是merged或者upper目录，则获得upper路径；否则，获得lower路径。
|-> type = ovl_path_real(file->f_path.dentry, &realpath);
| |-> type = ovl_path_type(dentry);
| | |-> type = __OVL_PATH_UPPER;
| | |-> type |= __OVL_PATH_MERGE;
| |
| |-> ovl_path_lower(dentry, path);
| | // < or >
| |-> ovl_path_upper(dentry, path);
| |-> path->mnt = ovl_upper_mnt(ofs);
| |-> path->dentry = ovl_dentry_upper(dentry);
|
| // 打开upper或者lower目录
|-> realfile = ovl_dir_open_realfile(file, &realpath);
| |-> struct file *res;
| |-> res = ovl_path_open(realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
| |-> dentry_open(path, flags, current_cred());
| |-> struct file *f;
| |-> f = alloc_empty_file(flags, cred);
| |-> vfs_open(path, f);
| |-> open = f->f_op->open(inode, f);
|
| // realfile作为ovldir_file的实际文件；ovl_dir_file又作为file的私有数据
|-> od->realfile = realfile;
|-> od->is_real = ovl_dir_is_real(file->f_path.dentry);
|-> od->is_upper = OVL_TYPE_UPPER(type);
|-> file->private_data = od;目录打开后形成如下结构：
struct file
+------------------------------+
|f_path: struct path; |
|f_inode: struct inode; |
|f_op: struct file_operations*;|
|private_data: void *; --------+-+
| ... | |
+------------------------------+ |
+-------+
Overlay fs: | struct ovl_dir_file
+->+-----------------------------+
|is_real: bool |
|is_upper: unsigned; |
|cache: struct ovl_dir_cache*;|
|cursor: struct list_head*; |
|realfile: struct file*; -----+-+
|upperfile: struct file*; | |
+-----------------------------+ |
+--------+
|
| struct file
upper: +->+------------------------------+
|f_path: struct path; |
|f_inode: struct inode; |
|f_op: struct file_operations*;|
|private_data: void *; |
| ... |
+------------------------------+（1）OVL目录内查找目标目录或者文件
每当在这样一个合并的目录中请求查找时，就会在每个实际的目录中进行查找，而合并的结果则缓存在属于覆盖文件系统的目录中。结果被缓存在属于叠加文件系统的目录中。如果两个实际的查询都能找到目录，那么这两个查询都会被存储起来，并且创建一个合并的目录。创建一个合并的目录，否则只存储一个：如果上层目录存在，则存储下层目录。只有来自目录的名称列表被合并。其他内容，如元数据和扩展属性，只报告给上层目录，下层目录的这些属性被隐藏。
ovl_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) // fs/overlayfs/dir.c
| // (1)在upper层父目录中查找目标文件或者目录：
| // (1.1)从ovl目录项的ovl_inode反向索引到实际文件系统的dentry
|-> upperdir = ovl_dentry_upper(dentry->d_parent);
| // (1.2)在upper层目录下搜索目标文件或者目录，若存在则使用局部目录项指针upperdentry指向目标目录项
|-> ovl_lookup_layer(upperdir, &d, &upperdentry, true); // 参数：实际文件系统中的父目录项、...、目标实际文件系统中目录项返回值、...
| |-> ovl_lookup_single(base, d, d->name.name, d->name.len,0, "", ret, drop_negative) // ret，即&upperdentry
| |-> this = ovl_lookup_positive_unlocked(name, base, namelen, drop_negative);
| | |-> lookup_one_len_unlocked(name, base, len);
| | |-> lookup_slow(&this, base, 0);
| | |-> __lookup_slow(name, dir, flags);
| | |-> dentry = d_alloc_parallel(dir, name, &wq);
| | |-> old = inode->i_op->lookup(inode, dentry, flags);
| | |-> dentry = old;
| |-> *ret = this;
|
| // (2)在所有lower层父目录中查找目标文件或者目录
|-> for (i = 0; !d.stop && i < poe->numlower; i++) {
|-> struct ovl_path lower = poe->lowerstack[i];
|-> d.last = i == poe->numlower - 1;
|-> ovl_lookup_layer(lower.dentry, &d, &this, false);
|-> ...
|-> stack[ctr].dentry = this;
|-> stack[ctr].layer = lower.layer;
|-> ctr++;
|-> ｝
|
| // (3)分配ovl_entry，其中包括根据堆叠层次确定的ovl_path层次数
|-> oe = ovl_alloc_entry(ctr);
|-> memcpy(oe->lowerstack, stack, sizeof(struct ovl_path) * ctr);
|-> dentry->d_fsdata = oe;
|
| // 关联实际文件系统denty和inode到虚拟文件系统inode
|-> inode = ovl_get_inode(dentry->d_sb, &oip);根据上面的流程，在overlayfs文件系统的一个目录下查询一个目录或者文件的流程为，通过ovl系统中父目录dentry找到实际文件系统的目录项和inode，找到目标目录或者文件对应的dentry和inode。
struct ovl_entry
+==>+-------------------------------+
H |lowerstack: struct ovl_path[]; |
H |numlower: unsigned; |
H | ... |
H +-------------------------------+
H
H struct ovl_inode
struct dentry(d_parent) H +---------------------------------+
+--------------------------------+H |redirect: char*; |
|d_parent: struct dentry *; |H |mnt_mountpoint: struct dentry*; |
|d_name: struct qstr; |H |vfs_inode: struct inode; |
|d_inode: struct inode*;==========H==>|+-struct inode -----------------+|
|d_op: struct dentry_operations*;|H ||i_op: struct inode_operations*;||
|d_sb: struct super_block; |H ||i_sb: struct super_block *; ||
|d_fsdata: void *; ==========(2)=H ||i_ino: unsigned long; ||
| ... | || ... ||
+--------------------------------+ |+-------------------------------+| struct dentry(base dir in realfs)
|__upperdentry: struct dentry*; ==(1)===>+--------------------------------+
|lower: struct inode*; | |d_parent: struct dentry *; |
|... | |d_name: struct qstr; |
+---------------------------------+ |d_inode: struct inode*; ============H
|d_op: struct dentry_operations*;| H
|d_sb: struct super_block; | H
| ... | H
+--------------------------------+ H
struct inode(lowerendtry->inode)V
+-------------------------------+
|i_op: struct inode_operations*;| ==> 实际文件系统的lookup()
|i_sb: struct super_block *; |
|i_ino: unsigned long; |
| ... |
+-------------------------------+查找到的目录或者文件，在ovl文件系统中表示如下：
struct ovl_entry
+->+------------------------------+
| | ... |
| |numlower: unsigned; |
| |lowerstack: struct ovl_path[];|
| +==============================+
| |struct ovl_path { |
| | layer: struct ovl_layer*; |
| | dentry: struct dentry*; |
| |}; |
| +------------------------------+
| |struct ovl_path { |
| | layer: struct ovl_layer*; |
| | dentry: struct dentry*; |
| |}; |
| +------------------------------+
| | |
| | |
|
| struct ovl_inode
struct dentry | +---------------------------------+
+--------------------------------+| |redirect: char*; |
|d_parent: struct dentry *; || |mnt_mountpoint: struct dentry*; |
|d_name: struct qstr; || |vfs_inode: struct inode; |
|d_inode: struct inode*;----------+->|+-struct inode -----------------+|
|d_op: struct dentry_operations*;|| ||i_op: struct inode_operations*;||
|d_sb: struct super_block; || ||i_sb: struct super_block *; ||
|d_fsdata: void *; -(3)----------+ ||i_ino: unsigned long; ||
| ... | || ... ||
+--------------------------------+ |+-------------------------------+|
|__upperdentry: struct dentry*; |
|lower: struct inode*; |
|... |
+---------------------------------+（2）打开目录
被打开的目录可能只来自于upper层或lower层，由或者是upper和lower目录merged而成，打开过程如下：
ovl_dir_open(struct inode *inode, struct file *file) // fs/overlayfs/readdir.c
|-> struct ovl_dir_file *od;
|-> od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
| // （1）获取目录路径：若目录是merged或者upper目录，则获得upper路径；否则，获得lower路径。
|-> type = ovl_path_real(file->f_path.dentry, &realpath);
| |-> type = ovl_path_type(dentry);
| | |-> type = __OVL_PATH_UPPER;
| | |-> type |= __OVL_PATH_MERGE;
| |
| |-> ovl_path_lower(dentry, path);
| | // < or >
| |-> ovl_path_upper(dentry, path);
| |-> path->mnt = ovl_upper_mnt(ofs);
| |-> path->dentry = ovl_dentry_upper(dentry);
|
| // （2）打开实际文件系统目录文件：upper或者lower目录
|-> realfile = ovl_dir_open_realfile(file, &realpath);
| |-> struct file *res;
| |-> res = ovl_path_open(realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
| |-> dentry_open(path, flags, current_cred());
| |-> struct file *f;
| |-> f = alloc_empty_file(flags, cred);
| |-> vfs_open(path, f);
| |-> open = f->f_op->open(inode, f);
|
| // （3）将overlayfs目录文件和实际文件系统目录文件关联：realfile作为ovldir_file的实际文件；ovl_dir_file又作为file的私有数据
|-> od->realfile = realfile;
|-> od->is_real = ovl_dir_is_real(file->f_path.dentry);
|-> od->is_upper = OVL_TYPE_UPPER(type);
|-> file->private_data = od;由上过程可知，Overlay fs目录文件被打开后，同时存在Overlay fs目录文件和实际文件系统目录文件，中间通过ovl_dir_file链接（ovl_dir_file作为Overlay fs目录文件的私有数据），他们直接的关系示意如下：
struct ovl_dir_file
+->+-----------------------------+
| |is_real: bool |
| |is_upper: unsigned; |
| |cache: struct ovl_dir_cache*;|
| |cursor: struct list_head*; |
| |realfile: struct file*; -----+-+
| |upperfile: struct file*; | |
| +-----------------------------+ |
+-------+ +-------+
struct file | | struct file
+------------------------------+ | +->+------------------------------+
|f_path: struct path; | | |f_path: struct path; |
|f_inode: struct inode; | | |f_inode: struct inode; |
|f_op: struct file_operations*;| | |f_op: struct file_operations*;|
|private_data: void *; --------+-+ |private_data: void *; |
| ... | | ... |
+------------------------------+ +------------------------------+
overlayfs目录文件 实际文件系统目录文件（3）遍历目录
遍历目录，即是所有各叠层相同目录下的所有文件列表。
（4）系统调用-遍历目录
YSCALL_DEFINE3(getdents, unsigned int, fd,,,,,) // fs/readdir.c
|-> struct fd f;
|-> struct getdents_callback buf = {
|-> .ctx.actor = filldir,
|-> .count = count,
|-> .current_dir = dirent
|-> };
|
|-> iterate_dir(f.file, &buf.ctx);
|-> file->f_op->iterate(file, ctx);（5）OVL函数iterate实现
假设目录是merged的目录：
struct ovl_dir_file
+=>+-----------------------------+
H |is_real: bool |
H |is_upper: unsigned; |
H |cache: struct ovl_dir_cache*;|
H |cursor: struct list_head*; |
H |realfile: struct file*; -----+-+
H +-----------------------------+ |
H |
struct file H struct file V
+------------------------------+ H +------------------------------+
|f_path: struct path; ==========+=+ |f_path: struct path; |
|f_inode: struct inode; | H H |f_inode: struct inode; |
|f_op: struct file_operations*;| H H |f_op: struct file_operations*;|
|private_data: void *; ========+=+ H |private_data: void *; |
| ... | H | ... |
+------------------------------+ H +------------------------------+
+================================+
V
struct path
+-----------------------+
|mnt: struct vfsmount*; |
|dentry: struct dentry*;|==+
+-----------------------+ H
+=========================+
V
struct dentry struct ovl_inode
+--------------------------------+ +---------------------------------+
|d_parent: struct dentry *; | |cache: struct ovl_dir_cache*; |
|d_name: struct qstr; | |flags: unsigned long; |
|d_inode: struct inode*;=========+==>|vfs_inode: struct inode; |
|d_op: struct dentry_operations*;| |+-struct inode -----------------+|
|d_sb: struct super_block; | ||i_op: struct inode_operations*;||
|d_fsdata: void *; =============++ ||i_sb: struct super_block *; ||
| ... |H ||i_ino: unsigned long; ||
+--------------------------------+H || ... ||
H |+-------------------------------+|
H |__upperdentry: struct dentry*; |
H |... |
H +---------------------------------+
H
H struct ovl_entry
+=>+------------------------------+
| ... |
|numlower: unsigned; |
|lowerstack: struct ovl_path[];|
+==============================+
|struct ovl_path { |
| layer: struct ovl_layer*; |
| dentry: struct dentry*; |
|}; |
+------------------------------+
| |
| |ovl_iterate(struct file *file, struct dir_context *ctx)
|-> struct ovl_dir_file *od = file->private_data;
|-> struct ovl_dir_cache *cache;
|
|-> cache = ovl_cache_get(dentry);
| |-> ovl_set_dir_cache(d_inode(dentry), NULL);
| |-> cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
| |-> cache->refcount = 1;
| |-> INIT_LIST_HEAD(&cache->entries);
| |-> cache->root = RB_ROOT;
| | // 遍历同名目录所在所有层次的文件列表，挂载文件列表到cache的红黑树节点
| |-> ovl_dir_read_merged(dentry, &cache->entries, &cache->root);
| | |-> for (idx = 0; idx != -1; idx = next) {
| | | next = ovl_path_next(idx, dentry, &realpath);
| | | |-> struct ovl_entry *oe = dentry->d_fsdata;
| | | |-> if (idx == 0) ovl_path_upper(dentry, path);
| | | |
| | | |-> path->dentry = oe->lowerstack[idx - 1].dentry;
| | | if (next != -1) {
| | | // 遍历某一层中目录下文件列表，结果存放在ctx中
| | | err = ovl_dir_read(&realpath, &rdd);
| | | |-> realfile = ovl_path_open(realpath, O_RDONLY | O_LARGEFILE);
| | | |-> iterate_dir(realfile, &rdd->ctx);
| | | } else {
| | | err = ovl_dir_read(&realpath, &rdd);
| | | }
| | |-> }
| |-> cache->root = RB_ROOT;
| |-> ovl_set_dir_cache(d_inode(dentry), cache);
|
|-> od->cache = cache;
|-> ovl_seek_cursor(od, ctx->pos);
| |-> od->cursor = p; 由上面的流程可以看出，若是目录是合成的，在遍历目录下文件列表时，是将各叠层目录下的所有文件列表保存在文件所有的红黑数上，结构如下：
struct ovl_dir_file
+=>+-----------------------------+
H |is_real: bool |
H |is_upper: unsigned; |
H |cache: struct ovl_dir_cache*;+--------+
H |cursor: struct list_head*; | |
H |realfile: struct file*; -----+-+ |
H +-----------------------------+ | |
H | |
struct file H struct file V |
+------------------------------+ H +------------------------------+ |
|f_path: struct path; ==========+=+ |f_path: struct path; | |
|f_inode: struct inode; | H H |f_inode: struct inode; | |
|f_op: struct file_operations*;| H H |f_op: struct file_operations*;| |
|private_data: void *; ========+=+ H |private_data: void *; | | 保存所有不同层次相同目录下的文件列表（红黑树）
| ... | H | ... | | /
+------------------------------+ H +------------------------------+ | V
+================================+ | struct ovl_dir_cache
V +->+------------------------+
struct path | |refcount: long; |
+-----------------------+ | |version: u64; |
|mnt: struct vfsmount*; | | |entries: struct entries;|
|dentry: struct dentry*;|==+ | |root: struct rb_root; |
+-----------------------+ H | +------------------------+
+=========================+ |
V |
struct dentry struct ovl_inode |
+--------------------------------+ +---------------------------------+ |
|d_parent: struct dentry *; | |cache: struct ovl_dir_cache*; ---+-+
|d_name: struct qstr; | |flags: unsigned long; |
|d_inode: struct inode*;=========+==>|vfs_inode: struct inode; |
|d_op: struct dentry_operations*;| |+-struct inode -----------------+|
|d_sb: struct super_block; | ||i_op: struct inode_operations*;||
|d_fsdata: void *; | ||i_sb: struct super_block *; ||
| ... | ||i_ino: unsigned long; ||
+--------------------------------+ || ... ||
|+-------------------------------+|
|__upperdentry: struct dentry*; |
|... |
+---------------------------------+总结
本文重点介绍Overlay FS的目录相关的关键数据结构和目录的相关接口，让读者能体会到所谓的Overlay文件系统的特点，很多实现细节是有别于传统的文件系统，均需考虑upper层和lower层不同情况，但是万变不离其宗，我们学习文件系统必须抓住文件系统的主要数据和接口的实现，这对C语言实现其它底层软件开发也是一种必备技能。
