Overlay FS聯(lián)合文件系統(tǒng)源碼解析系列—目錄接口詳解
作者：蔣衛(wèi)峰 鐘文清
本文重點(diǎn)介紹Overlay FS的目錄相關(guān)的關(guān)鍵數(shù)據(jù)結(jié)構(gòu)和目錄的相關(guān)接口，讓讀者能體會(huì)到所謂的Overlay文件系統(tǒng)的特點(diǎn)，很多實(shí)現(xiàn)細(xì)節(jié)是有別于傳統(tǒng)的文件系統(tǒng)，均需考慮upper層和lower層不同情況，但是萬(wàn)變不離其宗，我們學(xué)習(xí)文件系統(tǒng)必須抓住文件系統(tǒng)的主要數(shù)據(jù)和接口的實(shí)現(xiàn)，這對(duì)C語(yǔ)言實(shí)現(xiàn)其它底層軟件開(kāi)發(fā)也是一種必備技能。
想了解更多關(guān)于開(kāi)源的內(nèi)容，請(qǐng)?jiān)L問(wèn)：
引言
在前篇中介紹了Linux是如何掛載Overlay文件系統(tǒng)的，重點(diǎn)關(guān)注了Overlay的掛載流程的實(shí)現(xiàn)和關(guān)鍵數(shù)據(jù)結(jié)構(gòu)的關(guān)系，而在本文中主要介紹Overlay FS的目錄相關(guān)的關(guān)鍵數(shù)據(jù)結(jié)構(gòu)和目錄的相關(guān)接口，其中比較復(fù)雜的接口將會(huì)在下一篇中展開(kāi)介紹。
一、Overlay 關(guān)鍵數(shù)據(jù)結(jié)構(gòu)
1、目錄文件
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、目錄接口
// 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、目錄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、目錄操作
1、創(chuàng)建目錄
（1）創(chuàng)建目錄系統(tǒng)調(diào)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)創(chuàng)建目錄項(xiàng)：創(chuàng)建ovl_dentry，同時(shí)將其加入到其父目錄的子目錄項(xiàng)鏈表中
|-> dentry = user_path_create(dfd, pathname, &path, lookup_flags); // path保存父目錄路徑；dentry保存新建目錄項(xiàng)內(nèi)存對(duì)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)創(chuàng)建ovl inode：調(diào)用對(duì)應(yīng)文件系統(tǒng)目錄ops中的mkdir創(chuàng)建inode，并關(guān)聯(lián)inode到dentry
|-> vfs_mkdir(path.dentry->d_inode, dentry, mode);
|-> dir->i_op->mkdir(dir, dentry, mode) // 參數(shù)：父目錄inode、新目錄的目錄項(xiàng)、模式（2）Overlay FS中ovl_mkdir創(chuàng)建目錄的過(guò)程
在Overlay fs聯(lián)合文件系統(tǒng)下，創(chuàng)建目錄有2種情況：
（1）upper層和lower層均無(wú)同名目錄，則直接在upper層創(chuàng)建。
（2）<u>lower層有同名目錄，但是之后被刪除了</u>，則需要在已經(jīng)存在whiteout文件的情況下創(chuàng)建目錄。
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）創(chuàng)建ovl_inode節(jié)點(diǎn)
|-> inode = ovl_new_inode(dentry->d_sb, mode, rdev);
|-> inode->i_state |= I_CREATING;
|
| // （2）聯(lián)合文件系統(tǒng)調(diào)用（實(shí)際文件系統(tǒng)）創(chuàng)建節(jié)點(diǎn)，并連接（link）虛擬文件系統(tǒng)inode和實(shí)際文件系統(tǒng)inode
| // 實(shí)際文件系統(tǒng)創(chuàng)建新目錄，并將實(shí)際文件系統(tǒng)的inode賦值給虛擬文件系統(tǒng)的目錄項(xiàng)（dentry），同時(shí)虛擬文件系統(tǒng)的目錄項(xiàng)作為虛擬文件系統(tǒng)目錄項(xiàng)的一個(gè)別名
|-> ovl_create_or_link(dentry, inode, &attr, false) // 參數(shù)：虛擬文件系統(tǒng)目錄項(xiàng)、虛擬文件系統(tǒng)inode節(jié)點(diǎn)、屬性、非原始節(jié)點(diǎn)
| // 父目錄進(jìn)行copyup：將parent的lower目錄內(nèi)容copyup到upper層。若存在upper層，則直接返回。
|-> ovl_copy_up(parent);// copyup將會(huì)在后續(xù)的文章中詳細(xì)介紹
|-> if (!ovl_dentry_is_whiteout(dentry))
| // 不存在whiteout文件，則直接在upper層創(chuàng)建新目錄
|-> ovl_create_upper(dentry, inode, attr);
|-> else
| // 存在whiteout文件，則在upper層創(chuàng)建新目錄并..???..
|-> ovl_create_over_whiteout(dentry, inode, attr);（3）直接在upper層創(chuàng)建新目錄
參數(shù)inode為ovl_inode，而dentry則是與之對(duì)應(yīng)的overlayfs文件系統(tǒng)目錄節(jié)點(diǎn)的目錄項(xiàng)。
ovl_create_upper(struct dentry *dentry, struct inode *inode, struct ovl_cattr *attr)
| // 將overlayfs目錄轉(zhuǎn)換成源文件路徑
|-> upperdir = ovl_dentry_upper(dentry->d_parent); // 原始實(shí)際文件系統(tǒng)中的父目錄
|-> struct inode *udir = upperdir->d_inode;
|-> struct dentry *newdentry;
| // （1）創(chuàng)建父目錄的upper層創(chuàng)建新目錄
|-> 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);
|
| // 實(shí)例化overlayfs目錄項(xiàng)和ovl_inode:
| // (1)ovl_inode的i_private和__upperdentry分別指向upper的inode和目錄項(xiàng);(2)ovl目錄項(xiàng)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, // 實(shí)際文件系統(tǒng)目錄項(xiàng)
| .newinode = inode, // ovl_inode
| };
|
| // 返回ovl_inode的指針：ovl_inode的i_private和__upperdentry分別指向upper的inode和目錄項(xiàng)
|-> inode = ovl_get_inode(dentry->d_sb, &oip);
| | // 返回ovl_inode的指針,并將realinode作為ovl_inode私有數(shù)據(jù)
| |-> inode = ovl_iget5(sb, oip->newinode, key);
| | // 設(shè)置ovlinode的__upperdentry指向upper的目錄項(xiàng)upperdentry
| |-> ovl_inode_init(inode, oip, ino, fsid);
| |-> OVL_I(inode)->__upperdentry = oip->upperdentry;
|
| // 設(shè)置ovl目錄項(xiàng)的d_inode指向ovl_inode
|-> d_instantiate(dentry, inode); // overlayfs文件系統(tǒng)目錄項(xiàng)和inode
|-> __d_instantiate(entry, inode);
|-> __d_set_inode_and_type(dentry, inode, add_flags);
|-> dentry->d_inode = inode;ovl_create_upper()執(zhí)行完后，新創(chuàng)建的overlayfs目錄結(jié)構(gòu)，如下圖所示：
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）創(chuàng)建新目錄，但lower層存在被刪除的同名目錄
參數(shù)ovl_create_upper()一樣：inode為ovl_inode，而dentry則是與之對(duì)應(yīng)的overlayfs文件系統(tǒng)目錄節(jié)點(diǎn)的目錄項(xiàng)。
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父目錄項(xiàng)
|-> struct inode *udir = upperdir->d_inode; // upper父目錄inode
|-> struct dentry *upper; // ovl目錄項(xiàng)
|-> struct dentry *newdentry; // upper目錄項(xiàng)
|
| // (1)在upper層父目錄中查找目錄項(xiàng)（whiteout）
|-> upper = lookup_one_len(dentry->d_name.name, upperdir, dentry->d_name.len);
| // (2)在workdir目錄下創(chuàng)建臨時(shí)目錄
|-> newdentry = ovl_create_temp(workdir, cattr);
| |-> ovl_create_real(d_inode(workdir), ovl_lookup_temp(workdir), attr);
| // (3)設(shè)置worddir下新建臨時(shí)目錄屬性為opaque
|-> ovl_set_opaque(dentry, newdentry);
| // (4)workdir新建的臨時(shí)目錄與upper的whiteout目錄做交互rename
|-> ovl_do_rename(wdir, newdentry, udir, upper, RENAME_EXCHANGE);
| // (5)清除rename后workdir下的廢棄文件（whiteout）
|-> ovl_cleanup(wdir, upper);
|
| // 實(shí)例化overlayfs目錄項(xiàng)和ovl_inode：
| // (1)ovl_inode的i_private和__upperdentry分別指向upper的inode和目錄項(xiàng);(2)ovl目錄項(xiàng)的d_inode指向ovl_inode.
|-> ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);由上面流程可見(jiàn)，與ovl_create_upper()相比，ovl_create_over_whiteout()先在work目錄下創(chuàng)建一個(gè)臨時(shí)目錄，并設(shè)置目錄屬性為opaque，以屏蔽lower層同名目錄。然后將workdir目錄下臨時(shí)目錄與upper層的whiteout文件做重命名互換，并清除互換后的垃圾文件。
2、刪除目錄ovl_rmdir()
將要被刪除的聯(lián)合文件系統(tǒng)目錄主要有3種途徑：
（1）目錄只來(lái)自u(píng)pper層，則直接刪除即可；
（2）目錄只來(lái)自lower層，則需要建立對(duì)lower層同名目錄的whiteout文件；
（3）目錄是由uppwer層和lower層目錄合成，則既要?jiǎng)h除upper層目錄，又要建立lower層同名目錄的whiteout文件；
但是上面的情況，經(jīng)過(guò)簡(jiǎn)化合并為兩種情況：
- 不存在于lower層，則直接刪除即可
- 存在于lower層，需要生成whiteout文件屏蔽底層目錄。
下面代碼為Overlay fs聯(lián)合文件系統(tǒng)刪除目錄的過(guò)程，留意下列過(guò)程中根據(jù)lower_positive是否為true，即是否存在lower層，分別調(diào)用了函數(shù)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);
| | // 獲取目錄下的所有目錄和文件列掛到list鏈表
| |-> 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父目錄：若存在upper層，則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）刪除不存在lower層的目錄
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目錄
|-> upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,dentry->d_name.len)// 找到需要?jiǎng)h除的dengtry
|-> err = ovl_do_rmdir(ofs, dir, upper)
| |-> vfs_rmdir(ovl_upper_mnt_userns(ofs), dir, dentry) // 直接調(diào)用vfs_mkdir進(jìn)行刪除
| |...（2）刪除存在lower層的目錄
ovl_remove_and_whiteout(dentry, &list);
|-> struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
|-> workdir = ovl_workdir(dentry);
| // 返回opaque目錄：已經(jīng)完成了目標(biāo)目錄的刪除
|-> 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與upper的inode內(nèi)容互換
| |-> ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
| | // 遞歸刪除目標(biāo)目錄下的upper層內(nèi)容
| |-> ovl_cleanup_whiteouts(upper, list);
| | // 刪除目標(biāo)目錄
| |-> ovl_cleanup(wdir, upper);
| |-> ovl_do_rmdir(wdir, wdentry);
| |-> vfs_rmdir(dir, dentry);
| // 此時(shí)，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?jiǎng)h除的文件是upper層覆蓋lower層的文件，要?jiǎng)h除的目錄是上下層合并的目錄，其實(shí)是前兩個(gè)場(chǎng)景的合并，Overlay fs即需要?jiǎng)h除upper層對(duì)應(yīng)文件系統(tǒng)中的文件或目錄，也需要在對(duì)應(yīng)位置創(chuàng)建同名whiteout文件，讓upper層的文件被刪除后不至于lower層的文件被暴露出來(lái)。
3、ovl_dir_open打開(kāi)目錄
假設(shè)目錄是merged目錄：
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);
| // 獲取目錄路徑：若目錄是merged或者upper目錄，則獲得upper路徑；否則，獲得lower路徑。
|-> 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);
|
| // 打開(kāi)upper或者lower目錄
|-> 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作為ovldir_file的實(shí)際文件；ovl_dir_file又作為file的私有數(shù)據(jù)
|-> 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;目錄打開(kāi)后形成如下結(jié)構(gòu)：
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目錄內(nèi)查找目標(biāo)目錄或者文件
每當(dāng)在這樣一個(gè)合并的目錄中請(qǐng)求查找時(shí)，就會(huì)在每個(gè)實(shí)際的目錄中進(jìn)行查找，而合并的結(jié)果則緩存在屬于覆蓋文件系統(tǒng)的目錄中。結(jié)果被緩存在屬于疊加文件系統(tǒng)的目錄中。如果兩個(gè)實(shí)際的查詢都能找到目錄，那么這兩個(gè)查詢都會(huì)被存儲(chǔ)起來(lái)，并且創(chuàng)建一個(gè)合并的目錄。創(chuàng)建一個(gè)合并的目錄，否則只存儲(chǔ)一個(gè)：如果上層目錄存在，則存儲(chǔ)下層目錄。只有來(lái)自目錄的名稱列表被合并。其他內(nèi)容，如元數(shù)據(jù)和擴(kuò)展屬性，只報(bào)告給上層目錄，下層目錄的這些屬性被隱藏。
/*
DESCRIPTION: 在特定文件系統(tǒng)中，在目錄dir下查找dentry中包含
PARAMS:
@dir: 父目錄，在該目錄下查找目標(biāo)目錄或者文件，但是函數(shù)內(nèi)未使用，父目錄由dentry的d_parent獲得。
@dentry: 目標(biāo)目錄或者文件的目錄項(xiàng)，包含name，但是缺少inode
@flags*/
ovl_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) // fs/overlayfs/dir.c
| // (1)在upper層父目錄中查找目標(biāo)文件或者目錄：
| // (1.1)從ovl目錄項(xiàng)的ovl_inode反向索引到實(shí)際文件系統(tǒng)的dentry
|-> upperdir = ovl_dentry_upper(dentry->d_parent);
| // (1.2)在upper層目錄下搜索目標(biāo)文件或者目錄，若存在則使用局部目錄項(xiàng)指針upperdentry指向目標(biāo)目錄項(xiàng)
|-> ovl_lookup_layer(upperdir, &d, &upperdentry, true); // 參數(shù)：實(shí)際文件系統(tǒng)中的父目錄項(xiàng)、...、目標(biāo)實(shí)際文件系統(tǒng)中目錄項(xiàng)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l(wèi)ower層父目錄中查找目標(biāo)文件或者目錄
|-> 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，其中包括根據(jù)堆疊層次確定的ovl_path層次數(shù)
|-> oe = ovl_alloc_entry(ctr);
|-> memcpy(oe->lowerstack, stack, sizeof(struct ovl_path) * ctr);
|-> dentry->d_fsdata = oe;
|
| // 關(guān)聯(lián)實(shí)際文件系統(tǒng)denty和inode到虛擬文件系統(tǒng)inode
|-> inode = ovl_get_inode(dentry->d_sb, &oip);根據(jù)上面的流程，在overlayfs文件系統(tǒng)的一個(gè)目錄下查詢一個(gè)目錄或者文件的流程為，通過(guò)ovl系統(tǒng)中父目錄dentry找到實(shí)際文件系統(tǒng)的目錄項(xiàng)和inode，找到目標(biāo)目錄或者文件對(duì)應(yīng)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*;| ==> 實(shí)際文件系統(tǒng)的lookup()
|i_sb: struct super_block *; |
|i_ino: unsigned long; |
| ... |
+-------------------------------+查找到的目錄或者文件，在ovl文件系統(tǒng)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）打開(kāi)目錄
被打開(kāi)的目錄可能只來(lái)自于upper層或lower層，由或者是upper和lower目錄merged而成，打開(kāi)過(guò)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）獲取目錄路徑：若目錄是merged或者upper目錄，則獲得upper路徑；否則，獲得lower路徑。
|-> 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）打開(kāi)實(shí)際文件系統(tǒng)目錄文件：upper或者lower目錄
|-> 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）將overlayfs目錄文件和實(shí)際文件系統(tǒng)目錄文件關(guān)聯(lián)：realfile作為ovldir_file的實(shí)際文件；ovl_dir_file又作為file的私有數(shù)據(jù)
|-> 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;由上過(guò)程可知，Overlay fs目錄文件被打開(kāi)后，同時(shí)存在Overlay fs目錄文件和實(shí)際文件系統(tǒng)目錄文件，中間通過(guò)ovl_dir_file鏈接（ovl_dir_file作為Overlay fs目錄文件的私有數(shù)據(jù)），他們直接的關(guān)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目錄文件 實(shí)際文件系統(tǒng)目錄文件（3）遍歷目錄
遍歷目錄，即是所有各疊層相同目錄下的所有文件列表。
（4）系統(tǒng)調(diào)用-遍歷目錄
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函數(shù)iterate實(shí)現(xiàn)
假設(shè)目錄是merged的目錄：
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;
| | // 遍歷同名目錄所在所有層次的文件列表，掛載文件列表到cache的紅黑樹(shù)節(jié)點(diǎn)
| |-> 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);
| | | | <or>
| | | |-> path->dentry = oe->lowerstack[idx - 1].dentry;
| | | if (next != -1) {
| | | // 遍歷某一層中目錄下文件列表，結(jié)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，若是目錄是合成的，在遍歷目錄下文件列表時(shí)，是將各疊層目錄下的所有文件列表保存在文件所有的紅黑數(shù)上，結(jié)構(gòu)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 *; | | 保存所有不同層次相同目錄下的文件列表（紅黑樹(shù)）
| ... | 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*; |
|... |
+---------------------------------+總結(jié)
本文重點(diǎn)介紹Overlay FS的目錄相關(guān)的關(guān)鍵數(shù)據(jù)結(jié)構(gòu)和目錄的相關(guān)接口，讓讀者能體會(huì)到所謂的Overlay文件系統(tǒng)的特點(diǎn)，很多實(shí)現(xiàn)細(xì)節(jié)是有別于傳統(tǒng)的文件系統(tǒng)，均需考慮upper層和lower層不同情況，但是萬(wàn)變不離其宗，我們學(xué)習(xí)文件系統(tǒng)必須抓住文件系統(tǒng)的主要數(shù)據(jù)和接口的實(shí)現(xiàn)，這對(duì)C語(yǔ)言實(shí)現(xiàn)其它底層軟件開(kāi)發(fā)也是一種必備技能。
責(zé)任編輯：jianghua
來(lái)源：
51CTO 開(kāi)源基礎(chǔ)軟件社區(qū)
