taskfs fix up, minor refactoring

[lunaix-os.git] / lunaix-os / kernel / fs / vfs.c

diff --git a/lunaix-os/kernel/fs/vfs.c b/lunaix-os/kernel/fs/vfs.c
index e13b489226760544a232dacadaff09cda0387517..53c1efa93f822dffe16e4f468f1a3df8970d681e 100644 (file)
--- a/lunaix-os/kernel/fs/vfs.c
+++ b/lunaix-os/kernel/fs/vfs.c
@@ -9,22 +9,53 @@
  *
  */
 
  *
  */
 

+// Welcome to The Mountain O'Shit! :)
+
+/*
+ TODO vfs & device todos checklist
+
+    It is overseen by Twilight Sparkle ;)
+
+ 1. Get inodes hooked into lru (CHECKED)
+ 2. Get dnodes hooked into lru (CHECKED)
+ 3. Get inodes properly hashed so they can be reused by underling fs (CHECKED)
+ 4. (lru) Add a callback function (or destructor) for eviction. (CHECKED)
+        [good idea] or a constructor/destructor pattern in cake allocator ?
+ 5. (mount) Figure out a way to identify a busy mount point before unmount
+            maybe a unified mount_point structure that maintain a referencing
+            counter on any dnodes within the subtree? Such a counter will only
+            increament if a file is opened or a dnode is being used as working
+            directory and decreamenting conversely. (CHECKED)
+ 6. (mount) Ability to track all mount points (including sub-mounts)
+            so we can be confident to clean up everything when we
+            unmount. (CHECKED)
+ 7. (mount) Figure out a way to acquire the device represented by a dnode.
+            so it can be used to mount. (e.g. we wish to get `struct device*`
+            out of the dnode at /dev/sda)
+            [tip] we should pay attention at twifs and add a private_data field
+            under struct v_dnode? (CHECKED)
+ 8. (mount) Then, we should refactor on mount/unmount mechanism. (CHECKED)
+ 9. (mount) (future) Ability to mount any thing? e.g. Linux can mount a disk
+                    image file using a so called "loopback" pseudo device. Maybe
+                    we can do similar thing in Lunaix? A block device emulation
+                    above the regular file when we mount it on.
+ 10. (device) device number (dev_t) allocation
+            [good idea] <class>:<subclass>:<uniq_id> composition (CHECKED)
+*/
+

 #include <klibc/string.h>
 #include <klibc/string.h>

-#include <lunaix/dirent.h>

 #include <lunaix/foptions.h>
 #include <lunaix/fs.h>
 #include <lunaix/mm/cake.h>
 #include <lunaix/foptions.h>
 #include <lunaix/fs.h>
 #include <lunaix/mm/cake.h>

-#include <lunaix/mm/page.h>

 #include <lunaix/mm/valloc.h>
 #include <lunaix/process.h>
 #include <lunaix/spike.h>
 #include <lunaix/syscall.h>
 #include <lunaix/mm/valloc.h>
 #include <lunaix/process.h>
 #include <lunaix/spike.h>
 #include <lunaix/syscall.h>

+#include <lunaix/syscall_utils.h>
+
+#include <lunaix/fs/twifs.h>

 
 

-#define PATH_DELIM '/'
-#define DNODE_HASHTABLE_BITS 10
-#define DNODE_HASHTABLE_SIZE (1 << DNODE_HASHTABLE_BITS)
-#define DNODE_HASH_MASK (DNODE_HASHTABLE_SIZE - 1)
-#define DNODE_HASHBITS (32 - DNODE_HASHTABLE_BITS)
+#include <usr/lunaix/dirent_defs.h>

 
 static struct cake_pile* dnode_pile;
 static struct cake_pile* inode_pile;
 
 static struct cake_pile* dnode_pile;
 static struct cake_pile* inode_pile;


@@ -32,26 +63,19 @@ static struct cake_pile* file_pile;
 static struct cake_pile* superblock_pile;
 static struct cake_pile* fd_pile;
 
 static struct cake_pile* superblock_pile;
 static struct cake_pile* fd_pile;
 

-static struct v_superblock* root_sb;
-static struct hbucket* dnode_cache;
+struct v_dnode* vfs_sysroot;

 
 

-static int fs_id = 0;
+struct lru_zone *dnode_lru, *inode_lru;

 
 struct hstr vfs_ddot = HSTR("..", 2);
 struct hstr vfs_dot = HSTR(".", 1);
 struct hstr vfs_empty = HSTR("", 0);
 
 
 struct hstr vfs_ddot = HSTR("..", 2);
 struct hstr vfs_dot = HSTR(".", 1);
 struct hstr vfs_empty = HSTR("", 0);
 

-struct v_dnode*
-vfs_d_alloc();
-
-void
-vfs_d_free(struct v_dnode* dnode);
-
-struct v_superblock*
-vfs_sb_alloc();
+static int
+__vfs_try_evict_dnode(struct lru_node* obj);

 
 

-void
-vfs_sb_free(struct v_superblock* sb);
+static int
+__vfs_try_evict_inode(struct lru_node* obj);

 
 void
 vfs_init()
 
 void
 vfs_init()


@@ -64,25 +88,45 @@ vfs_init()
     superblock_pile =
       cake_new_pile("sb_cache", sizeof(struct v_superblock), 1, 0);
 
     superblock_pile =
       cake_new_pile("sb_cache", sizeof(struct v_superblock), 1, 0);
 

-    dnode_cache = vzalloc(DNODE_HASHTABLE_SIZE * sizeof(struct hbucket));
+    dnode_lru = lru_new_zone("vfs_dnode", __vfs_try_evict_dnode);
+    inode_lru = lru_new_zone("vfs_inode", __vfs_try_evict_inode);

 
     hstr_rehash(&vfs_ddot, HSTR_FULL_HASH);
     hstr_rehash(&vfs_dot, HSTR_FULL_HASH);
 
 
     hstr_rehash(&vfs_ddot, HSTR_FULL_HASH);
     hstr_rehash(&vfs_dot, HSTR_FULL_HASH);
 

-    // 创建一个根superblock，用来蕴含我们的根目录。
-    root_sb = vfs_sb_alloc();
-    root_sb->root = vfs_d_alloc();
-    root_sb->root->inode = vfs_i_alloc();
+    // 创建一个根dnode。
+    vfs_sysroot = vfs_d_alloc(NULL, &vfs_empty);
+    vfs_sysroot->parent = vfs_sysroot;
+    
+    vfs_ref_dnode(vfs_sysroot);

 }
 
 }
 

-inline struct hbucket*
-__dcache_get_bucket(struct v_dnode* parent, unsigned int hash)
+static inline struct hbucket*
+__dcache_hash(struct v_dnode* parent, u32_t* hash)

 {
 {

-    // 与parent的指针值做加法，来减小碰撞的可能性。
-    hash += (uint32_t)parent;
-    // 确保低位更加随机
-    hash = hash ^ (hash >> DNODE_HASHBITS);
-    return &dnode_cache[hash & DNODE_HASH_MASK];
+    struct hbucket* d_cache;
+    u32_t _hash;
+    
+    d_cache = parent->super_block->d_cache;
+    _hash = *hash;
+    _hash = _hash ^ (_hash >> VFS_HASHBITS);
+    _hash += (u32_t)__ptr(parent);
+
+    *hash = _hash;
+    return &d_cache[_hash & VFS_HASH_MASK];
+}
+
+static inline int
+__sync_inode_nolock(struct v_inode* inode)
+{
+    pcache_commit_all(inode);
+
+    int errno = ENOTSUP;
+    if (inode->ops->sync) {
+        errno = inode->ops->sync(inode);
+    }
+
+    return errno;

 }
 
 struct v_dnode*
 }
 
 struct v_dnode*


@@ -92,15 +136,16 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
         return parent;
 
     if (HSTR_EQ(str, &vfs_ddot)) {
         return parent;
 
     if (HSTR_EQ(str, &vfs_ddot)) {

-        return parent->parent ? parent->parent : parent;
+        return parent->parent;

     }
 
     }
 

-    struct hbucket* slot = __dcache_get_bucket(parent, str->hash);
+    u32_t hash = str->hash;
+    struct hbucket* slot = __dcache_hash(parent, &hash);

 
     struct v_dnode *pos, *n;
     hashtable_bucket_foreach(slot, pos, n, hash_list)
     {
 
     struct v_dnode *pos, *n;
     hashtable_bucket_foreach(slot, pos, n, hash_list)
     {

-        if (pos->name.hash == str->hash) {
+        if (pos->name.hash == hash && pos->parent == parent) {

             return pos;
         }
     }
             return pos;
         }
     }


@@ -110,292 +155,213 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
 void
 vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
 {
 void
 vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
 {

-    struct hbucket* bucket = __dcache_get_bucket(parent, dnode->name.hash);
+    assert(parent);
+
+    dnode->ref_count = 1;
+    dnode->parent = parent;
+    llist_append(&parent->children, &dnode->siblings);
+
+    struct hbucket* bucket = __dcache_hash(parent, &dnode->name.hash);

     hlist_add(&bucket->head, &dnode->hash_list);
 }
 
     hlist_add(&bucket->head, &dnode->hash_list);
 }
 

-int
-__vfs_walk(struct v_dnode* start,
-           const char* path,
-           struct v_dnode** dentry,
-           struct hstr* component,
-           int walk_options)
+void
+vfs_dcache_remove(struct v_dnode* dnode)

 {
 {

-    int errno = 0;
-    int i = 0, j = 0;
-
-    if (path[0] == PATH_DELIM || !start) {
-        if ((walk_options & VFS_WALK_FSRELATIVE) && start) {
-            start = start->super_block->root;
-        } else {
-            start = root_sb->root;
-        }
-        i++;
-    }
-
-    struct v_dnode* dnode;
-    struct v_dnode* current_level = start;
-
-    char name_content[VFS_NAME_MAXLEN];
-    struct hstr name = HSTR(name_content, 0);
-
-    char current = path[i++], lookahead;
-    while (current) {
-        lookahead = path[i++];
-        if (current != PATH_DELIM) {
-            if (j >= VFS_NAME_MAXLEN - 1) {
-                return ENAMETOOLONG;
-            }
-            if (!VFS_VALID_CHAR(current)) {
-                return EINVAL;
-            }
-            name_content[j++] = current;
-            if (lookahead) {
-                goto cont;
-            }
-        }
-
-        // handling cases like /^.*(\/+).*$/
-        if (lookahead == PATH_DELIM) {
-            goto cont;
-        }
-
-        name_content[j] = 0;
-        name.len = j;
-        hstr_rehash(&name, HSTR_FULL_HASH);
-
-        if (!lookahead && (walk_options & VFS_WALK_PARENT)) {
-            if (component) {
-                component->hash = name.hash;
-                component->len = j;
-                strcpy(component->value, name_content);
-            }
-            break;
-        }
-
-        dnode = vfs_dcache_lookup(current_level, &name);
+    assert(dnode);
+    assert(dnode->ref_count == 1);

 
 

-        if (!dnode) {
-            dnode = vfs_d_alloc();
-            dnode->name = HHSTR(valloc(VFS_NAME_MAXLEN), j, name.hash);
+    llist_delete(&dnode->siblings);
+    llist_delete(&dnode->aka_list);
+    hlist_delete(&dnode->hash_list);

 
 

-            strcpy(dnode->name.value, name_content);
+    dnode->parent = NULL;
+    dnode->ref_count = 0;
+}

 
 

-            errno =
-              current_level->inode->ops.dir_lookup(current_level->inode, dnode);
+void
+vfs_dcache_rehash(struct v_dnode* new_parent, struct v_dnode* dnode)
+{
+    assert(new_parent);

 
 

-            if (errno == ENOENT && (walk_options & VFS_WALK_MKPARENT)) {
-                if (!current_level->inode->ops.mkdir) {
-                    errno = ENOTSUP;
-                } else {
-                    errno = current_level->inode->ops.mkdir(
-                      current_level->inode, dnode);
-                }
-            }
+    hstr_rehash(&dnode->name, HSTR_FULL_HASH);
+    vfs_dcache_remove(dnode);
+    vfs_dcache_add(new_parent, dnode);
+}

 
 

-            if (errno) {
-                goto error;
-            }
+int
+vfs_open(struct v_dnode* dnode, struct v_file** file)
+{
+    if (!dnode->inode || !dnode->inode->ops->open) {
+        return ENOTSUP;
+    }

 
 

-            vfs_dcache_add(current_level, dnode);
+    struct v_inode* inode = dnode->inode;

 
 

-            dnode->parent = current_level;
-            llist_append(&current_level->children, &dnode->siblings);
-        }
+    lock_inode(inode);

 
 

-        j = 0;
-        current_level = dnode;
-    cont:
-        current = lookahead;
-    };
+    struct v_file* vfile = cake_grab(file_pile);
+    memset(vfile, 0, sizeof(*vfile));

 
 

-    *dentry = current_level;
-    return 0;
+    vfile->dnode = dnode;
+    vfile->inode = inode;
+    vfile->ref_count = 1;
+    vfile->ops = inode->default_fops;

 
 

-error:
-    vfree(dnode->name.value);
-    vfs_d_free(dnode);
-    *dentry = NULL;
-    return errno;
-}
+    if (check_regfile_node(inode) && !inode->pg_cache) {
+        struct pcache* pcache = vzalloc(sizeof(struct pcache));
+        pcache_init(pcache);
+        pcache->master = inode;
+        inode->pg_cache = pcache;
+    }

 
 

-#define VFS_MAX_SYMLINK 16
+    int errno = inode->ops->open(inode, vfile);
+    if (errno) {
+        cake_release(file_pile, vfile);
+    } else {
+        vfs_ref_dnode(dnode);
+        inode->open_count++;

 
 

-int
-vfs_walk(struct v_dnode* start,
-         const char* path,
-         struct v_dnode** dentry,
-         struct hstr* component,
-         int options)
-{
-    struct v_dnode* interim;
-    char* pathname = path;
-    int errno = __vfs_walk(start, path, &interim, component, options);
-    int counter = 0;
-
-    while (!errno) {
-        if (counter >= VFS_MAX_SYMLINK) {
-            errno = ELOOP;
-            continue;
-        }
-        if ((interim->inode->itype & VFS_IFSYMLINK) &&
-            !(options & VFS_WALK_NOFOLLOW) &&
-            interim->inode->ops.read_symlink) {
-            errno = interim->inode->ops.read_symlink(interim->inode, &pathname);
-            if (errno) {
-                break;
-            }
-        } else {
-            break;
-        }
-        errno = __vfs_walk(start, pathname, &interim, component, options);
-        counter++;
+        *file = vfile;

     }
 
     }
 

-    *dentry = errno ? 0 : interim;
+    unlock_inode(inode);

 
     return errno;
 }
 
 
     return errno;
 }
 

-int
-vfs_mount(const char* target, const char* fs_name, bdev_t device)
+void
+vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode)

 {
 {

-    int errno;
-    struct v_dnode* mnt;
-
-    if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
-        errno = vfs_mount_at(fs_name, device, mnt);
+    if (assign_to->inode) {
+        llist_delete(&assign_to->aka_list);
+        assign_to->inode->link_count--;

     }
 
     }
 

-    return errno;
+    llist_append(&inode->aka_dnodes, &assign_to->aka_list);
+    assign_to->inode = inode;
+    inode->link_count++;

 }
 
 int
 }
 
 int

-vfs_unmount(const char* target)
+vfs_link(struct v_dnode* to_link, struct v_dnode* name)

 {
     int errno;
 {
     int errno;

-    struct v_dnode* mnt;

 
 

-    if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
-        errno = vfs_unmount_at(mnt);
+    if ((errno = vfs_check_writable(to_link))) {
+        return errno;

     }
 
     }
 

-    return errno;
-}
-
-int
-vfs_mount_at(const char* fs_name, bdev_t device, struct v_dnode* mnt_point)
-{
-    struct filesystem* fs = fsm_get(fs_name);
-    if (!fs)
-        return ENODEV;
-    struct v_superblock* sb = vfs_sb_alloc();
-    sb->dev = device;
-    sb->fs_id = fs_id++;
-
-    int errno = 0;
-    if (!(errno = fs->mount(sb, mnt_point))) {
-        sb->fs = fs;
-        sb->root = mnt_point;
-        mnt_point->super_block = sb;
-        llist_append(&root_sb->sb_list, &sb->sb_list);
+    lock_inode(to_link->inode);
+    if (to_link->super_block->root != name->super_block->root) {
+        errno = EXDEV;
+    } else if (!to_link->inode->ops->link) {
+        errno = ENOTSUP;
+    } else if (!(errno = to_link->inode->ops->link(to_link->inode, name))) {
+        vfs_assign_inode(name, to_link->inode);

     }
     }

+    unlock_inode(to_link->inode);

 
     return errno;
 }
 
 int
 
     return errno;
 }
 
 int

-vfs_unmount_at(struct v_dnode* mnt_point)
+vfs_pclose(struct v_file* file, pid_t pid)

 {
 {

+    struct v_inode* inode;

     int errno = 0;
     int errno = 0;

-    struct v_superblock* sb = mnt_point->super_block;
-    if (!sb) {
-        return EINVAL;
-    }
-    if (!(errno = sb->fs->unmount(sb))) {
-        struct v_dnode* fs_root = sb->root;
-        llist_delete(&fs_root->siblings);
-        llist_delete(&sb->sb_list);
-        vfs_sb_free(sb);
-    }
-    return errno;
-}

 
 

-int
-vfs_open(struct v_dnode* dnode, struct v_file** file)
-{
-    if (!dnode->inode || !dnode->inode->ops.open) {
-        return ENOTSUP;
+    inode = file->inode;
+
+    /*
+     * Prevent dead lock.
+     * This happened when process is terminated while blocking on read.
+     * In that case, the process is still holding the inode lock and it
+         will never get released.
+     * The unlocking should also include ownership check.
+     *
+     * To see why, consider two process both open the same file both with
+     * fd=x.
+     *      Process A: busy on reading x
+     *      Process B: do nothing with x
+     * Assuming that, after a very short time, process B get terminated
+     * while process A is still busy in it's reading business. By this
+     * design, the inode lock of this file x is get released by B rather
+     * than A. And this will cause a probable race condition on A if other
+     * process is writing to this file later after B exit.
+    */
+
+    mutex_unlock_for(&inode->lock, pid);
+    
+    if (vfs_check_duped_file(file)) {
+        vfs_unref_file(file);
+        return 0;

     }
 
     }
 

-    struct v_inode* inode = dnode->inode;
-    struct v_file* vfile = cake_grab(file_pile);
-    memset(vfile, 0, sizeof(*vfile));
+    if ((errno = file->ops->close(file))) {
+        goto done;
+    }

 
 

-    vfile->dnode = dnode;
-    vfile->inode = inode;
-    vfile->ref_count = 1;
-    vfile->ops = inode->default_fops;
+    vfs_unref_dnode(file->dnode);
+    cake_release(file_pile, file);

 
 

-    if ((inode->itype & VFS_IFFILE) && !inode->pg_cache) {
-        struct pcache* pcache = vzalloc(sizeof(struct pcache));
-        pcache_init(pcache);
-        inode->pg_cache = pcache;
+    /*
+        if the current inode is not being locked by other 
+        threads that does not share same open context,
+        then we can try to do sync opportunistically
+    */
+    if (mutex_on_hold(&inode->lock)) {
+        goto done;

     }
     }

+    
+    lock_inode(inode);

 
 

-    int errno = inode->ops.open(inode, vfile);
-    if (errno) {
-        cake_release(file_pile, vfile);
-    } else {
-        dnode->ref_count++;
-        inode->open_count++;
-        *file = vfile;
+    pcache_commit_all(inode);
+    inode->open_count--;
+
+    if (!inode->open_count) {
+        __sync_inode_nolock(inode);

     }
 
     }
 

+    unlock_inode(inode);
+
+done:

     return errno;
 }
 
 int
     return errno;
 }
 
 int

-vfs_link(struct v_dnode* to_link, struct v_dnode* name)
+vfs_close(struct v_file* file)

 {
 {

-    int errno;
-    if (to_link->super_block->root != name->super_block->root) {
-        errno = EXDEV;
-    } else if (!to_link->inode->ops.link) {
-        errno = ENOTSUP;
-    } else if (!(errno = to_link->inode->ops.link(to_link->inode, name))) {
-        name->inode = to_link->inode;
-        to_link->inode->link_count++;
-    }
+    return vfs_pclose(file, __current->pid);
+}

 
 

-    return errno;
+void
+vfs_free_fd(struct v_fd* fd)
+{
+    cake_release(fd_pile, fd);

 }
 
 int
 }
 
 int

-vfs_close(struct v_file* file)
+vfs_isync(struct v_inode* inode)

 {
 {

-    int errno = 0;
-    if (!file->ops.close || !(errno = file->ops.close(file))) {
-        file->dnode->ref_count--;
-        file->inode->open_count--;
-        pcache_commit_all(file);
-        cake_release(file_pile, file);
-    }
+    lock_inode(inode);
+
+    int errno = __sync_inode_nolock(inode);
+
+    unlock_inode(inode);
+

     return errno;
 }
 
 int
 vfs_fsync(struct v_file* file)
 {
     return errno;
 }
 
 int
 vfs_fsync(struct v_file* file)
 {

-    int errno = ENOTSUP;
-    pcache_commit_all(file);
-    if (file->ops.sync) {
-        errno = file->ops.sync(file);
-    }
-    if (!errno && file->inode->ops.sync) {
-        return file->inode->ops.sync(file->inode);
+    int errno;
+    if ((errno = vfs_check_writable(file->dnode))) {
+        return errno;

     }
     }

-    return errno;
+
+    return vfs_isync(file->inode);

 }
 
 int
 }
 
 int


@@ -416,59 +382,213 @@ vfs_sb_alloc()
     struct v_superblock* sb = cake_grab(superblock_pile);
     memset(sb, 0, sizeof(*sb));
     llist_init_head(&sb->sb_list);
     struct v_superblock* sb = cake_grab(superblock_pile);
     memset(sb, 0, sizeof(*sb));
     llist_init_head(&sb->sb_list);

+    
+    sb->i_cache = vzalloc(VFS_HASHTABLE_SIZE * sizeof(struct hbucket));
+    sb->d_cache = vzalloc(VFS_HASHTABLE_SIZE * sizeof(struct hbucket));
+
+    sb->ref_count = 1;

     return sb;
 }
 
 void
     return sb;
 }
 
 void

-vfs_sb_free(struct v_superblock* sb)
+vfs_sb_ref(struct v_superblock* sb)
+{
+    sb->ref_count++;
+}
+
+void
+vfs_sb_unref(struct v_superblock* sb)

 {
 {

+    assert(sb->ref_count);
+
+    sb->ref_count--;
+    if (likely(sb->ref_count)) {
+        return;
+    }
+
+    if (sb->ops.release) {
+        sb->ops.release(sb);
+    }
+
+    vfree(sb->i_cache);
+    vfree(sb->d_cache);
+    

     cake_release(superblock_pile, sb);
 }
 
     cake_release(superblock_pile, sb);
 }
 

+static int
+__vfs_try_evict_dnode(struct lru_node* obj)
+{
+    struct v_dnode* dnode = container_of(obj, struct v_dnode, lru);
+
+    if (!dnode->ref_count) {
+        vfs_d_free(dnode);
+        return 1;
+    }
+    return 0;
+}
+
+static int
+__vfs_try_evict_inode(struct lru_node* obj)
+{
+    struct v_inode* inode = container_of(obj, struct v_inode, lru);
+
+    if (!inode->link_count && !inode->open_count) {
+        vfs_i_free(inode);
+        return 1;
+    }
+    return 0;
+}
+

 struct v_dnode*
 struct v_dnode*

-vfs_d_alloc()
+vfs_d_alloc(struct v_dnode* parent, struct hstr* name)

 {
     struct v_dnode* dnode = cake_grab(dnode_pile);
 {
     struct v_dnode* dnode = cake_grab(dnode_pile);

+    if (!dnode) {
+        lru_evict_half(dnode_lru);
+
+        if (!(dnode = cake_grab(dnode_pile))) {
+            return NULL;
+        }
+    }
+

     memset(dnode, 0, sizeof(*dnode));
     llist_init_head(&dnode->children);
     memset(dnode, 0, sizeof(*dnode));
     llist_init_head(&dnode->children);

-    dnode->name = vfs_empty;
+    llist_init_head(&dnode->siblings);
+    llist_init_head(&dnode->aka_list);
+    mutex_init(&dnode->lock);
+
+    dnode->name = HHSTR(vzalloc(VFS_NAME_MAXLEN), 0, 0);
+
+    hstrcpy(&dnode->name, name);
+
+    if (parent) {
+        vfs_d_assign_sb(dnode, parent->super_block);
+        dnode->mnt = parent->mnt;
+    }
+
+    lru_use_one(dnode_lru, &dnode->lru);
+

     return dnode;
 }
 
 void
 vfs_d_free(struct v_dnode* dnode)
 {
     return dnode;
 }
 
 void
 vfs_d_free(struct v_dnode* dnode)
 {

-    if (dnode->ops.destruct) {
-        dnode->ops.destruct(dnode);
+    assert(dnode->ref_count == 1);
+
+    if (dnode->inode) {
+        assert(dnode->inode->link_count > 0);
+        dnode->inode->link_count--;
+    }
+
+    vfs_dcache_remove(dnode);
+    // Make sure the children de-referencing their parent.
+    // With lru presented, the eviction will be propagated over the entire
+    // detached subtree eventually
+    struct v_dnode *pos, *n;
+    llist_for_each(pos, n, &dnode->children, siblings)
+    {
+        vfs_dcache_remove(pos);
+    }
+
+    if (dnode->destruct) {
+        dnode->destruct(dnode);

     }
     }

+
+    vfs_sb_unref(dnode->super_block);
+    vfree((void*)dnode->name.value);

     cake_release(dnode_pile, dnode);
 }
 
 struct v_inode*
     cake_release(dnode_pile, dnode);
 }
 
 struct v_inode*

-vfs_i_alloc()
+vfs_i_find(struct v_superblock* sb, u32_t i_id)
+{
+    struct hbucket* slot = &sb->i_cache[i_id & VFS_HASH_MASK];
+    struct v_inode *pos, *n;
+    hashtable_bucket_foreach(slot, pos, n, hash_list)
+    {
+        if (pos->id == i_id) {
+            lru_use_one(inode_lru, &pos->lru);
+            return pos;
+        }
+    }
+
+    return NULL;
+}
+
+void
+vfs_i_addhash(struct v_inode* inode)

 {
 {

-    struct v_inode* inode = cake_grab(inode_pile);
+    struct hbucket* slot = &inode->sb->i_cache[inode->id & VFS_HASH_MASK];
+
+    hlist_delete(&inode->hash_list);
+    hlist_add(&slot->head, &inode->hash_list);
+}
+
+struct v_inode*
+vfs_i_alloc(struct v_superblock* sb)
+{
+    assert(sb->ops.init_inode);
+
+    struct v_inode* inode;
+    if (!(inode = cake_grab(inode_pile))) {
+        lru_evict_half(inode_lru);
+        if (!(inode = cake_grab(inode_pile))) {
+            return NULL;
+        }
+    }
+

     memset(inode, 0, sizeof(*inode));
     memset(inode, 0, sizeof(*inode));

+    mutex_init(&inode->lock);
+    llist_init_head(&inode->xattrs);
+    llist_init_head(&inode->aka_dnodes);
+
+    sb->ops.init_inode(sb, inode);
+
+    inode->ctime = clock_unixtime();
+    inode->atime = inode->ctime;
+    inode->mtime = inode->ctime;
+
+    vfs_i_assign_sb(inode, sb);
+    lru_use_one(inode_lru, &inode->lru);

     return inode;
 }
 
 void
 vfs_i_free(struct v_inode* inode)
 {
     return inode;
 }
 
 void
 vfs_i_free(struct v_inode* inode)
 {

+    assert(inode->link_count == 0);
+    
+    if (inode->pg_cache) {
+        pcache_release(inode->pg_cache);
+        vfree(inode->pg_cache);
+    }
+    // we don't need to sync inode.
+    // If an inode can be free, then it must be properly closed.
+    // Hence it must be synced already!
+    if (inode->destruct) {
+        inode->destruct(inode);
+    }
+
+    vfs_sb_unref(inode->sb);
+    hlist_delete(&inode->hash_list);

     cake_release(inode_pile, inode);
 }
 
 /* ---- System call definition and support ---- */
 
     cake_release(inode_pile, inode);
 }
 
 /* ---- System call definition and support ---- */
 

-#define FLOCATE_CREATE_EMPTY 1
+// make a new name when not exists
+#define FLOC_MAYBE_MKNAME 1

 
 

-#define DO_STATUS(errno) SYSCALL_ESTATUS(__current->k_status = errno)
-#define DO_STATUS_OR_RETURN(errno) ({ errno < 0 ? DO_STATUS(errno) : errno; })
+// name must be non-exist and made.
+#define FLOC_MKNAME 2

 
 

-#define TEST_FD(fd) (fd >= 0 && fd < VFS_MAX_FD)
+// no follow symlink
+#define FLOC_NOFOLLOW 4

 
 int
 
 int

-__vfs_getfd(int fd, struct v_fd** fd_s)
+vfs_getfd(int fd, struct v_fd** fd_s)

 {
     if (TEST_FD(fd) && (*fd_s = __current->fdtable->fds[fd])) {
         return 0;
 {
     if (TEST_FD(fd) && (*fd_s = __current->fdtable->fds[fd])) {
         return 0;


@@ -476,72 +596,169 @@ __vfs_getfd(int fd, struct v_fd** fd_s)
     return EBADF;
 }
 
     return EBADF;
 }
 

-int
+static int
+__vfs_mknod(struct v_inode* parent, struct v_dnode* dnode, 
+            unsigned int itype, dev_t* dev)
+{
+    int errno;
+
+    errno = parent->ops->create(parent, dnode, itype);
+    if (errno) {
+        return errno;
+    }
+
+    return 0;
+}
+
+struct file_locator {
+    struct v_dnode* dir;
+    struct v_dnode* file;
+    bool fresh;
+};
+
+/**
+ * @brief unlock the file locator (floc) if possible.
+ *        If the file to be located if not exists, and
+ *        any FLOC_*MKNAME flag is set, then the parent
+ *        dnode will be locked until the file has been properly
+ *        finalised by subsequent logic.
+ * 
+ * @param floc 
+ */
+static inline void
+__floc_try_unlock(struct file_locator* floc)
+{
+    if (floc->fresh) {
+        assert(floc->dir);
+        unlock_dnode(floc->dir);
+    }
+}
+
+static int

 __vfs_try_locate_file(const char* path,
 __vfs_try_locate_file(const char* path,

-                      struct v_dnode** fdir,
-                      struct v_dnode** file,
+                      struct file_locator* floc,

                       int options)
 {
     char name_str[VFS_NAME_MAXLEN];
                       int options)
 {
     char name_str[VFS_NAME_MAXLEN];

+    struct v_dnode *fdir, *file;

     struct hstr name = HSTR(name_str, 0);
     struct hstr name = HSTR(name_str, 0);

-    int errno;
-    if ((errno =
-           vfs_walk(__current->cwd, path, fdir, &name, VFS_WALK_PARENT))) {
-        return errno;
+    int errno, woption = 0;
+
+    if ((options & FLOC_NOFOLLOW)) {
+        woption |= VFS_WALK_NOFOLLOW;
+        options &= ~FLOC_NOFOLLOW;

     }
 
     }
 

-    errno = vfs_walk(*fdir, name.value, file, NULL, 0);
-    if (errno != ENOENT || !(options & FLOCATE_CREATE_EMPTY)) {
+    floc->fresh = false;
+    name_str[0] = 0;
+    errno = vfs_walk_proc(path, &fdir, &name, woption | VFS_WALK_PARENT);
+    if (errno) {

         return errno;
     }
 
         return errno;
     }
 

-    if (!(errno = (*fdir)->inode->ops.create((*fdir)->inode))) {
-        struct v_dnode* file_new;
-        file_new = vfs_d_alloc();
-        file_new->name = HHSTR(valloc(VFS_NAME_MAXLEN), name.len, name.hash);
-        strcpy(file_new->name.value, name_str);
-        *file = file_new;
+    errno = vfs_walk(fdir, name.value, &file, NULL, woption);
+
+    if (errno && errno != ENOENT) {
+        goto done;
+    }
+    
+    if (!errno) {
+        if ((options & FLOC_MKNAME)) {
+            errno = EEXIST;
+        }
+        goto done;
+    }

 
 

-        llist_append(&(*fdir)->children, &file_new->siblings);
+    // errno == ENOENT
+    if (!options) {
+        goto done;

     }
 
     }
 

+    errno = vfs_check_writable(fdir);
+    if (errno) {
+        goto done;
+    }
+
+    floc->fresh = true;
+
+    file = vfs_d_alloc(fdir, &name);
+
+    if (!file) {
+        return ENOMEM;
+    }
+
+    lock_dnode(fdir);
+
+    vfs_dcache_add(fdir, file);
+
+done:
+    floc->dir   = fdir;
+    floc->file  = file;
+

     return errno;
 }
 
 int
 vfs_do_open(const char* path, int options)
 {
     return errno;
 }
 
 int
 vfs_do_open(const char* path, int options)
 {

-    int errno, fd;
+    int errno, fd, loptions = 0;

     struct v_dnode *dentry, *file;
     struct v_dnode *dentry, *file;

-    struct v_file* ofile = 0;
+    struct v_file* ofile = NULL;
+    struct file_locator floc;
+    struct v_inode* inode;

 
 

-    errno = __vfs_try_locate_file(path, &dentry, &file, 0);
+    if ((options & FO_CREATE)) {
+        loptions |= FLOC_MAYBE_MKNAME;
+    } else if ((options & FO_NOFOLLOW)) {
+        loptions |= FLOC_NOFOLLOW;
+    }

 
 

-    if (errno != ENOENT && (options & FO_CREATE)) {
-        errno = dentry->inode->ops.create(dentry->inode);
+    errno = __vfs_try_locate_file(path, &floc, loptions);
+
+    if (errno || (errno = vfs_alloc_fdslot(&fd))) {
+        return errno;

     }
 
     }
 

-    if (!errno && (errno = vfs_open(file, &ofile))) {
+    file   = floc.file;
+    dentry = floc.dir;
+
+    if (floc.fresh) {
+        errno = __vfs_mknod(dentry->inode, file, VFS_IFFILE, NULL);
+        if (errno) {
+            vfs_d_free(file);
+            __floc_try_unlock(&floc);
+            return errno;
+        }
+
+        __floc_try_unlock(&floc);
+    }
+
+
+    if ((errno = vfs_open(file, &ofile))) {

         return errno;
     }
 
         return errno;
     }
 

-    struct v_inode* o_inode = ofile->inode;
-    if (!(o_inode->itype & VFS_IFSEQDEV) && !(options & FO_DIRECT)) {
-        // XXX Change here accordingly when signature of pcache_r/w changed.
-        ofile->ops.read = pcache_read;
-        ofile->ops.write = pcache_write;
+    inode = ofile->inode;
+    lock_inode(inode);
+
+    struct v_fd* fd_s = cake_grab(fd_pile);
+    memset(fd_s, 0, sizeof(*fd_s));
+
+    if ((options & O_TRUNC)) {
+        file->inode->fsize = 0;   

     }
 
     }
 

-    if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
-        struct v_fd* fd_s = vzalloc(sizeof(*fd_s));
-        ofile->f_pos = ofile->inode->fsize & -((options & FO_APPEND) != 0);
-        fd_s->file = ofile;
-        fd_s->flags = options;
-        __current->fdtable->fds[fd] = fd_s;
-        return fd;
+    if (vfs_get_dtype(inode->itype) == DT_DIR) {
+        ofile->f_pos = 0;

     }
     }

+    
+    fd_s->file = ofile;
+    fd_s->flags = options;
+    __current->fdtable->fds[fd] = fd_s;

 
 

-    return errno;
+    unlock_inode(inode);
+    
+    return fd;

 }
 
 __DEFINE_LXSYSCALL2(int, open, const char*, path, int, options)
 }
 
 __DEFINE_LXSYSCALL2(int, open, const char*, path, int, options)


@@ -554,17 +771,15 @@ __DEFINE_LXSYSCALL1(int, close, int, fd)
 {
     struct v_fd* fd_s;
     int errno = 0;
 {
     struct v_fd* fd_s;
     int errno = 0;

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done_err;
     }
 
         goto done_err;
     }
 

-    if (fd_s->file->ref_count > 1) {
-        fd_s->file->ref_count--;
-    } else if ((errno = vfs_close(fd_s->file))) {
+    if ((errno = vfs_close(fd_s->file))) {

         goto done_err;
     }
 
         goto done_err;
     }
 

-    vfree(fd_s);
+    cake_release(fd_pile, fd_s);

     __current->fdtable->fds[fd] = 0;
 
 done_err:
     __current->fdtable->fds[fd] = 0;
 
 done_err:


@@ -577,101 +792,95 @@ __vfs_readdir_callback(struct dir_context* dctx,
                        const int len,
                        const int dtype)
 {
                        const int len,
                        const int dtype)
 {

-    struct dirent* dent = (struct dirent*)dctx->cb_data;
-    strncpy(dent->d_name, name, DIRENT_NAME_MAX_LEN);
-    dent->d_nlen = len;
+    struct lx_dirent* dent = (struct lx_dirent*)dctx->cb_data;
+    int len_ = MIN(len, DIRENT_NAME_MAX_LEN - 1);
+
+    strncpy(dent->d_name, name, len_);
+    dent->d_name[len_] = 0;
+
+    dent->d_nlen = len_;

     dent->d_type = dtype;
 }
 
     dent->d_type = dtype;
 }
 

-__DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
+__DEFINE_LXSYSCALL2(int, sys_readdir, int, fd, struct lx_dirent*, dent)

 {
     struct v_fd* fd_s;
     int errno;
 
 {
     struct v_fd* fd_s;
     int errno;
 

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
         goto done;
     }
 

-    if (!(fd_s->file->inode->itype & VFS_IFDIR)) {
+    struct v_inode* inode = fd_s->file->inode;
+
+    lock_inode(inode);
+
+    if (!check_directory_node(inode)) {

         errno = ENOTDIR;
         errno = ENOTDIR;

-    } else {
-        struct dir_context dctx =
-          (struct dir_context){ .cb_data = dent,
-                                .index = dent->d_offset,
-                                .read_complete_callback =
-                                  __vfs_readdir_callback };
-        if (dent->d_offset == 0) {
-            __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 0);
-        } else if (dent->d_offset == 1) {
-            __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0);
-        } else {
-            dctx.index -= 2;
-            if ((errno = fd_s->file->ops.readdir(fd_s->file, &dctx))) {
-                goto done;
-            }
-        }
-        errno = 0;
-        dent->d_offset++;
+        goto unlock;

     }
 
     }
 

-done:
-    return DO_STATUS(errno);
-}
+    struct dir_context dctx = (struct dir_context) {
+        .cb_data = dent,
+        .read_complete_callback = __vfs_readdir_callback
+    };

 
 

-__DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
-{
-    struct v_dnode *parent, *dir;
-    struct hstr component = HSTR(valloc(VFS_NAME_MAXLEN), 0);
-    int errno =
-      vfs_walk(__current->cwd, path, &parent, &component, VFS_WALK_PARENT);
-    if (errno) {
-        goto done;
+    if ((errno = fd_s->file->ops->readdir(fd_s->file, &dctx)) != 1) {
+        goto unlock;

     }
     }

+    dent->d_offset++;
+    fd_s->file->f_pos++;

 
 

-    if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
-        errno = ENOTSUP;
-    } else if (!parent->inode->ops.mkdir) {
-        errno = ENOTSUP;
-    } else if (!(parent->inode->itype & VFS_IFDIR)) {
-        errno = ENOTDIR;
-    } else {
-        dir = vfs_d_alloc();
-        dir->name = component;
-        if (!(errno = parent->inode->ops.mkdir(parent->inode, dir))) {
-            llist_append(&parent->children, &dir->siblings);
-        } else {
-            vfs_d_free(dir);
-            vfree(component.value);
-        }
-    }
+unlock:
+    unlock_inode(inode);

 
 done:
 
 done:

-    return DO_STATUS(errno);
+    return DO_STATUS_OR_RETURN(errno);
+}
+
+static inline bool
+check_pcache_eligibility(struct v_fd* fd_s)
+{
+    struct v_inode* inode;
+
+    inode = fd_s->file->inode;
+    return !check_seqdev_node(inode) \
+        && !fsm_check_pseudo_fs(inode->sb->fs) \
+        && !(fd_s->flags & FO_DIRECT);

 }
 
 __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
 {
     int errno = 0;
     struct v_fd* fd_s;
 }
 
 __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
 {
     int errno = 0;
     struct v_fd* fd_s;

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
     struct v_file* file = fd_s->file;
         goto done;
     }
 
     struct v_file* file = fd_s->file;

-    if ((file->inode->itype & VFS_IFDIR)) {
+    if (check_directory_node(file->inode)) {

         errno = EISDIR;
         goto done;
     }
 
         errno = EISDIR;
         goto done;
     }
 

-    cpu_enable_interrupt();
-    errno = file->ops.read(file, buf, count, file->f_pos);
-    cpu_disable_interrupt();
+    lock_inode(file->inode);
+
+    file->inode->atime = clock_unixtime();
+
+    if (!check_pcache_eligibility(fd_s)) {
+        errno = file->ops->read(file->inode, buf, count, file->f_pos);
+    } else {
+        errno = pcache_read(file->inode, buf, count, file->f_pos);
+    }

 
     if (errno > 0) {
         file->f_pos += errno;
 
     if (errno > 0) {
         file->f_pos += errno;

+        unlock_inode(file->inode);

         return errno;
     }
 
         return errno;
     }
 

+    unlock_inode(file->inode);
+

 done:
     return DO_STATUS(errno);
 }
 done:
     return DO_STATUS(errno);
 }


@@ -680,25 +889,46 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
 {
     int errno = 0;
     struct v_fd* fd_s;
 {
     int errno = 0;
     struct v_fd* fd_s;

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
         goto done;
     }
 

+    struct v_inode* inode;

     struct v_file* file = fd_s->file;
     struct v_file* file = fd_s->file;

-    if ((file->inode->itype & VFS_IFDIR)) {
+
+    if ((errno = vfs_check_writable(file->dnode))) {
+        goto done;
+    }
+
+    if (check_directory_node(file->inode)) {

         errno = EISDIR;
         goto done;
     }
 
         errno = EISDIR;
         goto done;
     }
 

-    cpu_enable_interrupt();
-    errno = file->ops.write(file, buf, count, file->f_pos);
-    cpu_disable_interrupt();
+    inode = file->inode;
+    lock_inode(inode);
+
+    inode->mtime = clock_unixtime();
+    if ((fd_s->flags & O_APPEND)) {
+        file->f_pos = inode->fsize;
+    }
+
+    if (!check_pcache_eligibility(fd_s)) {
+        errno = file->ops->write(inode, buf, count, file->f_pos);
+    } else {
+        errno = pcache_write(inode, buf, count, file->f_pos);
+    }

 
     if (errno > 0) {
         file->f_pos += errno;
 
     if (errno > 0) {
         file->f_pos += errno;

+        inode->fsize = MAX(inode->fsize, file->f_pos);
+
+        unlock_inode(inode);

         return errno;
     }
 
         return errno;
     }
 

+    unlock_inode(inode);
+

 done:
     return DO_STATUS(errno);
 }
 done:
     return DO_STATUS(errno);
 }


@@ -707,27 +937,48 @@ __DEFINE_LXSYSCALL3(int, lseek, int, fd, int, offset, int, options)
 {
     int errno = 0;
     struct v_fd* fd_s;
 {
     int errno = 0;
     struct v_fd* fd_s;

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
     struct v_file* file = fd_s->file;
         goto done;
     }
 
     struct v_file* file = fd_s->file;

-    size_t fpos = file->f_pos;
+    struct v_inode* inode = file->inode;
+
+    if (!file->ops->seek) {
+        errno = ENOTSUP;
+        goto done;
+    }
+
+    lock_inode(inode);
+
+    int overflow = 0;
+    int fpos = file->f_pos;
+
+    if (vfs_get_dtype(inode->itype) == DT_DIR) {
+        options = (options != FSEEK_END) ? options : FSEEK_SET;
+    }
+    

     switch (options) {
         case FSEEK_CUR:
     switch (options) {
         case FSEEK_CUR:

-            fpos = (size_t)((int)file->f_pos + offset);
+            overflow = sadd_of((int)file->f_pos, offset, &fpos);

             break;
         case FSEEK_END:
             break;
         case FSEEK_END:

-            fpos = (size_t)((int)file->inode->fsize + offset);
+            overflow = sadd_of((int)inode->fsize, offset, &fpos);

             break;
         case FSEEK_SET:
             fpos = offset;
             break;
     }
             break;
         case FSEEK_SET:
             fpos = offset;
             break;
     }

-    if (!file->ops.seek || !(errno = file->ops.seek(file, fpos))) {
-        file->f_pos = fpos;
+
+    if (overflow) {
+        errno = EOVERFLOW;
+    }
+    else {
+        errno = file->ops->seek(file, fpos);

     }
 
     }
 

+    unlock_inode(inode);
+

 done:
     return DO_STATUS(errno);
 }
 done:
     return DO_STATUS(errno);
 }


@@ -740,22 +991,28 @@ vfs_get_path(struct v_dnode* dnode, char* buf, size_t size, int depth)
     }
 
     if (depth > 64) {
     }
 
     if (depth > 64) {

-        return ELOOP;
+        return ENAMETOOLONG;

     }
 
     }
 

-    size_t len = vfs_get_path(dnode->parent, buf, size, depth + 1);
+    size_t len = 0;
+
+    if (dnode->parent != dnode) {
+        len = vfs_get_path(dnode->parent, buf, size, depth + 1);
+    }

 
     if (len >= size) {
         return len;
     }
 
 
     if (len >= size) {
         return len;
     }
 

+    if (!len || buf[len - 1] != VFS_PATH_DELIM) {
+        buf[len++] = VFS_PATH_DELIM;
+    }
+

     size_t cpy_size = MIN(dnode->name.len, size - len);
     strncpy(buf + len, dnode->name.value, cpy_size);
     size_t cpy_size = MIN(dnode->name.len, size - len);
     strncpy(buf + len, dnode->name.value, cpy_size);

-    len += cpy_size;
+    buf[len + cpy_size] = 0;

 
 

-    if (len < size) {
-        buf[len++] = PATH_DELIM;
-    }
+    len += cpy_size;

 
     return len;
 }
 
     return len;
 }


@@ -763,20 +1020,51 @@ vfs_get_path(struct v_dnode* dnode, char* buf, size_t size, int depth)
 int
 vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
 {
 int
 vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
 {

-    char* link;
-    if (dnode->inode->ops.read_symlink) {
-        int errno = dnode->inode->ops.read_symlink(dnode->inode, &link);
-        strncpy(buf, link, size);
-        return errno;
+    const char* link;
+    struct v_inode* inode = dnode->inode;
+
+    if (!check_symlink_node(inode)) {
+        return EINVAL;

     }
     }

-    return 0;
+
+    if (!inode->ops->read_symlink) {
+        return ENOTSUP;
+    }
+
+    lock_inode(inode);
+
+    int errno = inode->ops->read_symlink(inode, &link);
+    if (errno >= 0) {
+        strncpy(buf, link, MIN(size, (size_t)errno));
+    }
+
+    unlock_inode(inode);
+    return errno;
+}
+
+int
+vfs_get_dtype(int itype)
+{
+    int dtype = DT_FILE;
+    if (check_itype(itype, VFS_IFSYMLINK)) {
+        dtype |= DT_SYMLINK;
+    }
+    
+    if (check_itype(itype, VFS_IFDIR)) {
+        dtype |= DT_DIR;
+        return dtype;
+    }
+
+    // TODO other types
+    
+    return dtype;

 }
 
 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
 {
     int errno;
     struct v_fd* fd_s;
 }
 
 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
 {
     int errno;
     struct v_fd* fd_s;

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
         goto done;
     }
 


@@ -795,8 +1083,7 @@ __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size)
 {
     int errno;
     struct v_dnode* dnode;
 {
     int errno;
     struct v_dnode* dnode;

-    if (!(errno =
-            vfs_walk(__current->cwd, path, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
+    if (!(errno = vfs_walk_proc(path, &dnode, NULL, VFS_WALK_NOFOLLOW))) {

         errno = vfs_readlink(dnode, buf, size);
     }
 
         errno = vfs_readlink(dnode, buf, size);
     }
 


@@ -807,23 +1094,17 @@ __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size)
     return DO_STATUS(errno);
 }
 
     return DO_STATUS(errno);
 }
 

-__DEFINE_LXSYSCALL4(int,
-                    readlinkat,
-                    int,
-                    dirfd,
-                    const char*,
-                    pathname,
-                    char*,
-                    buf,
-                    size_t,
-                    size)
+__DEFINE_LXSYSCALL4(
+  int, readlinkat, int, dirfd, const char*, pathname, char*, buf, size_t, size)

 {
     int errno;
     struct v_fd* fd_s;
 {
     int errno;
     struct v_fd* fd_s;

-    if ((errno = __vfs_getfd(dirfd, &fd_s))) {
+    if ((errno = vfs_getfd(dirfd, &fd_s))) {

         goto done;
     }
 
         goto done;
     }
 

+    pathname = pathname ? pathname : "";
+

     struct v_dnode* dnode;
     if (!(errno = vfs_walk(
             fd_s->file->dnode, pathname, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
     struct v_dnode* dnode;
     if (!(errno = vfs_walk(
             fd_s->file->dnode, pathname, &dnode, NULL, VFS_WALK_NOFOLLOW))) {


@@ -838,49 +1119,149 @@ done:
     return DO_STATUS(errno);
 }
 
     return DO_STATUS(errno);
 }
 

+/*
+    NOTE
+    When we perform operation that could affect the layout of
+    directory (i.e., rename, mkdir, rmdir). We must lock the parent dir
+    whenever possible. This will blocking any ongoing path walking to reach
+    it hence avoid any partial state.
+*/
+

 __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
 {
     int errno;
     struct v_dnode* dnode;
 __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
 {
     int errno;
     struct v_dnode* dnode;

-    if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
+    if ((errno = vfs_walk_proc(pathname, &dnode, NULL, 0))) {
+        return DO_STATUS(errno);
+    }
+
+    lock_dnode(dnode);
+
+    if ((errno = vfs_check_writable(dnode))) {

         goto done;
     }
         goto done;
     }

+

     if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
         errno = EROFS;
         goto done;
     }
 
     if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
         errno = EROFS;
         goto done;
     }
 

-    if (dnode->inode->open_count) {
+    if (dnode->ref_count > 1 || dnode->inode->open_count) {

         errno = EBUSY;
         goto done;
     }
 
         errno = EBUSY;
         goto done;
     }
 

-    if ((dnode->inode->itype & VFS_IFDIR)) {
-        errno = dnode->inode->ops.rmdir(dnode->inode);
+    if (!llist_empty(&dnode->children)) {
+        errno = ENOTEMPTY;
+        goto done;
+    }
+
+    struct v_dnode* parent = dnode->parent;
+
+    if (!parent) {
+        errno = EINVAL;
+        goto done;
+    }
+
+    lock_dnode(parent);
+    lock_inode(parent->inode);
+
+    if (check_directory_node(dnode->inode)) {
+        errno = parent->inode->ops->rmdir(parent->inode, dnode);
+        if (!errno) {
+            vfs_dcache_remove(dnode);
+        }

     } else {
         errno = ENOTDIR;
     }
 
     } else {
         errno = ENOTDIR;
     }
 

+    unlock_inode(parent->inode);
+    unlock_dnode(parent);
+
+done:
+    unlock_dnode(dnode);
+    return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
+{
+    int errno = 0;
+    struct v_dnode *parent, *dir;
+    char name_value[VFS_NAME_MAXLEN];
+    struct hstr name = HHSTR(name_value, 0, 0);
+
+    if ((errno = vfs_walk_proc(path, &parent, &name, VFS_WALK_PARENT))) {
+        goto done;
+    }
+
+    if (!(errno = vfs_walk(parent, name_value, &dir, NULL, 0))) {
+        errno = EEXIST;
+        goto done;
+    }
+
+    if ((errno = vfs_check_writable(parent))) {
+        goto done;
+    }
+
+    if (!(dir = vfs_d_alloc(parent, &name))) {
+        errno = ENOMEM;
+        goto done;
+    }
+
+    struct v_inode* inode = parent->inode;
+
+    lock_dnode(parent);
+    lock_inode(inode);
+
+    if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
+        errno = ENOTSUP;
+    } else if (!inode->ops->mkdir) {
+        errno = ENOTSUP;
+    } else if (!check_directory_node(inode)) {
+        errno = ENOTDIR;
+    } else if (!(errno = inode->ops->mkdir(inode, dir))) {
+        vfs_dcache_add(parent, dir);
+        goto cleanup;
+    }
+
+    vfs_d_free(dir);
+
+cleanup:
+    unlock_inode(inode);
+    unlock_dnode(parent);

 done:
     return DO_STATUS(errno);
 }
 
 int
 done:
     return DO_STATUS(errno);
 }
 
 int

-__vfs_do_unlink(struct v_inode* inode)
+__vfs_do_unlink(struct v_dnode* dnode)

 {
     int errno;
 {
     int errno;

+    struct v_inode* inode = dnode->inode;
+
+    if (dnode->ref_count > 1) {
+        return EBUSY;
+    }
+
+    if ((errno = vfs_check_writable(dnode))) {
+        return errno;
+    }
+
+    lock_inode(inode);
+

     if (inode->open_count) {
         errno = EBUSY;
     if (inode->open_count) {
         errno = EBUSY;

-    } else if (!(inode->itype & VFS_IFDIR)) {
-        // TODO handle symbolic link and type other than regular file
-        errno = inode->ops.unlink(inode);
+    } else if (!check_directory_node(inode)) {
+        errno = inode->ops->unlink(inode, dnode);

         if (!errno) {
         if (!errno) {

-            inode->link_count--;
+            vfs_d_free(dnode);

         }
     } else {
         errno = EISDIR;
     }
 
         }
     } else {
         errno = EISDIR;
     }
 

+    unlock_inode(inode);
+

     return errno;
 }
 
     return errno;
 }
 


@@ -888,15 +1269,11 @@ __DEFINE_LXSYSCALL1(int, unlink, const char*, pathname)
 {
     int errno;
     struct v_dnode* dnode;
 {
     int errno;
     struct v_dnode* dnode;

-    if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
-        goto done;
-    }
-    if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
-        errno = EROFS;
+    if ((errno = vfs_walk_proc(pathname, &dnode, NULL, 0))) {

         goto done;
     }
 
         goto done;
     }
 

-    errno = __vfs_do_unlink(dnode->inode);
+    errno = __vfs_do_unlink(dnode);

 
 done:
     return DO_STATUS(errno);
 
 done:
     return DO_STATUS(errno);


@@ -906,13 +1283,13 @@ __DEFINE_LXSYSCALL2(int, unlinkat, int, fd, const char*, pathname)
 {
     int errno;
     struct v_fd* fd_s;
 {
     int errno;
     struct v_fd* fd_s;

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
     struct v_dnode* dnode;
     if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {
         goto done;
     }
 
     struct v_dnode* dnode;
     if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {

-        errno = __vfs_do_unlink(dnode->inode);
+        errno = __vfs_do_unlink(dnode);

     }
 
 done:
     }
 
 done:


@@ -922,18 +1299,30 @@ done:
 __DEFINE_LXSYSCALL2(int, link, const char*, oldpath, const char*, newpath)
 {
     int errno;
 __DEFINE_LXSYSCALL2(int, link, const char*, oldpath, const char*, newpath)
 {
     int errno;

-    struct v_dnode *dentry, *to_link, *name_dentry, *name_file;
+    struct file_locator floc;
+    struct v_dnode *to_link, *name_file;

 
 

-    errno = __vfs_try_locate_file(oldpath, &dentry, &to_link, 0);
+    errno = __vfs_try_locate_file(oldpath, &floc, 0);
+    if (errno) {
+        goto done;
+    }
+
+    __floc_try_unlock(&floc);
+
+    to_link = floc.file;
+    errno = __vfs_try_locate_file(newpath, &floc, FLOC_MKNAME);

     if (!errno) {
     if (!errno) {

-        errno = __vfs_try_locate_file(
-          newpath, &name_dentry, &name_file, FLOCATE_CREATE_EMPTY);
-        if (!errno) {
-            errno = EEXIST;
-        } else if (name_file) {
-            errno = vfs_link(to_link, name_file);
-        }
+        goto done;       

     }
     }

+
+    name_file = floc.file;
+    errno = vfs_link(to_link, name_file);
+    if (errno) {
+        vfs_d_free(name_file);
+    }
+
+done:
+    __floc_try_unlock(&floc);

     return DO_STATUS(errno);
 }
 
     return DO_STATUS(errno);
 }
 


@@ -941,7 +1330,8 @@ __DEFINE_LXSYSCALL1(int, fsync, int, fildes)
 {
     int errno;
     struct v_fd* fd_s;
 {
     int errno;
     struct v_fd* fd_s;

-    if (!(errno = __vfs_getfd(fildes, &fd_s))) {
+
+    if (!(errno = vfs_getfd(fildes, &fd_s))) {

         errno = vfs_fsync(fd_s->file);
     }
 
         errno = vfs_fsync(fd_s->file);
     }
 


@@ -955,7 +1345,8 @@ vfs_dup_fd(struct v_fd* old, struct v_fd** new)
     struct v_fd* copied = cake_grab(fd_pile);
 
     memcpy(copied, old, sizeof(struct v_fd));
     struct v_fd* copied = cake_grab(fd_pile);
 
     memcpy(copied, old, sizeof(struct v_fd));

-    old->file->ref_count++;
+
+    vfs_ref_file(old->file);

 
     *new = copied;
 
 
     *new = copied;
 


@@ -971,7 +1362,7 @@ vfs_dup2(int oldfd, int newfd)
 
     int errno;
     struct v_fd *oldfd_s, *newfd_s;
 
     int errno;
     struct v_fd *oldfd_s, *newfd_s;

-    if ((errno = __vfs_getfd(oldfd, &oldfd_s))) {
+    if ((errno = vfs_getfd(oldfd, &oldfd_s))) {

         goto done;
     }
 
         goto done;
     }
 


@@ -981,7 +1372,7 @@ vfs_dup2(int oldfd, int newfd)
     }
 
     newfd_s = __current->fdtable->fds[newfd];
     }
 
     newfd_s = __current->fdtable->fds[newfd];

-    if (newfd_s && (errno = vfs_close(newfd_s))) {
+    if (newfd_s && (errno = vfs_close(newfd_s->file))) {

         goto done;
     }
 
         goto done;
     }
 


@@ -1003,7 +1394,7 @@ __DEFINE_LXSYSCALL1(int, dup, int, oldfd)
 {
     int errno, newfd;
     struct v_fd *oldfd_s, *newfd_s;
 {
     int errno, newfd;
     struct v_fd *oldfd_s, *newfd_s;

-    if ((errno = __vfs_getfd(oldfd, &oldfd_s))) {
+    if ((errno = vfs_getfd(oldfd, &oldfd_s))) {

         goto done;
     }
 
         goto done;
     }
 


@@ -1017,50 +1408,73 @@ done:
     return DO_STATUS(errno);
 }
 
     return DO_STATUS(errno);
 }
 

-__DEFINE_LXSYSCALL2(int,
-                    symlink,
-                    const char*,
-                    pathname,
-                    const char*,
-                    link_target)
+__DEFINE_LXSYSCALL2(
+  int, symlink, const char*, pathname, const char*, link_target)

 {
     int errno;
 {
     int errno;

-    struct v_dnode* dnode;
-    if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
+    struct file_locator floc;
+    struct v_dnode *file;
+    struct v_inode *f_ino;
+    
+    errno = __vfs_try_locate_file(pathname, &floc, FLOC_MKNAME);
+    if (errno) {

         goto done;
     }
         goto done;
     }

-    if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
-        errno = EROFS;
+
+    file = floc.file;
+    errno = __vfs_mknod(floc.dir->inode, file, VFS_IFSYMLINK, NULL);
+    if (errno) {
+        vfs_d_free(file);

         goto done;
     }
         goto done;
     }

-    if (!dnode->inode->ops.symlink) {
+
+    f_ino = file->inode;
+
+    assert(f_ino);
+
+    errno = vfs_check_writable(file);
+    if (errno) {
+        goto done;
+    }
+
+    if (!f_ino->ops->set_symlink) {

         errno = ENOTSUP;
         goto done;
     }
 
         errno = ENOTSUP;
         goto done;
     }
 

-    errno = dnode->inode->ops.symlink(dnode->inode, link_target);
+    lock_inode(f_ino);
+
+    errno = f_ino->ops->set_symlink(f_ino, link_target);
+
+    unlock_inode(f_ino);

 
 done:
 
 done:

+    __floc_try_unlock(&floc);

     return DO_STATUS(errno);
 }
 
 int
     return DO_STATUS(errno);
 }
 
 int

-__vfs_do_chdir(struct v_dnode* dnode)
+vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode)

 {
     int errno = 0;
 {
     int errno = 0;

-    if (!(dnode->inode->itype & VFS_IFDIR)) {
+
+    lock_dnode(dnode);
+
+    if (!check_directory_node(dnode->inode)) {

         errno = ENOTDIR;
         goto done;
     }
 
         errno = ENOTDIR;
         goto done;
     }
 

-    if (__current->cwd) {
-        __current->cwd->ref_count--;
+    if (proc->cwd) {
+        vfs_unref_dnode(proc->cwd);

     }
 
     }
 

-    dnode->ref_count++;
-    __current->cwd = dnode;
+    vfs_ref_dnode(dnode);
+    proc->cwd = dnode;

 
 done:
 
 done:

+    unlock_dnode(dnode);
+

     return errno;
 }
 
     return errno;
 }
 


@@ -1069,11 +1483,11 @@ __DEFINE_LXSYSCALL1(int, chdir, const char*, path)
     struct v_dnode* dnode;
     int errno = 0;
 
     struct v_dnode* dnode;
     int errno = 0;
 

-    if ((errno = vfs_walk(__current->cwd, path, &dnode, NULL, 0))) {
+    if ((errno = vfs_walk_proc(path, &dnode, NULL, 0))) {

         goto done;
     }
 
         goto done;
     }
 

-    errno = __vfs_do_chdir(dnode);
+    errno = vfs_do_chdir((struct proc_info*)__current, dnode);

 
 done:
     return DO_STATUS(errno);
 
 done:
     return DO_STATUS(errno);


@@ -1084,11 +1498,11 @@ __DEFINE_LXSYSCALL1(int, fchdir, int, fd)
     struct v_fd* fd_s;
     int errno = 0;
 
     struct v_fd* fd_s;
     int errno = 0;
 

-    if ((errno = __vfs_getfd(fd, &fd_s))) {
+    if ((errno = vfs_getfd(fd, &fd_s))) {

         goto done;
     }
 
         goto done;
     }
 

-    errno = __vfs_do_chdir(fd_s->file->dnode);
+    errno = vfs_do_chdir((struct proc_info*)__current, fd_s->file->dnode);

 
 done:
     return DO_STATUS(errno);
 
 done:
     return DO_STATUS(errno);


@@ -1103,22 +1517,162 @@ __DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size)
         goto done;
     }
 
         goto done;
     }
 

+    size_t len = 0;
+

     if (!__current->cwd) {
     if (!__current->cwd) {

-        *buf = PATH_DELIM;
+        *buf = VFS_PATH_DELIM;
+        len = 1;
+    } else {
+        len = vfs_get_path(__current->cwd, buf, size, 0);
+        if (len == size) {
+            errno = ERANGE;
+            goto done;
+        }
+    }
+
+    buf[len] = '\0';
+
+    ret_ptr = buf;
+
+done:
+    syscall_result(errno);
+    return ret_ptr;
+}
+
+int
+vfs_do_rename(struct v_dnode* current, struct v_dnode* target)
+{
+    int errno = 0;
+    if (current->inode->id == target->inode->id) {
+        // hard link
+        return 0;
+    }
+
+    if ((errno = vfs_check_writable(current))) {
+        return errno;
+    }
+
+    if (current->ref_count > 1 || target->ref_count > 1) {
+        return EBUSY;
+    }
+
+    if (current->super_block != target->super_block) {
+        return EXDEV;
+    }
+
+    struct v_dnode* oldparent = current->parent;
+    struct v_dnode* newparent = target->parent;
+
+    lock_dnode(current);
+    lock_dnode(target);
+    if (oldparent)
+        lock_dnode(oldparent);
+    if (newparent)
+        lock_dnode(newparent);
+
+    if (!llist_empty(&target->children)) {
+        errno = ENOTEMPTY;
+        unlock_dnode(target);
+        goto cleanup;
+    }
+
+    if ((errno =
+           current->inode->ops->rename(current->inode, current, target))) {
+        unlock_dnode(target);
+        goto cleanup;
+    }
+
+    // re-position current
+    hstrcpy(&current->name, &target->name);
+    vfs_dcache_rehash(newparent, current);
+
+    // detach target
+    vfs_d_free(target);
+
+    unlock_dnode(target);
+
+cleanup:
+    unlock_dnode(current);
+    if (oldparent)
+        unlock_dnode(oldparent);
+    if (newparent)
+        unlock_dnode(newparent);
+
+    return errno;
+}
+
+__DEFINE_LXSYSCALL2(int, rename, const char*, oldpath, const char*, newpath)
+{
+    struct v_dnode *cur, *target_parent, *target;
+    struct hstr name = HSTR(valloc(VFS_NAME_MAXLEN), 0);
+    int errno = 0;
+
+    if ((errno = vfs_walk_proc(oldpath, &cur, NULL, 0))) {

         goto done;
     }
 
         goto done;
     }
 

-    size_t len = vfs_get_path(__current->cwd, buf, size, 0);
-    if (len == size) {
-        errno = ERANGE;
+    if ((errno = vfs_walk(
+           __current->cwd, newpath, &target_parent, &name, VFS_WALK_PARENT))) {

         goto done;
     }
 
         goto done;
     }
 

-    buf[len + 1] = '\0';
+    errno = vfs_walk(target_parent, name.value, &target, NULL, 0);
+    if (errno == ENOENT) {
+        target = vfs_d_alloc(target_parent, &name);
+        vfs_dcache_add(target_parent, target);
+    } else if (errno) {
+        goto done;
+    }

 
 

-    ret_ptr = buf;
+    if (!target) {
+        errno = ENOMEM;
+        goto done;
+    }
+
+    errno = vfs_do_rename(cur, target);

 
 done:
 
 done:

-    __current->k_status = errno;
-    return ret_ptr;
+    vfree((void*)name.value);
+    return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL2(int, fstat, int, fd, struct file_stat*, stat)
+{
+    int errno = 0;
+    struct v_fd* fds;
+
+    if ((errno = vfs_getfd(fd, &fds))) {
+        goto done;
+    }
+
+    struct v_inode* vino = fds->file->inode;
+    struct device* fdev = vino->sb->dev;
+
+    *stat = (struct file_stat){.st_ino = vino->id,
+                               .st_blocks = vino->lb_usage,
+                               .st_size = vino->fsize,
+                               .mode = vino->itype,
+                               .st_ioblksize = PAGE_SIZE,
+                               .st_blksize = vino->sb->blksize};
+
+    if (check_device_node(vino)) {
+        struct device* rdev = resolve_device(vino->data);
+        if (!rdev) {
+            errno = EINVAL;
+            goto done;
+        }
+
+        stat->st_rdev = (dev_t){.meta = rdev->ident.fn_grp,
+                                .unique = rdev->ident.unique,
+                                .index = dev_uid(rdev) };
+    }
+
+    if (fdev) {
+        stat->st_dev = (dev_t){.meta = fdev->ident.fn_grp,
+                               .unique = fdev->ident.unique,
+                               .index = dev_uid(fdev) };
+    }
+
+done:
+    return DO_STATUS(errno);

 }
\ No newline at end of file
 }
\ No newline at end of file