Merge branch 'master' into iso-9660

[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 82f409a6c9996019ccf724fb7ce88ee2f069621e..d0fbb1cbb6ded93406770d38894873da4faf4f49 100644 (file)
--- a/lunaix-os/kernel/fs/vfs.c
+++ b/lunaix-os/kernel/fs/vfs.c
@@ -9,6 +9,40 @@
  *
  */
 
  *
  */
 

+// 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
+*/
+

 #include <klibc/string.h>
 #include <lunaix/dirent.h>
 #include <lunaix/foptions.h>
 #include <klibc/string.h>
 #include <lunaix/dirent.h>
 #include <lunaix/foptions.h>


@@ -20,17 +54,7 @@
 #include <lunaix/spike.h>
 #include <lunaix/syscall.h>
 
 #include <lunaix/spike.h>
 #include <lunaix/syscall.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)
-
-#define lock_inode(inode) mutex_lock(&inode->lock)
-#define unlock_inode(inode) mutex_unlock(&inode->lock)
-
-#define lock_dnode(dnode) mutex_lock(&dnode->lock)
-#define unlock_dnode(dnode) mutex_unlock(&dnode->lock)
+#include <lunaix/fs/twifs.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;


@@ -38,27 +62,27 @@ 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;
+struct v_dnode* vfs_sysroot;

 static struct hbucket* dnode_cache;
 
 static struct hbucket* dnode_cache;
 

-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();
 
 void
 vfs_sb_free(struct v_superblock* sb);
 
 struct v_superblock*
 vfs_sb_alloc();
 
 void
 vfs_sb_free(struct v_superblock* sb);
 

+static int
+__vfs_try_evict_dnode(struct lru_node* obj);
+
+static int
+__vfs_try_evict_inode(struct lru_node* obj);
+

 void
 vfs_init()
 {
 void
 vfs_init()
 {


@@ -70,25 +94,30 @@ 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_cache = vzalloc(VFS_HASHTABLE_SIZE * sizeof(struct hbucket));
+
+    dnode_lru = lru_new_zone(__vfs_try_evict_dnode);
+    inode_lru = lru_new_zone(__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;
+    atomic_fetch_add(&vfs_sysroot->ref_count, 1);

 }
 
 inline struct hbucket*
 }
 
 inline struct hbucket*

-__dcache_get_bucket(struct v_dnode* parent, unsigned int hash)
+__dcache_hash(struct v_dnode* parent, uint32_t* hash)

 {
 {

-    // 与parent的指针值做加法，来减小碰撞的可能性。
-    hash += (uint32_t)parent;
+    uint32_t _hash = *hash;

     // 确保低位更加随机
     // 确保低位更加随机

-    hash = hash ^ (hash >> DNODE_HASHBITS);
-    return &dnode_cache[hash & DNODE_HASH_MASK];
+    _hash = _hash ^ (_hash >> VFS_HASHBITS);
+    // 与parent的指针值做加法，来减小碰撞的可能性。
+    _hash += (uint32_t)parent;
+    *hash = _hash;
+    return &dnode_cache[_hash & VFS_HASH_MASK];

 }
 
 struct v_dnode*
 }
 
 struct v_dnode*


@@ -98,15 +127,16 @@ 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);
+    uint32_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) {

             return pos;
         }
     }
             return pos;
         }
     }


@@ -116,236 +146,51 @@ 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);
-    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)
-{
-    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;
-        }
-
-        lock_dnode(current_level);
-
-        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);
-            }
-            unlock_dnode(current_level);
-            break;
-        }
+    assert(parent);

 
 

-        dnode = vfs_dcache_lookup(current_level, &name);
-
-        if (!dnode) {
-            dnode = vfs_d_alloc();
-            dnode->name = HHSTR(valloc(VFS_NAME_MAXLEN), j, name.hash);
-
-            strcpy(dnode->name.value, name_content);
-
-            lock_inode(current_level->inode);
-
-            errno =
-              current_level->inode->ops.dir_lookup(current_level->inode, dnode);
-
-            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);
-                }
-            }
-
-            unlock_inode(current_level->inode);
-
-            if (errno) {
-                unlock_dnode(current_level);
-                vfree(dnode->name.value);
-                goto error;
-            }
-
-            vfs_dcache_add(current_level, dnode);
-
-            dnode->parent = current_level;
-            llist_append(&current_level->children, &dnode->siblings);
-        }
-
-        unlock_dnode(current_level);
-
-        j = 0;
-        current_level = dnode;
-    cont:
-        current = lookahead;
-    };
-
-    *dentry = current_level;
-    return 0;
-
-error:
-    vfree(dnode->name.value);
-    vfs_d_free(dnode);
-    *dentry = NULL;
-    return errno;
-}
-
-#define VFS_MAX_SYMLINK 16
-
-int
-vfs_walk(struct v_dnode* start,
-         const char* path,
-         struct v_dnode** dentry,
-         struct hstr* component,
-         int options)
-{
-    struct v_dnode* interim;
-    const 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++;
-    }
-
-    *dentry = errno ? 0 : interim;
-
-    return errno;
-}
-
-int
-vfs_mount(const char* target, const char* fs_name, bdev_t device)
-{
-    int errno;
-    struct v_dnode* mnt;
-
-    if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
-        errno = vfs_mount_at(fs_name, device, mnt);
-    }
+    atomic_fetch_add(&dnode->ref_count, 1);
+    dnode->parent = parent;
+    llist_append(&parent->children, &dnode->siblings);

 
 

-    return errno;
+    struct hbucket* bucket = __dcache_hash(parent, &dnode->name.hash);
+    hlist_add(&bucket->head, &dnode->hash_list);

 }
 
 }
 

-int
-vfs_unmount(const char* target)
+void
+vfs_dcache_remove(struct v_dnode* dnode)

 {
 {

-    int errno;
-    struct v_dnode* mnt;
+    assert(dnode);
+    assert(dnode->ref_count == 1);

 
 

-    if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
-        errno = vfs_unmount_at(mnt);
-    }
+    llist_delete(&dnode->siblings);
+    llist_delete(&dnode->aka_list);
+    hlist_delete(&dnode->hash_list);

 
 

-    return errno;
+    dnode->parent = NULL;
+    atomic_fetch_sub(&dnode->ref_count, 1);

 }
 
 }
 

-int
-vfs_mount_at(const char* fs_name, bdev_t device, struct v_dnode* mnt_point)
+void
+vfs_dcache_rehash(struct v_dnode* new_parent, struct v_dnode* dnode)

 {
 {

-    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++;
+    assert(new_parent);

 
 

-    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);
-    }
-
-    return errno;
-}
-
-int
-vfs_unmount_at(struct v_dnode* mnt_point)
-{
-    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;
+    hstr_rehash(&dnode->name, HSTR_FULL_HASH);
+    vfs_dcache_remove(dnode);
+    vfs_dcache_add(new_parent, dnode);

 }
 
 int
 vfs_open(struct v_dnode* dnode, struct v_file** file)
 {
 }
 
 int
 vfs_open(struct v_dnode* dnode, struct v_file** file)
 {

-    if (!dnode->inode || !dnode->inode->ops.open) {
+    if (!dnode->inode || !dnode->inode->ops->open) {

         return ENOTSUP;
     }
 
     struct v_inode* inode = dnode->inode;
         return ENOTSUP;
     }
 
     struct v_inode* inode = dnode->inode;

+
+    lock_inode(inode);
+

     struct v_file* vfile = cake_grab(file_pile);
     memset(vfile, 0, sizeof(*vfile));
 
     struct v_file* vfile = cake_grab(file_pile);
     memset(vfile, 0, sizeof(*vfile));
 


@@ -361,32 +206,50 @@ vfs_open(struct v_dnode* dnode, struct v_file** file)
         inode->pg_cache = pcache;
     }
 
         inode->pg_cache = pcache;
     }
 

-    int errno = inode->ops.open(inode, vfile);
+    int errno = inode->ops->open(inode, vfile);

     if (errno) {
         cake_release(file_pile, vfile);
     } else {
         atomic_fetch_add(&dnode->ref_count, 1);
         inode->open_count++;
     if (errno) {
         cake_release(file_pile, vfile);
     } else {
         atomic_fetch_add(&dnode->ref_count, 1);
         inode->open_count++;

+        mnt_mkbusy(dnode->mnt);

 
         *file = vfile;
     }
 
 
         *file = vfile;
     }
 

+    unlock_inode(inode);
+

     return errno;
 }
 
     return errno;
 }
 

+void
+vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode)
+{
+    if (assign_to->inode) {
+        llist_delete(&assign_to->aka_list);
+        assign_to->inode->link_count--;
+    }
+    llist_append(&inode->aka_dnodes, &assign_to->aka_list);
+    assign_to->inode = inode;
+    inode->link_count++;
+}
+

 int
 vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 {
     int errno;
 
 int
 vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 {
     int errno;
 

+    if ((errno = vfs_check_writable(to_link))) {
+        return errno;
+    }
+

     lock_inode(to_link->inode);
     if (to_link->super_block->root != name->super_block->root) {
         errno = EXDEV;
     lock_inode(to_link->inode);
     if (to_link->super_block->root != name->super_block->root) {
         errno = EXDEV;

-    } else if (!to_link->inode->ops.link) {
+    } else if (!to_link->inode->ops->link) {

         errno = ENOTSUP;
         errno = ENOTSUP;

-    } else if (!(errno = to_link->inode->ops.link(to_link->inode, name))) {
-        name->inode = to_link->inode;
-        atomic_fetch_add(&to_link->inode->link_count, 1);
+    } else if (!(errno = to_link->inode->ops->link(to_link->inode, name))) {
+        vfs_assign_inode(name, to_link->inode);

     }
     unlock_inode(to_link->inode);
 
     }
     unlock_inode(to_link->inode);
 


@@ -394,28 +257,64 @@ vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 }
 
 int
 }
 
 int

-vfs_close(struct v_file* file)
+vfs_pclose(struct v_file* file, pid_t pid)

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

-    if (!file->ops.close || !(errno = file->ops.close(file))) {
+    if (file->ref_count > 1) {
+        atomic_fetch_sub(&file->ref_count, 1);
+    } else if (!(errno = file->ops->close(file))) {

         atomic_fetch_sub(&file->dnode->ref_count, 1);
         file->inode->open_count--;
 
         atomic_fetch_sub(&file->dnode->ref_count, 1);
         file->inode->open_count--;
 

+        // 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.
+         */
+        if (mutex_on_hold(&file->inode->lock)) {
+            mutex_unlock_for(&file->inode->lock, pid);
+        }
+        mnt_chillax(file->dnode->mnt);
+

         pcache_commit_all(file->inode);
         cake_release(file_pile, file);
     }
     return errno;
 }
 
         pcache_commit_all(file->inode);
         cake_release(file_pile, file);
     }
     return errno;
 }
 

+int
+vfs_close(struct v_file* file)
+{
+    return vfs_pclose(file, __current->pid);
+}
+

 int
 vfs_fsync(struct v_file* file)
 {
 int
 vfs_fsync(struct v_file* file)
 {

+    int errno;
+    if ((errno = vfs_check_writable(file->dnode))) {
+        return errno;
+    }
+

     lock_inode(file->inode);
 
     lock_inode(file->inode);
 

-    int errno = ENOTSUP;

     pcache_commit_all(file->inode);
     pcache_commit_all(file->inode);

-    if (file->ops.sync) {
-        errno = file->ops.sync(file->inode);
+
+    errno = ENOTSUP;
+    if (file->ops->sync) {
+        errno = file->ops->sync(file);

     }
 
     unlock_inode(file->inode);
     }
 
     unlock_inode(file->inode);


@@ -441,51 +340,162 @@ 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));

     return sb;
 }
 
 void
 vfs_sb_free(struct v_superblock* sb)
 {
     return sb;
 }
 
 void
 vfs_sb_free(struct v_superblock* sb)
 {

+    vfree(sb->i_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);

+    llist_init_head(&dnode->siblings);
+    llist_init_head(&dnode->aka_list);

     mutex_init(&dnode->lock);
 
     dnode->ref_count = ATOMIC_VAR_INIT(0);
     mutex_init(&dnode->lock);
 
     dnode->ref_count = ATOMIC_VAR_INIT(0);

+    dnode->name = HHSTR(vzalloc(VFS_NAME_MAXLEN), 0, 0);
+
+    hstrcpy(&dnode->name, name);
+
+    if (parent) {
+        dnode->super_block = parent->super_block;
+        dnode->mnt = parent->mnt;
+    }
+
+    lru_use_one(dnode_lru, &dnode->lru);

 
 

-    dnode->name = vfs_empty;

     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);

     }
     }

+
+    vfree(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, uint32_t i_id)

 {
 {

-    struct v_inode* inode = cake_grab(inode_pile);
+    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 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));
     mutex_init(&inode->lock);
     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->sb = sb;
+    inode->ctime = clock_unixtime();
+    inode->atime = inode->ctime;
+    inode->mtime = inode->ctime;

 
 

+done:
+    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)
 {

+    if (inode->pg_cache) {
+        pcache_release(inode->pg_cache);
+        vfree(inode->pg_cache);
+    }
+    inode->ops->sync(inode);
+    hlist_delete(&inode->hash_list);

     cake_release(inode_pile, inode);
 }
 
     cake_release(inode_pile, inode);
 }
 


@@ -493,13 +503,8 @@ vfs_i_free(struct v_inode* inode)
 
 #define FLOCATE_CREATE_EMPTY 1
 
 
 #define FLOCATE_CREATE_EMPTY 1
 

-#define DO_STATUS(errno) SYSCALL_ESTATUS(__current->k_status = errno)
-#define DO_STATUS_OR_RETURN(errno) ({ errno < 0 ? DO_STATUS(errno) : errno; })
-
-#define TEST_FD(fd) (fd >= 0 && fd < VFS_MAX_FD)
-

 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;


@@ -516,8 +521,9 @@ __vfs_try_locate_file(const char* path,
     char name_str[VFS_NAME_MAXLEN];
     struct hstr name = HSTR(name_str, 0);
     int errno;
     char name_str[VFS_NAME_MAXLEN];
     struct hstr name = HSTR(name_str, 0);
     int errno;

-    if ((errno =
-           vfs_walk(__current->cwd, path, fdir, &name, VFS_WALK_PARENT))) {
+
+    name_str[0] = 0;
+    if ((errno = vfs_walk_proc(path, fdir, &name, VFS_WALK_PARENT))) {

         return errno;
     }
 
         return errno;
     }
 


@@ -527,20 +533,23 @@ __vfs_try_locate_file(const char* path,
     }
 
     struct v_dnode* parent = *fdir;
     }
 
     struct v_dnode* parent = *fdir;

-    struct v_dnode* file_new = vfs_d_alloc();
-    file_new->name = HHSTR(valloc(VFS_NAME_MAXLEN), name.len, name.hash);
-    strcpy(file_new->name.value, name_str);
+    struct v_dnode* file_new = vfs_d_alloc(parent, &name);

 
 

-    if (!(errno = parent->inode->ops.create(parent->inode, file_new))) {
-        *file = file_new;
+    if (!file_new) {
+        return ENOMEM;
+    }
+
+    lock_dnode(parent);

 
 

+    if (!(errno = parent->inode->ops->create(parent->inode, file_new))) {

         vfs_dcache_add(parent, file_new);
         vfs_dcache_add(parent, file_new);

-        llist_append(&parent->children, &file_new->siblings);
+        *file = file_new;

     } else {
     } else {

-        vfree(file_new->name.value);

         vfs_d_free(file_new);
     }
 
         vfs_d_free(file_new);
     }
 

+    unlock_dnode(parent);
+

     return errno;
 }
 
     return errno;
 }
 


@@ -559,11 +568,6 @@ vfs_do_open(const char* path, int options)
     }
 
     struct v_inode* o_inode = ofile->inode;
     }
 
     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;
-    }

 
     if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
         struct v_fd* fd_s = vzalloc(sizeof(*fd_s));
 
     if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
         struct v_fd* fd_s = vzalloc(sizeof(*fd_s));


@@ -587,13 +591,11 @@ __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;
     }
 


@@ -621,7 +623,7 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct 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;
     }
 


@@ -629,7 +631,7 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
 
     lock_inode(inode);
 
 
     lock_inode(inode);
 

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

         errno = ENOTDIR;
     } else {
         struct dir_context dctx =
         errno = ENOTDIR;
     } else {
         struct dir_context dctx =


@@ -637,32 +639,32 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
                                 .index = dent->d_offset,
                                 .read_complete_callback =
                                   __vfs_readdir_callback };
                                 .index = dent->d_offset,
                                 .read_complete_callback =
                                   __vfs_readdir_callback };

+        errno = 1;

         if (dent->d_offset == 0) {
         if (dent->d_offset == 0) {

-            __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 0);
+            __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, DT_DIR);

         } else if (dent->d_offset == 1) {
         } else if (dent->d_offset == 1) {

-            __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0);
+            __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, DT_DIR);

         } else {
             dctx.index -= 2;
         } else {
             dctx.index -= 2;

-            if ((errno = fd_s->file->ops.readdir(inode, &dctx))) {
+            if ((errno = fd_s->file->ops->readdir(fd_s->file, &dctx)) != 1) {

                 unlock_inode(inode);
                 goto done;
             }
         }
                 unlock_inode(inode);
                 goto done;
             }
         }

-        errno = 0;

         dent->d_offset++;
     }
 
     unlock_inode(inode);
 
 done:
         dent->d_offset++;
     }
 
     unlock_inode(inode);
 
 done:

-    return DO_STATUS(errno);
+    return DO_STATUS_OR_RETURN(errno);

 }
 
 __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;
     }
 
         goto done;
     }
 


@@ -676,16 +678,20 @@ __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
 
     file->inode->atime = clock_unixtime();
 
 
     file->inode->atime = clock_unixtime();
 

-    __SYSCALL_INTERRUPTIBLE(
-      { errno = file->ops.read(file->inode, buf, count, file->f_pos); })
-
-    unlock_inode(file->inode);
+    if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
+        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);
 }


@@ -694,11 +700,16 @@ __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;
     }
 
     struct v_file* file = fd_s->file;
         goto done;
     }
 
     struct v_file* file = fd_s->file;

+
+    if ((errno = vfs_check_writable(file->dnode))) {
+        goto done;
+    }
+

     if ((file->inode->itype & VFS_IFDIR)) {
         errno = EISDIR;
         goto done;
     if ((file->inode->itype & VFS_IFDIR)) {
         errno = EISDIR;
         goto done;


@@ -708,16 +719,20 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
 
     file->inode->mtime = clock_unixtime();
 
 
     file->inode->mtime = clock_unixtime();
 

-    __SYSCALL_INTERRUPTIBLE(
-      { errno = file->ops.write(file->inode, buf, count, file->f_pos); })
-
-    unlock_inode(file->inode);
+    if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
+        errno = file->ops->write(file->inode, buf, count, file->f_pos);
+    } else {
+        errno = pcache_write(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);
 }


@@ -726,27 +741,36 @@ __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;
 

+    if (!file->ops->seek) {
+        errno = ENOTSUP;
+        goto done;
+    }
+

     lock_inode(file->inode);
 
     lock_inode(file->inode);
 

-    size_t fpos = file->f_pos;
+    int overflow = 0;
+    int fpos = file->f_pos;

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

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

             break;
         case FSEEK_END:
             break;
         case FSEEK_END:

-            fpos = (size_t)((int)file->inode->fsize + offset);
+            overflow =
+              __builtin_sadd_overflow((int)file->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->inode, fpos))) {
+    if (overflow) {
+        errno = EOVERFLOW;
+    } else if (!(errno = file->ops->seek(file->inode, fpos))) {

         file->f_pos = fpos;
     }
 
         file->f_pos = fpos;
     }
 


@@ -764,23 +788,27 @@ 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);
     len += cpy_size;
 
     size_t cpy_size = MIN(dnode->name.len, size - len);
     strncpy(buf + len, dnode->name.value, cpy_size);
     len += cpy_size;
 

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

     return len;
 }
 
     return len;
 }
 


@@ -789,10 +817,10 @@ vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
 {
     const char* link;
     struct v_inode* inode = dnode->inode;
 {
     const char* link;
     struct v_inode* inode = dnode->inode;

-    if (inode->ops.read_symlink) {
+    if (inode->ops->read_symlink) {

         lock_inode(inode);
 
         lock_inode(inode);
 

-        int errno = inode->ops.read_symlink(inode, &link);
+        int errno = inode->ops->read_symlink(inode, &link);

         strncpy(buf, link, size);
 
         unlock_inode(inode);
         strncpy(buf, link, size);
 
         unlock_inode(inode);


@@ -801,11 +829,24 @@ vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
     return 0;
 }
 
     return 0;
 }
 

+int
+vfs_get_dtype(int itype)
+{
+    switch (itype) {
+        case VFS_IFDIR:
+            return DT_DIR;
+        case VFS_IFSYMLINK:
+            return DT_SYMLINK;
+        default:
+            return DT_PIPE;
+    }
+}
+

 __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;
     }
 


@@ -824,8 +865,7 @@ __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);
     }
 


@@ -849,7 +889,7 @@ __DEFINE_LXSYSCALL4(int,
 {
     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;
     }
 


@@ -879,21 +919,22 @@ __DEFINE_LXSYSCALL1(int, rmdir, 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))) {
+    if ((errno = vfs_walk_proc(pathname, &dnode, NULL, 0))) {

         return DO_STATUS(errno);
     }
 
     lock_dnode(dnode);
 
         return DO_STATUS(errno);
     }
 
     lock_dnode(dnode);
 

-    if (dnode->parent)
-        lock_dnode(dnode->parent);
+    if ((errno = vfs_check_writable(dnode))) {
+        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->ref_count || dnode->inode->open_count) {
+    if (dnode->ref_count > 1 || dnode->inode->open_count) {

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


@@ -903,31 +944,50 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
         goto done;
     }
 
         goto done;
     }
 

-    lock_inode(dnode->inode);
+    struct v_dnode* parent = dnode->parent;
+
+    if (!parent) {
+        errno = EINVAL;
+        goto done;
+    }
+
+    lock_dnode(parent);
+    lock_inode(parent->inode);

 
     if ((dnode->inode->itype & VFS_IFDIR)) {
 
     if ((dnode->inode->itype & VFS_IFDIR)) {

-        errno = dnode->inode->ops.rmdir(dnode->inode);
-        // FIXME remove the dnode from cache & parent.
+        errno = parent->inode->ops->rmdir(parent->inode, dnode);
+        if (!errno) {
+            vfs_dcache_remove(dnode);
+        }

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

-    unlock_inode(dnode->inode);
+    unlock_inode(parent->inode);
+    unlock_dnode(parent);

 
 done:
     unlock_dnode(dnode);
 
 done:
     unlock_dnode(dnode);

-    if (dnode->parent)
-        unlock_dnode(dnode->parent);

     return DO_STATUS(errno);
 }
 
 __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
 {
     return DO_STATUS(errno);
 }
 
 __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
 {

+    int errno = 0;

     struct v_dnode *parent, *dir;
     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) {
+    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_check_writable(parent))) {
+        goto done;
+    }
+
+    if (!(dir = vfs_d_alloc(parent, &name))) {
+        errno = ENOMEM;

         goto done;
     }
 
         goto done;
     }
 


@@ -936,24 +996,20 @@ __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
 
     if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
         errno = ENOTSUP;
 
     if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
         errno = ENOTSUP;

-    } else if (!parent->inode->ops.mkdir) {
+    } else if (!parent->inode->ops->mkdir) {

         errno = ENOTSUP;
     } else if (!(parent->inode->itype & VFS_IFDIR)) {
         errno = ENOTDIR;
         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);
-        }
+    } else if (!(errno = parent->inode->ops->mkdir(parent->inode, dir))) {
+        vfs_dcache_add(parent, dir);
+        goto cleanup;

     }
 
     }
 

+    vfs_d_free(dir);
+
+cleanup:

     unlock_inode(parent->inode);
     unlock_dnode(parent);
     unlock_inode(parent->inode);
     unlock_dnode(parent);

-

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


@@ -961,24 +1017,27 @@ done:
 int
 __vfs_do_unlink(struct v_dnode* dnode)
 {
 int
 __vfs_do_unlink(struct v_dnode* dnode)
 {

+    int errno;

     struct v_inode* inode = dnode->inode;
 
     struct v_inode* inode = dnode->inode;
 

-    if (dnode->ref_count) {
+    if (dnode->ref_count > 1) {

         return EBUSY;
     }
 
         return EBUSY;
     }
 

+    if ((errno = vfs_check_writable(dnode))) {
+        return errno;
+    }
+

     lock_inode(inode);
 
     lock_inode(inode);
 

-    int errno;

     if (inode->open_count) {
         errno = EBUSY;
     } else if (!(inode->itype & VFS_IFDIR)) {
         // The underlying unlink implementation should handle
         //  symlink case
     if (inode->open_count) {
         errno = EBUSY;
     } else if (!(inode->itype & VFS_IFDIR)) {
         // The underlying unlink implementation should handle
         //  symlink case

-        errno = inode->ops.unlink(inode);
+        errno = inode->ops->unlink(inode);

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

-            inode->link_count--;
-            // FIXME remove the dnode from cache & parent
+            vfs_d_free(dnode);

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


@@ -993,11 +1052,7 @@ __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;
     }
 


@@ -1011,7 +1066,7 @@ __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;
     }
 
         goto done;
     }
 


@@ -1046,7 +1101,8 @@ __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);
     }
 


@@ -1077,7 +1133,7 @@ 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;
     }
 


@@ -1109,7 +1165,7 @@ __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;
     }
 


@@ -1132,21 +1188,22 @@ __DEFINE_LXSYSCALL2(int,
 {
     int errno;
     struct v_dnode* dnode;
 {
     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))) {

         goto done;
     }
         goto done;
     }

-    if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
-        errno = EROFS;
+
+    if (errno = vfs_check_writable(dnode)) {

         goto done;
     }
         goto done;
     }

-    if (!dnode->inode->ops.symlink) {
+
+    if (!dnode->inode->ops->set_symlink) {

         errno = ENOTSUP;
         goto done;
     }
 
     lock_inode(dnode->inode);
 
         errno = ENOTSUP;
         goto done;
     }
 
     lock_inode(dnode->inode);
 

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

 
     unlock_inode(dnode->inode);
 
 
     unlock_inode(dnode->inode);
 


@@ -1154,8 +1211,22 @@ done:
     return DO_STATUS(errno);
 }
 
     return DO_STATUS(errno);
 }
 

+void
+vfs_ref_dnode(struct v_dnode* dnode)
+{
+    atomic_fetch_add(&dnode->ref_count, 1);
+    mnt_mkbusy(dnode->mnt);
+}
+
+void
+vfs_unref_dnode(struct v_dnode* dnode)
+{
+    atomic_fetch_sub(&dnode->ref_count, 1);
+    mnt_chillax(dnode->mnt);
+}
+

 int
 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;
 


@@ -1166,12 +1237,12 @@ __vfs_do_chdir(struct v_dnode* dnode)
         goto done;
     }
 
         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;

 
     unlock_dnode(dnode);
 
 
     unlock_dnode(dnode);
 


@@ -1184,11 +1255,11 @@ __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(__current, dnode);

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


@@ -1199,11 +1270,11 @@ __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(__current, fd_s->file->dnode);

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


@@ -1221,7 +1292,7 @@ __DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size)
     size_t len = 0;
 
     if (!__current->cwd) {
     size_t len = 0;
 
     if (!__current->cwd) {

-        *buf = PATH_DELIM;
+        *buf = VFS_PATH_DELIM;

         len = 1;
     } else {
         len = vfs_get_path(__current->cwd, buf, size, 0);
         len = 1;
     } else {
         len = vfs_get_path(__current->cwd, buf, size, 0);


@@ -1240,4 +1311,99 @@ done:
     return ret_ptr;
 }
 
     return ret_ptr;
 }
 

-// TODO rename syscall
\ No newline at end of file
+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;
+    }
+
+    if ((errno = vfs_walk(
+           __current->cwd, newpath, &target_parent, &name, VFS_WALK_PARENT))) {
+        goto done;
+    }
+
+    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;
+    }
+
+    if (!target) {
+        errno = ENOMEM;
+        goto done;
+    }
+
+    errno = vfs_do_rename(cur, target);
+
+done:
+    vfree(name.value);
+    return DO_STATUS(errno);
+}
\ No newline at end of file