X-Git-Url: https://scm.lunaixsky.com/lunaix-os.git/blobdiff_plain/b58b151ff3f69422ba1d553f66455700cd3c204b..21ffd87b228fa0bd0c25a09b710c6aeac8b81281:/lunaix-os/kernel/fs/vfs.c

diff --git a/lunaix-os/kernel/fs/vfs.c b/lunaix-os/kernel/fs/vfs.c
index fcceb70..e219c52 100644
--- 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>
@@ -20,11 +54,23 @@
 #include <lunaix/spike.h>
 #include <lunaix/syscall.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)
+
+#define unlock_inode(inode) mutex_unlock(&inode->lock)
+#define lock_inode(inode)                                                      \
+    ({                                                                         \
+        mutex_lock(&inode->lock);                                              \
+        lru_use_one(inode_lru, &inode->lru);                                   \
+    })
+
+#define unlock_dnode(dnode) mutex_unlock(&dnode->lock)
+#define lock_dnode(dnode)                                                      \
+    ({                                                                         \
+        mutex_lock(&dnode->lock);                                              \
+        lru_use_one(dnode_lru, &dnode->lru);                                   \
+    })
 
 static struct cake_pile* dnode_pile;
 static struct cake_pile* inode_pile;
@@ -32,27 +78,27 @@ static struct cake_pile* file_pile;
 static struct cake_pile* superblock_pile;
 static struct cake_pile* fd_pile;
 
-static struct v_superblock* root_sb;
+static struct v_dnode* sysroot;
 static struct hbucket* dnode_cache;
 
-static int fs_id = 0;
+static 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 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);
 
+static int
+__vfs_try_evict_dnode(struct lru_node* obj);
+
+static int
+__vfs_try_evict_inode(struct lru_node* obj);
+
 void
 vfs_init()
 {
@@ -64,24 +110,29 @@ vfs_init()
     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);
 
-    // 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();
+    // 创建一个根dnode。
+    sysroot = vfs_d_alloc(NULL, &vfs_empty);
+    atomic_fetch_add(&sysroot->ref_count, 1);
 }
 
 inline struct hbucket*
-__dcache_get_bucket(struct v_dnode* parent, unsigned int hash)
+__dcache_hash(struct v_dnode* parent, uint32_t* hash)
 {
+    uint32_t _hash = *hash;
     // 与parent的指针值做加法，来减小碰撞的可能性。
-    hash += (uint32_t)parent;
+    _hash += (uint32_t)parent;
     // 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);
+    *hash = _hash;
+    return &dnode_cache[_hash & VFS_HASH_MASK];
 }
 
 struct v_dnode*
@@ -94,12 +145,13 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
         return parent->parent ? parent->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)
     {
-        if (pos->name.hash == str->hash) {
+        if (pos->name.hash == hash) {
             return pos;
         }
     }
@@ -109,10 +161,34 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
 void
 vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
 {
-    struct hbucket* bucket = __dcache_get_bucket(parent, dnode->name.hash);
+    atomic_fetch_add(&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);
 }
 
+void
+vfs_dcache_remove(struct v_dnode* dnode)
+{
+    assert(dnode->ref_count == 1);
+
+    llist_delete(&dnode->siblings);
+    hlist_delete(&dnode->hash_list);
+
+    dnode->parent = NULL;
+    atomic_fetch_sub(&dnode->ref_count, 1);
+}
+
+void
+vfs_dcache_rehash(struct v_dnode* new_parent, struct v_dnode* dnode)
+{
+    hstr_rehash(&dnode->name, HSTR_FULL_HASH);
+    vfs_dcache_remove(dnode);
+    vfs_dcache_add(new_parent, dnode);
+}
+
 int
 __vfs_walk(struct v_dnode* start,
            const char* path,
@@ -127,7 +203,7 @@ __vfs_walk(struct v_dnode* start,
         if ((walk_options & VFS_WALK_FSRELATIVE) && start) {
             start = start->super_block->root;
         } else {
-            start = root_sb->root;
+            start = sysroot;
         }
         i++;
     }
@@ -139,14 +215,14 @@ __vfs_walk(struct v_dnode* start,
     struct hstr name = HSTR(name_content, 0);
 
     char current = path[i++], lookahead;
-    while (current) {
+    while (current && current_level) {
         lookahead = path[i++];
         if (current != PATH_DELIM) {
             if (j >= VFS_NAME_MAXLEN - 1) {
                 return ENAMETOOLONG;
             }
             if (!VFS_VALID_CHAR(current)) {
-                return VFS_EINVLD;
+                return EINVAL;
             }
             name_content[j++] = current;
             if (lookahead) {
@@ -159,6 +235,8 @@ __vfs_walk(struct v_dnode* start,
             goto cont;
         }
 
+        lock_dnode(current_level);
+
         name_content[j] = 0;
         name.len = j;
         hstr_rehash(&name, HSTR_FULL_HASH);
@@ -169,16 +247,21 @@ __vfs_walk(struct v_dnode* start,
                 component->len = j;
                 strcpy(component->value, name_content);
             }
+            unlock_dnode(current_level);
             break;
         }
 
         dnode = vfs_dcache_lookup(current_level, &name);
 
         if (!dnode) {
-            dnode = vfs_d_alloc();
-            dnode->name = HHSTR(valloc(VFS_NAME_MAXLEN), j, name.hash);
+            dnode = vfs_d_alloc(current_level, &name);
 
-            strcpy(dnode->name.value, name_content);
+            if (!dnode) {
+                errno = ENOMEM;
+                goto error;
+            }
+
+            lock_inode(current_level->inode);
 
             errno =
               current_level->inode->ops.dir_lookup(current_level->inode, dnode);
@@ -192,16 +275,17 @@ __vfs_walk(struct v_dnode* start,
                 }
             }
 
-            if (errno) {
-                goto error;
-            }
-
             vfs_dcache_add(current_level, dnode);
+            unlock_inode(current_level->inode);
 
-            dnode->parent = current_level;
-            llist_append(&current_level->children, &dnode->siblings);
+            if (errno) {
+                unlock_dnode(current_level);
+                goto cleanup;
+            }
         }
 
+        unlock_dnode(current_level);
+
         j = 0;
         current_level = dnode;
     cont:
@@ -211,9 +295,9 @@ __vfs_walk(struct v_dnode* start,
     *dentry = current_level;
     return 0;
 
-error:
-    vfree(dnode->name.value);
+cleanup:
     vfs_d_free(dnode);
+error:
     *dentry = NULL;
     return errno;
 }
@@ -228,27 +312,29 @@ vfs_walk(struct v_dnode* start,
          int options)
 {
     struct v_dnode* interim;
-    char* pathname = path;
+    const char* pathname = path;
     int errno = __vfs_walk(start, path, &interim, component, options);
     int counter = 0;
 
-    while (!errno) {
+    while (!errno && interim->inode && (options & VFS_WALK_NOFOLLOW)) {
         if (counter >= VFS_MAX_SYMLINK) {
             errno = ELOOP;
             continue;
         }
-        if ((interim->inode->itype & VFS_INODE_TYPE_SYMLINK) &&
-            !(options & VFS_WALK_NOFOLLOW) &&
+        if ((interim->inode->itype & VFS_IFSYMLINK) &&
             interim->inode->ops.read_symlink) {
+
+            lock_inode(interim->inode);
             errno = interim->inode->ops.read_symlink(interim->inode, &pathname);
+            unlock_inode(interim->inode);
+
             if (errno) {
                 break;
             }
         } else {
             break;
         }
-        errno =
-          __vfs_walk_internal(start, pathname, &interim, component, options);
+        errno = __vfs_walk(start, pathname, &interim, component, options);
         counter++;
     }
 
@@ -257,70 +343,6 @@ vfs_walk(struct v_dnode* start,
     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(NULL, target, &mnt, NULL, 0))) {
-        errno = vfs_mount_at(fs_name, device, mnt);
-    }
-
-    return errno;
-}
-
-int
-vfs_unmount(const char* target)
-{
-    int errno;
-    struct v_dnode* mnt;
-
-    if (!(errno = vfs_walk(NULL, target, &mnt, NULL, 0))) {
-        errno = vfs_unmount_at(mnt);
-    }
-
-    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 VFS_ENOFS;
-    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);
-    }
-
-    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 VFS_EBADMNT;
-    }
-    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)
 {
@@ -328,34 +350,65 @@ vfs_open(struct v_dnode* dnode, struct v_file** file)
         return ENOTSUP;
     }
 
+    struct v_inode* inode = dnode->inode;
+
+    lock_inode(inode);
+
     struct v_file* vfile = cake_grab(file_pile);
     memset(vfile, 0, sizeof(*vfile));
 
     vfile->dnode = dnode;
-    vfile->inode = dnode->inode;
-    dnode->inode->open_count++;
+    vfile->inode = inode;
+    vfile->ref_count = ATOMIC_VAR_INIT(1);
+    vfile->ops = inode->default_fops;
+
+    if ((inode->itype & VFS_IFFILE) && !inode->pg_cache) {
+        struct pcache* pcache = vzalloc(sizeof(struct pcache));
+        pcache_init(pcache);
+        pcache->master = inode;
+        inode->pg_cache = pcache;
+    }
 
-    int errno = dnode->inode->ops.open(dnode->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++;
+        mnt_mkbusy(dnode->mnt);
+
         *file = vfile;
     }
+
+    unlock_inode(inode);
+
     return errno;
 }
 
+void
+vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode)
+{
+    if (assign_to->inode) {
+        assign_to->inode->link_count--;
+    }
+    assign_to->inode = inode;
+    inode->link_count++;
+}
+
 int
 vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 {
     int errno;
+
+    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))) {
-        name->inode = to_link->inode;
-        to_link->inode->link_count++;
+        vfs_assign_inode(name, to_link->inode);
     }
+    unlock_inode(to_link->inode);
 
     return errno;
 }
@@ -363,15 +416,13 @@ vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 int
 vfs_close(struct v_file* file)
 {
-    if (!file->ops.close) {
-        return ENOTSUP;
-    }
+    int errno = 0;
+    if (!file->ops.close || !(errno = file->ops.close(file))) {
+        atomic_fetch_sub(&file->dnode->ref_count, 1);
+        file->inode->open_count--;
+        mnt_chillax(file->dnode->mnt);
 
-    int errno = file->ops.close(file);
-    if (!errno) {
-        if (file->inode->open_count) {
-            file->inode->open_count--;
-        }
+        pcache_commit_all(file->inode);
         cake_release(file_pile, file);
     }
     return errno;
@@ -380,13 +431,16 @@ vfs_close(struct v_file* file)
 int
 vfs_fsync(struct v_file* file)
 {
+    lock_inode(file->inode);
+
     int errno = ENOTSUP;
+    pcache_commit_all(file->inode);
     if (file->ops.sync) {
-        errno = file->ops.sync(file);
-    }
-    if (!errno && file->inode->ops.sync) {
-        return file->inode->ops.sync(file->inode);
+        errno = file->ops.sync(file->inode);
     }
+
+    unlock_inode(file->inode);
+
     return errno;
 }
 
@@ -408,46 +462,144 @@ vfs_sb_alloc()
     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)
 {
+    vfree(sb->i_cache);
     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*
-vfs_d_alloc()
+vfs_d_alloc(struct v_dnode* parent, struct hstr* name)
 {
     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);
-    dnode->name = vfs_empty;
+    llist_init_head(&dnode->siblings);
+    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;
+    }
+
+    lru_use_one(dnode_lru, &dnode->lru);
+
     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*
-vfs_i_alloc()
+vfs_i_alloc(struct v_superblock* sb,
+            uint32_t inode_id,
+            void (*init)(struct v_inode* inode, void* data),
+            void* data)
 {
-    struct v_inode* inode = cake_grab(inode_pile);
-    memset(inode, 0, sizeof(*inode));
+    // 每个超级块儿维护一个inode缓存哈希表。
+    // 他们的hash value自然就是inode id了。
+    struct hbucket* slot = &sb->i_cache[inode_id & VFS_HASH_MASK];
+    struct v_inode *pos, *n;
+    hashtable_bucket_foreach(slot, pos, n, hash_list)
+    {
+        if (pos->id == inode_id) {
+            goto done;
+        }
+    }
+
+    if (!(pos = cake_grab(inode_pile))) {
+        lru_evict_half(inode_lru);
+        if (!(pos = cake_grab(inode_pile))) {
+            return NULL;
+        }
+    }
+
+    memset(pos, 0, sizeof(*pos));
+
+    pos->id = inode_id;
 
-    return inode;
+    mutex_init(&pos->lock);
+
+    init(pos, data);
+
+    hlist_add(&slot->head, &pos->hash_list);
+
+done:
+    lru_use_one(inode_lru, &pos->lru);
+    return pos;
 }
 
 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);
 }
 
@@ -456,6 +608,18 @@ vfs_i_free(struct v_inode* inode)
 #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
+__vfs_getfd(int fd, struct v_fd** fd_s)
+{
+    if (TEST_FD(fd) && (*fd_s = __current->fdtable->fds[fd])) {
+        return 0;
+    }
+    return EBADF;
+}
 
 int
 __vfs_try_locate_file(const char* path,
@@ -466,71 +630,94 @@ __vfs_try_locate_file(const char* path,
     char name_str[VFS_NAME_MAXLEN];
     struct hstr name = HSTR(name_str, 0);
     int errno;
-    if (!(errno = vfs_walk(NULL, path, fdir, &name, VFS_WALK_PARENT))) {
-        errno = vfs_walk(*fdir, name.value, file, NULL, 0);
-        if (errno == ENOENT && (options & FLOCATE_CREATE_EMPTY)) {
-            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;
-
-            llist_append(&(*fdir)->children, &file_new->siblings);
-        }
+    if ((errno =
+           vfs_walk(__current->cwd, path, fdir, &name, VFS_WALK_PARENT))) {
+        return errno;
     }
 
+    errno = vfs_walk(*fdir, name.value, file, NULL, 0);
+    if (errno != ENOENT || !(options & FLOCATE_CREATE_EMPTY)) {
+        return errno;
+    }
+
+    struct v_dnode* parent = *fdir;
+    struct v_dnode* file_new = vfs_d_alloc(parent, &name);
+
+    if (!file_new) {
+        return ENOMEM;
+    }
+
+    lock_dnode(parent);
+
+    if (!(errno = parent->inode->ops.create(parent->inode, file_new))) {
+        vfs_dcache_add(parent, file_new);
+        *file = file_new;
+    } else {
+        vfs_d_free(file_new);
+    }
+
+    unlock_dnode(parent);
+
     return errno;
 }
 
 int
-__vfs_do_open(struct v_file** file_out, const char* path, int options)
+vfs_do_open(const char* path, int options)
 {
-    int errno;
+    int errno, fd;
     struct v_dnode *dentry, *file;
-    struct v_file* opened_file = 0;
+    struct v_file* ofile = 0;
 
-    errno = __vfs_try_locate_file(path, &dentry, &file, 0);
+    errno = __vfs_try_locate_file(
+      path, &dentry, &file, (options & FO_CREATE) ? FLOCATE_CREATE_EMPTY : 0);
 
-    if (errno != ENOENT && (options & FO_CREATE)) {
-        errno = dentry->inode->ops.create(dentry->inode, opened_file);
-    } else if (!errno) {
-        errno = vfs_open(file, &opened_file);
+    if (errno || (errno = vfs_open(file, &ofile))) {
+        return errno;
     }
 
-    *file_out = opened_file;
-    return errno;
-}
-
-__DEFINE_LXSYSCALL2(int, open, const char*, path, int, options)
-{
-    struct v_file* opened_file;
-    int errno = __vfs_do_open(&opened_file, path, options), fd;
+    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));
-        fd_s->file = opened_file;
-        fd_s->pos = opened_file->inode->fsize & -((options & FO_APPEND) != 0);
+        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;
     }
 
-    return DO_STATUS(errno);
+    return errno;
+}
+
+__DEFINE_LXSYSCALL2(int, open, const char*, path, int, options)
+{
+    int errno = vfs_do_open(path, options);
+    return DO_STATUS_OR_RETURN(errno);
 }
-#define TEST_FD(fd) (fd >= 0 && fd < VFS_MAX_FD)
-#define GET_FD(fd, fd_s) (TEST_FD(fd) && (fd_s = __current->fdtable->fds[fd]))
 
 __DEFINE_LXSYSCALL1(int, close, int, fd)
 {
     struct v_fd* fd_s;
-    int errno;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else if (!(errno = vfs_close(fd_s->file))) {
-        vfree(fd_s);
-        __current->fdtable->fds[fd] = 0;
+    int errno = 0;
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
+        goto done_err;
     }
 
+    if (fd_s->file->ref_count > 1) {
+        fd_s->file->ref_count--;
+    } else if ((errno = vfs_close(fd_s->file))) {
+        goto done_err;
+    }
+
+    vfree(fd_s);
+    __current->fdtable->fds[fd] = 0;
+
+done_err:
     return DO_STATUS(errno);
 }
 
@@ -550,9 +737,16 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
 {
     struct v_fd* fd_s;
     int errno;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else if (!(fd_s->file->inode->itype & VFS_INODE_TYPE_DIR)) {
+
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
+        goto done;
+    }
+
+    struct v_inode* inode = fd_s->file->inode;
+
+    lock_inode(inode);
+
+    if (!(fd_s->file->inode->itype & VFS_IFDIR)) {
         errno = ENOTDIR;
     } else {
         struct dir_context dctx =
@@ -566,7 +760,8 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
             __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))) {
+            if ((errno = fd_s->file->ops.readdir(inode, &dctx))) {
+                unlock_inode(inode);
                 goto done;
             }
         }
@@ -574,54 +769,42 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
         dent->d_offset++;
     }
 
+    unlock_inode(inode);
+
 done:
     return DO_STATUS(errno);
 }
 
-__DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
+__DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
 {
-    struct v_dnode *parent, *dir;
-    struct hstr component = HSTR(valloc(VFS_NAME_MAXLEN), 0);
-    int errno = vfs_walk(NULL, path, &parent, &component, VFS_WALK_PARENT);
-    if (errno) {
+    int errno = 0;
+    struct v_fd* fd_s;
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
         goto done;
     }
 
-    if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
-        errno = ENOTSUP;
-    } else if (!parent->inode->ops.mkdir) {
-        errno = ENOTSUP;
-    } else if (!(parent->inode->itype & VFS_INODE_TYPE_DIR)) {
-        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);
-        }
+    struct v_file* file = fd_s->file;
+    if ((file->inode->itype & VFS_IFDIR)) {
+        errno = EISDIR;
+        goto done;
     }
 
-done:
-    return DO_STATUS(errno);
-}
+    lock_inode(file->inode);
 
-__DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
-{
-    int errno = 0;
-    struct v_fd* fd_s;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else {
-        struct v_file* file = fd_s->file;
-        file->f_pos = fd_s->pos;
-        if ((errno = file->ops.read(file, buf, count)) >= 0) {
-            fd_s->pos += errno;
-        }
+    file->inode->atime = clock_unixtime();
+
+    __SYSCALL_INTERRUPTIBLE(
+      { errno = file->ops.read(file->inode, buf, count, file->f_pos); })
+
+    if (errno > 0) {
+        file->f_pos += errno;
+        unlock_inode(file->inode);
+        return errno;
     }
 
+    unlock_inode(file->inode);
+
+done:
     return DO_STATUS(errno);
 }
 
@@ -629,16 +812,32 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
 {
     int errno = 0;
     struct v_fd* fd_s;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else {
-        struct v_file* file = fd_s->file;
-        file->f_pos = fd_s->pos;
-        if ((errno = file->ops.write(file, buf, count)) >= 0) {
-            fd_s->pos += errno;
-        }
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
+        goto done;
+    }
+
+    struct v_file* file = fd_s->file;
+    if ((file->inode->itype & VFS_IFDIR)) {
+        errno = EISDIR;
+        goto done;
+    }
+
+    lock_inode(file->inode);
+
+    file->inode->mtime = clock_unixtime();
+
+    __SYSCALL_INTERRUPTIBLE(
+      { errno = file->ops.write(file->inode, buf, count, file->f_pos); })
+
+    if (errno > 0) {
+        file->f_pos += errno;
+        unlock_inode(file->inode);
+        return errno;
     }
 
+    unlock_inode(file->inode);
+
+done:
     return DO_STATUS(errno);
 }
 
@@ -646,27 +845,33 @@ __DEFINE_LXSYSCALL3(int, lseek, int, fd, int, offset, int, options)
 {
     int errno = 0;
     struct v_fd* fd_s;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else {
-        size_t fpos = fd_s->file->f_pos;
-        switch (options) {
-            case FSEEK_CUR:
-                fpos = (size_t)((int)fd_s->file->f_pos + offset);
-                break;
-            case FSEEK_END:
-                fpos = (size_t)((int)fd_s->file->inode->fsize + offset);
-                break;
-            case FSEEK_SET:
-                fpos = offset;
-                break;
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
+        goto done;
+    }
 
-            default:
-                break;
-        }
-        fd_s->pos = fpos;
+    struct v_file* file = fd_s->file;
+
+    lock_inode(file->inode);
+
+    size_t fpos = file->f_pos;
+    switch (options) {
+        case FSEEK_CUR:
+            fpos = (size_t)((int)file->f_pos + offset);
+            break;
+        case FSEEK_END:
+            fpos = (size_t)((int)file->inode->fsize + offset);
+            break;
+        case FSEEK_SET:
+            fpos = offset;
+            break;
+    }
+    if (!file->ops.seek || !(errno = file->ops.seek(file->inode, fpos))) {
+        file->f_pos = fpos;
     }
 
+    unlock_inode(file->inode);
+
+done:
     return DO_STATUS(errno);
 }
 
@@ -701,10 +906,15 @@ 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)
 {
-    char* link;
-    if (dnode->inode->ops.read_symlink) {
-        int errno = dnode->inode->ops.read_symlink(dnode->inode, &link);
+    const char* link;
+    struct v_inode* inode = dnode->inode;
+    if (inode->ops.read_symlink) {
+        lock_inode(inode);
+
+        int errno = inode->ops.read_symlink(inode, &link);
         strncpy(buf, link, size);
+
+        unlock_inode(inode);
         return errno;
     }
     return 0;
@@ -714,17 +924,18 @@ __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
 {
     int errno;
     struct v_fd* fd_s;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else {
-        struct v_dnode* dnode;
-        errno = vfs_get_path(fd_s->file->dnode, buf, size, 0);
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
+        goto done;
     }
 
+    struct v_dnode* dnode;
+    errno = vfs_get_path(fd_s->file->dnode, buf, size, 0);
+
     if (errno >= 0) {
         return errno;
     }
 
+done:
     return DO_STATUS(errno);
 }
 
@@ -732,7 +943,8 @@ __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size)
 {
     int errno;
     struct v_dnode* dnode;
-    if (!(errno = vfs_walk(NULL, path, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
+    if (!(errno =
+            vfs_walk(__current->cwd, path, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
         errno = vfs_readlink(dnode, buf, size);
     }
 
@@ -756,69 +968,153 @@ __DEFINE_LXSYSCALL4(int,
 {
     int errno;
     struct v_fd* fd_s;
-    if (!GET_FD(dirfd, fd_s)) {
-        errno = EBADF;
-    } else {
-        struct v_dnode* dnode;
-        if (!(errno = vfs_walk(fd_s->file->dnode,
-                               pathname,
-                               &dnode,
-                               NULL,
-                               VFS_WALK_NOFOLLOW))) {
-            errno = vfs_readlink(fd_s->file->dnode, buf, size);
-        }
+    if ((errno = __vfs_getfd(dirfd, &fd_s))) {
+        goto done;
+    }
+
+    struct v_dnode* dnode;
+    if (!(errno = vfs_walk(
+            fd_s->file->dnode, pathname, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
+        errno = vfs_readlink(fd_s->file->dnode, buf, size);
     }
 
     if (errno >= 0) {
         return errno;
     }
 
+done:
     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;
-    if ((errno = vfs_walk(NULL, pathname, &dnode, NULL, 0))) {
-        goto done;
+    if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
+        return DO_STATUS(errno);
     }
+
+    lock_dnode(dnode);
+
     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;
     }
 
-    if ((dnode->inode->itype & VFS_INODE_TYPE_DIR)) {
-        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 ((dnode->inode->itype & VFS_IFDIR)) {
+        errno = parent->inode->ops.rmdir(parent->inode, dnode);
+        if (!errno) {
+            vfs_dcache_remove(dnode);
+        }
     } 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 (!dir) {
+        errno = ENOMEM;
+        goto done;
+    }
+
+    if ((errno =
+           vfs_walk(__current->cwd, path, &parent, &name, VFS_WALK_PARENT))) {
+        goto done;
+    }
+
+    dir = vfs_d_alloc(parent, &name);
+
+    lock_dnode(parent);
+    lock_inode(parent->inode);
+
+    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 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);
 done:
     return DO_STATUS(errno);
 }
 
 int
-__vfs_do_unlink(struct v_inode* inode)
+__vfs_do_unlink(struct v_dnode* dnode)
 {
+    struct v_inode* inode = dnode->inode;
+
+    if (dnode->ref_count > 1) {
+        return EBUSY;
+    }
+
+    lock_inode(inode);
+
     int errno;
     if (inode->open_count) {
         errno = EBUSY;
-    } else if (!(inode->itype & VFS_INODE_TYPE_DIR)) {
-        // TODO handle symbolic link and type other than regular file
+    } else if (!(inode->itype & VFS_IFDIR)) {
+        // The underlying unlink implementation should handle
+        //  symlink case
         errno = inode->ops.unlink(inode);
         if (!errno) {
-            inode->link_count--;
+            vfs_d_free(dnode);
         }
     } else {
         errno = EISDIR;
     }
 
+    unlock_inode(inode);
+
     return errno;
 }
 
@@ -826,7 +1122,7 @@ __DEFINE_LXSYSCALL1(int, unlink, const char*, pathname)
 {
     int errno;
     struct v_dnode* dnode;
-    if ((errno = vfs_walk(NULL, pathname, &dnode, NULL, 0))) {
+    if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
         goto done;
     }
     if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
@@ -834,7 +1130,7 @@ __DEFINE_LXSYSCALL1(int, unlink, const char*, pathname)
         goto done;
     }
 
-    errno = __vfs_do_unlink(dnode->inode);
+    errno = __vfs_do_unlink(dnode);
 
 done:
     return DO_STATUS(errno);
@@ -844,13 +1140,13 @@ __DEFINE_LXSYSCALL2(int, unlinkat, int, fd, const char*, pathname)
 {
     int errno;
     struct v_fd* fd_s;
-    if (!GET_FD(fd, fd_s)) {
-        errno = EBADF;
-    } else {
-        struct v_dnode* dnode;
-        if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {
-            errno = __vfs_do_unlink(dnode->inode);
-        }
+    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))) {
+        errno = __vfs_do_unlink(dnode);
     }
 
 done:
@@ -879,9 +1175,7 @@ __DEFINE_LXSYSCALL1(int, fsync, int, fildes)
 {
     int errno;
     struct v_fd* fd_s;
-    if (!GET_FD(fildes, fd_s)) {
-        errno = EBADF;
-    } else {
+    if (!(errno = __vfs_getfd(fildes, &fd_s))) {
         errno = vfs_fsync(fd_s->file);
     }
 
@@ -889,21 +1183,22 @@ __DEFINE_LXSYSCALL1(int, fsync, int, fildes)
 }
 
 int
-__vfs_dup_fd(struct v_fd* old, struct v_fd** new)
+vfs_dup_fd(struct v_fd* old, struct v_fd** new)
 {
     int errno = 0;
-    struct v_file* newopened;
-    if (!(errno = vfs_open(old->file->dnode, &newopened))) {
-        *new = cake_grab(fd_pile);
-        **new = (struct v_fd){ .file = newopened,
-                               .pos = old->pos,
-                               .flags = old->flags };
-    }
+    struct v_fd* copied = cake_grab(fd_pile);
+
+    memcpy(copied, old, sizeof(struct v_fd));
+
+    atomic_fetch_add(&old->file->ref_count, 1);
+
+    *new = copied;
 
     return errno;
 }
 
-__DEFINE_LXSYSCALL2(int, dup2, int, oldfd, int, newfd)
+int
+vfs_dup2(int oldfd, int newfd)
 {
     if (newfd == oldfd) {
         return newfd;
@@ -911,16 +1206,21 @@ __DEFINE_LXSYSCALL2(int, dup2, int, oldfd, int, newfd)
 
     int errno;
     struct v_fd *oldfd_s, *newfd_s;
-    if (!GET_FD(oldfd, oldfd_s) || TEST_FD(newfd)) {
+    if ((errno = __vfs_getfd(oldfd, &oldfd_s))) {
+        goto done;
+    }
+
+    if (!TEST_FD(newfd)) {
         errno = EBADF;
         goto done;
     }
+
     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;
     }
 
-    if (!(errno = __vfs_dup_fd(oldfd_s, &newfd_s))) {
+    if (!(errno = vfs_dup_fd(oldfd_s, &newfd_s))) {
         __current->fdtable->fds[newfd] = newfd_s;
         return newfd;
     }
@@ -929,18 +1229,26 @@ done:
     return DO_STATUS(errno);
 }
 
+__DEFINE_LXSYSCALL2(int, dup2, int, oldfd, int, newfd)
+{
+    return vfs_dup2(oldfd, newfd);
+}
+
 __DEFINE_LXSYSCALL1(int, dup, int, oldfd)
 {
     int errno, newfd;
     struct v_fd *oldfd_s, *newfd_s;
-    if (!GET_FD(oldfd, oldfd_s)) {
-        errno = EBADF;
-    } else if (!(errno = vfs_alloc_fdslot(&newfd)) &&
-               !(errno = __vfs_dup_fd(oldfd_s, &newfd_s))) {
+    if ((errno = __vfs_getfd(oldfd, &oldfd_s))) {
+        goto done;
+    }
+
+    if (!(errno = vfs_alloc_fdslot(&newfd)) &&
+        !(errno = vfs_dup_fd(oldfd_s, &newfd_s))) {
         __current->fdtable->fds[newfd] = newfd_s;
         return newfd;
     }
 
+done:
     return DO_STATUS(errno);
 }
 
@@ -953,20 +1261,250 @@ __DEFINE_LXSYSCALL2(int,
 {
     int errno;
     struct v_dnode* dnode;
-    if ((errno = vfs_walk(NULL, pathname, &dnode, NULL, 0))) {
+    if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
         goto done;
     }
     if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
         errno = EROFS;
         goto done;
     }
-    if (!dnode->inode->ops.symlink) {
+    if (!dnode->inode->ops.set_symlink) {
         errno = ENOTSUP;
         goto done;
     }
 
-    errno = dnode->inode->ops.symlink(dnode->inode, link_target);
+    lock_inode(dnode->inode);
+
+    errno = dnode->inode->ops.set_symlink(dnode->inode, link_target);
+
+    unlock_inode(dnode->inode);
 
 done:
     return DO_STATUS(errno);
+}
+
+int
+__vfs_do_chdir(struct v_dnode* dnode)
+{
+    int errno = 0;
+
+    lock_dnode(dnode);
+
+    if (!(dnode->inode->itype & VFS_IFDIR)) {
+        errno = ENOTDIR;
+        goto done;
+    }
+
+    if (__current->cwd) {
+        atomic_fetch_sub(&__current->cwd->ref_count, 1);
+        mnt_chillax(__current->cwd->mnt);
+    }
+
+    atomic_fetch_add(&dnode->ref_count, 1);
+    mnt_mkbusy(dnode->mnt);
+    __current->cwd = dnode;
+
+    unlock_dnode(dnode);
+
+done:
+    return errno;
+}
+
+__DEFINE_LXSYSCALL1(int, chdir, const char*, path)
+{
+    struct v_dnode* dnode;
+    int errno = 0;
+
+    if ((errno = vfs_walk(__current->cwd, path, &dnode, NULL, 0))) {
+        goto done;
+    }
+
+    errno = __vfs_do_chdir(dnode);
+
+done:
+    return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL1(int, fchdir, int, fd)
+{
+    struct v_fd* fd_s;
+    int errno = 0;
+
+    if ((errno = __vfs_getfd(fd, &fd_s))) {
+        goto done;
+    }
+
+    errno = __vfs_do_chdir(fd_s->file->dnode);
+
+done:
+    return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size)
+{
+    int errno = 0;
+    char* ret_ptr = 0;
+    if (size < 2) {
+        errno = ERANGE;
+        goto done;
+    }
+
+    size_t len = 0;
+
+    if (!__current->cwd) {
+        *buf = PATH_DELIM;
+        len = 1;
+    } else {
+        len = vfs_get_path(__current->cwd, buf, size, 0);
+        if (len == size) {
+            errno = ERANGE;
+            goto done;
+        }
+    }
+
+    buf[len + 1] = '\0';
+
+    ret_ptr = buf;
+
+done:
+    __current->k_status = errno;
+    return ret_ptr;
+}
+
+int
+vfs_do_rename(struct v_dnode* current, struct v_dnode* target)
+{
+    if (current->inode->id == target->inode->id) {
+        // hard link
+        return 0;
+    }
+
+    if (current->ref_count > 1 || target->ref_count > 1) {
+        return EBUSY;
+    }
+
+    if (current->super_block != target->super_block) {
+        return EXDEV;
+    }
+
+    int errno = 0;
+
+    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(__current->cwd, 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);
+}
+
+__DEFINE_LXSYSCALL3(int,
+                    mount,
+                    const char*,
+                    source,
+                    const char*,
+                    target,
+                    const char*,
+                    fstype)
+{
+    struct v_dnode *dev, *mnt;
+    int errno = 0;
+
+    if ((errno = vfs_walk(__current->cwd, source, &dev, NULL, 0))) {
+        goto done;
+    }
+
+    if ((errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
+        goto done;
+    }
+
+    if (mnt->ref_count > 1) {
+        errno = EBUSY;
+        goto done;
+    }
+
+    // By our convention.
+    // XXX could we do better?
+    struct device* device = (struct device*)dev->data;
+
+    if (!(dev->inode->itype & VFS_IFVOLDEV) || !device) {
+        errno = ENOTDEV;
+        goto done;
+    }
+
+    errno = vfs_mount_at(fstype, device, mnt);
+
+done:
+    return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL1(int, unmount, const char*, target)
+{
+    return vfs_unmount(target);
 }
\ No newline at end of file