Fix: stale dnode caching instance cause locked-up upon accessing (#52)

[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 eb4239639e84082ff89f13effb20da6a4e00e054..1696d6e961ace4b3ac092809db67cd9dddd42183 100644 (file)
--- a/lunaix-os/kernel/fs/vfs.c
+++ b/lunaix-os/kernel/fs/vfs.c
@@ -64,7 +64,6 @@ static struct cake_pile* superblock_pile;
 static struct cake_pile* fd_pile;
 
 struct v_dnode* vfs_sysroot;
 static struct cake_pile* fd_pile;
 
 struct v_dnode* vfs_sysroot;

-static struct hbucket* dnode_cache;

 
 struct lru_zone *dnode_lru, *inode_lru;
 
 
 struct lru_zone *dnode_lru, *inode_lru;
 


@@ -72,12 +71,6 @@ struct hstr vfs_ddot = HSTR("..", 2);
 struct hstr vfs_dot = HSTR(".", 1);
 struct hstr vfs_empty = HSTR("", 0);
 
 struct hstr vfs_dot = HSTR(".", 1);
 struct hstr vfs_empty = HSTR("", 0);
 

-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_dnode(struct lru_node* obj);
 


@@ -95,10 +88,8 @@ 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(VFS_HASHTABLE_SIZE * sizeof(struct hbucket));
-
-    dnode_lru = lru_new_zone(__vfs_try_evict_dnode);
-    inode_lru = lru_new_zone(__vfs_try_evict_inode);
+    dnode_lru = lru_new_zone("vfs_dnode", __vfs_try_evict_dnode);
+    inode_lru = lru_new_zone("vfs_inode", __vfs_try_evict_inode);

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


@@ -106,19 +97,36 @@ vfs_init()
     // 创建一个根dnode。
     vfs_sysroot = vfs_d_alloc(NULL, &vfs_empty);
     vfs_sysroot->parent = vfs_sysroot;
     // 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);
+    
+    vfs_ref_dnode(vfs_sysroot);

 }
 
 static inline struct hbucket*
 __dcache_hash(struct v_dnode* parent, u32_t* hash)
 {
 }
 
 static inline struct hbucket*
 __dcache_hash(struct v_dnode* parent, u32_t* hash)
 {

-    u32_t _hash = *hash;
-    // 确保低位更加随机
+    struct hbucket* d_cache;
+    u32_t _hash;
+    
+    d_cache = parent->super_block->d_cache;
+    _hash = *hash;

     _hash = _hash ^ (_hash >> VFS_HASHBITS);
     _hash = _hash ^ (_hash >> VFS_HASHBITS);

-    // 与parent的指针值做加法，来减小碰撞的可能性。
-    _hash += (u32_t)parent;
+    _hash += (u32_t)__ptr(parent);
+

     *hash = _hash;
     *hash = _hash;

-    return &dnode_cache[_hash & VFS_HASH_MASK];
+    return &d_cache[_hash & VFS_HASH_MASK];
+}
+
+static inline int
+__sync_inode_nolock(struct v_inode* inode)
+{
+    pcache_commit_all(inode);
+
+    int errno = ENOTSUP;
+    if (inode->ops->sync) {
+        errno = inode->ops->sync(inode);
+    }
+
+    return errno;

 }
 
 struct v_dnode*
 }
 
 struct v_dnode*


@@ -137,7 +145,7 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
     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 == hash) {
+        if (pos->name.hash == hash && pos->parent == parent) {

             return pos;
         }
     }
             return pos;
         }
     }


@@ -149,7 +157,7 @@ vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
 {
     assert(parent);
 
 {
     assert(parent);
 

-    atomic_fetch_add(&dnode->ref_count, 1);
+    dnode->ref_count = 1;

     dnode->parent = parent;
     llist_append(&parent->children, &dnode->siblings);
 
     dnode->parent = parent;
     llist_append(&parent->children, &dnode->siblings);
 


@@ -168,7 +176,7 @@ vfs_dcache_remove(struct v_dnode* dnode)
     hlist_delete(&dnode->hash_list);
 
     dnode->parent = NULL;
     hlist_delete(&dnode->hash_list);
 
     dnode->parent = NULL;

-    atomic_fetch_sub(&dnode->ref_count, 1);
+    dnode->ref_count = 0;

 }
 
 void
 }
 
 void


@@ -197,10 +205,10 @@ vfs_open(struct v_dnode* dnode, struct v_file** file)
 
     vfile->dnode = dnode;
     vfile->inode = inode;
 
     vfile->dnode = dnode;
     vfile->inode = inode;

-    vfile->ref_count = ATOMIC_VAR_INIT(1);
+    vfile->ref_count = 1;

     vfile->ops = inode->default_fops;
 
     vfile->ops = inode->default_fops;
 

-    if ((inode->itype & F_MFILE) && !inode->pg_cache) {
+    if (check_regfile_node(inode) && !inode->pg_cache) {

         struct pcache* pcache = vzalloc(sizeof(struct pcache));
         pcache_init(pcache);
         pcache->master = inode;
         struct pcache* pcache = vzalloc(sizeof(struct pcache));
         pcache_init(pcache);
         pcache->master = inode;


@@ -211,9 +219,8 @@ vfs_open(struct v_dnode* dnode, struct v_file** file)
     if (errno) {
         cake_release(file_pile, vfile);
     } else {
     if (errno) {
         cake_release(file_pile, vfile);
     } else {

-        atomic_fetch_add(&dnode->ref_count, 1);
+        vfs_ref_dnode(dnode);

         inode->open_count++;
         inode->open_count++;

-        mnt_mkbusy(dnode->mnt);

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


@@ -230,6 +237,7 @@ vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode)
         llist_delete(&assign_to->aka_list);
         assign_to->inode->link_count--;
     }
         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++;
     llist_append(&inode->aka_dnodes, &assign_to->aka_list);
     assign_to->inode = inode;
     inode->link_count++;


@@ -260,38 +268,64 @@ vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 int
 vfs_pclose(struct v_file* file, pid_t pid)
 {
 int
 vfs_pclose(struct v_file* file, pid_t pid)
 {

+    struct v_inode* inode;

     int errno = 0;
     int errno = 0;

-    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--;
-
-        /*
-         * 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);
+    inode = file->inode;
+
+    /*
+     * Prevent dead lock.
+     * This happened when process is terminated while blocking on read.
+     * In that case, the process is still holding the inode lock and it
+         will never get released.
+     * The unlocking should also include ownership check.
+     *
+     * To see why, consider two process both open the same file both with
+     * fd=x.
+     *      Process A: busy on reading x
+     *      Process B: do nothing with x
+     * Assuming that, after a very short time, process B get terminated
+     * while process A is still busy in it's reading business. By this
+     * design, the inode lock of this file x is get released by B rather
+     * than A. And this will cause a probable race condition on A if other
+     * process is writing to this file later after B exit.
+    */
+
+    mutex_unlock_for(&inode->lock, pid);
+    
+    if (vfs_check_duped_file(file)) {
+        vfs_unref_file(file);
+        return 0;
+    }
+
+    if ((errno = file->ops->close(file))) {
+        goto done;
+    }
+
+    vfs_unref_dnode(file->dnode);
+    cake_release(file_pile, file);
+
+    /*
+        if the current inode is not being locked by other 
+        threads that does not share same open context,
+        then we can try to do sync opportunistically
+    */
+    if (mutex_on_hold(&inode->lock)) {
+        goto done;

     }
     }

+    
+    lock_inode(inode);
+
+    pcache_commit_all(inode);
+    inode->open_count--;
+
+    if (!inode->open_count) {
+        __sync_inode_nolock(inode);
+    }
+
+    unlock_inode(inode);
+
+done:

     return errno;
 }
 
     return errno;
 }
 


@@ -307,6 +341,18 @@ vfs_free_fd(struct v_fd* fd)
     cake_release(fd_pile, fd);
 }
 
     cake_release(fd_pile, fd);
 }
 

+int
+vfs_isync(struct v_inode* inode)
+{
+    lock_inode(inode);
+
+    int errno = __sync_inode_nolock(inode);
+
+    unlock_inode(inode);
+
+    return errno;
+}
+

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


@@ -315,18 +361,7 @@ vfs_fsync(struct v_file* file)
         return errno;
     }
 
         return errno;
     }
 

-    lock_inode(file->inode);
-
-    pcache_commit_all(file->inode);
-
-    errno = ENOTSUP;
-    if (file->ops->sync) {
-        errno = file->ops->sync(file);
-    }
-
-    unlock_inode(file->inode);
-
-    return errno;
+    return vfs_isync(file->inode);

 }
 
 int
 }
 
 int


@@ -347,14 +382,37 @@ vfs_sb_alloc()
     struct v_superblock* sb = cake_grab(superblock_pile);
     memset(sb, 0, sizeof(*sb));
     llist_init_head(&sb->sb_list);
     struct v_superblock* sb = cake_grab(superblock_pile);
     memset(sb, 0, sizeof(*sb));
     llist_init_head(&sb->sb_list);

+    

     sb->i_cache = vzalloc(VFS_HASHTABLE_SIZE * sizeof(struct hbucket));
     sb->i_cache = vzalloc(VFS_HASHTABLE_SIZE * sizeof(struct hbucket));

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

     return sb;
 }
 
 void
     return sb;
 }
 
 void

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

 {
 {

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

     vfree(sb->i_cache);
     vfree(sb->i_cache);

+    vfree(sb->d_cache);
+    

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


@@ -400,13 +458,12 @@ vfs_d_alloc(struct v_dnode* parent, struct hstr* name)
     llist_init_head(&dnode->aka_list);
     mutex_init(&dnode->lock);
 
     llist_init_head(&dnode->aka_list);
     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->name = HHSTR(vzalloc(VFS_NAME_MAXLEN), 0, 0);
 
     hstrcpy(&dnode->name, name);
 
     if (parent) {

-        dnode->super_block = parent->super_block;
+        vfs_d_assign_sb(dnode, parent->super_block);

         dnode->mnt = parent->mnt;
     }
 
         dnode->mnt = parent->mnt;
     }
 


@@ -435,6 +492,11 @@ vfs_d_free(struct v_dnode* dnode)
         vfs_dcache_remove(pos);
     }
 
         vfs_dcache_remove(pos);
     }
 

+    if (dnode->destruct) {
+        dnode->destruct(dnode);
+    }
+
+    vfs_sb_unref(dnode->super_block);

     vfree((void*)dnode->name.value);
     cake_release(dnode_pile, dnode);
 }
     vfree((void*)dnode->name.value);
     cake_release(dnode_pile, dnode);
 }


@@ -484,11 +546,11 @@ vfs_i_alloc(struct v_superblock* sb)
 
     sb->ops.init_inode(sb, inode);
 
 
     sb->ops.init_inode(sb, inode);
 

-    inode->sb = sb;

     inode->ctime = clock_unixtime();
     inode->atime = inode->ctime;
     inode->mtime = inode->ctime;
 
     inode->ctime = clock_unixtime();
     inode->atime = inode->ctime;
     inode->mtime = inode->ctime;
 

+    vfs_i_assign_sb(inode, sb);

     lru_use_one(inode_lru, &inode->lru);
     return inode;
 }
     lru_use_one(inode_lru, &inode->lru);
     return inode;
 }


@@ -506,15 +568,22 @@ vfs_i_free(struct v_inode* inode)
     if (inode->destruct) {
         inode->destruct(inode);
     }
     if (inode->destruct) {
         inode->destruct(inode);
     }

+
+    vfs_sb_unref(inode->sb);

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

-#define FLOCATE_CREATE_EMPTY 1
-#define FLOCATE_CREATE_ONLY 2
-#define FLOCATE_NOFOLLOW 4
+// make a new name when not exists
+#define FLOC_MAYBE_MKNAME 1
+
+// name must be non-exist and made.
+#define FLOC_MKNAME 2
+
+// no follow symlink
+#define FLOC_NOFOLLOW 4

 
 int
 vfs_getfd(int fd, struct v_fd** fd_s)
 
 int
 vfs_getfd(int fd, struct v_fd** fd_s)


@@ -525,53 +594,104 @@ vfs_getfd(int fd, struct v_fd** fd_s)
     return EBADF;
 }
 
     return EBADF;
 }
 

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

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

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

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

+    struct v_dnode *fdir, *file;

     struct hstr name = HSTR(name_str, 0);
     int errno, woption = 0;
 
     struct hstr name = HSTR(name_str, 0);
     int errno, woption = 0;
 

-    if ((options & FLOCATE_NOFOLLOW)) {
+    if ((options & FLOC_NOFOLLOW)) {

         woption |= VFS_WALK_NOFOLLOW;
         woption |= VFS_WALK_NOFOLLOW;

+        options &= ~FLOC_NOFOLLOW;

     }
 
     }
 

+    floc->fresh = false;

     name_str[0] = 0;
     name_str[0] = 0;

-    if ((errno = vfs_walk_proc(path, fdir, &name, woption | VFS_WALK_PARENT))) {
+    errno = vfs_walk_proc(path, &fdir, &name, woption | VFS_WALK_PARENT);
+    if (errno) {

         return errno;
     }
 
         return errno;
     }
 

-    errno = vfs_walk(*fdir, name.value, file, NULL, woption);
+    errno = vfs_walk(fdir, name.value, &file, NULL, woption);
+
+    if (errno && errno != ENOENT) {
+        goto done;
+    }
+    
+    if (!errno) {
+        if ((options & FLOC_MKNAME)) {
+            errno = EEXIST;
+        }
+        goto done;
+    }

 
 

-    if (errno != ENOENT && (options & FLOCATE_CREATE_ONLY)) {
-        return EEXIST;
+    // errno == ENOENT
+    if (!options) {
+        goto done;

     }
 
     }
 

-    if (errno != ENOENT ||
-        !(options & (FLOCATE_CREATE_EMPTY | FLOCATE_CREATE_ONLY))) {
-        return errno;
+    errno = vfs_check_writable(fdir);
+    if (errno) {
+        goto done;

     }
 
     }
 

-    struct v_dnode* parent = *fdir;
-    struct v_dnode* file_new = vfs_d_alloc(parent, &name);
+    floc->fresh = true;
+
+    file = vfs_d_alloc(fdir, &name);

 
 

-    if (!file_new) {
+    if (!file) {

         return ENOMEM;
     }
 
         return ENOMEM;
     }
 

-    lock_dnode(parent);
+    lock_dnode(fdir);

 
 

-    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);
-    }
+    vfs_dcache_add(fdir, file);

 
 

-    unlock_dnode(parent);
+done:
+    floc->dir   = fdir;
+    floc->file  = file;

 
     return errno;
 }
 
     return errno;
 }


@@ -582,32 +702,61 @@ vfs_do_open(const char* path, int options)
     int errno, fd, loptions = 0;
     struct v_dnode *dentry, *file;
     struct v_file* ofile = NULL;
     int errno, fd, loptions = 0;
     struct v_dnode *dentry, *file;
     struct v_file* ofile = NULL;

+    struct file_locator floc;
+    struct v_inode* inode;

 
     if ((options & FO_CREATE)) {
 
     if ((options & FO_CREATE)) {

-        loptions |= FLOCATE_CREATE_EMPTY;
+        loptions |= FLOC_MAYBE_MKNAME;

     } else if ((options & FO_NOFOLLOW)) {
     } else if ((options & FO_NOFOLLOW)) {

-        loptions |= FLOCATE_NOFOLLOW;
+        loptions |= FLOC_NOFOLLOW;

     }
 
     }
 

-    errno = __vfs_try_locate_file(path, &dentry, &file, loptions);
+    errno = __vfs_try_locate_file(path, &floc, loptions);
+
+    if (errno || (errno = vfs_alloc_fdslot(&fd))) {
+        return errno;
+    }

 
 

-    if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
+    file   = floc.file;
+    dentry = floc.dir;

 
 

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

             return errno;
         }
 
             return errno;
         }
 

-        struct v_fd* fd_s = cake_grab(fd_pile);
-        memset(fd_s, 0, sizeof(*fd_s));
+        __floc_try_unlock(&floc);
+    }
+

 
 

-        ofile->f_pos = ofile->inode->fsize & -((options & FO_APPEND) != 0);
-        fd_s->file = ofile;
-        fd_s->flags = options;
-        __current->fdtable->fds[fd] = fd_s;
-        return fd;
+    if ((errno = vfs_open(file, &ofile))) {
+        return errno;

     }
 
     }
 

-    return errno;
+    inode = ofile->inode;
+    lock_inode(inode);
+
+    struct v_fd* fd_s = cake_grab(fd_pile);
+    memset(fd_s, 0, sizeof(*fd_s));
+
+    if ((options & O_TRUNC)) {
+        file->inode->fsize = 0;   
+    }
+
+    if (vfs_get_dtype(inode->itype) == DT_DIR) {
+        ofile->f_pos = 0;
+    }
+    
+    fd_s->file = ofile;
+    fd_s->flags = options;
+    __current->fdtable->fds[fd] = fd_s;
+
+    unlock_inode(inode);
+    
+    return fd;

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


@@ -642,7 +791,7 @@ __vfs_readdir_callback(struct dir_context* dctx,
                        const int dtype)
 {
     struct lx_dirent* dent = (struct lx_dirent*)dctx->cb_data;
                        const int dtype)
 {
     struct lx_dirent* dent = (struct lx_dirent*)dctx->cb_data;

-    strncpy(dent->d_name, name, DIRENT_NAME_MAX_LEN);
+    strncpy(dent->d_name, name, MIN(len, DIRENT_NAME_MAX_LEN));

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


@@ -660,28 +809,23 @@ __DEFINE_LXSYSCALL2(int, sys_readdir, int, fd, struct lx_dirent*, dent)
 
     lock_inode(inode);
 
 
     lock_inode(inode);
 

-    if ((inode->itype & F_FILE)) {
+    if (!check_directory_node(inode)) {

         errno = ENOTDIR;
         errno = ENOTDIR;

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

     }
     }

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

 
 

+unlock:

     unlock_inode(inode);
 
 done:
     unlock_inode(inode);
 
 done:


@@ -697,7 +841,7 @@ __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
     }
 
     struct v_file* file = fd_s->file;
     }
 
     struct v_file* file = fd_s->file;

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

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


@@ -706,7 +850,7 @@ __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
 
     file->inode->atime = clock_unixtime();
 
 
     file->inode->atime = clock_unixtime();
 

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


@@ -732,34 +876,41 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
         goto done;
     }
 
         goto done;
     }
 

+    struct v_inode* inode;

     struct v_file* file = fd_s->file;
 
     if ((errno = vfs_check_writable(file->dnode))) {
         goto done;
     }
 
     struct v_file* file = fd_s->file;
 
     if ((errno = vfs_check_writable(file->dnode))) {
         goto done;
     }
 

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

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

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

 
 

-    file->inode->mtime = clock_unixtime();
+    inode->mtime = clock_unixtime();
+    if ((fd_s->flags & O_APPEND)) {
+        file->f_pos = inode->fsize;
+    }

 
 

-    if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
-        errno = file->ops->write(file->inode, buf, count, file->f_pos);
+    if (check_seqdev_node(inode) || (fd_s->flags & FO_DIRECT)) {
+        errno = file->ops->write(inode, buf, count, file->f_pos);

     } else {
     } else {

-        errno = pcache_write(file->inode, buf, count, file->f_pos);
+        errno = pcache_write(inode, buf, count, file->f_pos);

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

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

         return errno;
     }
 
         return errno;
     }
 

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

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


@@ -774,34 +925,42 @@ __DEFINE_LXSYSCALL3(int, lseek, int, fd, int, offset, int, options)
     }
 
     struct v_file* file = fd_s->file;
     }
 
     struct v_file* file = fd_s->file;

+    struct v_inode* inode = file->inode;

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

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

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

+
+    if (vfs_get_dtype(inode->itype) == DT_DIR) {
+        options = (options != FSEEK_END) ? options : FSEEK_SET;
+    }
+    

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

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

             break;
         case FSEEK_END:
             break;
         case FSEEK_END:

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

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

+

     if (overflow) {
         errno = EOVERFLOW;
     if (overflow) {
         errno = EOVERFLOW;

-    } else if (!(errno = file->ops->seek(file->inode, fpos))) {
-        file->f_pos = fpos;
+    }
+    else {
+        errno = file->ops->seek(file, fpos);

     }
 
     }
 

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

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


@@ -844,28 +1003,42 @@ 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) {
-        lock_inode(inode);

 
 

-        int errno = inode->ops->read_symlink(inode, &link);
-        strncpy(buf, link, size);
+    if (!check_symlink_node(inode)) {
+        return EINVAL;
+    }

 
 

-        unlock_inode(inode);
-        return errno;
+    if (!inode->ops->read_symlink) {
+        return ENOTSUP;

     }
     }

-    return 0;
+
+    lock_inode(inode);
+
+    int errno = inode->ops->read_symlink(inode, &link);
+    if (errno >= 0) {
+        strncpy(buf, link, MIN(size, (size_t)errno));
+    }
+
+    unlock_inode(inode);
+    return errno;

 }
 
 int
 vfs_get_dtype(int itype)
 {
 }
 
 int
 vfs_get_dtype(int itype)
 {

-    if ((itype & VFS_IFSYMLINK) == VFS_IFSYMLINK) {
-        return DT_SYMLINK;
-    } else if (!(itype & VFS_IFFILE)) {
-        return DT_DIR;
-    } else {
-        return DT_FILE;
+    int dtype = DT_FILE;
+    if (check_itype(itype, VFS_IFSYMLINK)) {
+        dtype |= DT_SYMLINK;
+    }
+    
+    if (check_itype(itype, VFS_IFDIR)) {
+        dtype |= DT_DIR;
+        return dtype;

     }
     }

+
+    // TODO other types
+    
+    return dtype;

 }
 
 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
 }
 
 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)


@@ -974,7 +1147,7 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
     lock_dnode(parent);
     lock_inode(parent->inode);
 
     lock_dnode(parent);
     lock_inode(parent->inode);
 

-    if (!(dnode->inode->itype & F_MFILE)) {
+    if (check_directory_node(dnode->inode)) {

         errno = parent->inode->ops->rmdir(parent->inode, dnode);
         if (!errno) {
             vfs_dcache_remove(dnode);
         errno = parent->inode->ops->rmdir(parent->inode, dnode);
         if (!errno) {
             vfs_dcache_remove(dnode);


@@ -1016,16 +1189,18 @@ __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
         goto done;
     }
 
         goto done;
     }
 

+    struct v_inode* inode = parent->inode;
+

     lock_dnode(parent);
     lock_dnode(parent);

-    lock_inode(parent->inode);
+    lock_inode(inode);

 
     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 (!inode->ops->mkdir) {

         errno = ENOTSUP;
         errno = ENOTSUP;

-    } else if ((parent->inode->itype & F_FILE)) {
+    } else if (!check_directory_node(inode)) {

         errno = ENOTDIR;
         errno = ENOTDIR;

-    } else if (!(errno = parent->inode->ops->mkdir(parent->inode, dir))) {
+    } else if (!(errno = inode->ops->mkdir(inode, dir))) {

         vfs_dcache_add(parent, dir);
         goto cleanup;
     }
         vfs_dcache_add(parent, dir);
         goto cleanup;
     }


@@ -1033,7 +1208,7 @@ __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
     vfs_d_free(dir);
 
 cleanup:
     vfs_d_free(dir);
 
 cleanup:

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

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


@@ -1057,8 +1232,8 @@ __vfs_do_unlink(struct v_dnode* dnode)
 
     if (inode->open_count) {
         errno = EBUSY;
 
     if (inode->open_count) {
         errno = EBUSY;

-    } else if ((inode->itype & F_MFILE)) {
-        errno = inode->ops->unlink(inode);
+    } else if (!check_directory_node(inode)) {
+        errno = inode->ops->unlink(inode, dnode);

         if (!errno) {
             vfs_d_free(dnode);
         }
         if (!errno) {
             vfs_d_free(dnode);
         }


@@ -1105,16 +1280,30 @@ done:
 __DEFINE_LXSYSCALL2(int, link, const char*, oldpath, const char*, newpath)
 {
     int errno;
 __DEFINE_LXSYSCALL2(int, link, const char*, oldpath, const char*, newpath)
 {
     int errno;

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

 
 

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

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

-        errno = __vfs_try_locate_file(
-          newpath, &name_dentry, &name_file, FLOCATE_CREATE_ONLY);
-        if (!errno) {
-            errno = vfs_link(to_link, name_file);
-        }
+        goto done;       
+    }
+
+    name_file = floc.file;
+    errno = vfs_link(to_link, name_file);
+    if (errno) {
+        vfs_d_free(name_file);

     }
     }

+
+done:
+    __floc_try_unlock(&floc);

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


@@ -1138,7 +1327,7 @@ vfs_dup_fd(struct v_fd* old, struct v_fd** new)
 
     memcpy(copied, old, sizeof(struct v_fd));
 
 
     memcpy(copied, old, sizeof(struct v_fd));
 

-    atomic_fetch_add(&old->file->ref_count, 1);
+    vfs_ref_file(old->file);

 
     *new = copied;
 
 
     *new = copied;
 


@@ -1204,56 +1393,47 @@ __DEFINE_LXSYSCALL2(
   int, symlink, const char*, pathname, const char*, link_target)
 {
     int errno;
   int, symlink, const char*, pathname, const char*, link_target)
 {
     int errno;

-    struct v_dnode *dnode, *file;
-    if ((errno = __vfs_try_locate_file(
-           pathname, &dnode, &file, FLOCATE_CREATE_ONLY))) {
+    struct file_locator floc;
+    struct v_dnode *file;
+    struct v_inode *f_ino;
+    
+    errno = __vfs_try_locate_file(pathname, &floc, FLOC_MKNAME);
+    if (errno) {

         goto done;
     }
 
         goto done;
     }
 

-    if ((errno = vfs_check_writable(file))) {
+    file = floc.file;
+    errno = __vfs_mknod(floc.dir->inode, file, VFS_IFSYMLINK, NULL);
+    if (errno) {
+        vfs_d_free(file);

         goto done;
     }
 
         goto done;
     }
 

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

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

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

 
 

-    errno = file->inode->ops->set_symlink(file->inode, link_target);
+    errno = f_ino->ops->set_symlink(f_ino, link_target);

 
 

-    unlock_inode(file->inode);
+    unlock_inode(f_ino);

 
 done:
 
 done:

+    __floc_try_unlock(&floc);

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

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

 int
 vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode)
 {
 int
 vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode)
 {


@@ -1261,7 +1441,7 @@ vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode)
 
     lock_dnode(dnode);
 
 
     lock_dnode(dnode);
 

-    if ((dnode->inode->itype & F_FILE)) {
+    if (!check_directory_node(dnode->inode)) {

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


@@ -1456,22 +1636,22 @@ __DEFINE_LXSYSCALL2(int, fstat, int, fd, struct file_stat*, stat)
                                .st_ioblksize = PAGE_SIZE,
                                .st_blksize = vino->sb->blksize};
 
                                .st_ioblksize = PAGE_SIZE,
                                .st_blksize = vino->sb->blksize};
 

-    if (VFS_DEVFILE(vino->itype)) {
+    if (check_device_node(vino)) {

         struct device* rdev = resolve_device(vino->data);
         struct device* rdev = resolve_device(vino->data);

-        if (!rdev || rdev->magic != DEV_STRUCT_MAGIC) {
+        if (!rdev) {

             errno = EINVAL;
             goto done;
         }
 
         stat->st_rdev = (dev_t){.meta = rdev->ident.fn_grp,
                                 .unique = rdev->ident.unique,
             errno = EINVAL;
             goto done;
         }
 
         stat->st_rdev = (dev_t){.meta = rdev->ident.fn_grp,
                                 .unique = rdev->ident.unique,

-                                .index = rdev->dev_uid};
+                                .index = dev_uid(rdev) };

     }
 
     if (fdev) {
         stat->st_dev = (dev_t){.meta = fdev->ident.fn_grp,
                                .unique = fdev->ident.unique,
     }
 
     if (fdev) {
         stat->st_dev = (dev_t){.meta = fdev->ident.fn_grp,
                                .unique = fdev->ident.unique,

-                               .index = fdev->dev_uid};
+                               .index = dev_uid(fdev) };

     }
 
 done:
     }
 
 done: