X-Git-Url: https://scm.lunaixsky.com/lunaix-os.git/blobdiff_plain/e6dbd781772925dfc07b2facfa0a1350e756ad8f..fb2d2e17e3ffc4bddf944dc67a6d86fe9453bb5a:/lunaix-os/kernel/fs/vfs.c

diff --git a/lunaix-os/kernel/fs/vfs.c b/lunaix-os/kernel/fs/vfs.c
index 207427f..8edc9fd 100644
--- a/lunaix-os/kernel/fs/vfs.c
+++ b/lunaix-os/kernel/fs/vfs.c
@@ -9,6 +9,39 @@
  *
  */
 
+// 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.
+ 6. (mount) Ability to track all mount points (including sub-mounts)
+            so we can be confident to clean up everything when we unmount.
+ 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?
+ 8. (mount) Then, we should refactor on mount/unmount mechanism.
+ 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>
@@ -23,16 +56,24 @@
 #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 HASHTABLE_BITS 10
+#define HASHTABLE_SIZE (1 << HASHTABLE_BITS)
+#define HASH_MASK (HASHTABLE_SIZE - 1)
+#define HASHBITS (32 - HASHTABLE_BITS)
 
-#define lock_inode(inode) mutex_lock(&inode->lock)
 #define unlock_inode(inode) mutex_unlock(&inode->lock)
+#define lock_inode(inode)                                                      \
+    ({                                                                         \
+        mutex_lock(&inode->lock);                                              \
+        lru_use_one(inode_lru, &inode->lru);                                   \
+    })
 
-#define lock_dnode(dnode) mutex_lock(&dnode->lock)
 #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;
@@ -41,26 +82,26 @@ static struct cake_pile* superblock_pile;
 static struct cake_pile* fd_pile;
 
 static struct v_superblock* root_sb;
-static struct hbucket* dnode_cache;
+static struct hbucket *dnode_cache, *inode_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()
 {
@@ -72,7 +113,11 @@ 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(HASHTABLE_SIZE * sizeof(struct hbucket));
+    inode_cache = vzalloc(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);
@@ -80,7 +125,6 @@ vfs_init()
     // 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();
 }
 
 inline struct hbucket*
@@ -90,9 +134,9 @@ __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;
     // 确保低位更加随机
-    _hash = _hash ^ (_hash >> DNODE_HASHBITS);
+    _hash = _hash ^ (_hash >> HASHBITS);
     *hash = _hash;
-    return &dnode_cache[_hash & DNODE_HASH_MASK];
+    return &dnode_cache[_hash & HASH_MASK];
 }
 
 struct v_dnode*
@@ -121,16 +165,31 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
 void
 vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
 {
+    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_rehash(struct v_dnode* parent, struct v_dnode* dnode)
+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_add(parent, dnode);
+    vfs_dcache_remove(dnode);
+    vfs_dcache_add(new_parent, dnode);
 }
 
 int
@@ -200,6 +259,11 @@ __vfs_walk(struct v_dnode* start,
         if (!dnode) {
             dnode = vfs_d_alloc();
 
+            if (!dnode) {
+                errno = ENOMEM;
+                goto error;
+            }
+
             hstrcpy(&dnode->name, &name);
 
             lock_inode(current_level->inode);
@@ -221,13 +285,10 @@ __vfs_walk(struct v_dnode* start,
             if (errno) {
                 unlock_dnode(current_level);
                 vfree(dnode->name.value);
-                goto error;
+                goto cleanup;
             }
 
             vfs_dcache_add(current_level, dnode);
-
-            dnode->parent = current_level;
-            llist_append(&current_level->children, &dnode->siblings);
         }
 
         unlock_dnode(current_level);
@@ -241,8 +302,9 @@ __vfs_walk(struct v_dnode* start,
     *dentry = current_level;
     return 0;
 
-error:
+cleanup:
     vfs_d_free(dnode);
+error:
     *dentry = NULL;
     return errno;
 }
@@ -261,15 +323,18 @@ vfs_walk(struct v_dnode* start,
     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_IFSYMLINK) &&
-            !(options & VFS_WALK_NOFOLLOW) &&
             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;
             }
@@ -316,7 +381,7 @@ vfs_mount_at(const char* fs_name,
              struct device* device,
              struct v_dnode* mnt_point)
 {
-    if (!(mnt_point->inode->itype & VFS_IFDIR)) {
+    if (mnt_point->inode && !(mnt_point->inode->itype & VFS_IFDIR)) {
         return ENOTDIR;
     }
 
@@ -327,7 +392,7 @@ vfs_mount_at(const char* fs_name,
 
     struct v_superblock* sb = vfs_sb_alloc();
     sb->dev = device;
-    sb->fs_id = fs_id++;
+    sb->fs_id = fs->fs_id;
 
     int errno = 0;
     if (!(errno = fs->mount(sb, mnt_point))) {
@@ -343,17 +408,22 @@ vfs_mount_at(const char* fs_name,
 int
 vfs_unmount_at(struct v_dnode* mnt_point)
 {
-    // FIXME mnt point check & deal with the detached dcache subtree
+    // FIXME deal with the detached dcache subtree
     int errno = 0;
     struct v_superblock* sb = mnt_point->super_block;
     if (!sb) {
         return EINVAL;
     }
+
+    if (sb->root != mnt_point) {
+        return EINVAL;
+    }
+
     if (!(errno = sb->fs->unmount(sb))) {
         struct v_dnode* fs_root = sb->root;
-        llist_delete(&fs_root->siblings);
+        vfs_dcache_remove(fs_root);
+
         llist_delete(&sb->sb_list);
-        hlist_delete(&fs_root->hash_list);
         vfs_sb_free(sb);
         vfs_d_free(fs_root);
     }
@@ -396,6 +466,16 @@ vfs_open(struct v_dnode* dnode, struct v_file** file)
     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)
 {
@@ -407,8 +487,7 @@ vfs_link(struct v_dnode* to_link, struct v_dnode* name)
     } 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;
-        atomic_fetch_add(&to_link->inode->link_count, 1);
+        vfs_assign_inode(name, to_link->inode);
     }
     unlock_inode(to_link->inode);
 
@@ -472,10 +551,42 @@ vfs_sb_free(struct v_superblock* 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*
 vfs_d_alloc()
 {
     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);
     llist_init_head(&dnode->siblings);
@@ -484,33 +595,78 @@ vfs_d_alloc()
     dnode->ref_count = ATOMIC_VAR_INIT(0);
     dnode->name = HHSTR(vzalloc(VFS_NAME_MAXLEN), 0, 0);
 
+    lru_use_one(dnode_lru, &dnode->lru);
+
     return dnode;
 }
 
 void
 vfs_d_free(struct v_dnode* dnode)
 {
-    if (dnode->inode && dnode->inode->link_count) {
+    assert(dnode->ref_count == 0);
+
+    if (dnode->inode) {
+        assert(dnode->inode->link_count > 0);
         dnode->inode->link_count--;
     }
+
+    // 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(dev_t device_id, uint32_t inode_id)
 {
-    struct v_inode* inode = cake_grab(inode_pile);
-    memset(inode, 0, sizeof(*inode));
-    inode->link_count = 1;
-    mutex_init(&inode->lock);
+    // 我们这里假设每个文件系统与设备是一一对应（毕竟一个分区不可能有两个不同的文件系统）
+    // 而每个文件系统所产生的 v_inode 缓存必须要和其他文件系统产生的区分开来。
+    // 这也就是说，每个 v_inode 的 id
+    // 必须要由设备ID，和该虚拟inode缓存所对应的物理inode
+    // 相对于其所在的文件系统的id，进行组成！
+    inode_id = hash_32(inode_id ^ (-device_id), HASH_SIZE_BITS);
+    inode_id = (inode_id >> HASHBITS) ^ inode_id;
+
+    struct hbucket* slot = &inode_cache[inode_id & HASH_MASK];
+    struct v_inode *pos, *n;
+    hashtable_bucket_foreach(slot, pos, n, hash_list)
+    {
+        if (pos->id == inode_id) {
+            goto done;
+        }
+    }
 
-    return inode;
+    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;
+
+    mutex_init(&pos->lock);
+
+    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)
 {
+    hlist_delete(&inode->hash_list);
     cake_release(inode_pile, inode);
 }
 
@@ -553,8 +709,15 @@ __vfs_try_locate_file(const char* path,
 
     struct v_dnode* parent = *fdir;
     struct v_dnode* file_new = vfs_d_alloc();
+
+    if (!file_new) {
+        return ENOMEM;
+    }
+
     hstrcpy(&file_new->name, &name);
 
+    lock_dnode(parent);
+
     if (!(errno = parent->inode->ops.create(parent->inode, file_new))) {
         *file = file_new;
 
@@ -564,6 +727,8 @@ __vfs_try_locate_file(const char* path,
         vfs_d_free(file_new);
     }
 
+    unlock_dnode(parent);
+
     return errno;
 }
 
@@ -702,13 +867,14 @@ __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
     __SYSCALL_INTERRUPTIBLE(
       { errno = file->ops.read(file->inode, buf, count, file->f_pos); })
 
-    unlock_inode(file->inode);
-
     if (errno > 0) {
         file->f_pos += errno;
+        unlock_inode(file->inode);
         return errno;
     }
 
+    unlock_inode(file->inode);
+
 done:
     return DO_STATUS(errno);
 }
@@ -734,13 +900,14 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
     __SYSCALL_INTERRUPTIBLE(
       { errno = file->ops.write(file->inode, buf, count, file->f_pos); })
 
-    unlock_inode(file->inode);
-
     if (errno > 0) {
         file->f_pos += errno;
+        unlock_inode(file->inode);
         return errno;
     }
 
+    unlock_inode(file->inode);
+
 done:
     return DO_STATUS(errno);
 }
@@ -908,15 +1075,12 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
 
     lock_dnode(dnode);
 
-    if (dnode->parent)
-        lock_dnode(dnode->parent);
-
     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;
     }
@@ -926,37 +1090,45 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
         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)) {
-        errno = dnode->inode->ops.rmdir(dnode->inode);
+        errno = parent->inode->ops.rmdir(parent->inode, dnode);
         if (!errno) {
-            llist_delete(&dnode->siblings);
-            hlist_delete(&dnode->hash_list);
-            unlock_inode(dnode->inode);
-            vfs_d_free(dnode);
-
-            goto done;
+            vfs_dcache_remove(dnode);
         }
     } else {
         errno = ENOTDIR;
     }
 
-    unlock_inode(dnode->inode);
+    unlock_inode(parent->inode);
+    unlock_dnode(parent);
 
 done:
     unlock_dnode(dnode);
-    if (dnode->parent)
-        unlock_dnode(dnode->parent);
     return DO_STATUS(errno);
 }
 
 __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
 {
+    int errno = 0;
     struct v_dnode *parent, *dir = vfs_d_alloc();
-    int errno =
-      vfs_walk(__current->cwd, path, &parent, &dir->name, VFS_WALK_PARENT);
-    if (errno) {
+
+    if (!dir) {
+        errno = ENOMEM;
+        goto done;
+    }
+
+    if ((errno = vfs_walk(
+           __current->cwd, path, &parent, &dir->name, VFS_WALK_PARENT))) {
         goto done;
     }
 
@@ -988,7 +1160,7 @@ __vfs_do_unlink(struct v_dnode* dnode)
 {
     struct v_inode* inode = dnode->inode;
 
-    if (dnode->ref_count) {
+    if (dnode->ref_count > 1) {
         return EBUSY;
     }
 
@@ -1002,9 +1174,7 @@ __vfs_do_unlink(struct v_dnode* dnode)
         //  symlink case
         errno = inode->ops.unlink(inode);
         if (!errno) {
-            inode->link_count--;
-            llist_delete(&dnode->siblings);
-            hlist_delete(&dnode->hash_list);
+            vfs_dcache_remove(dnode);
             vfs_d_free(dnode);
         }
     } else {
@@ -1166,14 +1336,14 @@ __DEFINE_LXSYSCALL2(int,
         errno = EROFS;
         goto done;
     }
-    if (!dnode->inode->ops.symlink) {
+    if (!dnode->inode->ops.set_symlink) {
         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);
 
@@ -1194,10 +1364,10 @@ __vfs_do_chdir(struct v_dnode* dnode)
     }
 
     if (__current->cwd) {
-        __current->cwd->ref_count--;
+        atomic_fetch_add(&__current->cwd->ref_count, 1);
     }
 
-    dnode->ref_count--;
+    atomic_fetch_sub(&dnode->ref_count, 1);
     __current->cwd = dnode;
 
     unlock_dnode(dnode);
@@ -1275,7 +1445,7 @@ vfs_do_rename(struct v_dnode* current, struct v_dnode* target)
         return 0;
     }
 
-    if (current->ref_count || target->ref_count) {
+    if (current->ref_count > 1 || target->ref_count > 1) {
         return EBUSY;
     }
 
@@ -1307,15 +1477,11 @@ vfs_do_rename(struct v_dnode* current, struct v_dnode* target)
     }
 
     // re-position current
-    current->parent = newparent;
     hstrcpy(&current->name, &target->name);
-    llist_delete(&current->siblings);
-    llist_append(&newparent->children, &current->siblings);
     vfs_dcache_rehash(newparent, current);
 
     // detach target
-    llist_delete(&target->siblings);
-    hlist_delete(&target->hash_list);
+    vfs_dcache_remove(target);
 
     unlock_dnode(target);
 
@@ -1347,11 +1513,17 @@ __DEFINE_LXSYSCALL2(int, rename, const char*, oldpath, const char*, newpath)
     errno = vfs_walk(target_parent, name.value, &target, NULL, 0);
     if (errno == ENOENT) {
         target = vfs_d_alloc();
-        hstrcpy(&target->name, &name);
     } else if (errno) {
         goto done;
     }
 
+    if (!target) {
+        errno = ENOMEM;
+        goto done;
+    }
+
+    hstrcpy(&target->name, &name);
+
     if (!(errno = vfs_do_rename(cur, target))) {
         vfs_d_free(target);
     }
@@ -1386,6 +1558,11 @@ __DEFINE_LXSYSCALL3(int,
         goto done;
     }
 
+    if (mnt->ref_count > 1) {
+        errno = EBUSY;
+        goto done;
+    }
+
     // FIXME should not touch the underlying fs!
     struct device* device =
       (struct device*)((struct twifs_node*)dev->inode->data)->data;