X-Git-Url: https://scm.lunaixsky.com/lunaix-os.git/blobdiff_plain/eb1d71eb06e977f9ea61f6f43e5ce65342faf1cc..bcc25888b3299758ad36721530cca3b899b7166c:/lunaix-os/kernel/fs/vfs.c diff --git a/lunaix-os/kernel/fs/vfs.c b/lunaix-os/kernel/fs/vfs.c index 958c761..8bcb644 100644 --- a/lunaix-os/kernel/fs/vfs.c +++ b/lunaix-os/kernel/fs/vfs.c @@ -16,56 +16,49 @@ It is overseen by Twilight Sparkle ;) - 1. Get inodes hooked into lru - 2. Get dnodes hooked into lru + 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. + 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. + 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. + 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? - 8. (mount) Then, we should refactor on mount/unmount mechanism. + 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] :: composition + [good idea] :: composition (CHECKED) */ #include -#include #include #include #include -#include #include #include #include #include +#include #include -#define PATH_DELIM '/' -#define HASHTABLE_BITS 10 -#define HASHTABLE_SIZE (1 << HASHTABLE_BITS) -#define HASH_MASK (HASHTABLE_SIZE - 1) -#define HASHBITS (32 - HASHTABLE_BITS) +#include -#define lock_inode(inode) mutex_lock(&inode->lock) -#define unlock_inode(inode) mutex_unlock(&inode->lock) - -#define lock_dnode(dnode) mutex_lock(&dnode->lock) -#define unlock_dnode(dnode) mutex_unlock(&dnode->lock) +#define INODE_ACCESSED 0 +#define INODE_MODIFY 1 static struct cake_pile* dnode_pile; static struct cake_pile* inode_pile; @@ -73,18 +66,19 @@ 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 hbucket *dnode_cache, *inode_cache; +struct v_dnode* vfs_sysroot = NULL; + +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_superblock* -vfs_sb_alloc(); +static int +__vfs_try_evict_dnode(struct lru_node* obj); -void -vfs_sb_free(struct v_superblock* sb); +static int +__vfs_try_evict_inode(struct lru_node* obj); void vfs_init() @@ -97,406 +91,376 @@ vfs_init() superblock_pile = cake_new_pile("sb_cache", sizeof(struct v_superblock), 1, 0); - dnode_cache = vzalloc(HASHTABLE_SIZE * sizeof(struct hbucket)); - inode_cache = vzalloc(HASHTABLE_SIZE * sizeof(struct hbucket)); + dnode_lru = lru_new_zone("vfs_dnode", __vfs_try_evict_dnode); + inode_lru = lru_new_zone("vfs_inode", __vfs_try_evict_inode); hstr_rehash(&vfs_ddot, HSTR_FULL_HASH); hstr_rehash(&vfs_dot, HSTR_FULL_HASH); - // 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(root_sb, 0); + // 创建一个根dnode。 + vfs_sysroot = vfs_d_alloc(NULL, &vfs_empty); + vfs_sysroot->parent = vfs_sysroot; + + vfs_ref_dnode(vfs_sysroot); + lru_remove(dnode_lru, &vfs_sysroot->lru); +} + +void +vfs_vncache_init(struct vncache* cache) +{ + cache->pool = vzalloc(VFS_HASHTABLE_SIZE * sizeof(struct hbucket)); + rwlock_init(&cache->lock); +} + +void +vfs_vncache_free(struct vncache* cache) +{ + // clear all other reader/writer + rwlock_begin_write(&cache->lock); + vfree(cache->pool); + + // already freed, so as the lock +} + +void +vfs_vncache_add(struct vncache* cache, size_t key, struct hlist_node* node) +{ + struct hbucket* slot; + + cache_atomic_write(cache, + { + slot = &cache->pool[key & VFS_HASH_MASK]; + hlist_delete(node); + hlist_add(&slot->head, node); + }); } -inline struct hbucket* -__dcache_hash(struct v_dnode* parent, uint32_t* hash) +static inline struct hbucket* +__dcache_hash_nolock(struct v_dnode* parent, u32_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; - // 确保低位更加随机 - _hash = _hash ^ (_hash >> HASHBITS); + struct v_superblock* sb; + struct hbucket* d_cache; + u32_t _hash; + + sb = parent->super_block; + + _hash = *hash; + _hash = _hash ^ (_hash >> VFS_HASHBITS); + _hash += (u32_t)__ptr(parent); + *hash = _hash; - return &dnode_cache[_hash & HASH_MASK]; + return &sb->d_cache.pool[_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* vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str) { + u32_t hash; + struct hbucket* slot; + struct v_dnode *pos, *n; + struct vncache *dcache; + if (!str->len || HSTR_EQ(str, &vfs_dot)) return parent; if (HSTR_EQ(str, &vfs_ddot)) { - return parent->parent ? parent->parent : parent; + return parent->parent; } - uint32_t hash = str->hash; - struct hbucket* slot = __dcache_hash(parent, &hash); + hash = str->hash; + dcache = dnode_cache(parent); + + vncache_lock_read(dcache); - struct v_dnode *pos, *n; + slot = __dcache_hash_nolock(parent, &hash); hashtable_bucket_foreach(slot, pos, n, hash_list) { - if (pos->name.hash == hash) { - return pos; + if (pos->name.hash != hash || pos->parent != parent) { + continue; } + + vncache_unlock_read(dcache); + return pos; } + + vncache_unlock_read(dcache); return NULL; } +static void +__vfs_touch_inode(struct v_inode* inode, const int type) +{ + if (type == INODE_MODIFY) { + inode->mtime = clock_unixtime(); + } + + else if (type == INODE_ACCESSED) { + inode->atime = clock_unixtime(); + } + + lru_use_one(inode_lru, &inode->lru); +} + void vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode) { - atomic_fetch_add(&dnode->ref_count, 1); + struct hbucket* bucket; + struct vncache* cache; + + assert(parent); + assert(locked_node(parent)); + + dnode->ref_count = 1; dnode->parent = parent; llist_append(&parent->children, &dnode->siblings); - struct hbucket* bucket = __dcache_hash(parent, &dnode->name.hash); - hlist_add(&bucket->head, &dnode->hash_list); + + cache_atomic_write(dnode_cache(parent), + { + bucket = __dcache_hash_nolock(parent, &dnode->name.hash); + hlist_add(&bucket->head, &dnode->hash_list); + }); } void vfs_dcache_remove(struct v_dnode* dnode) { + assert(dnode); assert(dnode->ref_count == 1); llist_delete(&dnode->siblings); - hlist_delete(&dnode->hash_list); + llist_delete(&dnode->aka_list); + lru_remove(dnode_lru, &dnode->lru); + + cache_atomic_write(dnode_cache(dnode), + { + hlist_delete(&dnode->hash_list); + }); dnode->parent = NULL; - atomic_fetch_sub(&dnode->ref_count, 1); + dnode->ref_count = 0; } 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); + assert(new_parent); + assert(locked_node(new_parent)); + + dnode_atomic(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, - struct v_dnode** dentry, - struct hstr* component, - int walk_options) +vfs_open(struct v_dnode* dnode, struct v_file** file) { - int errno = 0; - int i = 0, j = 0; - - if (path[0] == PATH_DELIM || !start) { - if ((walk_options & VFS_WALK_FSRELATIVE) && start) { - start = start->super_block->root; - } else { - start = root_sb->root; - } - i++; + struct v_inode* inode = dnode->inode; + + if (!inode || !inode->ops->open) { + return ENOTSUP; } - struct v_dnode* dnode; - struct v_dnode* current_level = start; - - char name_content[VFS_NAME_MAXLEN]; - struct hstr name = HSTR(name_content, 0); - - char current = path[i++], lookahead; - while (current) { - lookahead = path[i++]; - if (current != PATH_DELIM) { - if (j >= VFS_NAME_MAXLEN - 1) { - return ENAMETOOLONG; - } - if (!VFS_VALID_CHAR(current)) { - return EINVAL; - } - name_content[j++] = current; - if (lookahead) { - goto cont; - } - } - - // handling cases like /^.*(\/+).*$/ - if (lookahead == PATH_DELIM) { - goto cont; - } - - lock_dnode(current_level); - - name_content[j] = 0; - name.len = j; - hstr_rehash(&name, HSTR_FULL_HASH); - - if (!lookahead && (walk_options & VFS_WALK_PARENT)) { - if (component) { - component->hash = name.hash; - component->len = j; - strcpy(component->value, name_content); - } - unlock_dnode(current_level); - break; - } - - dnode = vfs_dcache_lookup(current_level, &name); - - if (!dnode) { - dnode = vfs_d_alloc(); - - hstrcpy(&dnode->name, &name); - - lock_inode(current_level->inode); - - errno = - current_level->inode->ops.dir_lookup(current_level->inode, dnode); - - if (errno == ENOENT && (walk_options & VFS_WALK_MKPARENT)) { - if (!current_level->inode->ops.mkdir) { - errno = ENOTSUP; - } else { - errno = current_level->inode->ops.mkdir( - current_level->inode, dnode); - } - } - - unlock_inode(current_level->inode); - - if (errno) { - unlock_dnode(current_level); - vfree(dnode->name.value); - goto error; - } - - vfs_dcache_add(current_level, dnode); - } - - unlock_dnode(current_level); + lock_inode(inode); - j = 0; - current_level = dnode; - cont: - current = lookahead; - }; + struct v_file* vfile = cake_grab(file_pile); + memset(vfile, 0, sizeof(*vfile)); - *dentry = current_level; - return 0; + vfile->dnode = dnode; + vfile->inode = inode; + vfile->ref_count = 1; + vfile->ops = inode->default_fops; -error: - vfs_d_free(dnode); - *dentry = NULL; - return errno; -} + if (check_regfile_node(inode) && !inode->pg_cache) { + struct pcache* pcache = vzalloc(sizeof(struct pcache)); + pcache_init(pcache); + pcache->master = inode; + inode->pg_cache = pcache; + } -#define VFS_MAX_SYMLINK 16 + int errno = inode->ops->open(inode, vfile); + if (errno) { + cake_release(file_pile, vfile); + } else { + vfs_ref_dnode(dnode); + inode->open_count++; -int -vfs_walk(struct v_dnode* start, - const char* path, - struct v_dnode** dentry, - struct hstr* component, - int options) -{ - struct v_dnode* interim; - const char* pathname = path; - int errno = __vfs_walk(start, path, &interim, component, options); - int counter = 0; - - while (!errno) { - if (counter >= VFS_MAX_SYMLINK) { - errno = ELOOP; - continue; - } - if ((interim->inode->itype & VFS_IFSYMLINK) && - !(options & VFS_WALK_NOFOLLOW) && - interim->inode->ops.read_symlink) { - errno = interim->inode->ops.read_symlink(interim->inode, &pathname); - if (errno) { - break; - } - } else { - break; - } - errno = __vfs_walk(start, pathname, &interim, component, options); - counter++; + *file = vfile; } - *dentry = errno ? 0 : interim; + unlock_inode(inode); return errno; } -int -vfs_mount(const char* target, const char* fs_name, struct device* device) +void +vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode) { - int errno; - struct v_dnode* mnt; + lock_dnode(assign_to); - if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) { - errno = vfs_mount_at(fs_name, device, mnt); + if (assign_to->inode) { + llist_delete(&assign_to->aka_list); + assign_to->inode->link_count--; } - return errno; + llist_append(&inode->aka_dnodes, &assign_to->aka_list); + assign_to->inode = inode; + inode->link_count++; + + unlock_dnode(assign_to); } int -vfs_unmount(const char* target) +vfs_link(struct v_dnode* to_link, struct v_dnode* name) { int errno; - struct v_dnode* mnt; - - if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) { - errno = vfs_unmount_at(mnt); - } - - return errno; -} + struct v_inode* inode; -int -vfs_mount_at(const char* fs_name, - struct device* device, - struct v_dnode* mnt_point) -{ - if (!(mnt_point->inode->itype & VFS_IFDIR)) { - return ENOTDIR; - } + inode = to_link->inode; - struct filesystem* fs = fsm_get(fs_name); - if (!fs) { - return ENODEV; + if ((errno = vfs_check_writable(to_link))) { + return errno; } - struct v_superblock* sb = vfs_sb_alloc(); - sb->dev = device; - sb->fs_id = fs->fs_id; + lock_inode(inode); - 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); + if (to_link->super_block->root != name->super_block->root) { + errno = EXDEV; + } else if (!inode->ops->link) { + errno = ENOTSUP; + } else if (!(errno = inode->ops->link(inode, name))) { + vfs_assign_inode(name, inode); } + unlock_inode(inode); + return errno; } int -vfs_unmount_at(struct v_dnode* mnt_point) +vfs_pclose(struct v_file* file, pid_t pid) { - // FIXME deal with the detached dcache subtree + struct v_inode* inode; 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; - vfs_dcache_remove(fs_root); + inode = file->inode; - llist_delete(&sb->sb_list); - vfs_sb_free(sb); - vfs_d_free(fs_root); - } - return errno; -} - -int -vfs_open(struct v_dnode* dnode, struct v_file** file) -{ - if (!dnode->inode || !dnode->inode->ops.open) { - return ENOTSUP; + if (vfs_check_duped_file(file)) { + vfs_unref_file(file); + return 0; } - struct v_inode* inode = dnode->inode; - struct v_file* vfile = cake_grab(file_pile); - memset(vfile, 0, sizeof(*vfile)); + /* + * 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); + + // now regain lock for inode syncing - vfile->dnode = dnode; - vfile->inode = inode; - vfile->ref_count = ATOMIC_VAR_INIT(1); - vfile->ops = inode->default_fops; + lock_inode(inode); - 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; + if ((errno = file->ops->close(file))) { + goto done; } - 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++; + vfs_unref_dnode(file->dnode); + cake_release(file_pile, file); - *file = vfile; + pcache_commit_all(inode); + inode->open_count--; + + if (!inode->open_count) { + __sync_inode_nolock(inode); } +done: + unlock_inode(inode); return errno; } int -vfs_link(struct v_dnode* to_link, struct v_dnode* name) +vfs_close(struct v_file* file) { - 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++; - } - unlock_inode(to_link->inode); + return vfs_pclose(file, __current->pid); +} - return errno; +void +vfs_free_fd(struct v_fd* fd) +{ + cake_release(fd_pile, fd); } int -vfs_close(struct v_file* file) +vfs_isync(struct v_inode* inode) { - int errno = 0; - if (!file->ops.close || !(errno = file->ops.close(file))) { - atomic_fetch_sub(&file->dnode->ref_count, 1); - file->inode->open_count--; + lock_inode(inode); + + int errno = __sync_inode_nolock(inode); + + unlock_inode(inode); - pcache_commit_all(file->inode); - cake_release(file_pile, file); - } return errno; } 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->inode); + int errno; + if ((errno = vfs_check_writable(file->dnode))) { + return errno; } - unlock_inode(file->inode); - - return errno; + return vfs_isync(file->inode); } int vfs_alloc_fdslot(int* fd) { + struct v_fdtable* fdtab; + + fdtab = __current->fdtable; + lock_fdtable(fdtab); + for (size_t i = 0; i < VFS_MAX_FD; i++) { - if (!__current->fdtable->fds[i]) { - *fd = i; - return 0; + if (__current->fdtable->fds[i]) { + continue; } + + *fd = i; + unlock_fdtable(fdtab); + return 0; } + + unlock_fdtable(fdtab); return EMFILE; } @@ -506,41 +470,119 @@ vfs_sb_alloc() struct v_superblock* sb = cake_grab(superblock_pile); memset(sb, 0, sizeof(*sb)); llist_init_head(&sb->sb_list); + + vfs_vncache_init(&sb->i_cache); + vfs_vncache_init(&sb->d_cache); + + sb->ref_count = 1; 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); + } + + vfs_vncache_free(&sb->i_cache); + vfs_vncache_free(&sb->d_cache); + cake_release(superblock_pile, sb); } +static inline bool +__dnode_evictable(struct v_dnode* dnode) +{ + return dnode->ref_count == 1 + && llist_empty(&dnode->children); +} + +static bool +__vfs_try_evict_dnode(struct lru_node* obj) +{ + struct v_dnode* dnode = container_of(obj, struct v_dnode, lru); + + if (mutex_on_hold(&dnode->lock)) + return false; + + if (!__dnode_evictable(dnode)) { + return false; + } + + vfs_d_free(dnode); + return true; +} + +static bool +__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); llist_init_head(&dnode->siblings); + 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) { + vfs_d_assign_sb(dnode, parent->super_block); + dnode->mnt = parent->mnt; + } + + lru_use_one(dnode_lru, &dnode->lru); + return dnode; } void vfs_d_free(struct v_dnode* dnode) { - if (dnode->ref_count) { - // it can be only freed if no one is refering - return; - } - if (dnode->inode && dnode->inode->link_count) { + 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 @@ -550,134 +592,326 @@ vfs_d_free(struct v_dnode* dnode) vfs_dcache_remove(pos); } - vfree(dnode->name.value); + if (dnode->destruct) { + dnode->destruct(dnode); + } + + vfs_sb_unref(dnode->super_block); + + vfree((void*)dnode->name.value); cake_release(dnode_pile, dnode); } struct v_inode* -vfs_i_alloc(dev_t device_id, uint32_t inode_id) -{ - // 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) +vfs_i_find(struct v_superblock* sb, u32_t i_id) +{ + struct hbucket* slot; + struct v_inode *pos, *n, *found = NULL; + + cache_atomic_read(&sb->i_cache, { - if (pos->id == inode_id) { - return pos; + slot = &sb->i_cache.pool[i_id & VFS_HASH_MASK]; + + hashtable_bucket_foreach(slot, pos, n, hash_list) + { + if (pos->id != i_id) { + continue; + } + + lru_use_one(inode_lru, &pos->lru); + found = pos; + break; } - } + }); - pos = cake_grab(inode_pile); - memset(pos, 0, sizeof(*pos)); + return found; +} - pos->id = inode_id; - pos->link_count = 1; +void +vfs_i_addhash(struct v_inode* inode) +{ + vfs_vncache_add(inode_cache(inode), inode->id, &inode->hash_list); +} - mutex_init(&pos->lock); +struct v_inode* +vfs_i_alloc(struct v_superblock* sb) +{ + assert(sb->ops.init_inode); - hlist_add(&slot->head, &pos->hash_list); + struct v_inode* inode; + if (!(inode = cake_grab(inode_pile))) { + lru_evict_half(inode_lru); + if (!(inode = cake_grab(inode_pile))) { + return NULL; + } + } - return pos; + memset(inode, 0, sizeof(*inode)); + mutex_init(&inode->lock); + llist_init_head(&inode->xattrs); + llist_init_head(&inode->aka_dnodes); + + sb->ops.init_inode(sb, inode); + + inode->ctime = clock_unixtime(); + inode->atime = inode->ctime; + inode->mtime = inode->ctime; + + vfs_i_assign_sb(inode, sb); + lru_use_one(inode_lru, &inode->lru); + + return inode; } void vfs_i_free(struct v_inode* inode) { + if (inode->pg_cache) { + pcache_release(inode->pg_cache); + vfree(inode->pg_cache); + } + + // we don't need to sync inode. + // If an inode can be free, then it must be properly closed. + // Hence it must be synced already! + if (inode->destruct) { + inode->destruct(inode); + } + + vfs_sb_unref(inode->sb); + hlist_delete(&inode->hash_list); + lru_remove(inode_lru, &inode->lru); + cake_release(inode_pile, inode); } /* ---- System call definition and support ---- */ -#define FLOCATE_CREATE_EMPTY 1 +// make a new name when not exists +#define FLOC_MAYBE_MKNAME 1 -#define DO_STATUS(errno) SYSCALL_ESTATUS(__current->k_status = errno) -#define DO_STATUS_OR_RETURN(errno) ({ errno < 0 ? DO_STATUS(errno) : errno; }) +// name must be non-exist and made. +#define FLOC_MKNAME 2 -#define TEST_FD(fd) (fd >= 0 && fd < VFS_MAX_FD) +// no follow symlink +#define FLOC_NOFOLLOW 4 int -__vfs_getfd(int fd, struct v_fd** fd_s) +vfs_getfd(int fd, struct v_fd** fd_s) { - if (TEST_FD(fd) && (*fd_s = __current->fdtable->fds[fd])) { - return 0; + struct v_fdtable* fdtab; + + if (!TEST_FD(fd)) { + return EBADF; } - return EBADF; + + fdtab = __current->fdtable; + + lock_fdtable(fdtab); + *fd_s = __current->fdtable->fds[fd]; + unlock_fdtable(fdtab); + + return !*fd_s ? EBADF : 0; } -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, - struct v_dnode** fdir, - struct v_dnode** file, + struct file_locator* floc, int options) { char name_str[VFS_NAME_MAXLEN]; + struct v_dnode *fdir, *file; struct hstr name = HSTR(name_str, 0); - int errno; - if ((errno = - vfs_walk(__current->cwd, path, fdir, &name, VFS_WALK_PARENT))) { - return errno; + int errno, woption = 0; + + if ((options & FLOC_NOFOLLOW)) { + woption |= VFS_WALK_NOFOLLOW; + options &= ~FLOC_NOFOLLOW; } - errno = vfs_walk(*fdir, name.value, file, NULL, 0); - if (errno != ENOENT || !(options & FLOCATE_CREATE_EMPTY)) { + floc->fresh = false; + name_str[0] = 0; + errno = vfs_walk_proc(path, &fdir, &name, woption | VFS_WALK_PARENT); + if (errno) { return errno; } - struct v_dnode* parent = *fdir; - struct v_dnode* file_new = vfs_d_alloc(); - hstrcpy(&file_new->name, &name); + lock_dnode(fdir); - if (!(errno = parent->inode->ops.create(parent->inode, file_new))) { - *file = file_new; + errno = vfs_walk(fdir, name.value, &file, NULL, woption); - vfs_dcache_add(parent, file_new); - llist_append(&parent->children, &file_new->siblings); - } else { - vfs_d_free(file_new); + if (errno && errno != ENOENT) { + goto error; + } + + if (!errno && (options & FLOC_MKNAME)) { + errno = EEXIST; + goto error; + } + + if (!errno) { + // the file present, no need to hold the directory lock + unlock_dnode(fdir); + goto done; + } + + // errno == ENOENT + if (!options) { + goto error; + } + + errno = vfs_check_writable(fdir); + if (errno) { + goto error; } + floc->fresh = true; + + file = vfs_d_alloc(fdir, &name); + + if (!file) { + errno = ENOMEM; + goto error; + } + + vfs_dcache_add(fdir, file); + +done: + floc->dir = fdir; + floc->file = file; + + return errno; + +error: + unlock_dnode(fdir); return errno; } + +static bool +__check_unlinkable(struct v_dnode* dnode) +{ + int acl; + bool wr_self, wr_parent; + struct v_dnode* parent; + + parent = dnode->parent; + acl = dnode->inode->acl; + + wr_self = check_allow_write(dnode->inode); + wr_parent = check_allow_write(parent->inode); + + if (!fsacl_test(acl, svtx)) { + return wr_self; + } + + if (current_euid() == dnode->inode->uid) { + return true; + } + + return wr_self && wr_parent; +} + int vfs_do_open(const char* path, int options) { - int errno, fd; + int errno, fd, loptions = 0; struct v_dnode *dentry, *file; - struct v_file* ofile = 0; + struct v_file* ofile = NULL; + struct file_locator floc; + struct v_inode* inode; + + if ((options & FO_CREATE)) { + loptions |= FLOC_MAYBE_MKNAME; + } else if ((options & FO_NOFOLLOW)) { + loptions |= FLOC_NOFOLLOW; + } + + errno = __vfs_try_locate_file(path, &floc, loptions); + + if (errno || (errno = vfs_alloc_fdslot(&fd))) { + return errno; + } + + file = floc.file; + dentry = floc.dir; + + if (floc.fresh) { + errno = __vfs_mknod(dentry->inode, file, VFS_IFFILE, NULL); + if (errno) { + vfs_d_free(file); + __floc_try_unlock(&floc); + return errno; + } + + __floc_try_unlock(&floc); + } - errno = __vfs_try_locate_file( - path, &dentry, &file, (options & FO_CREATE) ? FLOCATE_CREATE_EMPTY : 0); - if (errno || (errno = vfs_open(file, &ofile))) { + if ((errno = vfs_open(file, &ofile))) { return errno; } - struct v_inode* o_inode = ofile->inode; - if (!(o_inode->itype & VFS_IFSEQDEV) && !(options & FO_DIRECT)) { - // XXX Change here accordingly when signature of pcache_r/w changed. - ofile->ops.read = pcache_read; - ofile->ops.write = pcache_write; + inode = ofile->inode; + lock_inode(inode); + + struct v_fd* fd_s = cake_grab(fd_pile); + memset(fd_s, 0, sizeof(*fd_s)); + + if ((options & O_TRUNC)) { + file->inode->fsize = 0; } - if (!errno && !(errno = vfs_alloc_fdslot(&fd))) { - struct v_fd* fd_s = vzalloc(sizeof(*fd_s)); - ofile->f_pos = ofile->inode->fsize & -((options & FO_APPEND) != 0); - fd_s->file = ofile; - fd_s->flags = options; - __current->fdtable->fds[fd] = fd_s; - return fd; + if (vfs_get_dtype(inode->itype) == DT_DIR) { + ofile->f_pos = 0; } + + fd_s->file = ofile; + fd_s->flags = options; + __current->fdtable->fds[fd] = fd_s; - return errno; + unlock_inode(inode); + + return fd; } __DEFINE_LXSYSCALL2(int, open, const char*, path, int, options) @@ -690,17 +924,15 @@ __DEFINE_LXSYSCALL1(int, close, int, fd) { struct v_fd* fd_s; int errno = 0; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { goto done_err; } - if (fd_s->file->ref_count > 1) { - fd_s->file->ref_count--; - } else if ((errno = vfs_close(fd_s->file))) { + if ((errno = vfs_close(fd_s->file))) { goto done_err; } - vfree(fd_s); + cake_release(fd_pile, fd_s); __current->fdtable->fds[fd] = 0; done_err: @@ -713,18 +945,18 @@ __vfs_readdir_callback(struct dir_context* dctx, const int len, const int dtype) { - struct dirent* dent = (struct dirent*)dctx->cb_data; - strncpy(dent->d_name, name, DIRENT_NAME_MAX_LEN); + struct lx_dirent* dent = (struct lx_dirent*)dctx->cb_data; + strncpy(dent->d_name, name, MIN(len, DIRENT_NAME_MAX_LEN)); dent->d_nlen = len; dent->d_type = dtype; } -__DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent) +__DEFINE_LXSYSCALL2(int, sys_readdir, int, fd, struct lx_dirent*, dent) { struct v_fd* fd_s; int errno; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { goto done; } @@ -732,63 +964,75 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent) lock_inode(inode); - if (!(fd_s->file->inode->itype & VFS_IFDIR)) { + if (!check_directory_node(inode)) { errno = ENOTDIR; - } else { - struct dir_context dctx = - (struct dir_context){ .cb_data = dent, - .index = dent->d_offset, - .read_complete_callback = - __vfs_readdir_callback }; - if (dent->d_offset == 0) { - __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 0); - } else if (dent->d_offset == 1) { - __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0); - } else { - dctx.index -= 2; - if ((errno = fd_s->file->ops.readdir(inode, &dctx))) { - unlock_inode(inode); - goto done; - } - } - errno = 0; - dent->d_offset++; + goto unlock; + } + + if (!check_allow_read(inode)) { + errno = EPERM; + 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: - return DO_STATUS(errno); + return DO_STATUS_OR_RETURN(errno); } __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count) { int errno = 0; struct v_fd* fd_s; - if ((errno = __vfs_getfd(fd, &fd_s))) { + struct v_inode* inode; + + if ((errno = vfs_getfd(fd, &fd_s))) { goto done; } struct v_file* file = fd_s->file; - if ((file->inode->itype & VFS_IFDIR)) { + if (check_directory_node(file->inode)) { errno = EISDIR; goto done; } - lock_inode(file->inode); + if (!check_allow_read(file->inode)) { + errno = EPERM; + goto done; + } - file->inode->atime = clock_unixtime(); + inode = file->inode; + lock_inode(inode); - __SYSCALL_INTERRUPTIBLE( - { errno = file->ops.read(file->inode, buf, count, file->f_pos); }) + __vfs_touch_inode(inode, INODE_ACCESSED); - unlock_inode(file->inode); + if (check_seqdev_node(inode) || (fd_s->flags & FO_DIRECT)) { + errno = file->ops->read(inode, buf, count, file->f_pos); + } else { + errno = pcache_read(inode, buf, count, file->f_pos); + } if (errno > 0) { file->f_pos += errno; + unlock_inode(inode); return errno; } + unlock_inode(inode); + done: return DO_STATUS(errno); } @@ -797,30 +1041,46 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count) { int errno = 0; struct v_fd* fd_s; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { goto done; } + struct v_inode* inode; struct v_file* file = fd_s->file; - if ((file->inode->itype & VFS_IFDIR)) { - errno = EISDIR; + + if ((errno = vfs_check_writable(file->dnode))) { goto done; } - lock_inode(file->inode); + if (check_directory_node(file->inode)) { + errno = EISDIR; + goto done; + } - file->inode->mtime = clock_unixtime(); + inode = file->inode; + lock_inode(inode); - __SYSCALL_INTERRUPTIBLE( - { errno = file->ops.write(file->inode, buf, count, file->f_pos); }) + __vfs_touch_inode(inode, INODE_MODIFY); + if ((fd_s->flags & O_APPEND)) { + file->f_pos = inode->fsize; + } - unlock_inode(file->inode); + if (check_seqdev_node(inode) || (fd_s->flags & FO_DIRECT)) { + errno = file->ops->write(inode, buf, count, file->f_pos); + } else { + errno = pcache_write(inode, buf, count, file->f_pos); + } if (errno > 0) { file->f_pos += errno; + inode->fsize = MAX(inode->fsize, file->f_pos); + + unlock_inode(inode); return errno; } + unlock_inode(inode); + done: return DO_STATUS(errno); } @@ -829,31 +1089,52 @@ __DEFINE_LXSYSCALL3(int, lseek, int, fd, int, offset, int, options) { int errno = 0; struct v_fd* fd_s; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { goto done; } struct v_file* file = fd_s->file; + struct v_inode* inode = file->inode; - lock_inode(file->inode); + if (!file->ops->seek) { + errno = ENOTSUP; + goto done; + } + + if (!check_allow_read(inode)) { + errno = EPERM; + goto done; + } + + lock_inode(inode); - size_t 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: - fpos = (size_t)((int)file->f_pos + offset); + overflow = sadd_of((int)file->f_pos, offset, &fpos); break; case FSEEK_END: - fpos = (size_t)((int)file->inode->fsize + offset); + overflow = sadd_of((int)inode->fsize, offset, &fpos); break; case FSEEK_SET: fpos = offset; break; } - if (!file->ops.seek || !(errno = file->ops.seek(file->inode, fpos))) { - file->f_pos = fpos; + + if (overflow) { + errno = EOVERFLOW; + } + else { + errno = file->ops->seek(file, fpos); } - unlock_inode(file->inode); + unlock_inode(inode); done: return DO_STATUS(errno); @@ -867,23 +1148,27 @@ vfs_get_path(struct v_dnode* dnode, char* buf, size_t size, int depth) } if (depth > 64) { - return ELOOP; + return ENAMETOOLONG; } - size_t len = vfs_get_path(dnode->parent, buf, size, depth + 1); + size_t len = 0; + + if (dnode->parent != dnode) { + len = vfs_get_path(dnode->parent, buf, size, depth + 1); + } if (len >= size) { return len; } + if (!len || buf[len - 1] != VFS_PATH_DELIM) { + buf[len++] = VFS_PATH_DELIM; + } + size_t cpy_size = MIN(dnode->name.len, size - len); strncpy(buf + len, dnode->name.value, cpy_size); len += cpy_size; - if (len < size) { - buf[len++] = PATH_DELIM; - } - return len; } @@ -892,32 +1177,99 @@ vfs_readlink(struct v_dnode* dnode, char* buf, size_t size) { 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; + + if (!check_allow_read(inode)) { + return EPERM; + } + + lock_inode(inode); + + int errno = inode->ops->read_symlink(inode, &link); + if (errno >= 0) { + strncpy(buf, link, MIN(size, (size_t)errno)); + } + + unlock_inode(inode); + return errno; +} + +int +vfs_get_dtype(int itype) +{ + int dtype = DT_FILE; + if (check_itype(itype, VFS_IFSYMLINK)) { + dtype |= DT_SYMLINK; + } + + if (check_itype(itype, VFS_IFDIR)) { + dtype |= DT_DIR; + return dtype; + } + + // TODO other types + + return dtype; +} + +struct v_fdtable* +fdtable_create() +{ + struct v_fdtable* fdtab; + + fdtab = vzalloc(sizeof(struct v_fdtable)); + mutex_init(&fdtab->lock); + + return fdtab; +} + +void +fdtable_copy(struct v_fdtable* dest, struct v_fdtable* src) +{ + lock_fdtable(dest); + lock_fdtable(src); + + for (size_t i = 0; i < VFS_MAX_FD; i++) { + struct v_fd* fd = src->fds[i]; + if (!fd) + continue; + vfs_dup_fd(fd, &dest->fds[i]); + } + + unlock_fdtable(dest); + unlock_fdtable(src); +} + +void +fdtable_free(struct v_fdtable* table) +{ + assert(!mutex_on_hold(&table->lock)); + + vfree(table); } __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size) { int errno; struct v_fd* fd_s; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { goto done; } struct v_dnode* dnode; - errno = vfs_get_path(fd_s->file->dnode, buf, size, 0); - if (errno >= 0) { - return errno; - } + dnode = fd_s->file->dnode; + + lock_dnode(dnode); + errno = vfs_get_path(dnode, buf, size, 0); + unlock_dnode(dnode); done: return DO_STATUS(errno); @@ -927,8 +1279,7 @@ __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size) { int errno; struct v_dnode* dnode; - if (!(errno = - vfs_walk(__current->cwd, path, &dnode, NULL, VFS_WALK_NOFOLLOW))) { + if (!(errno = vfs_walk_proc(path, &dnode, NULL, VFS_WALK_NOFOLLOW))) { errno = vfs_readlink(dnode, buf, size); } @@ -939,23 +1290,17 @@ __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size) return DO_STATUS(errno); } -__DEFINE_LXSYSCALL4(int, - readlinkat, - int, - dirfd, - const char*, - pathname, - char*, - buf, - size_t, - size) +__DEFINE_LXSYSCALL4( + int, readlinkat, int, dirfd, const char*, pathname, char*, buf, size_t, size) { int errno; struct v_fd* fd_s; - if ((errno = __vfs_getfd(dirfd, &fd_s))) { + if ((errno = vfs_getfd(dirfd, &fd_s))) { goto done; } + pathname = pathname ? pathname : ""; + struct v_dnode* dnode; if (!(errno = vfs_walk( fd_s->file->dnode, pathname, &dnode, NULL, VFS_WALK_NOFOLLOW))) { @@ -982,21 +1327,27 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname) { int errno; struct v_dnode* dnode; - if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) { + if ((errno = vfs_walk_proc(pathname, &dnode, NULL, 0))) { return DO_STATUS(errno); } lock_dnode(dnode); - if (dnode->parent) - lock_dnode(dnode->parent); + if (!__check_unlinkable(dnode)) { + errno = EPERM; + goto done; + } + + if ((errno = vfs_check_writable(dnode))) { + goto done; + } if ((dnode->super_block->fs->types & FSTYPE_ROFS)) { errno = EROFS; goto done; } - if (dnode->ref_count || dnode->inode->open_count) { + if (dnode->ref_count > 1 || dnode->inode->open_count) { errno = EBUSY; goto done; } @@ -1006,83 +1357,111 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname) goto done; } - lock_inode(dnode->inode); + struct v_dnode* parent = dnode->parent; - if ((dnode->inode->itype & VFS_IFDIR)) { - errno = dnode->inode->ops.rmdir(dnode->inode); + if (!parent) { + errno = EINVAL; + goto done; + } + + lock_dnode(parent); + lock_inode(parent->inode); + + if (check_directory_node(dnode->inode)) { + errno = parent->inode->ops->rmdir(parent->inode, dnode); if (!errno) { vfs_dcache_remove(dnode); - unlock_inode(dnode->inode); - vfs_d_free(dnode); - - goto done; } } 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) { - struct v_dnode *parent, *dir = vfs_d_alloc(); - int errno = - vfs_walk(__current->cwd, path, &parent, &dir->name, VFS_WALK_PARENT); - if (errno) { + int errno; + struct hstr name; + struct v_inode* inode; + struct v_dnode *parent, *dir; + char name_value[VFS_NAME_MAXLEN]; + + name = HHSTR(name_value, 0, 0); + + if ((errno = vfs_walk_proc(path, &parent, &name, VFS_WALK_PARENT))) { + goto done; + } + + if (!(errno = vfs_walk(parent, name_value, &dir, NULL, 0))) { + errno = EEXIST; + goto done; + } + + if ((errno = vfs_check_writable(parent))) { + goto done; + } + + if (!(dir = vfs_d_alloc(parent, &name))) { + errno = ENOMEM; goto done; } + inode = parent->inode; + lock_dnode(parent); - lock_inode(parent->inode); + lock_inode(inode); if ((parent->super_block->fs->types & FSTYPE_ROFS)) { errno = ENOTSUP; - } else if (!parent->inode->ops.mkdir) { + } else if (!inode->ops->mkdir) { errno = ENOTSUP; - } else if (!(parent->inode->itype & VFS_IFDIR)) { + } else if (!check_directory_node(inode)) { errno = ENOTDIR; - } else if (!(errno = parent->inode->ops.mkdir(parent->inode, dir))) { - llist_append(&parent->children, &dir->siblings); + } else if (!(errno = inode->ops->mkdir(inode, dir))) { + vfs_dcache_add(parent, dir); goto cleanup; } vfs_d_free(dir); cleanup: - unlock_inode(parent->inode); + unlock_inode(inode); unlock_dnode(parent); done: return DO_STATUS(errno); } -int +static int __vfs_do_unlink(struct v_dnode* dnode) { + int errno; struct v_inode* inode = dnode->inode; if (dnode->ref_count > 1) { return EBUSY; } + if (!__check_unlinkable(dnode)) { + return EPERM; + } + + if ((errno = vfs_check_writable(dnode))) { + return errno; + } + lock_inode(inode); - int errno; if (inode->open_count) { errno = EBUSY; - } else if (!(inode->itype & VFS_IFDIR)) { - // The underlying unlink implementation should handle - // symlink case - errno = inode->ops.unlink(inode); + } else if (!check_directory_node(inode)) { + errno = inode->ops->unlink(inode, dnode); if (!errno) { - inode->link_count--; - vfs_dcache_remove(dnode); vfs_d_free(dnode); } } else { @@ -1098,11 +1477,7 @@ __DEFINE_LXSYSCALL1(int, unlink, const char*, pathname) { int errno; struct v_dnode* dnode; - if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) { - goto done; - } - if ((dnode->super_block->fs->types & FSTYPE_ROFS)) { - errno = EROFS; + if ((errno = vfs_walk_proc(pathname, &dnode, NULL, 0))) { goto done; } @@ -1116,7 +1491,7 @@ __DEFINE_LXSYSCALL2(int, unlinkat, int, fd, const char*, pathname) { int errno; struct v_fd* fd_s; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { goto done; } @@ -1132,18 +1507,30 @@ done: __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, &floc, 0); + if (errno) { + goto done; + } + + __floc_try_unlock(&floc); - errno = __vfs_try_locate_file(oldpath, &dentry, &to_link, 0); + to_link = floc.file; + errno = __vfs_try_locate_file(newpath, &floc, FLOC_MKNAME); if (!errno) { - errno = __vfs_try_locate_file( - newpath, &name_dentry, &name_file, FLOCATE_CREATE_EMPTY); - if (!errno) { - errno = EEXIST; - } else if (name_file) { - errno = vfs_link(to_link, name_file); - } + goto done; } + + name_file = floc.file; + errno = vfs_link(to_link, name_file); + if (errno) { + vfs_d_free(name_file); + } + +done: + __floc_try_unlock(&floc); return DO_STATUS(errno); } @@ -1151,7 +1538,8 @@ __DEFINE_LXSYSCALL1(int, fsync, int, fildes) { int errno; struct v_fd* fd_s; - if (!(errno = __vfs_getfd(fildes, &fd_s))) { + + if (!(errno = vfs_getfd(fildes, &fd_s))) { errno = vfs_fsync(fd_s->file); } @@ -1166,7 +1554,7 @@ vfs_dup_fd(struct v_fd* old, struct v_fd** new) memcpy(copied, old, sizeof(struct v_fd)); - atomic_fetch_add(&old->file->ref_count, 1); + vfs_ref_file(old->file); *new = copied; @@ -1176,13 +1564,15 @@ vfs_dup_fd(struct v_fd* old, struct v_fd** new) int vfs_dup2(int oldfd, int newfd) { + int errno; + struct v_fdtable* fdtab; + struct v_fd *oldfd_s, *newfd_s; + if (newfd == oldfd) { return newfd; } - int errno; - struct v_fd *oldfd_s, *newfd_s; - if ((errno = __vfs_getfd(oldfd, &oldfd_s))) { + if ((errno = vfs_getfd(oldfd, &oldfd_s))) { goto done; } @@ -1191,16 +1581,26 @@ vfs_dup2(int oldfd, int newfd) goto done; } - newfd_s = __current->fdtable->fds[newfd]; + fdtab = __current->fdtable; + lock_fdtable(fdtab); + + newfd_s = fdtab->fds[newfd]; if (newfd_s && (errno = vfs_close(newfd_s->file))) { - goto done; + goto unlock_and_done; } - if (!(errno = vfs_dup_fd(oldfd_s, &newfd_s))) { - __current->fdtable->fds[newfd] = newfd_s; - return newfd; + if ((errno = vfs_dup_fd(oldfd_s, &newfd_s))) { + goto unlock_and_done; } + fdtab->fds[newfd] = newfd_s; + + unlock_fdtable(fdtab); + return newfd; + +unlock_and_done: + unlock_fdtable(fdtab); + done: return DO_STATUS(errno); } @@ -1214,7 +1614,7 @@ __DEFINE_LXSYSCALL1(int, dup, int, oldfd) { int errno, newfd; struct v_fd *oldfd_s, *newfd_s; - if ((errno = __vfs_getfd(oldfd, &oldfd_s))) { + if ((errno = vfs_getfd(oldfd, &oldfd_s))) { goto done; } @@ -1228,59 +1628,79 @@ done: return DO_STATUS(errno); } -__DEFINE_LXSYSCALL2(int, - symlink, - const char*, - pathname, - const char*, - link_target) +__DEFINE_LXSYSCALL2( + int, symlink, const char*, pathname, const char*, link_target) { int errno; - struct v_dnode* dnode; - if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) { + struct file_locator floc; + struct v_dnode *file; + struct v_inode *f_ino; + + errno = __vfs_try_locate_file(pathname, &floc, FLOC_MKNAME); + if (errno) { goto done; } - if ((dnode->super_block->fs->types & FSTYPE_ROFS)) { - errno = EROFS; + + file = floc.file; + errno = __vfs_mknod(floc.dir->inode, file, VFS_IFSYMLINK, NULL); + if (errno) { + vfs_d_free(file); + goto done; + } + + f_ino = file->inode; + + assert(f_ino); + + errno = vfs_check_writable(file); + if (errno) { goto done; } - if (!dnode->inode->ops.symlink) { + + if (!f_ino->ops->set_symlink) { errno = ENOTSUP; goto done; } - lock_inode(dnode->inode); + lock_inode(f_ino); - errno = dnode->inode->ops.symlink(dnode->inode, link_target); + errno = f_ino->ops->set_symlink(f_ino, link_target); - unlock_inode(dnode->inode); + unlock_inode(f_ino); done: + __floc_try_unlock(&floc); return DO_STATUS(errno); } -int -__vfs_do_chdir(struct v_dnode* dnode) +static int +vfs_do_chdir_nolock(struct proc_info* proc, struct v_dnode* dnode) { - int errno = 0; - - lock_dnode(dnode); - - if (!(dnode->inode->itype & VFS_IFDIR)) { - errno = ENOTDIR; - goto done; + if (!check_directory_node(dnode->inode)) { + return ENOTDIR; } - if (__current->cwd) { - atomic_fetch_add(&__current->cwd->ref_count, 1); + if (proc->cwd) { + vfs_unref_dnode(proc->cwd); } - atomic_fetch_sub(&dnode->ref_count, 1); - __current->cwd = dnode; + vfs_ref_dnode(dnode); + proc->cwd = dnode; + + return 0; +} + +static int +vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode) +{ + int errno = 0; + + lock_dnode(dnode); + + errno = vfs_do_chdir_nolock(proc, dnode); unlock_dnode(dnode); -done: return errno; } @@ -1289,11 +1709,11 @@ __DEFINE_LXSYSCALL1(int, chdir, const char*, path) struct v_dnode* dnode; int errno = 0; - if ((errno = vfs_walk(__current->cwd, path, &dnode, NULL, 0))) { + if ((errno = vfs_walk_proc(path, &dnode, NULL, 0))) { goto done; } - errno = __vfs_do_chdir(dnode); + errno = vfs_do_chdir((struct proc_info*)__current, dnode); done: return DO_STATUS(errno); @@ -1304,11 +1724,36 @@ __DEFINE_LXSYSCALL1(int, fchdir, int, fd) struct v_fd* fd_s; int errno = 0; - if ((errno = __vfs_getfd(fd, &fd_s))) { + if ((errno = vfs_getfd(fd, &fd_s))) { + goto done; + } + + errno = vfs_do_chdir((struct proc_info*)__current, fd_s->file->dnode); + +done: + return DO_STATUS(errno); +} + + +__DEFINE_LXSYSCALL1(int, chroot, const char*, path) +{ + int errno; + struct v_dnode* dnode; + if ((errno = vfs_walk_proc(path, &dnode, NULL, 0))) { + return errno; + } + + lock_dnode(dnode); + + errno = vfs_do_chdir_nolock(__current, dnode); + if (errno) { + unlock_dnode(dnode); goto done; } - errno = __vfs_do_chdir(fd_s->file->dnode); + __current->root = dnode; + + unlock_dnode(dnode); done: return DO_STATUS(errno); @@ -1326,7 +1771,7 @@ __DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size) size_t len = 0; if (!__current->cwd) { - *buf = PATH_DELIM; + *buf = VFS_PATH_DELIM; len = 1; } else { len = vfs_get_path(__current->cwd, buf, size, 0); @@ -1336,23 +1781,28 @@ __DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size) } } - buf[len + 1] = '\0'; + buf[len] = '\0'; ret_ptr = buf; done: - __current->k_status = errno; + syscall_result(errno); return ret_ptr; } int vfs_do_rename(struct v_dnode* current, struct v_dnode* target) { + int errno = 0; if (current->inode->id == target->inode->id) { // hard link return 0; } + if ((errno = vfs_check_writable(current))) { + return errno; + } + if (current->ref_count > 1 || target->ref_count > 1) { return EBUSY; } @@ -1361,13 +1811,12 @@ vfs_do_rename(struct v_dnode* current, struct v_dnode* target) 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) @@ -1379,7 +1828,8 @@ vfs_do_rename(struct v_dnode* current, struct v_dnode* target) goto cleanup; } - if ((errno = current->inode->ops.rename(current->inode, current, target))) { + if ((errno = + current->inode->ops->rename(current->inode, current, target))) { unlock_dnode(target); goto cleanup; } @@ -1389,12 +1839,13 @@ vfs_do_rename(struct v_dnode* current, struct v_dnode* target) vfs_dcache_rehash(newparent, current); // detach target - vfs_dcache_remove(target); + vfs_d_free(target); unlock_dnode(target); cleanup: unlock_dnode(current); + if (oldparent) unlock_dnode(oldparent); if (newparent) @@ -1409,7 +1860,7 @@ __DEFINE_LXSYSCALL2(int, rename, const char*, oldpath, const char*, newpath) struct hstr name = HSTR(valloc(VFS_NAME_MAXLEN), 0); int errno = 0; - if ((errno = vfs_walk(__current->cwd, oldpath, &cur, NULL, 0))) { + if ((errno = vfs_walk_proc(oldpath, &cur, NULL, 0))) { goto done; } @@ -1420,62 +1871,172 @@ __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); + target = vfs_d_alloc(target_parent, &name); + vfs_dcache_add(target_parent, target); } else if (errno) { goto done; } - if (!(errno = vfs_do_rename(cur, target))) { - vfs_d_free(target); + if (!target) { + errno = ENOMEM; + goto done; } + errno = vfs_do_rename(cur, target); + done: - vfree(name.value); + vfree((void*)name.value); return DO_STATUS(errno); } -__DEFINE_LXSYSCALL3(int, - mount, - const char*, - source, - const char*, - target, - const char*, - fstype) +__DEFINE_LXSYSCALL2(int, fstat, int, fd, struct file_stat*, stat) { - struct v_dnode *dev, *mnt; int errno = 0; + struct v_fd* fds; - if ((errno = vfs_walk(__current->cwd, source, &dev, NULL, 0))) { + if ((errno = vfs_getfd(fd, &fds))) { goto done; } - if ((errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) { - goto done; + struct v_inode* vino = fds->file->inode; + struct device* fdev = vino->sb->dev; + + stat->st_ino = vino->id; + stat->st_blocks = vino->lb_usage; + stat->st_size = vino->fsize; + stat->st_blksize = vino->sb->blksize; + stat->st_nlink = vino->link_count; + stat->st_uid = vino->uid; + stat->st_gid = vino->gid; + + stat->st_ctim = vino->ctime; + stat->st_atim = vino->atime; + stat->st_mtim = vino->mtime; + + stat->st_mode = (vino->itype << 16) | vino->acl; + + stat->st_ioblksize = PAGE_SIZE; + + if (check_device_node(vino)) { + struct device* rdev = resolve_device(vino->data); + if (!rdev) { + errno = EINVAL; + goto done; + } + + stat->st_rdev = (dev_t){.meta = rdev->ident.fn_grp, + .unique = rdev->ident.unique, + .index = dev_uid(rdev) }; + } + + if (fdev) { + stat->st_dev = (dev_t){.meta = fdev->ident.fn_grp, + .unique = fdev->ident.unique, + .index = dev_uid(fdev) }; } - if (!(dev->inode->itype & VFS_IFVOLDEV)) { - errno = ENOTDEV; +done: + return DO_STATUS(errno); +} + +__DEFINE_LXSYSCALL4(int, fchmodat, int, fd, + const char*, path, int, mode, int, flags) +{ + int errno; + struct v_dnode *dnode; + struct v_inode* inode; + + errno = vfs_walkat(fd, path, flags, &dnode); + if (errno) { goto done; } - if (mnt->ref_count > 1) { - errno = EBUSY; + errno = vfs_check_writable(dnode); + if (errno) { + return errno; + } + + inode = dnode->inode; + lock_inode(inode); + + if (!current_is_root()) { + mode = mode & FSACL_RWXMASK; + } + + inode->acl = mode; + __vfs_touch_inode(inode, INODE_MODIFY); + + unlock_inode(inode); + +done: + return DO_STATUS(errno); +} + +__DEFINE_LXSYSCALL5(int, fchownat, int, fd, + const char*, path, uid_t, uid, gid_t, gid, int, flags) +{ + int errno; + struct v_dnode *dnode; + struct v_inode *inode; + + errno = vfs_walkat(fd, path, flags, &dnode); + if (errno) { goto done; } - // FIXME should not touch the underlying fs! - struct device* device = - (struct device*)((struct twifs_node*)dev->inode->data)->data; + errno = vfs_check_writable(dnode); + if (errno) { + return errno; + } + + inode = dnode->inode; + lock_inode(inode); + + inode->uid = uid; + inode->gid = gid; + __vfs_touch_inode(inode, INODE_MODIFY); - errno = vfs_mount_at(fstype, device, mnt); + unlock_inode(inode); done: return DO_STATUS(errno); } -__DEFINE_LXSYSCALL1(int, unmount, const char*, target) +__DEFINE_LXSYSCALL4(int, faccessat, int, fd, + const char*, path, int, amode, int, flags) { - return vfs_unmount(target); + int errno, acl; + struct v_dnode *dnode; + struct v_inode *inode; + struct user_scope* uscope; + + uid_t tuid; + gid_t tgid; + + errno = vfs_walkat(fd, path, flags, &dnode); + if (errno) { + goto done; + } + + if ((flags & AT_EACCESS)) { + tuid = current_euid(); + tgid = current_egid(); + } + else { + uscope = current_user_scope(); + tuid = uscope->ruid; + tgid = uscope->rgid; + } + + inode = dnode->inode; + + acl = inode->acl; + acl &= amode; + acl &= check_acl_between(inode->uid, inode->gid, tuid, tgid); + if (!acl) { + errno = EACCESS; + } + +done: + return DO_STATUS(errno); } \ No newline at end of file