#include <lunaix/spike.h>
#include <lunaix/syscall.h>
+#include <lunaix/fs/twifs.h>
+
#define PATH_DELIM '/'
#define DNODE_HASHTABLE_BITS 10
#define DNODE_HASHTABLE_SIZE (1 << DNODE_HASHTABLE_BITS)
#define DNODE_HASH_MASK (DNODE_HASHTABLE_SIZE - 1)
#define DNODE_HASHBITS (32 - DNODE_HASHTABLE_BITS)
+#define 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)
+
static struct cake_pile* dnode_pile;
static struct cake_pile* inode_pile;
static struct cake_pile* file_pile;
}
inline struct hbucket*
-__dcache_get_bucket(struct v_dnode* parent, unsigned int hash)
+__dcache_hash(struct v_dnode* parent, uint32_t* hash)
{
+ uint32_t _hash = *hash;
// 与parent的指针值做加法，来减小碰撞的可能性。
- hash += (uint32_t)parent;
+ _hash += (uint32_t)parent;
// 确保低位更加随机
- hash = hash ^ (hash >> DNODE_HASHBITS);
- return &dnode_cache[hash & DNODE_HASH_MASK];
+ _hash = _hash ^ (_hash >> DNODE_HASHBITS);
+ *hash = _hash;
+ return &dnode_cache[_hash & DNODE_HASH_MASK];
}
struct v_dnode*
return parent->parent ? parent->parent : parent;
}
- struct hbucket* slot = __dcache_get_bucket(parent, str->hash);
+ uint32_t hash = str->hash;
+ struct hbucket* slot = __dcache_hash(parent, &hash);
struct v_dnode *pos, *n;
hashtable_bucket_foreach(slot, pos, n, hash_list)
{
- if (pos->name.hash == str->hash) {
+ if (pos->name.hash == hash) {
return pos;
}
}
void
vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
{
- struct hbucket* bucket = __dcache_get_bucket(parent, dnode->name.hash);
+ 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)
+{
+ hlist_delete(&dnode->hash_list);
+ hstr_rehash(&dnode->name, HSTR_FULL_HASH);
+ vfs_dcache_add(parent, dnode);
+}
+
int
__vfs_walk(struct v_dnode* start,
const char* path,
goto cont;
}
+ lock_dnode(current_level);
+
name_content[j] = 0;
name.len = j;
hstr_rehash(&name, HSTR_FULL_HASH);
component->len = j;
strcpy(component->value, name_content);
}
+ unlock_dnode(current_level);
break;
}
if (!dnode) {
dnode = vfs_d_alloc();
- dnode->name = HHSTR(valloc(VFS_NAME_MAXLEN), j, name.hash);
- strcpy(dnode->name.value, name_content);
+ hstrcpy(&dnode->name, &name);
+
+ lock_inode(current_level->inode);
errno =
current_level->inode->ops.dir_lookup(current_level->inode, dnode);
}
}
+ unlock_inode(current_level->inode);
+
if (errno) {
+ unlock_dnode(current_level);
+ vfree(dnode->name.value);
goto error;
}
llist_append(¤t_level->children, &dnode->siblings);
}
+ unlock_dnode(current_level);
+
j = 0;
current_level = dnode;
cont:
return 0;
error:
- vfree(dnode->name.value);
vfs_d_free(dnode);
*dentry = NULL;
return errno;
int options)
{
struct v_dnode* interim;
- char* pathname = path;
+ const char* pathname = path;
int errno = __vfs_walk(start, path, &interim, component, options);
int counter = 0;
}
int
-vfs_mount(const char* target, const char* fs_name, bdev_t device)
+vfs_mount(const char* target, const char* fs_name, struct device* device)
{
int errno;
struct v_dnode* mnt;
}
int
-vfs_mount_at(const char* fs_name, bdev_t device, struct v_dnode* mnt_point)
+vfs_mount_at(const char* fs_name,
+ struct device* device,
+ struct v_dnode* mnt_point)
{
+ if (!(mnt_point->inode->itype & VFS_IFDIR)) {
+ return ENOTDIR;
+ }
+
struct filesystem* fs = fsm_get(fs_name);
- if (!fs)
+ if (!fs) {
return ENODEV;
+ }
+
struct v_superblock* sb = vfs_sb_alloc();
sb->dev = device;
sb->fs_id = fs_id++;
int
vfs_unmount_at(struct v_dnode* mnt_point)
{
+ // FIXME mnt point check & deal with the detached dcache subtree
int errno = 0;
struct v_superblock* sb = mnt_point->super_block;
if (!sb) {
struct v_dnode* fs_root = sb->root;
llist_delete(&fs_root->siblings);
llist_delete(&sb->sb_list);
+ hlist_delete(&fs_root->hash_list);
vfs_sb_free(sb);
+ vfs_d_free(fs_root);
}
return errno;
}
vfile->dnode = dnode;
vfile->inode = inode;
- vfile->ref_count = 1;
+ vfile->ref_count = ATOMIC_VAR_INIT(1);
vfile->ops = inode->default_fops;
if ((inode->itype & VFS_IFFILE) && !inode->pg_cache) {
struct pcache* pcache = vzalloc(sizeof(struct pcache));
pcache_init(pcache);
+ pcache->master = inode;
inode->pg_cache = pcache;
}
if (errno) {
cake_release(file_pile, vfile);
} else {
- dnode->ref_count++;
+ atomic_fetch_add(&dnode->ref_count, 1);
inode->open_count++;
+
*file = vfile;
}
vfs_link(struct v_dnode* to_link, struct v_dnode* name)
{
int errno;
+
+ lock_inode(to_link->inode);
if (to_link->super_block->root != name->super_block->root) {
errno = EXDEV;
} else if (!to_link->inode->ops.link) {
errno = ENOTSUP;
} else if (!(errno = to_link->inode->ops.link(to_link->inode, name))) {
name->inode = to_link->inode;
- to_link->inode->link_count++;
+ atomic_fetch_add(&to_link->inode->link_count, 1);
}
+ unlock_inode(to_link->inode);
return errno;
}
{
int errno = 0;
if (!file->ops.close || !(errno = file->ops.close(file))) {
- file->dnode->ref_count--;
+ atomic_fetch_sub(&file->dnode->ref_count, 1);
file->inode->open_count--;
- pcache_commit_all(file);
+
+ 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);
+ pcache_commit_all(file->inode);
if (file->ops.sync) {
- errno = file->ops.sync(file);
- }
- if (!errno && file->inode->ops.sync) {
- return file->inode->ops.sync(file->inode);
+ errno = file->ops.sync(file->inode);
}
+
+ unlock_inode(file->inode);
+
return errno;
}
struct v_dnode* dnode = cake_grab(dnode_pile);
memset(dnode, 0, sizeof(*dnode));
llist_init_head(&dnode->children);
- dnode->name = vfs_empty;
+ llist_init_head(&dnode->siblings);
+ mutex_init(&dnode->lock);
+
+ dnode->ref_count = ATOMIC_VAR_INIT(0);
+ dnode->name = HHSTR(vzalloc(VFS_NAME_MAXLEN), 0, 0);
+
return dnode;
}
void
vfs_d_free(struct v_dnode* dnode)
{
- if (dnode->ops.destruct) {
- dnode->ops.destruct(dnode);
+ if (dnode->inode && dnode->inode->link_count) {
+ dnode->inode->link_count--;
}
+ vfree(dnode->name.value);
cake_release(dnode_pile, dnode);
}
{
struct v_inode* inode = cake_grab(inode_pile);
memset(inode, 0, sizeof(*inode));
+ inode->link_count = 1;
+ mutex_init(&inode->lock);
+
return inode;
}
return errno;
}
- if (!(errno = (*fdir)->inode->ops.create((*fdir)->inode))) {
- struct v_dnode* file_new;
- file_new = vfs_d_alloc();
- file_new->name = HHSTR(valloc(VFS_NAME_MAXLEN), name.len, name.hash);
- strcpy(file_new->name.value, name_str);
+ struct v_dnode* parent = *fdir;
+ struct v_dnode* file_new = vfs_d_alloc();
+ hstrcpy(&file_new->name, &name);
+
+ if (!(errno = parent->inode->ops.create(parent->inode, file_new))) {
*file = file_new;
- llist_append(&(*fdir)->children, &file_new->siblings);
+ vfs_dcache_add(parent, file_new);
+ llist_append(&parent->children, &file_new->siblings);
+ } else {
+ vfs_d_free(file_new);
}
return errno;
struct v_dnode *dentry, *file;
struct v_file* ofile = 0;
- errno = __vfs_try_locate_file(path, &dentry, &file, 0);
+ errno = __vfs_try_locate_file(
+ path, &dentry, &file, (options & FO_CREATE) ? FLOCATE_CREATE_EMPTY : 0);
- if (errno != ENOENT && (options & FO_CREATE)) {
- errno = dentry->inode->ops.create(dentry->inode);
- }
-
- if (!errno && (errno = vfs_open(file, &ofile))) {
+ if (errno || (errno = vfs_open(file, &ofile))) {
return errno;
}
goto done;
}
+ struct v_inode* inode = fd_s->file->inode;
+
+ lock_inode(inode);
+
if (!(fd_s->file->inode->itype & VFS_IFDIR)) {
errno = ENOTDIR;
} else {
__vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0);
} else {
dctx.index -= 2;
- if ((errno = fd_s->file->ops.readdir(fd_s->file, &dctx))) {
+ if ((errno = fd_s->file->ops.readdir(inode, &dctx))) {
+ unlock_inode(inode);
goto done;
}
}
dent->d_offset++;
}
-done:
- return DO_STATUS(errno);
-}
-
-__DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
-{
- struct v_dnode *parent, *dir;
- struct hstr component = HSTR(valloc(VFS_NAME_MAXLEN), 0);
- int errno =
- vfs_walk(__current->cwd, path, &parent, &component, VFS_WALK_PARENT);
- if (errno) {
- goto done;
- }
-
- if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
- errno = ENOTSUP;
- } else if (!parent->inode->ops.mkdir) {
- errno = ENOTSUP;
- } else if (!(parent->inode->itype & VFS_IFDIR)) {
- errno = ENOTDIR;
- } else {
- dir = vfs_d_alloc();
- dir->name = component;
- if (!(errno = parent->inode->ops.mkdir(parent->inode, dir))) {
- llist_append(&parent->children, &dir->siblings);
- } else {
- vfs_d_free(dir);
- vfree(component.value);
- }
- }
+ unlock_inode(inode);
done:
return DO_STATUS(errno);
goto done;
}
+ lock_inode(file->inode);
+
+ file->inode->atime = clock_unixtime();
+
__SYSCALL_INTERRUPTIBLE(
- { errno = file->ops.read(file, buf, count, file->f_pos); })
+ { errno = file->ops.read(file->inode, buf, count, file->f_pos); })
+
+ unlock_inode(file->inode);
if (errno > 0) {
file->f_pos += errno;
goto done;
}
+ lock_inode(file->inode);
+
+ file->inode->mtime = clock_unixtime();
+
__SYSCALL_INTERRUPTIBLE(
- { errno = file->ops.write(file, buf, count, file->f_pos); })
+ { errno = file->ops.write(file->inode, buf, count, file->f_pos); })
+
+ unlock_inode(file->inode);
if (errno > 0) {
file->f_pos += errno;
}
struct v_file* file = fd_s->file;
+
+ lock_inode(file->inode);
+
size_t fpos = file->f_pos;
switch (options) {
case FSEEK_CUR:
fpos = offset;
break;
}
- if (!file->ops.seek || !(errno = file->ops.seek(file, fpos))) {
+ if (!file->ops.seek || !(errno = file->ops.seek(file->inode, fpos))) {
file->f_pos = fpos;
}
+ unlock_inode(file->inode);
+
done:
return DO_STATUS(errno);
}
int
vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
{
- char* link;
- if (dnode->inode->ops.read_symlink) {
- int errno = dnode->inode->ops.read_symlink(dnode->inode, &link);
+ const char* link;
+ struct v_inode* inode = dnode->inode;
+ if (inode->ops.read_symlink) {
+ lock_inode(inode);
+
+ int errno = inode->ops.read_symlink(inode, &link);
strncpy(buf, link, size);
+
+ unlock_inode(inode);
return errno;
}
return 0;
return DO_STATUS(errno);
}
+/*
+ NOTE
+ When we perform operation that could affect the layout of
+ directory (i.e., rename, mkdir, rmdir). We must lock the parent dir
+ whenever possible. This will blocking any ongoing path walking to reach
+ it hence avoid any partial state.
+*/
+
__DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
{
int errno;
struct v_dnode* dnode;
if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
- goto done;
+ return DO_STATUS(errno);
}
+
+ lock_dnode(dnode);
+
+ if (dnode->parent)
+ lock_dnode(dnode->parent);
+
if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
errno = EROFS;
goto done;
}
- if (dnode->inode->open_count) {
+ if (dnode->ref_count || dnode->inode->open_count) {
errno = EBUSY;
goto done;
}
+ if (!llist_empty(&dnode->children)) {
+ errno = ENOTEMPTY;
+ goto done;
+ }
+
+ lock_inode(dnode->inode);
+
if ((dnode->inode->itype & VFS_IFDIR)) {
errno = dnode->inode->ops.rmdir(dnode->inode);
+ if (!errno) {
+ llist_delete(&dnode->siblings);
+ hlist_delete(&dnode->hash_list);
+ unlock_inode(dnode->inode);
+ vfs_d_free(dnode);
+
+ goto done;
+ }
} else {
errno = ENOTDIR;
}
+ unlock_inode(dnode->inode);
+
+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) {
+ goto done;
+ }
+
+ lock_dnode(parent);
+ lock_inode(parent->inode);
+
+ if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
+ errno = ENOTSUP;
+ } else if (!parent->inode->ops.mkdir) {
+ errno = ENOTSUP;
+ } else if (!(parent->inode->itype & VFS_IFDIR)) {
+ errno = ENOTDIR;
+ } else if (!(errno = parent->inode->ops.mkdir(parent->inode, dir))) {
+ llist_append(&parent->children, &dir->siblings);
+ goto cleanup;
+ }
+
+ vfs_d_free(dir);
+
+cleanup:
+ unlock_inode(parent->inode);
+ unlock_dnode(parent);
done:
return DO_STATUS(errno);
}
int
-__vfs_do_unlink(struct v_inode* inode)
+__vfs_do_unlink(struct v_dnode* dnode)
{
+ struct v_inode* inode = dnode->inode;
+
+ if (dnode->ref_count) {
+ return EBUSY;
+ }
+
+ lock_inode(inode);
+
int errno;
if (inode->open_count) {
errno = EBUSY;
} else if (!(inode->itype & VFS_IFDIR)) {
- // TODO handle symbolic link and type other than regular file
+ // The underlying unlink implementation should handle
+ // symlink case
errno = inode->ops.unlink(inode);
if (!errno) {
inode->link_count--;
+ llist_delete(&dnode->siblings);
+ hlist_delete(&dnode->hash_list);
+ vfs_d_free(dnode);
}
} else {
errno = EISDIR;
}
+ unlock_inode(inode);
+
return errno;
}
goto done;
}
- errno = __vfs_do_unlink(dnode->inode);
+ errno = __vfs_do_unlink(dnode);
done:
return DO_STATUS(errno);
struct v_dnode* dnode;
if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {
- errno = __vfs_do_unlink(dnode->inode);
+ errno = __vfs_do_unlink(dnode);
}
done:
struct v_fd* copied = cake_grab(fd_pile);
memcpy(copied, old, sizeof(struct v_fd));
- old->file->ref_count++;
+
+ atomic_fetch_add(&old->file->ref_count, 1);
*new = copied;
}
newfd_s = __current->fdtable->fds[newfd];
- if (newfd_s && (errno = vfs_close(newfd_s))) {
+ if (newfd_s && (errno = vfs_close(newfd_s->file))) {
goto done;
}
goto done;
}
+ lock_inode(dnode->inode);
+
errno = dnode->inode->ops.symlink(dnode->inode, link_target);
+ unlock_inode(dnode->inode);
+
done:
return DO_STATUS(errno);
}
__vfs_do_chdir(struct v_dnode* dnode)
{
int errno = 0;
+
+ lock_dnode(dnode);
+
if (!(dnode->inode->itype & VFS_IFDIR)) {
errno = ENOTDIR;
goto done;
__current->cwd->ref_count--;
}
- dnode->ref_count++;
+ dnode->ref_count--;
__current->cwd = dnode;
+ unlock_dnode(dnode);
+
done:
return errno;
}
done:
__current->k_status = errno;
return ret_ptr;
+}
+
+int
+vfs_do_rename(struct v_dnode* current, struct v_dnode* target)
+{
+ if (current->inode->id == target->inode->id) {
+ // hard link
+ return 0;
+ }
+
+ if (current->ref_count || target->ref_count) {
+ return EBUSY;
+ }
+
+ if (current->super_block != target->super_block) {
+ return EXDEV;
+ }
+
+ int errno = 0;
+
+ struct v_dnode* oldparent = current->parent;
+ struct v_dnode* newparent = target->parent;
+
+ lock_dnode(current);
+ lock_dnode(target);
+ if (oldparent)
+ lock_dnode(oldparent);
+ if (newparent)
+ lock_dnode(newparent);
+
+ if (!llist_empty(&target->children)) {
+ errno = ENOTEMPTY;
+ unlock_dnode(target);
+ goto cleanup;
+ }
+
+ if ((errno = current->inode->ops.rename(current->inode, current, target))) {
+ unlock_dnode(target);
+ goto cleanup;
+ }
+
+ // re-position current
+ current->parent = newparent;
+ hstrcpy(¤t->name, &target->name);
+ llist_delete(¤t->siblings);
+ llist_append(&newparent->children, ¤t->siblings);
+ vfs_dcache_rehash(newparent, current);
+
+ // detach target
+ llist_delete(&target->siblings);
+ hlist_delete(&target->hash_list);
+
+ unlock_dnode(target);
+
+cleanup:
+ unlock_dnode(current);
+ if (oldparent)
+ unlock_dnode(oldparent);
+ if (newparent)
+ unlock_dnode(newparent);
+
+ return errno;
+}
+
+__DEFINE_LXSYSCALL2(int, rename, const char*, oldpath, const char*, newpath)
+{
+ struct v_dnode *cur, *target_parent, *target;
+ struct hstr name = HSTR(valloc(VFS_NAME_MAXLEN), 0);
+ int errno = 0;
+
+ if ((errno = vfs_walk(__current->cwd, oldpath, &cur, NULL, 0))) {
+ goto done;
+ }
+
+ if ((errno = vfs_walk(
+ __current->cwd, newpath, &target_parent, &name, VFS_WALK_PARENT))) {
+ goto done;
+ }
+
+ errno = vfs_walk(target_parent, name.value, &target, NULL, 0);
+ if (errno == ENOENT) {
+ target = vfs_d_alloc();
+ hstrcpy(&target->name, &name);
+ } else if (errno) {
+ goto done;
+ }
+
+ if (!(errno = vfs_do_rename(cur, target))) {
+ vfs_d_free(target);
+ }
+
+done:
+ vfree(name.value);
+ return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL3(int,
+ mount,
+ const char*,
+ source,
+ const char*,
+ target,
+ const char*,
+ fstype)
+{
+ struct v_dnode *dev, *mnt;
+ int errno = 0;
+
+ if ((errno = vfs_walk(__current->cwd, source, &dev, NULL, 0))) {
+ goto done;
+ }
+
+ if ((errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
+ goto done;
+ }
+
+ if (!(dev->inode->itype & VFS_IFVOLDEV)) {
+ errno = ENOTDEV;
+ goto done;
+ }
+
+ // FIXME should not touch the underlying fs!
+ struct device* device =
+ (struct device*)((struct twifs_node*)dev->inode->data)->data;
+
+ errno = vfs_mount_at(fstype, device, mnt);
+
+done:
+ return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL1(int, unmount, const char*, target)
+{
+ return vfs_unmount(target);
}
\ No newline at end of file
git://scm.lunaixsky.com / lunaix-os.git / blobdiff