*/
#include <klibc/string.h>
-#include <lunaix/dirent.h>
#include <lunaix/foptions.h>
#include <lunaix/fs.h>
#include <lunaix/mm/cake.h>
#include <lunaix/process.h>
#include <lunaix/spike.h>
#include <lunaix/syscall.h>
+#include <lunaix/syscall_utils.h>
#include <lunaix/fs/twifs.h>
+#include <usr/lunaix/dirent_defs.h>
+
static struct cake_pile* dnode_pile;
static struct cake_pile* inode_pile;
static struct cake_pile* file_pile;
// 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);
}
-inline struct hbucket*
-__dcache_hash(struct v_dnode* parent, uint32_t* hash)
+static inline struct hbucket*
+__dcache_hash(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;
+ u32_t _hash = *hash;
// 确保低位更加随机
_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 = _hash;
return &dnode_cache[_hash & VFS_HASH_MASK];
}
return parent;
if (HSTR_EQ(str, &vfs_ddot)) {
- return parent->parent ? parent->parent : parent;
+ return parent->parent;
}
- uint32_t hash = str->hash;
+ u32_t hash = str->hash;
struct hbucket* slot = __dcache_hash(parent, &hash);
struct v_dnode *pos, *n;
assert(dnode->ref_count == 1);
llist_delete(&dnode->siblings);
+ llist_delete(&dnode->aka_list);
hlist_delete(&dnode->hash_list);
dnode->parent = NULL;
vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode)
{
if (assign_to->inode) {
+ 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++;
}
{
int errno;
+ if ((errno = vfs_check_writable(to_link))) {
+ return errno;
+ }
+
lock_inode(to_link->inode);
if (to_link->super_block->root != name->super_block->root) {
errno = EXDEV;
}
int
-vfs_close(struct v_file* file)
+vfs_pclose(struct v_file* file, pid_t pid)
{
int errno = 0;
- if (!(errno = file->ops->close(file))) {
+ 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);
return errno;
}
+int
+vfs_close(struct v_file* file)
+{
+ return vfs_pclose(file, __current->pid);
+}
+
+void
+vfs_free_fd(struct v_fd* fd)
+{
+ cake_release(fd_pile, fd);
+}
+
int
vfs_fsync(struct v_file* file)
{
+ int errno;
+ if ((errno = vfs_check_writable(file->dnode))) {
+ return errno;
+ }
+
lock_inode(file->inode);
- int errno = ENOTSUP;
pcache_commit_all(file->inode);
+ errno = ENOTSUP;
if (file->ops->sync) {
errno = file->ops->sync(file);
}
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);
if (parent) {
dnode->super_block = parent->super_block;
+ dnode->mnt = parent->mnt;
}
lru_use_one(dnode_lru, &dnode->lru);
vfs_dcache_remove(pos);
}
- vfree(dnode->name.value);
+ vfree((void*)dnode->name.value);
cake_release(dnode_pile, dnode);
}
struct v_inode*
-vfs_i_find(struct v_superblock* sb, uint32_t i_id)
+vfs_i_find(struct v_superblock* sb, u32_t i_id)
{
struct hbucket* slot = &sb->i_cache[i_id & VFS_HASH_MASK];
struct v_inode *pos, *n;
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->atime = inode->ctime;
inode->mtime = inode->ctime;
-done:
lru_use_one(inode_lru, &inode->lru);
return inode;
}
pcache_release(inode->pg_cache);
vfree(inode->pg_cache);
}
- inode->ops->sync(inode);
+ // 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);
+ }
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
int
vfs_getfd(int fd, struct v_fd** fd_s)
char name_str[VFS_NAME_MAXLEN];
struct hstr name = HSTR(name_str, 0);
int errno;
- if ((errno =
- vfs_walk(__current->cwd, path, fdir, &name, VFS_WALK_PARENT))) {
+
+ name_str[0] = 0;
+ if ((errno = vfs_walk_proc(path, fdir, &name, VFS_WALK_PARENT))) {
return errno;
}
errno = vfs_walk(*fdir, name.value, file, NULL, 0);
- if (errno != ENOENT || !(options & FLOCATE_CREATE_EMPTY)) {
+
+ if (errno != ENOENT && (options & FLOCATE_CREATE_ONLY)) {
+ return EEXIST;
+ }
+
+ if (errno != ENOENT ||
+ !(options & (FLOCATE_CREATE_EMPTY | FLOCATE_CREATE_ONLY))) {
return errno;
}
{
int errno, fd;
struct v_dnode *dentry, *file;
- struct v_file* ofile = 0;
+ struct v_file* ofile = NULL;
errno = __vfs_try_locate_file(
path, &dentry, &file, (options & FO_CREATE) ? FLOCATE_CREATE_EMPTY : 0);
- if (errno || (errno = vfs_open(file, &ofile))) {
- return errno;
- }
+ if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
- struct v_inode* o_inode = ofile->inode;
+ if (errno || (errno = vfs_open(file, &ofile))) {
+ return errno;
+ }
+
+ struct v_fd* fd_s = cake_grab(fd_pile);
+ memset(fd_s, 0, sizeof(*fd_s));
- 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;
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:
const int len,
const int dtype)
{
- struct dirent* dent = (struct dirent*)dctx->cb_data;
+ struct lx_dirent* dent = (struct lx_dirent*)dctx->cb_data;
strncpy(dent->d_name, name, 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;
__vfs_readdir_callback };
errno = 1;
if (dent->d_offset == 0) {
- __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 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, 0);
+ __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) {
file->inode->atime = clock_unixtime();
- __SYSCALL_INTERRUPTIBLE({
- if ((file->inode->itype & VFS_IFSEQDEV) || (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);
- }
- })
+ if ((file->inode->itype & VFS_IFSEQDEV) || (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);
+ }
if (errno > 0) {
file->f_pos += errno;
}
struct v_file* file = fd_s->file;
+
+ if ((errno = vfs_check_writable(file->dnode))) {
+ goto done;
+ }
+
if ((file->inode->itype & VFS_IFDIR)) {
errno = EISDIR;
goto done;
file->inode->mtime = clock_unixtime();
- __SYSCALL_INTERRUPTIBLE({
- if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
- errno = file->ops->write(file->inode, buf, count, file->f_pos);
- } else {
- errno = pcache_write(file->inode, buf, count, file->f_pos);
- }
- })
+ if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
+ errno = file->ops->write(file->inode, buf, count, file->f_pos);
+ } else {
+ errno = pcache_write(file->inode, buf, count, file->f_pos);
+ }
if (errno > 0) {
file->f_pos += errno;
struct v_file* file = fd_s->file;
+ if (!file->ops->seek) {
+ errno = ENOTSUP;
+ goto done;
+ }
+
lock_inode(file->inode);
- size_t fpos = file->f_pos;
+ int overflow = 0;
+ int fpos = file->f_pos;
switch (options) {
case FSEEK_CUR:
- fpos = (size_t)((int)file->f_pos + offset);
+ overflow = __builtin_sadd_overflow((int)file->f_pos, offset, &fpos);
break;
case FSEEK_END:
- fpos = (size_t)((int)file->inode->fsize + offset);
+ overflow =
+ __builtin_sadd_overflow((int)file->inode->fsize, offset, &fpos);
break;
case FSEEK_SET:
fpos = offset;
break;
}
- if (!(errno = file->ops->seek(file->inode, fpos))) {
+ if (overflow) {
+ errno = EOVERFLOW;
+ } else if (!(errno = file->ops->seek(file->inode, fpos))) {
file->f_pos = fpos;
}
}
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++] = VFS_PATH_DELIM;
- }
-
return len;
}
return 0;
}
+int
+vfs_get_dtype(int itype)
+{
+ switch (itype) {
+ case VFS_IFDIR:
+ return DT_DIR;
+ case VFS_IFSYMLINK:
+ return DT_SYMLINK;
+ default:
+ return DT_PIPE;
+ }
+}
+
__DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
{
int errno;
{
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);
}
{
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 ((errno = vfs_check_writable(dnode))) {
+ goto done;
+ }
+
if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
errno = EROFS;
goto done;
char name_value[VFS_NAME_MAXLEN];
struct hstr name = HHSTR(name_value, 0, 0);
- if (!dir) {
- errno = ENOMEM;
+ if ((errno = vfs_walk_proc(path, &parent, &name, VFS_WALK_PARENT))) {
goto done;
}
- if ((errno =
- vfs_walk(__current->cwd, path, &parent, &name, VFS_WALK_PARENT))) {
+ if ((errno = vfs_check_writable(parent))) {
goto done;
}
- dir = vfs_d_alloc(parent, &name);
+ if (!(dir = vfs_d_alloc(parent, &name))) {
+ errno = ENOMEM;
+ goto done;
+ }
lock_dnode(parent);
lock_inode(parent->inode);
int
__vfs_do_unlink(struct v_dnode* dnode)
{
+ int errno;
struct v_inode* inode = dnode->inode;
if (dnode->ref_count > 1) {
return EBUSY;
}
+ 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);
if (!errno) {
vfs_d_free(dnode);
{
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;
}
errno = __vfs_try_locate_file(oldpath, &dentry, &to_link, 0);
if (!errno) {
errno = __vfs_try_locate_file(
- newpath, &name_dentry, &name_file, FLOCATE_CREATE_EMPTY);
+ newpath, &name_dentry, &name_file, FLOCATE_CREATE_ONLY);
if (!errno) {
- errno = EEXIST;
- } else if (name_file) {
errno = vfs_link(to_link, name_file);
}
}
{
int errno;
struct v_fd* fd_s;
+
if (!(errno = vfs_getfd(fildes, &fd_s))) {
errno = vfs_fsync(fd_s->file);
}
link_target)
{
int errno;
- struct v_dnode* dnode;
- if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
+ struct v_dnode *dnode, *file;
+ if ((errno = __vfs_try_locate_file(
+ pathname, &dnode, &file, FLOCATE_CREATE_ONLY))) {
goto done;
}
- if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
- errno = EROFS;
+
+ if ((errno = vfs_check_writable(file))) {
goto done;
}
- if (!dnode->inode->ops->set_symlink) {
+
+ if (!file->inode->ops->set_symlink) {
errno = ENOTSUP;
goto done;
}
- lock_inode(dnode->inode);
+ lock_inode(file->inode);
- errno = dnode->inode->ops->set_symlink(dnode->inode, link_target);
+ errno = file->inode->ops->set_symlink(file->inode, link_target);
- unlock_inode(dnode->inode);
+ unlock_inode(file->inode);
done:
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);
+ mnt_mkbusy(dnode->mnt);
+}
+
+void
+vfs_unref_dnode(struct v_dnode* dnode)
+{
+ atomic_fetch_sub(&dnode->ref_count, 1);
+ mnt_chillax(dnode->mnt);
+}
+
int
-__vfs_do_chdir(struct v_dnode* dnode)
+vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode)
{
int errno = 0;
goto done;
}
- if (__current->cwd) {
- atomic_fetch_sub(&__current->cwd->ref_count, 1);
- mnt_chillax(__current->cwd->mnt);
+ if (proc->cwd) {
+ vfs_unref_dnode(proc->cwd);
}
- atomic_fetch_add(&dnode->ref_count, 1);
- mnt_mkbusy(dnode->mnt);
- __current->cwd = dnode;
+ vfs_ref_dnode(dnode);
+ proc->cwd = dnode;
unlock_dnode(dnode);
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);
goto done;
}
- errno = __vfs_do_chdir(fd_s->file->dnode);
+ errno = vfs_do_chdir((struct proc_info*)__current, fd_s->file->dnode);
done:
return DO_STATUS(errno);
}
}
- buf[len + 1] = '\0';
+ buf[len] = '\0';
ret_ptr = buf;
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;
}
return EXDEV;
}
- int errno = 0;
-
struct v_dnode* oldparent = current->parent;
struct v_dnode* newparent = target->parent;
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;
}
errno = vfs_do_rename(cur, target);
done:
- vfree(name.value);
+ vfree((void*)name.value);
return DO_STATUS(errno);
}
\ No newline at end of file