#include <lunaix/fs/twifs.h>
-#define PATH_DELIM '/'
-
-#define unlock_inode(inode) mutex_unlock(&inode->lock)
-#define lock_inode(inode) \
- ({ \
- mutex_lock(&inode->lock); \
- lru_use_one(inode_lru, &inode->lru); \
- })
-
-#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;
static struct cake_pile* file_pile;
struct v_dnode* vfs_sysroot;
static struct hbucket* dnode_cache;
-static struct lru_zone *dnode_lru, *inode_lru;
+struct lru_zone *dnode_lru, *inode_lru;
struct hstr vfs_ddot = HSTR("..", 2);
struct hstr vfs_dot = HSTR(".", 1);
// 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);
}
__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;
// 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 += (uint32_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;
void
vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
{
+ assert(parent);
+
atomic_fetch_add(&dnode->ref_count, 1);
dnode->parent = parent;
llist_append(&parent->children, &dnode->siblings);
void
vfs_dcache_remove(struct v_dnode* dnode)
{
+ assert(dnode);
assert(dnode->ref_count == 1);
llist_delete(&dnode->siblings);
+ llist_delete(&dnode->aka_list);
hlist_delete(&dnode->hash_list);
dnode->parent = NULL;
void
vfs_dcache_rehash(struct v_dnode* new_parent, struct v_dnode* dnode)
{
+ assert(new_parent);
+
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)
-{
- 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 = vfs_sysroot;
- if (!vfs_sysroot->mnt) {
- panick("vfs: no root");
- }
- }
- i++;
- }
-
- 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 && current_level) {
- 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(current_level, &name);
-
- if (!dnode) {
- errno = ENOMEM;
- goto error;
- }
-
- struct v_inode* current_inode = current_level->inode;
-
- lock_inode(current_inode);
-
- errno = current_inode->ops->dir_lookup(current_inode, dnode);
-
- if (errno == ENOENT && (walk_options & VFS_WALK_MKPARENT)) {
- if (!current_inode->ops->mkdir) {
- errno = ENOTSUP;
- } else {
- errno = current_inode->ops->mkdir(current_inode, dnode);
- }
- }
-
- vfs_dcache_add(current_level, dnode);
- unlock_inode(current_inode);
-
- if (errno) {
- unlock_dnode(current_level);
- goto cleanup;
- }
- }
-
- unlock_dnode(current_level);
-
- j = 0;
- current_level = dnode;
- cont:
- current = lookahead;
- };
-
- *dentry = current_level;
- return 0;
-
-cleanup:
- vfs_d_free(dnode);
-error:
- *dentry = NULL;
- return errno;
-}
-
-#define VFS_MAX_SYMLINK 16
-
-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;
-
- // FIXME This is NOT a correct way to resolve symlink!
- while (!errno && interim->inode && (options & VFS_WALK_NOFOLLOW)) {
- if (counter >= VFS_MAX_SYMLINK) {
- errno = ELOOP;
- continue;
- }
- if ((interim->inode->itype & VFS_IFSYMLINK) &&
- 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;
- }
- } else {
- break;
- }
- errno = __vfs_walk(start, pathname, &interim, component, options);
- counter++;
- }
-
- *dentry = errno ? 0 : interim;
-
- return errno;
-}
-
int
vfs_open(struct v_dnode* dnode, struct v_file** file)
{
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);
+}
+
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);
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);
#define FLOCATE_CREATE_EMPTY 1
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;
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;
}
{
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;
}
struct v_fd* fd_s;
int errno;
- if ((errno = __vfs_getfd(fd, &fd_s))) {
+ if ((errno = vfs_getfd(fd, &fd_s))) {
goto done;
}
__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) {
{
int errno = 0;
struct v_fd* fd_s;
- if ((errno = __vfs_getfd(fd, &fd_s))) {
+ if ((errno = vfs_getfd(fd, &fd_s))) {
goto done;
}
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;
{
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;
+
+ 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;
{
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;
+ 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++] = 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;
struct v_fd* fd_s;
- if ((errno = __vfs_getfd(fd, &fd_s))) {
+ if ((errno = vfs_getfd(fd, &fd_s))) {
goto done;
}
{
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_fd* fd_s;
- if ((errno = __vfs_getfd(dirfd, &fd_s))) {
+ if ((errno = vfs_getfd(dirfd, &fd_s))) {
goto done;
}
{
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)) {
{
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;
}
{
int errno;
struct v_fd* fd_s;
- if ((errno = __vfs_getfd(fd, &fd_s))) {
+ if ((errno = vfs_getfd(fd, &fd_s))) {
goto done;
}
{
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);
}
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;
}
{
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;
}
{
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))) {
goto done;
}
- if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
- errno = EROFS;
+
+ if (errno = vfs_check_writable(dnode)) {
goto done;
}
+
if (!dnode->inode->ops->set_symlink) {
errno = ENOTSUP;
goto done;
return DO_STATUS(errno);
}
+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(__current, dnode);
done:
return DO_STATUS(errno);
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(fd_s->file->dnode);
+ errno = vfs_do_chdir(__current, fd_s->file->dnode);
done:
return DO_STATUS(errno);
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);
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;
}
git://scm.lunaixsky.com / lunaix-os.git / blobdiff