3 * @author Lunaixsky (zelong56@gmail.com)
4 * @brief Lunaix virtual file system - an abstraction layer for all file system.
8 * @copyright Copyright (c) 2022
12 #include <klibc/string.h>
13 #include <lunaix/dirent.h>
14 #include <lunaix/foptions.h>
15 #include <lunaix/fs.h>
16 #include <lunaix/mm/cake.h>
17 #include <lunaix/mm/page.h>
18 #include <lunaix/mm/valloc.h>
19 #include <lunaix/process.h>
20 #include <lunaix/spike.h>
21 #include <lunaix/syscall.h>
23 #define PATH_DELIM '/'
24 #define DNODE_HASHTABLE_BITS 10
25 #define DNODE_HASHTABLE_SIZE (1 << DNODE_HASHTABLE_BITS)
26 #define DNODE_HASH_MASK (DNODE_HASHTABLE_SIZE - 1)
27 #define DNODE_HASHBITS (32 - DNODE_HASHTABLE_BITS)
29 static struct cake_pile* dnode_pile;
30 static struct cake_pile* inode_pile;
31 static struct cake_pile* file_pile;
32 static struct cake_pile* superblock_pile;
33 static struct cake_pile* fd_pile;
35 static struct v_superblock* root_sb;
36 static struct hbucket* dnode_cache;
40 struct hstr vfs_ddot = HSTR("..", 2);
41 struct hstr vfs_dot = HSTR(".", 1);
42 struct hstr vfs_empty = HSTR("", 0);
48 vfs_d_free(struct v_dnode* dnode);
54 vfs_sb_free(struct v_superblock* sb);
59 // 为他们专门创建一个蛋糕堆,而不使用valloc,这样我们可以最小化内碎片的产生
60 dnode_pile = cake_new_pile("dnode_cache", sizeof(struct v_dnode), 1, 0);
61 inode_pile = cake_new_pile("inode_cache", sizeof(struct v_inode), 1, 0);
62 file_pile = cake_new_pile("file_cache", sizeof(struct v_file), 1, 0);
63 fd_pile = cake_new_pile("fd_cache", sizeof(struct v_fd), 1, 0);
65 cake_new_pile("sb_cache", sizeof(struct v_superblock), 1, 0);
67 dnode_cache = vzalloc(DNODE_HASHTABLE_SIZE * sizeof(struct hbucket));
69 hstr_rehash(&vfs_ddot, HSTR_FULL_HASH);
70 hstr_rehash(&vfs_dot, HSTR_FULL_HASH);
72 // 创建一个根superblock,用来蕴含我们的根目录。
73 root_sb = vfs_sb_alloc();
74 root_sb->root = vfs_d_alloc();
77 inline struct hbucket*
78 __dcache_get_bucket(struct v_dnode* parent, unsigned int hash)
80 // 与parent的指针值做加法,来减小碰撞的可能性。
81 hash += (uint32_t)parent;
83 hash = hash ^ (hash >> DNODE_HASHBITS);
84 return &dnode_cache[hash & DNODE_HASH_MASK];
88 vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
90 if (!str->len || HSTR_EQ(str, &vfs_dot))
93 if (HSTR_EQ(str, &vfs_ddot)) {
94 return parent->parent ? parent->parent : parent;
97 struct hbucket* slot = __dcache_get_bucket(parent, str->hash);
99 struct v_dnode *pos, *n;
100 hashtable_bucket_foreach(slot, pos, n, hash_list)
102 if (pos->name.hash == str->hash) {
110 vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
112 struct hbucket* bucket = __dcache_get_bucket(parent, dnode->name.hash);
113 hlist_add(&bucket->head, &dnode->hash_list);
117 vfs_walk(struct v_dnode* start,
119 struct v_dnode** dentry,
120 struct hstr* component,
126 if (path[0] == PATH_DELIM || !start) {
127 if ((walk_options & VFS_WALK_FSRELATIVE) && start) {
128 start = start->super_block->root;
130 start = root_sb->root;
135 struct v_dnode* dnode;
136 struct v_dnode* current_level = start;
138 char name_content[VFS_NAME_MAXLEN];
139 struct hstr name = HSTR(name_content, 0);
141 char current = path[i++], lookahead;
143 lookahead = path[i++];
144 if (current != PATH_DELIM) {
145 if (j >= VFS_NAME_MAXLEN - 1) {
148 if (!VFS_VALID_CHAR(current)) {
151 name_content[j++] = current;
157 // handling cases like /^.*(\/+).*$/
158 if (lookahead == PATH_DELIM) {
164 hstr_rehash(&name, HSTR_FULL_HASH);
166 if (!lookahead && (walk_options & VFS_WALK_PARENT)) {
168 component->hash = name.hash;
170 strcpy(component->value, name_content);
175 dnode = vfs_dcache_lookup(current_level, &name);
178 dnode = vfs_d_alloc();
179 dnode->name = HSTR(valloc(VFS_NAME_MAXLEN), j);
180 dnode->name.hash = name.hash;
182 strcpy(dnode->name.value, name_content);
185 current_level->inode->ops.dir_lookup(current_level->inode, dnode);
187 if (errno == ENOENT && (walk_options & VFS_WALK_MKPARENT)) {
188 if (!current_level->inode->ops.mkdir) {
191 errno = current_level->inode->ops.mkdir(
192 current_level->inode, dnode);
200 vfs_dcache_add(current_level, dnode);
202 dnode->parent = current_level;
203 llist_append(¤t_level->children, &dnode->siblings);
207 current_level = dnode;
212 *dentry = current_level;
216 vfree(dnode->name.value);
223 vfs_mount(const char* target, const char* fs_name, bdev_t device)
228 if (!(errno = vfs_walk(NULL, target, &mnt, NULL, 0))) {
229 errno = vfs_mount_at(fs_name, device, mnt);
236 vfs_unmount(const char* target)
241 if (!(errno = vfs_walk(NULL, target, &mnt, NULL, 0))) {
242 errno = vfs_unmount_at(mnt);
249 vfs_mount_at(const char* fs_name, bdev_t device, struct v_dnode* mnt_point)
251 struct filesystem* fs = fsm_get(fs_name);
254 struct v_superblock* sb = vfs_sb_alloc();
259 if (!(errno = fs->mount(sb, mnt_point))) {
261 sb->root = mnt_point;
262 mnt_point->super_block = sb;
263 llist_append(&root_sb->sb_list, &sb->sb_list);
270 vfs_unmount_at(struct v_dnode* mnt_point)
273 struct v_superblock* sb = mnt_point->super_block;
277 if (!(errno = sb->fs->unmount(sb))) {
278 struct v_dnode* fs_root = sb->root;
279 llist_delete(&fs_root->siblings);
280 llist_delete(&sb->sb_list);
287 vfs_open(struct v_dnode* dnode, struct v_file** file)
289 if (!dnode->inode || !dnode->inode->ops.open) {
293 struct v_file* vfile = cake_grab(file_pile);
294 memset(vfile, 0, sizeof(*vfile));
296 vfile->dnode = dnode;
297 vfile->inode = dnode->inode;
298 dnode->inode->open_count++;
300 int errno = dnode->inode->ops.open(dnode->inode, vfile);
302 cake_release(file_pile, vfile);
310 vfs_close(struct v_file* file)
312 if (!file->ops.close) {
316 int errno = file->ops.close(file);
318 if (file->inode->open_count) {
319 file->inode->open_count--;
321 cake_release(file_pile, file);
327 vfs_fsync(struct v_file* file)
330 if (file->ops.sync) {
331 errno = file->ops.sync(file);
333 if (!errno && file->inode->ops.sync) {
334 return file->inode->ops.sync(file->inode);
340 vfs_alloc_fdslot(int* fd)
342 for (size_t i = 0; i < VFS_MAX_FD; i++) {
343 if (!__current->fdtable->fds[i]) {
354 struct v_superblock* sb = cake_grab(superblock_pile);
355 memset(sb, 0, sizeof(*sb));
356 llist_init_head(&sb->sb_list);
361 vfs_sb_free(struct v_superblock* sb)
363 cake_release(superblock_pile, sb);
369 struct v_dnode* dnode = cake_grab(dnode_pile);
370 memset(dnode, 0, sizeof(*dnode));
371 llist_init_head(&dnode->children);
372 dnode->name = vfs_empty;
377 vfs_d_free(struct v_dnode* dnode)
379 if (dnode->ops.destruct) {
380 dnode->ops.destruct(dnode);
382 cake_release(dnode_pile, dnode);
388 struct v_inode* inode = cake_grab(inode_pile);
389 memset(inode, 0, sizeof(*inode));
395 vfs_i_free(struct v_inode* inode)
397 cake_release(inode_pile, inode);
400 /* ---- System call definition and support ---- */
403 __vfs_do_open(struct v_file** file_out, const char* path, int options)
405 char name_str[VFS_NAME_MAXLEN];
406 struct hstr name = HSTR(name_str, 0);
407 struct v_dnode *dentry, *file;
409 if ((errno = vfs_walk(NULL, path, &dentry, &name, VFS_WALK_PARENT))) {
413 vfs_walk(dentry, name.value, &file, NULL, 0);
415 struct v_file* opened_file = 0;
417 if ((options & FO_CREATE)) {
418 errno = dentry->inode->ops.create(dentry->inode, opened_file);
423 errno = vfs_open(file, &opened_file);
426 *file_out = opened_file;
430 __DEFINE_LXSYSCALL2(int, open, const char*, path, int, options)
432 struct v_file* opened_file;
433 int errno = __vfs_do_open(&opened_file, path, options), fd;
435 if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
436 struct v_fd* fd_s = vzalloc(sizeof(*fd_s));
437 fd_s->file = opened_file;
438 fd_s->pos = opened_file->inode->fsize & -((options & FO_APPEND) != 0);
439 __current->fdtable->fds[fd] = fd_s;
443 __current->k_status = errno;
444 return SYSCALL_ESTATUS(errno);
447 #define GET_FD(fd, fd_s) \
448 (fd >= 0 && fd < VFS_MAX_FD && (fd_s = __current->fdtable->fds[fd]))
450 __DEFINE_LXSYSCALL1(int, close, int, fd)
454 if (!GET_FD(fd, fd_s)) {
456 } else if (!(errno = vfs_close(fd_s->file))) {
458 __current->fdtable->fds[fd] = 0;
461 __current->k_status = errno;
463 return SYSCALL_ESTATUS(errno);
467 __vfs_readdir_callback(struct dir_context* dctx,
472 struct dirent* dent = (struct dirent*)dctx->cb_data;
473 strncpy(dent->d_name, name, DIRENT_NAME_MAX_LEN);
475 dent->d_type = dtype;
478 __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
482 if (!GET_FD(fd, fd_s)) {
484 } else if (!(fd_s->file->inode->itype & VFS_INODE_TYPE_DIR)) {
487 struct dir_context dctx =
488 (struct dir_context){ .cb_data = dent,
489 .index = dent->d_offset,
490 .read_complete_callback =
491 __vfs_readdir_callback };
492 if (dent->d_offset == 0) {
493 __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 0);
494 } else if (dent->d_offset == 1) {
495 __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0);
498 if ((errno = fd_s->file->ops.readdir(fd_s->file, &dctx))) {
507 __current->k_status = errno;
508 return SYSCALL_ESTATUS(errno);
511 __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
513 struct v_dnode *parent, *dir;
514 struct hstr component = HSTR(valloc(VFS_NAME_MAXLEN), 0);
515 int errno = vfs_walk(NULL, path, &parent, &component, VFS_WALK_PARENT);
520 if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
522 } else if (!parent->inode->ops.mkdir) {
524 } else if (!(parent->inode->itype & VFS_INODE_TYPE_DIR)) {
528 dir->name = component;
529 if (!(errno = parent->inode->ops.mkdir(parent->inode, dir))) {
530 llist_append(&parent->children, &dir->siblings);
533 vfree(component.value);
538 __current->k_status = errno;
539 return SYSCALL_ESTATUS(errno);
542 __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
546 if (!GET_FD(fd, fd_s)) {
549 struct v_file* file = fd_s->file;
550 file->f_pos = fd_s->pos;
551 if ((errno = file->ops.read(file, buf, count)) >= 0) {
556 __current->k_status = errno;
557 return SYSCALL_ESTATUS(errno);
560 __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
564 if (!GET_FD(fd, fd_s)) {
567 struct v_file* file = fd_s->file;
568 file->f_pos = fd_s->pos;
569 if ((errno = file->ops.write(file, buf, count)) >= 0) {
574 __current->k_status = errno;
575 return SYSCALL_ESTATUS(errno);
578 __DEFINE_LXSYSCALL3(int, lseek, int, fd, int, offset, int, options)
582 if (!GET_FD(fd, fd_s)) {
585 size_t fpos = fd_s->file->f_pos;
588 fpos = (size_t)((int)fd_s->file->f_pos + offset);
591 fpos = (size_t)((int)fd_s->file->inode->fsize + offset);
603 __current->k_status = errno;
604 return SYSCALL_ESTATUS(errno);
608 vfs_readlink(struct v_dnode* dnode, char* buf, size_t size, int depth)
618 size_t len = vfs_readlink(dnode->parent, buf, size, depth + 1);
624 size_t cpy_size = MIN(dnode->name.len, size - len);
625 strncpy(buf + len, dnode->name.value, cpy_size);
629 buf[len++] = PATH_DELIM;
635 __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size)
638 struct v_dnode* dnode;
639 if (!(errno = vfs_walk(NULL, path, &dnode, NULL, 0))) {
640 errno = vfs_readlink(dnode, buf, size, 0);
647 __current->k_status = errno;
648 return SYSCALL_ESTATUS(errno);
651 __DEFINE_LXSYSCALL4(int,
664 if (!GET_FD(dirfd, fd_s)) {
667 struct v_dnode* dnode;
668 if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {
669 errno = vfs_readlink(fd_s->file->dnode, buf, size, 0);
677 __current->k_status = errno;
678 return SYSCALL_ESTATUS(errno);
681 __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
684 struct v_dnode* dnode;
685 if ((errno = vfs_walk(NULL, pathname, &dnode, NULL, 0))) {
688 if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
693 if (dnode->inode->open_count) {
698 if (dnode->inode->itype == VFS_INODE_TYPE_DIR) {
699 errno = dnode->inode->ops.rmdir(dnode->inode);
705 __current->k_status = errno;
706 return SYSCALL_ESTATUS(errno);
710 __vfs_do_unlink(struct v_inode* inode)
713 if (inode->open_count) {
715 } else if (inode->itype != VFS_INODE_TYPE_DIR) {
716 // TODO handle symbolic link and type other than regular file
717 errno = inode->ops.unlink(inode);
725 __DEFINE_LXSYSCALL1(int, unlink, const char*, pathname)
728 struct v_dnode* dnode;
729 if ((errno = vfs_walk(NULL, pathname, &dnode, NULL, 0))) {
732 if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
737 errno = __vfs_do_unlink(dnode->inode);
740 __current->k_status = errno;
741 return SYSCALL_ESTATUS(errno);
744 __DEFINE_LXSYSCALL2(int, unlinkat, int, fd, const char*, pathname)
748 if (!GET_FD(fd, fd_s)) {
751 struct v_dnode* dnode;
752 if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {
753 errno = __vfs_do_unlink(dnode->inode);
758 __current->k_status = errno;
759 return SYSCALL_ESTATUS(errno);