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();
75 root_sb->root->inode = vfs_i_alloc();
78 inline struct hbucket*
79 __dcache_get_bucket(struct v_dnode* parent, unsigned int hash)
81 // 与parent的指针值做加法,来减小碰撞的可能性。
82 hash += (uint32_t)parent;
84 hash = hash ^ (hash >> DNODE_HASHBITS);
85 return &dnode_cache[hash & DNODE_HASH_MASK];
89 vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
91 if (!str->len || HSTR_EQ(str, &vfs_dot))
94 if (HSTR_EQ(str, &vfs_ddot)) {
95 return parent->parent ? parent->parent : parent;
98 struct hbucket* slot = __dcache_get_bucket(parent, str->hash);
100 struct v_dnode *pos, *n;
101 hashtable_bucket_foreach(slot, pos, n, hash_list)
103 if (pos->name.hash == str->hash) {
111 vfs_dcache_add(struct v_dnode* parent, struct v_dnode* dnode)
113 struct hbucket* bucket = __dcache_get_bucket(parent, dnode->name.hash);
114 hlist_add(&bucket->head, &dnode->hash_list);
118 __vfs_walk(struct v_dnode* start,
120 struct v_dnode** dentry,
121 struct hstr* component,
127 if (path[0] == PATH_DELIM || !start) {
128 if ((walk_options & VFS_WALK_FSRELATIVE) && start) {
129 start = start->super_block->root;
131 start = root_sb->root;
136 struct v_dnode* dnode;
137 struct v_dnode* current_level = start;
139 char name_content[VFS_NAME_MAXLEN];
140 struct hstr name = HSTR(name_content, 0);
142 char current = path[i++], lookahead;
144 lookahead = path[i++];
145 if (current != PATH_DELIM) {
146 if (j >= VFS_NAME_MAXLEN - 1) {
149 if (!VFS_VALID_CHAR(current)) {
152 name_content[j++] = current;
158 // handling cases like /^.*(\/+).*$/
159 if (lookahead == PATH_DELIM) {
165 hstr_rehash(&name, HSTR_FULL_HASH);
167 if (!lookahead && (walk_options & VFS_WALK_PARENT)) {
169 component->hash = name.hash;
171 strcpy(component->value, name_content);
176 dnode = vfs_dcache_lookup(current_level, &name);
179 dnode = vfs_d_alloc();
180 dnode->name = HHSTR(valloc(VFS_NAME_MAXLEN), j, 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);
222 #define VFS_MAX_SYMLINK 16
225 vfs_walk(struct v_dnode* start,
227 struct v_dnode** dentry,
228 struct hstr* component,
231 struct v_dnode* interim;
232 const char* pathname = path;
233 int errno = __vfs_walk(start, path, &interim, component, options);
237 if (counter >= VFS_MAX_SYMLINK) {
241 if ((interim->inode->itype & VFS_IFSYMLINK) &&
242 !(options & VFS_WALK_NOFOLLOW) &&
243 interim->inode->ops.read_symlink) {
244 errno = interim->inode->ops.read_symlink(interim->inode, &pathname);
251 errno = __vfs_walk(start, pathname, &interim, component, options);
255 *dentry = errno ? 0 : interim;
261 vfs_mount(const char* target, const char* fs_name, bdev_t device)
266 if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
267 errno = vfs_mount_at(fs_name, device, mnt);
274 vfs_unmount(const char* target)
279 if (!(errno = vfs_walk(__current->cwd, target, &mnt, NULL, 0))) {
280 errno = vfs_unmount_at(mnt);
287 vfs_mount_at(const char* fs_name, bdev_t device, struct v_dnode* mnt_point)
289 struct filesystem* fs = fsm_get(fs_name);
292 struct v_superblock* sb = vfs_sb_alloc();
297 if (!(errno = fs->mount(sb, mnt_point))) {
299 sb->root = mnt_point;
300 mnt_point->super_block = sb;
301 llist_append(&root_sb->sb_list, &sb->sb_list);
308 vfs_unmount_at(struct v_dnode* mnt_point)
311 struct v_superblock* sb = mnt_point->super_block;
315 if (!(errno = sb->fs->unmount(sb))) {
316 struct v_dnode* fs_root = sb->root;
317 llist_delete(&fs_root->siblings);
318 llist_delete(&sb->sb_list);
325 vfs_open(struct v_dnode* dnode, struct v_file** file)
327 if (!dnode->inode || !dnode->inode->ops.open) {
331 struct v_inode* inode = dnode->inode;
332 struct v_file* vfile = cake_grab(file_pile);
333 memset(vfile, 0, sizeof(*vfile));
335 vfile->dnode = dnode;
336 vfile->inode = inode;
337 vfile->ref_count = 1;
338 vfile->ops = inode->default_fops;
340 if ((inode->itype & VFS_IFFILE) && !inode->pg_cache) {
341 struct pcache* pcache = vzalloc(sizeof(struct pcache));
343 pcache->master = inode;
344 inode->pg_cache = pcache;
347 int errno = inode->ops.open(inode, vfile);
349 cake_release(file_pile, vfile);
360 vfs_link(struct v_dnode* to_link, struct v_dnode* name)
363 if (to_link->super_block->root != name->super_block->root) {
365 } else if (!to_link->inode->ops.link) {
367 } else if (!(errno = to_link->inode->ops.link(to_link->inode, name))) {
368 name->inode = to_link->inode;
369 to_link->inode->link_count++;
376 vfs_close(struct v_file* file)
379 if (!file->ops.close || !(errno = file->ops.close(file))) {
380 file->dnode->ref_count--;
381 file->inode->open_count--;
382 pcache_commit_all(file->inode);
383 cake_release(file_pile, file);
389 vfs_fsync(struct v_file* file)
392 pcache_commit_all(file->inode);
393 if (file->ops.sync) {
394 errno = file->ops.sync(file->inode);
400 vfs_alloc_fdslot(int* fd)
402 for (size_t i = 0; i < VFS_MAX_FD; i++) {
403 if (!__current->fdtable->fds[i]) {
414 struct v_superblock* sb = cake_grab(superblock_pile);
415 memset(sb, 0, sizeof(*sb));
416 llist_init_head(&sb->sb_list);
421 vfs_sb_free(struct v_superblock* sb)
423 cake_release(superblock_pile, sb);
429 struct v_dnode* dnode = cake_grab(dnode_pile);
430 memset(dnode, 0, sizeof(*dnode));
431 llist_init_head(&dnode->children);
432 dnode->name = vfs_empty;
437 vfs_d_free(struct v_dnode* dnode)
439 if (dnode->ops.destruct) {
440 dnode->ops.destruct(dnode);
442 cake_release(dnode_pile, dnode);
448 struct v_inode* inode = cake_grab(inode_pile);
449 memset(inode, 0, sizeof(*inode));
454 vfs_i_free(struct v_inode* inode)
456 cake_release(inode_pile, inode);
459 /* ---- System call definition and support ---- */
461 #define FLOCATE_CREATE_EMPTY 1
463 #define DO_STATUS(errno) SYSCALL_ESTATUS(__current->k_status = errno)
464 #define DO_STATUS_OR_RETURN(errno) ({ errno < 0 ? DO_STATUS(errno) : errno; })
466 #define TEST_FD(fd) (fd >= 0 && fd < VFS_MAX_FD)
469 __vfs_getfd(int fd, struct v_fd** fd_s)
471 if (TEST_FD(fd) && (*fd_s = __current->fdtable->fds[fd])) {
478 __vfs_try_locate_file(const char* path,
479 struct v_dnode** fdir,
480 struct v_dnode** file,
483 char name_str[VFS_NAME_MAXLEN];
484 struct hstr name = HSTR(name_str, 0);
487 vfs_walk(__current->cwd, path, fdir, &name, VFS_WALK_PARENT))) {
491 errno = vfs_walk(*fdir, name.value, file, NULL, 0);
492 if (errno != ENOENT || !(options & FLOCATE_CREATE_EMPTY)) {
496 if (!(errno = (*fdir)->inode->ops.create((*fdir)->inode))) {
497 struct v_dnode* file_new;
498 file_new = vfs_d_alloc();
499 file_new->name = HHSTR(valloc(VFS_NAME_MAXLEN), name.len, name.hash);
500 strcpy(file_new->name.value, name_str);
503 llist_append(&(*fdir)->children, &file_new->siblings);
510 vfs_do_open(const char* path, int options)
513 struct v_dnode *dentry, *file;
514 struct v_file* ofile = 0;
516 errno = __vfs_try_locate_file(path, &dentry, &file, 0);
518 if (errno != ENOENT && (options & FO_CREATE)) {
519 errno = dentry->inode->ops.create(dentry->inode);
522 if (!errno && (errno = vfs_open(file, &ofile))) {
526 struct v_inode* o_inode = ofile->inode;
527 if (!(o_inode->itype & VFS_IFSEQDEV) && !(options & FO_DIRECT)) {
528 // XXX Change here accordingly when signature of pcache_r/w changed.
529 ofile->ops.read = pcache_read;
530 ofile->ops.write = pcache_write;
533 if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
534 struct v_fd* fd_s = vzalloc(sizeof(*fd_s));
535 ofile->f_pos = ofile->inode->fsize & -((options & FO_APPEND) != 0);
537 fd_s->flags = options;
538 __current->fdtable->fds[fd] = fd_s;
545 __DEFINE_LXSYSCALL2(int, open, const char*, path, int, options)
547 int errno = vfs_do_open(path, options);
548 return DO_STATUS_OR_RETURN(errno);
551 __DEFINE_LXSYSCALL1(int, close, int, fd)
555 if ((errno = __vfs_getfd(fd, &fd_s))) {
559 if (fd_s->file->ref_count > 1) {
560 fd_s->file->ref_count--;
561 } else if ((errno = vfs_close(fd_s->file))) {
566 __current->fdtable->fds[fd] = 0;
569 return DO_STATUS(errno);
573 __vfs_readdir_callback(struct dir_context* dctx,
578 struct dirent* dent = (struct dirent*)dctx->cb_data;
579 strncpy(dent->d_name, name, DIRENT_NAME_MAX_LEN);
581 dent->d_type = dtype;
584 __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
589 if ((errno = __vfs_getfd(fd, &fd_s))) {
593 if (!(fd_s->file->inode->itype & VFS_IFDIR)) {
596 struct dir_context dctx =
597 (struct dir_context){ .cb_data = dent,
598 .index = dent->d_offset,
599 .read_complete_callback =
600 __vfs_readdir_callback };
601 if (dent->d_offset == 0) {
602 __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 0);
603 } else if (dent->d_offset == 1) {
604 __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0);
607 if ((errno = fd_s->file->ops.readdir(fd_s->file->inode, &dctx))) {
616 return DO_STATUS(errno);
619 __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
621 struct v_dnode *parent, *dir;
622 struct hstr component = HSTR(valloc(VFS_NAME_MAXLEN), 0);
624 vfs_walk(__current->cwd, path, &parent, &component, VFS_WALK_PARENT);
629 if ((parent->super_block->fs->types & FSTYPE_ROFS)) {
631 } else if (!parent->inode->ops.mkdir) {
633 } else if (!(parent->inode->itype & VFS_IFDIR)) {
637 dir->name = component;
638 if (!(errno = parent->inode->ops.mkdir(parent->inode, dir))) {
639 llist_append(&parent->children, &dir->siblings);
642 vfree(component.value);
647 return DO_STATUS(errno);
650 __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
654 if ((errno = __vfs_getfd(fd, &fd_s))) {
658 struct v_file* file = fd_s->file;
659 if ((file->inode->itype & VFS_IFDIR)) {
664 __SYSCALL_INTERRUPTIBLE(
665 { errno = file->ops.read(file->inode, buf, count, file->f_pos); })
668 file->f_pos += errno;
673 return DO_STATUS(errno);
676 __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
680 if ((errno = __vfs_getfd(fd, &fd_s))) {
684 struct v_file* file = fd_s->file;
685 if ((file->inode->itype & VFS_IFDIR)) {
690 __SYSCALL_INTERRUPTIBLE(
691 { errno = file->ops.write(file->inode, buf, count, file->f_pos); })
694 file->f_pos += errno;
699 return DO_STATUS(errno);
702 __DEFINE_LXSYSCALL3(int, lseek, int, fd, int, offset, int, options)
706 if ((errno = __vfs_getfd(fd, &fd_s))) {
710 struct v_file* file = fd_s->file;
711 size_t fpos = file->f_pos;
714 fpos = (size_t)((int)file->f_pos + offset);
717 fpos = (size_t)((int)file->inode->fsize + offset);
723 if (!file->ops.seek || !(errno = file->ops.seek(file->inode, fpos))) {
728 return DO_STATUS(errno);
732 vfs_get_path(struct v_dnode* dnode, char* buf, size_t size, int depth)
742 size_t len = vfs_get_path(dnode->parent, buf, size, depth + 1);
748 size_t cpy_size = MIN(dnode->name.len, size - len);
749 strncpy(buf + len, dnode->name.value, cpy_size);
753 buf[len++] = PATH_DELIM;
760 vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
763 if (dnode->inode->ops.read_symlink) {
764 int errno = dnode->inode->ops.read_symlink(dnode->inode, &link);
765 strncpy(buf, link, size);
771 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
775 if ((errno = __vfs_getfd(fd, &fd_s))) {
779 struct v_dnode* dnode;
780 errno = vfs_get_path(fd_s->file->dnode, buf, size, 0);
787 return DO_STATUS(errno);
790 __DEFINE_LXSYSCALL3(int, readlink, const char*, path, char*, buf, size_t, size)
793 struct v_dnode* dnode;
795 vfs_walk(__current->cwd, path, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
796 errno = vfs_readlink(dnode, buf, size);
803 return DO_STATUS(errno);
806 __DEFINE_LXSYSCALL4(int,
819 if ((errno = __vfs_getfd(dirfd, &fd_s))) {
823 struct v_dnode* dnode;
824 if (!(errno = vfs_walk(
825 fd_s->file->dnode, pathname, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
826 errno = vfs_readlink(fd_s->file->dnode, buf, size);
834 return DO_STATUS(errno);
837 __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
840 struct v_dnode* dnode;
841 if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
844 if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
849 if (dnode->inode->open_count) {
854 if ((dnode->inode->itype & VFS_IFDIR)) {
855 errno = dnode->inode->ops.rmdir(dnode->inode);
861 return DO_STATUS(errno);
865 __vfs_do_unlink(struct v_inode* inode)
868 if (inode->open_count) {
870 } else if (!(inode->itype & VFS_IFDIR)) {
871 // TODO handle symbolic link and type other than regular file
872 errno = inode->ops.unlink(inode);
883 __DEFINE_LXSYSCALL1(int, unlink, const char*, pathname)
886 struct v_dnode* dnode;
887 if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
890 if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
895 errno = __vfs_do_unlink(dnode->inode);
898 return DO_STATUS(errno);
901 __DEFINE_LXSYSCALL2(int, unlinkat, int, fd, const char*, pathname)
905 if ((errno = __vfs_getfd(fd, &fd_s))) {
909 struct v_dnode* dnode;
910 if (!(errno = vfs_walk(fd_s->file->dnode, pathname, &dnode, NULL, 0))) {
911 errno = __vfs_do_unlink(dnode->inode);
915 return DO_STATUS(errno);
918 __DEFINE_LXSYSCALL2(int, link, const char*, oldpath, const char*, newpath)
921 struct v_dnode *dentry, *to_link, *name_dentry, *name_file;
923 errno = __vfs_try_locate_file(oldpath, &dentry, &to_link, 0);
925 errno = __vfs_try_locate_file(
926 newpath, &name_dentry, &name_file, FLOCATE_CREATE_EMPTY);
929 } else if (name_file) {
930 errno = vfs_link(to_link, name_file);
933 return DO_STATUS(errno);
936 __DEFINE_LXSYSCALL1(int, fsync, int, fildes)
940 if (!(errno = __vfs_getfd(fildes, &fd_s))) {
941 errno = vfs_fsync(fd_s->file);
944 return DO_STATUS(errno);
948 vfs_dup_fd(struct v_fd* old, struct v_fd** new)
951 struct v_fd* copied = cake_grab(fd_pile);
953 memcpy(copied, old, sizeof(struct v_fd));
954 old->file->ref_count++;
962 vfs_dup2(int oldfd, int newfd)
964 if (newfd == oldfd) {
969 struct v_fd *oldfd_s, *newfd_s;
970 if ((errno = __vfs_getfd(oldfd, &oldfd_s))) {
974 if (!TEST_FD(newfd)) {
979 newfd_s = __current->fdtable->fds[newfd];
980 if (newfd_s && (errno = vfs_close(newfd_s->file))) {
984 if (!(errno = vfs_dup_fd(oldfd_s, &newfd_s))) {
985 __current->fdtable->fds[newfd] = newfd_s;
990 return DO_STATUS(errno);
993 __DEFINE_LXSYSCALL2(int, dup2, int, oldfd, int, newfd)
995 return vfs_dup2(oldfd, newfd);
998 __DEFINE_LXSYSCALL1(int, dup, int, oldfd)
1001 struct v_fd *oldfd_s, *newfd_s;
1002 if ((errno = __vfs_getfd(oldfd, &oldfd_s))) {
1006 if (!(errno = vfs_alloc_fdslot(&newfd)) &&
1007 !(errno = vfs_dup_fd(oldfd_s, &newfd_s))) {
1008 __current->fdtable->fds[newfd] = newfd_s;
1013 return DO_STATUS(errno);
1016 __DEFINE_LXSYSCALL2(int,
1024 struct v_dnode* dnode;
1025 if ((errno = vfs_walk(__current->cwd, pathname, &dnode, NULL, 0))) {
1028 if ((dnode->super_block->fs->types & FSTYPE_ROFS)) {
1032 if (!dnode->inode->ops.symlink) {
1037 errno = dnode->inode->ops.symlink(dnode->inode, link_target);
1040 return DO_STATUS(errno);
1044 __vfs_do_chdir(struct v_dnode* dnode)
1047 if (!(dnode->inode->itype & VFS_IFDIR)) {
1052 if (__current->cwd) {
1053 __current->cwd->ref_count--;
1057 __current->cwd = dnode;
1063 __DEFINE_LXSYSCALL1(int, chdir, const char*, path)
1065 struct v_dnode* dnode;
1068 if ((errno = vfs_walk(__current->cwd, path, &dnode, NULL, 0))) {
1072 errno = __vfs_do_chdir(dnode);
1075 return DO_STATUS(errno);
1078 __DEFINE_LXSYSCALL1(int, fchdir, int, fd)
1083 if ((errno = __vfs_getfd(fd, &fd_s))) {
1087 errno = __vfs_do_chdir(fd_s->file->dnode);
1090 return DO_STATUS(errno);
1093 __DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size)
1104 if (!__current->cwd) {
1108 len = vfs_get_path(__current->cwd, buf, size, 0);
1115 buf[len + 1] = '\0';
1120 __current->k_status = errno;