system-servers/tmpfs/tmpfs.c

/*
 * Copyright (c) 2023 Institute of Parallel And Distributed Systems (IPADS), Shanghai Jiao Tong University (SJTU)
 * Licensed under the Mulan PSL v2.
 * You can use this software according to the terms and conditions of the Mulan PSL v2.
 * You may obtain a copy of Mulan PSL v2 at:
 *     http://license.coscl.org.cn/MulanPSL2
 * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY OR FIT FOR A PARTICULAR
 * PURPOSE.
 * See the Mulan PSL v2 for more details.
 */
#include "chcore/error.h"
#include "chcore/ipc.h"
#include "fcntl.h"
#include "fs_vnode.h"
#include "pthread.h"
#include "sys/stat.h"
#include "sys/mman.h"
#include <chcore/container/radix.h>
#include <fs_wrapper_defs.h>
#include <chcore/falloc.h>
#include "defs.h"
#include "namei.h"
#include <limits.h>
#include <stddef.h>
#include <string.h>
#include <stdint.h>

/*
 * This file contains interfaces that fs_base will invoke
 */

/* File operations */

/*
 * POSIX says:
 * The creat() function shall behave as if it is implemented as follows:
 *
 * int creat(const char *path, mode_t mode)
 * {
 *      return open(path, O_WRONLY|O_CREAT|O_TRUNC, mode);
 * }
 *
 * however, this should be done at the fs_base level,
 * when dealing with creat() syscall
 * we still implement a separate creat() function, and try to
 * obey POSIX in tmpfs
 */
int tmpfs_creat(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int err;
    struct nameidata nd;
    struct dentry *dentry;

    char *pathname = fr->creat.pathname;

    err = path_openat(&nd, pathname, O_CREAT, 0, &dentry);
    if (err) {
        goto error;
    }

    /* dentry->inode must be a regular file at this point */
    dentry->inode->f_ops->truncate(dentry->inode, 0);

error:
    return err;
}

int tmpfs_open(char *path, int o_flags, int mode, ino_t *vnode_id,
               off_t *vnode_size, int *vnode_type, void **vnode_private)
{
    /*
     * For monitoring memory usage in tmpfs
     *
     * The current naive implementation use open() to
     * trigger memory usage collecting and show usage info.
     * Details can be found in defs.h
     */
    if (!strcmp(path, "/tmpfs_mem_usage")) {
#if DEBUG_MEM_USAGE
        tmpfs_get_mem_usage();
#endif
        return -ENOENT; /* this should fail */
    }

    int err;
    struct nameidata nd;
    struct dentry *dentry;

    err = path_openat(&nd, path, o_flags, 0, &dentry);

    if (!err) {
        struct inode *inode = dentry->inode;

        *vnode_id = (ino_t)(uintptr_t)inode;
        *vnode_size = inode->size;
        *vnode_private = inode;

        /* vnode type can not be symlink? */
        switch (inode->type) {
        case FS_REG:
            *vnode_type = FS_NODE_REG;
            break;
        case FS_DIR:
            *vnode_type = FS_NODE_DIR;
            break;
        }

        inode->base_ops->open(inode);
    }

    return err;
}

int tmpfs_close(void *operator, bool is_dir /* not handled */, bool do_close)
{
    struct inode *inode = (struct inode *)operator;
    if (do_close) {
        inode->base_ops->close(inode);
    }

    return 0;
}

ssize_t tmpfs_read(void *operator, off_t offset, size_t size, char * buf)
{
    struct inode *inode = (struct inode *)operator;
    BUG_ON(inode->type != FS_REG);

    return inode->f_ops->read(inode, buf, size, offset);
}

ssize_t tmpfs_write(void *operator, off_t offset, size_t size, const char * buf)
{
    struct inode *inode = (struct inode *)operator;
    BUG_ON(inode->type != FS_REG);
    return inode->f_ops->write(inode, buf, size, offset);
}

int tmpfs_ftruncate(void *operator, off_t len)
{
    struct inode *inode = (struct inode *)operator;

    if (inode->type == FS_DIR)
        return -EISDIR;

    if (inode->type != FS_REG)
        return -EINVAL;

    return inode->f_ops->truncate(inode, len);
}

int tmpfs_fallocate(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int fd = fr->fallocate.fd;
    mode_t mode = fr->fallocate.mode;
    off_t offset = fr->fallocate.offset;
    off_t len = fr->fallocate.len;
    int keep_size, ret;
    struct inode *inode;
    struct fs_vnode *vnode;

    vnode = server_entrys[fd]->vnode;
    inode = (struct inode *)vnode->private;

    if (offset < 0 || len <= 0)
        return -EINVAL;

    if (inode->type != FS_REG) {
        return -EBADF;
    }

    pthread_rwlock_wrlock(&vnode->rwlock);

    /* return error if mode is not supported */
    if (mode
        & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE
            | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE
            | FALLOC_FL_INSERT_RANGE)) {
        ret = -EOPNOTSUPP;
        goto out;
    }

    if (mode & FALLOC_FL_PUNCH_HOLE) {
        ret = inode->f_ops->punch_hole(inode, offset, len);
    } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
        ret = inode->f_ops->collapse_range(inode, offset, len);
    } else if (mode & FALLOC_FL_ZERO_RANGE) {
        ret = inode->f_ops->zero_range(inode, offset, len, mode);
    } else if (mode & FALLOC_FL_INSERT_RANGE) {
        ret = inode->f_ops->insert_range(inode, offset, len);
    } else {
        keep_size = mode & FALLOC_FL_KEEP_SIZE ? 1 : 0;

        ret = inode->f_ops->allocate(inode, offset, len, keep_size);
    }

    /**
     * FALLOC_FL_KEEP_SIZE is handled in handlers above, i.e., if size
     * should be kept, the inode->size won't change, vice versa.
     */
    vnode->size = inode->size;
out:
    pthread_rwlock_unlock(&vnode->rwlock);
    return ret;
}

/* Directory operations */
int tmpfs_mkdir(const char *path, mode_t mode)
{
    int err;
    struct nameidata nd;
    struct dentry *d_parent, *d_new_dir;
    struct inode *i_parent;

    err = path_parentat(&nd, path, 0, &d_parent);
    if (err) {
        goto error;
    }
    /* only slashes */
    if (nd.last.str == NULL) {
        return -EBUSY;
    }

    /* creating the new dir */
    i_parent = d_parent->inode;
    d_new_dir = i_parent->d_ops->dirlookup(i_parent, nd.last.str, nd.last.len);
    if (d_new_dir) {
        err = -EEXIST;
        goto error;
    }

    d_new_dir = i_parent->d_ops->alloc_dentry();
    if (CHCORE_IS_ERR(d_new_dir)) {
        err = CHCORE_PTR_ERR(d_new_dir);
        goto error;
    }

    err = i_parent->d_ops->add_dentry(
        i_parent, d_new_dir, nd.last.str, nd.last.len);
    if (err) {
        goto free_dent;
    }

    err = i_parent->d_ops->mkdir(i_parent, d_new_dir, mode);
    if (err) {
        goto remove_dent;
    }

    free_string(&nd.last);
    return 0;

remove_dent:
    i_parent->d_ops->remove_dentry(i_parent, d_new_dir);
free_dent:
    i_parent->d_ops->free_dentry(d_new_dir);
error:
    free_string(&nd.last);
    return err;
}

int tmpfs_rmdir(const char *path, int flags)
{
    int err;
    struct nameidata nd;
    struct dentry *d_parent, *d_to_remove;
    struct inode *i_parent;

    err = path_parentat(&nd, path, 0, &d_parent);
    if (err) {
        goto error;
    }
    /* only slashes */
    if (nd.last.str == NULL) {
        return -EBUSY;
    }
    i_parent = d_parent->inode;

    d_to_remove =
        i_parent->d_ops->dirlookup(i_parent, nd.last.str, nd.last.len);
    if (!d_to_remove) {
        err = -ENOENT;
        goto error;
    }

    if (d_to_remove->inode->type != FS_DIR) {
        err = -ENOTDIR;
        goto error;
    }

    err = i_parent->d_ops->rmdir(i_parent, d_to_remove);

error:
    free_string(&nd.last);
    return err;
}

int tmpfs_getdents(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int fd = fr->getdents64.fd;
    unsigned int count = fr->getdents64.count;
    char *buff = ipc_get_msg_data(ipc_msg);

    struct inode *inode = (struct inode *)server_entrys[fd]->vnode->private;
    int ret = 0, read_bytes;
    if (inode) {
        if (inode->type == FS_DIR) {
            ret = inode->d_ops->scan(inode,
                                     server_entrys[fd]->offset,
                                     buff,
                                     buff + count,
                                     &read_bytes);

            server_entrys[fd]->offset += ret;
            ret = read_bytes;
        } else {
            ret = -ENOTDIR;
        }
    } else {
        ret = -ENOENT;
    }
    return ret;
}

/* Symbolic link operations */

int tmpfs_symlinkat(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int err;
    struct nameidata nd;
    struct dentry *d_parent, *d_symlink;
    struct inode *i_parent, *i_symlink;
    char *target, *path;

    target = fr->symlinkat.target;
    path = fr->symlinkat.linkpath;

    err = path_parentat(&nd, path, 0, &d_parent);
    if (err) {
        goto error;
    }
    /* only slashes */
    if (nd.last.str == NULL) {
        return -EBUSY;
    }
    i_parent = d_parent->inode;

    d_symlink = i_parent->d_ops->dirlookup(i_parent, nd.last.str, nd.last.len);
    if (d_symlink) {
        err = -EEXIST;
        goto error;
    }

    /* the path contains trailing slashes, return ENOTDIR per POSIX */
    if (nd.last.str[nd.last.len]) {
        err = -ENOTDIR;
        goto error;
    }

    d_symlink = i_parent->d_ops->alloc_dentry();
    if (CHCORE_IS_ERR(d_symlink)) {
        err = CHCORE_PTR_ERR(d_symlink);
        goto error;
    }

    err = i_parent->d_ops->add_dentry(
        i_parent, d_symlink, nd.last.str, nd.last.len);
    if (err) {
        goto free_dent;
    }

    err = i_parent->d_ops->mknod(i_parent, d_symlink, 0, FS_SYM);
    if (err) {
        goto remove_dent;
    }
    i_symlink = d_symlink->inode;

    ssize_t len =
        i_symlink->sym_ops->write_link(i_symlink, target, strlen(target));
    if (len != strlen(target)) {
        err = (int)len;
        goto free_node;
    }

    free_string(&nd.last);
    return 0;

free_node:
    i_symlink->base_ops->free(i_symlink);
remove_dent:
    i_parent->d_ops->remove_dentry(i_parent, d_symlink);
free_dent:
    i_parent->d_ops->free_dentry(d_symlink);
error:
    free_string(&nd.last);
    return err;
}

ssize_t tmpfs_readlinkat(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int err;
    struct nameidata nd;
    struct dentry *d_symlink;
    struct inode *i_symlink;

    const char *path;
    char *buff;
    ssize_t len;

    path = fr->readlinkat.pathname;
    buff = fr->readlinkat.buf;
    len = (ssize_t)fr->readlinkat.bufsiz;

    err = path_lookupat(&nd, path, 0, &d_symlink);
    if (err) {
        return err;
    }

    i_symlink = d_symlink->inode;
    if (i_symlink->type != FS_SYM) {
        return -EINVAL;
    }
    len = i_symlink->sym_ops->read_link(i_symlink, buff, len);

    return len;
}

/* File and directory operations */
int tmpfs_unlink(const char *path, int flags)
{
    int err;
    struct nameidata nd;
    struct dentry *d_parent, *d_unlink;
    struct inode *i_parent;

    if (flags & (~AT_SYMLINK_NOFOLLOW)) {
        return -EINVAL;
    }

    err = path_parentat(&nd, path, 0, &d_parent);
    if (err) {
        goto error;
    }
    /* only slashes */
    if (nd.last.str == NULL) {
        return -EBUSY;
    }

    i_parent = d_parent->inode;

    d_unlink = i_parent->d_ops->dirlookup(i_parent, nd.last.str, nd.last.len);
    if (!d_unlink) {
        err = -ENOENT;
        goto error;
    }

    if (d_unlink->inode->type == FS_DIR) {
        err = -EISDIR;
        goto error;
    }

    i_parent->d_ops->unlink(i_parent, d_unlink);

error:
    free_string(&nd.last);
    return err;
}

/*
 * Some facts is ensured when tmpfs_rename is called by fs_base:
 *     1. the two paths is not ended with dots
 *     2. oldpath actually exists
 *     3. oldpath is not an ancestor of new
 *     4. newpath has a valid, existed prefix and it's a dir
 *     5. if newpath exists, the two paths' types match
 *     6. if newpath exists, the newpath is removed.
 *
 * NOTE: dependency on fs_base
 * The implementation relies on these facts so it can be simple,
 * but if the implementation of fs_base changed, the implementation
 * here may also need modification
 */
int tmpfs_rename(const char *oldpath, const char *newpath)
{
    int err;
    struct nameidata nd;
    struct dentry *d_old_parent, *d_new_parent, *d_old, *d_new;
    struct inode *i_old_parent, *i_new_parent;

    err = path_parentat(&nd, oldpath, 0, &d_old_parent);
    if (err) {
        error("oldpath should exist!\n");
        goto error;
    }
    /* only slashes */
    if (nd.last.str == NULL) {
        return -EBUSY;
    }

    i_old_parent = d_old_parent->inode;

    d_old =
        i_old_parent->d_ops->dirlookup(i_old_parent, nd.last.str, nd.last.len);

    free_string(&nd.last);

    err = path_parentat(&nd, newpath, 0, &d_new_parent);
    if (err) {
        error("newpath prefix should exist!\n");
        goto error;
    }
    /* only slashes */
    if (nd.last.str == NULL) {
        return -EBUSY;
    }

    i_new_parent = d_new_parent->inode;

    d_new = i_new_parent->d_ops->alloc_dentry();
    if (CHCORE_IS_ERR(d_new)) {
        err = CHCORE_PTR_ERR(d_new);
        goto error;
    }

    err = i_new_parent->d_ops->add_dentry(
        i_new_parent, d_new, nd.last.str, nd.last.len);
    if (err) {
        goto free_dent;
    }

    i_old_parent->d_ops->rename(i_old_parent, d_old, i_new_parent, d_new);

    free_string(&nd.last);
    return 0;

free_dent:
    i_new_parent->d_ops->free_dentry(d_new);
error:
    free_string(&nd.last);
    return err;
}

/* File metadata operations */

int tmpfs_fstatat(const char *path, struct stat *st, int flags)
{
    int err;
    struct nameidata nd;
    struct dentry *dentry;
    struct inode *inode;

    /*
     * POSIX says we should follow the trailing symlink,
     * except AT_SYMLINK_NOFOLLOW is set in @flags
     */
    err = path_lookupat(
        &nd, path, flags & AT_SYMLINK_NOFOLLOW ? 0 : ND_FOLLOW, &dentry);
    if (err) {
        return err;
    }

    inode = dentry->inode;
    inode->base_ops->stat(inode, st);
    return 0;
}

int tmpfs_fstat(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    struct inode *inode;

    int fd = fr->stat.fd;
    struct stat *stat = (struct stat *)ipc_get_msg_data(ipc_msg);

    BUG_ON(!server_entrys[fd]);
    inode = (struct inode *)server_entrys[fd]->vnode->private;

    inode->base_ops->stat(inode, stat);

    return 0;
}

int tmpfs_faccessat(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int err;
    struct nameidata nd;
    struct dentry *dentry;

    const char *path = fr->faccessat.pathname;
    /* mode_t mode = fr->faccessat.mode; // again, mode is not handled */
    int flags = fr->faccessat.flags;

    if (flags & (~AT_EACCESS)) {
        return -EINVAL;
    }

    /* look the path up does all the job */
    err = path_lookupat(&nd, path, 0, &dentry);

    return err;
}

/* File system metadata operations */

int tmpfs_statfs(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int err;
    struct nameidata nd;
    struct dentry *dentry;

    /*
     * dirfd is actually guaranteed to be root dir,
     * so we are not using it in tmpfs
     */
    const char *path = fr->stat.pathname;
    struct statfs *stat = (struct statfs *)ipc_get_msg_data(ipc_msg);

    err = path_lookupat(&nd, path, 0, &dentry);
    if (!err) {
        tmpfs_fs_stat(stat);
    }

    return err;
}

int tmpfs_fstatfs(ipc_msg_t *ipc_msg, struct fs_request *fr)
{
    int fd = fr->stat.fd;
    struct statfs *stat = (struct statfs *)ipc_get_msg_data(ipc_msg);

    BUG_ON(!server_entrys[fd]);

    tmpfs_fs_stat(stat);

    return 0;
}

/* Additional functionalities */
int tmpfs_fcntl(void *operator, int fd, int fcntl_cmd, int fcntl_arg)
{
    struct server_entry *entry;
    int ret = 0;

    if ((entry = server_entrys[fd]) == NULL)
        return -EBADF;

    switch (fcntl_cmd) {
    case F_GETFL:
        ret = entry->flags;
        break;
    case F_SETFL: {
        // file access mode and the file creation flags
        // should be ignored, per POSIX
        int effective_bits =
            (~O_ACCMODE & ~O_CLOEXEC & ~O_CREAT & ~O_DIRECTORY & ~O_EXCL
             & ~O_NOCTTY & ~O_NOFOLLOW & ~O_TRUNC & ~O_TTY_INIT);

        entry->flags = (fcntl_arg & effective_bits)
                       | (entry->flags & ~effective_bits);
        break;
    }
    case F_DUPFD:
        break;

    case F_GETOWN:
        /* Fall through */
    case F_SETOWN:
    case F_GETLK:
    case F_SETLK:
    case F_SETLKW:
    default:
        error("unsupported fcntl cmd\n");
        ret = -1;
        break;
    }

    return ret;
}

#ifdef CHCORE_ENABLE_FMAP
vaddr_t tmpfs_get_page_addr(void *operator, off_t offset)
{
    struct inode *inode;
    void *page;
    off_t page_no;

    inode = (struct inode *)operator;
    page_no = offset / PAGE_SIZE;
    page = radix_get(&inode->data, page_no);

    /* the case of truncated to a larger size but not yet allocated */
    if (!page && offset < inode->size) {
        page = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
        if (!page) {
            return -ENOMEM;
        }
#if DEBUG_MEM_USAGE
        tmpfs_record_mem_usage(page, PAGE_SIZE, DATA_PAGE);
#endif

        /* set page to 0 and cause page to be actually mapped */
        memset(page, 0, PAGE_SIZE);
        radix_add(&inode->data, page_no, page);
    }

    return (vaddr_t)page;
}
#endif

struct fs_server_ops server_ops = {
    /* Unimplemented */
    .mount = default_server_operation,
    .umount = default_server_operation,

    /* File Operations */
    .creat = tmpfs_creat,
    .open = tmpfs_open,
    .close = tmpfs_close,
    .read = tmpfs_read,
    .write = tmpfs_write,
    .ftruncate = tmpfs_ftruncate,
    .fallocate = tmpfs_fallocate,

    /* Directory Operations */
    .mkdir = tmpfs_mkdir,
    .rmdir = tmpfs_rmdir,
    .getdents64 = tmpfs_getdents,

    /* Symbolic Link Operations */
    .symlinkat = tmpfs_symlinkat,
    .readlinkat = tmpfs_readlinkat,

    /* File and Directory Operations */
    .unlink = tmpfs_unlink,
    .rename = tmpfs_rename,

    /* File Metadata Operations */
    .fstatat = tmpfs_fstatat,
    .fstat = tmpfs_fstat,
    .faccessat = tmpfs_faccessat,

    /* File System Metadata Operations */
    .statfs = tmpfs_statfs,
    .fstatfs = tmpfs_fstatfs,

    /* Additional Functionalities */
    .fcntl = tmpfs_fcntl,
#ifdef CHCORE_ENABLE_FMAP
    .fmap_get_page_addr = tmpfs_get_page_addr,
#endif
};