从zhong duan 到 zhong duan

这是ucore step by step tutorial的最后一节: 实现一个简单的终端(shell)。

可以说，我们的操作系统之旅，从zhong duan(中断)开始， 也在zhong duan(终端）结束。

我们的终端需要实现这样的功能: 根据输入的程序名称, 从文件系统里加载对应的程序并执行。我们采取fork() exec()的方式来加载执行程序，exec()的一系列接口都需要重写来使用文件系统。以do_execve()为例，

以前的函数原型从内存的某个位置加载程序

int do_execve(const char *name, size_t len, unsigned char *binary, size_t size) ;

现在则调用文件系统接口加载程序：

// kern/process/proc.c
// do_execve - call exit_mmap(mm)&put_pgdir(mm) to reclaim memory space of current process
//           - call load_icode to setup new memory space accroding binary prog.
int
do_execve(const char *name, int argc, const char **argv) {
    static_assert(EXEC_MAX_ARG_LEN >= FS_MAX_FPATH_LEN);
    struct mm_struct *mm = current->mm;
    if (!(argc >= 1 && argc <= EXEC_MAX_ARG_NUM)) {
        return -E_INVAL;
    }

    char local_name[PROC_NAME_LEN + 1];
    memset(local_name, 0, sizeof(local_name));

    char *kargv[EXEC_MAX_ARG_NUM];
    const char *path;

    int ret = -E_INVAL;

    lock_mm(mm);
    if (name == NULL) {
        snprintf(local_name, sizeof(local_name), "<null> %d", current->pid);
    }
    else {
        if (!copy_string(mm, local_name, name, sizeof(local_name))) {
            unlock_mm(mm);
            return ret;
        }
    }
    if ((ret = copy_kargv(mm, argc, kargv, argv)) != 0) {
        unlock_mm(mm);
        return ret;
    }
    path = argv[0];
    unlock_mm(mm);
    files_closeall(current->filesp);

    /* sysfile_open will check the first argument path, thus we have to use a user-space pointer, and argv[0] may be incorrect */
    int fd;
    if ((ret = fd = sysfile_open(path, O_RDONLY)) < 0) {
        goto execve_exit;
    }
    if (mm != NULL) {
        lcr3(boot_cr3);
        if (mm_count_dec(mm) == 0) {
            exit_mmap(mm);
            put_pgdir(mm);
            mm_destroy(mm);
        }
        current->mm = NULL;
    }
    ret= -E_NO_MEM;;
    if ((ret = load_icode(fd, argc, kargv)) != 0) {
        goto execve_exit;
    }
    put_kargv(argc, kargv);
    set_proc_name(current, local_name);
    return 0;

execve_exit:
    put_kargv(argc, kargv);
    do_exit(ret);
    panic("already exit: %e.\n", ret);
}

我们还要看一下终端程序的实现。可以发现终端程序需要对命令进行词法和语法分析。

// user/sh.c
#include <ulib.h>
#include <stdio.h>
#include <string.h>
#include <dir.h>
#include <file.h>
#include <error.h>
#include <unistd.h>

#define printf(...)                     fprintf(1, __VA_ARGS__)
#define putc(c)                         printf("%c", c)

#define BUFSIZE                         4096
#define WHITESPACE                      " \t\r\n"
#define SYMBOLS                         "<|>&;"

char shcwd[BUFSIZE];

int
gettoken(char **p1, char **p2) {
    char *s;
    if ((s = *p1) == NULL) {
        return 0;
    }
    while (strchr(WHITESPACE, *s) != NULL) {
        *s ++ = '\0';
    }
    if (*s == '\0') {
        return 0;
    }

    *p2 = s;
    int token = 'w';
    if (strchr(SYMBOLS, *s) != NULL) {
        token = *s, *s ++ = '\0';
    }
    else {
        bool flag = 0;
        while (*s != '\0' && (flag || strchr(WHITESPACE SYMBOLS, *s) == NULL)) {
            if (*s == '"') {
                *s = ' ', flag = !flag;
            }
            s ++;
        }
    }
    *p1 = (*s != '\0' ? s : NULL);
    return token;
}

char * readline(const char *prompt) {
    static char buffer[BUFSIZE];
    if (prompt != NULL) {
        printf("%s", prompt);
    }
    int ret, i = 0;
    while (1) {
        char c;
        if ((ret = read(0, &c, sizeof(char))) < 0) {
            return NULL;
        }
        else if (ret == 0) {
            if (i > 0) {
                buffer[i] = '\0';
                break;
            }
            return NULL;
        }

        if (c == 3) {
            return NULL;
        }
        else if (c >= ' ' && i < BUFSIZE - 1) {
            putc(c);
            buffer[i ++] = c;
        }
        else if (c == '\b' && i > 0) {
            putc(c);
            i --;
        }
        else if (c == '\n' || c == '\r') {
            putc(c);
            buffer[i] = '\0';
            break;
        }
    }
    return buffer;
}

void
usage(void) {
    printf("usage: sh [command-file]\n");
}

int
reopen(int fd2, const char *filename, uint32_t open_flags) {
    int ret, fd1;
    close(fd2);
    if ((ret = open(filename, open_flags)) >= 0 && ret != fd2) {
        close(fd2);
        fd1 = ret, ret = dup2(fd1, fd2);
        close(fd1);
    }
    return ret < 0 ? ret : 0;
}

int
testfile(const char *name) {
    int ret;
    if ((ret = open(name, O_RDONLY)) < 0) {
        return ret;
    }
    close(ret);
    return 0;
}

int
runcmd(char *cmd) {
    static char argv0[BUFSIZE];
    static const char *argv[EXEC_MAX_ARG_NUM + 1];//must be static!
    char *t;
    int argc, token, ret, p[2];
again:
    argc = 0;
    while (1) {
        switch (token = gettoken(&cmd, &t)) {
        case 'w':
            if (argc == EXEC_MAX_ARG_NUM) {
                printf("sh error: too many arguments\n");
                return -1;
            }
            argv[argc ++] = t;
            break;
        case '<':
            if (gettoken(&cmd, &t) != 'w') {
                printf("sh error: syntax error: < not followed by word\n");
                return -1;
            }
            if ((ret = reopen(0, t, O_RDONLY)) != 0) {
                return ret;
            }
            break;
        case '>':
            if (gettoken(&cmd, &t) != 'w') {
                printf("sh error: syntax error: > not followed by word\n");
                return -1;
            }
            if ((ret = reopen(1, t, O_RDWR | O_TRUNC | O_CREAT)) != 0) {
                return ret;
            }
            break;
        case '|':
          //  if ((ret = pipe(p)) != 0) {
          //      return ret;
          //  }
            if ((ret = fork()) == 0) {
                close(0);
                if ((ret = dup2(p[0], 0)) < 0) {
                    return ret;
                }
                close(p[0]), close(p[1]);
                goto again;
            }
            else {
                if (ret < 0) {
                    return ret;
                }
                close(1);
                if ((ret = dup2(p[1], 1)) < 0) {
                    return ret;
                }
                close(p[0]), close(p[1]);
                goto runit;
            }
            break;
        case 0:
            goto runit;
        case ';':
            if ((ret = fork()) == 0) {
                goto runit;
            }
            else {
                if (ret < 0) {
                    return ret;
                }
                waitpid(ret, NULL);
                goto again;
            }
            break;
        default:
            printf("sh error: bad return %d from gettoken\n", token);
            return -1;
        }
    }

runit:
    if (argc == 0) {
        return 0;
    }
    else if (strcmp(argv[0], "cd") == 0) {
        if (argc != 2) {
            return -1;
        }
        strcpy(shcwd, argv[1]);
        return 0;
    }
    if ((ret = testfile(argv[0])) != 0) {
        if (ret != -E_NOENT) {
            return ret;
        }
        snprintf(argv0, sizeof(argv0), "/%s", argv[0]);
        argv[0] = argv0;
    }
    argv[argc] = NULL;
    return __exec(argv[0], argv);
}

int main(int argc, char **argv) {
    cputs("user sh is running!!!");
    int ret, interactive = 1;
    if (argc == 2) {
        if ((ret = reopen(0, argv[1], O_RDONLY)) != 0) {
            return ret;
        }
        interactive = 0;
    }
    else if (argc > 2) {
        usage();
        return -1;
    }
    //shcwd = malloc(BUFSIZE);
    assert(shcwd != NULL);

    char *buffer;
    while ((buffer = readline((interactive) ? "$ " : NULL)) != NULL) {
        shcwd[0] = '\0';
        int pid;
        if ((pid = fork()) == 0) {
            ret = runcmd(buffer);
            exit(ret);
        }
        assert(pid >= 0);
        if (waitpid(pid, &ret) == 0) {
            if (ret == 0 && shcwd[0] != '\0') {
                ret = 0;
            }
            if (ret != 0) {
                printf("error: %d - %e\n", ret, ret);
            }
        }
    }
    return 0;
}

如果我们能够把终端运行起来，并能输入命令执行用户程序，就说明程序运行正常。

目前的代码可以在这里找到。