exec.c

unix/linux 编程实践一书的所有源代码

/*
 *  linux/fs/exec.c
 *
 *  Copyright (C) 1991, 1992  Linus Torvalds
 */

/*
 * #!-checking implemented by tytso.
 */
/*
 * Demand-loading implemented 01.12.91 - no need to read anything but
 * the header into memory. The inode of the executable is put into
 * "current->executable", and page faults do the actual loading. Clean.
 *
 * Once more I can proudly say that linux stood up to being changed: it
 * was less than 2 hours work to get demand-loading completely implemented.
 *
 * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
 * current->executable is only used by the procfs.  This allows a dispatch
 * table to check for several different types  of binary formats.  We keep
 * trying until we recognize the file or we run out of supported binary
 * formats. 
 */

#include <linux/fs.h>
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/a.out.h>
#include <linux/errno.h>
#include <linux/signal.h>
#include <linux/string.h>
#include <linux/stat.h>
#include <linux/fcntl.h>
#include <linux/ptrace.h>
#include <linux/user.h>
#include <linux/malloc.h>
#include <linux/binfmts.h>
#include <linux/personality.h>

#include <asm/system.h>
#include <asm/segment.h>
#include <asm/pgtable.h>

#include <linux/config.h>
#ifdef CONFIG_KERNELD
#include <linux/kerneld.h>
#endif

asmlinkage int sys_exit(int exit_code);
asmlinkage int sys_brk(unsigned long);

/*
 * Here are the actual binaries that will be accepted:
 * add more with "register_binfmt()" if using modules...
 *
 * These are defined again for the 'real' modules if you are using a
 * module definition for these routines.
 */

static struct linux_binfmt *formats = (struct linux_binfmt *) NULL;

void binfmt_setup(void)
{
#ifdef CONFIG_BINFMT_ELF
	init_elf_binfmt();
#endif

#ifdef CONFIG_BINFMT_AOUT
	init_aout_binfmt();
#endif

#ifdef CONFIG_BINFMT_JAVA
	init_java_binfmt();
#endif
	/* This cannot be configured out of the kernel */
	init_script_binfmt();
}

int register_binfmt(struct linux_binfmt * fmt)
{
	struct linux_binfmt ** tmp = &formats;

	if (!fmt)
		return -EINVAL;
	if (fmt->next)
		return -EBUSY;
	while (*tmp) {
		if (fmt == *tmp)
			return -EBUSY;
		tmp = &(*tmp)->next;
	}
	fmt->next = formats;
	formats = fmt;
	return 0;	
}

#ifdef CONFIG_MODULES
int unregister_binfmt(struct linux_binfmt * fmt)
{
	struct linux_binfmt ** tmp = &formats;

	while (*tmp) {
		if (fmt == *tmp) {
			*tmp = fmt->next;
			return 0;
		}
		tmp = &(*tmp)->next;
	}
	return -EINVAL;
}
#endif	/* CONFIG_MODULES */

int open_inode(struct inode * inode, int mode)
{
	int fd;

	if (!inode->i_op || !inode->i_op->default_file_ops)
		return -EINVAL;
	fd = get_unused_fd();
	if (fd >= 0) {
		struct file * f = get_empty_filp();
		if (!f) {
			put_unused_fd(fd);
			return -ENFILE;
		}
		f->f_flags = mode;
		f->f_mode = (mode+1) & O_ACCMODE;
		f->f_inode = inode;
		f->f_pos = 0;
		f->f_reada = 0;
		f->f_op = inode->i_op->default_file_ops;
		if (f->f_op->open) {
			int error = f->f_op->open(inode,f);
			if (error) {
				f->f_count--;
				put_unused_fd(fd);
				return error;
			}
		}
		current->files->fd[fd] = f;
		inode->i_count++;
	}
	return fd;
}

/*
 * Note that a shared library must be both readable and executable due to
 * security reasons.
 *
 * Also note that we take the address to load from from the file itself.
 */
asmlinkage int sys_uselib(const char * library)
{
	int fd, retval;
	struct file * file;
	struct linux_binfmt * fmt;

	fd = sys_open(library, 0, 0);
	if (fd < 0)
		return fd;
	file = current->files->fd[fd];
	retval = -ENOEXEC;
	if (file && file->f_inode && file->f_op && file->f_op->read) {
		for (fmt = formats ; fmt ; fmt = fmt->next) {
			int (*fn)(int) = fmt->load_shlib;
			if (!fn)
				continue;
			retval = fn(fd);
			if (retval != -ENOEXEC)
				break;
		}
	}
	sys_close(fd);
  	return retval;
}

/*
 * count() counts the number of arguments/envelopes
 *
 * We also do some limited EFAULT checking: this isn't complete, but
 * it does cover most cases. I'll have to do this correctly some day..
 */
static int count(char ** argv)
{
	int error, i = 0;
	char ** tmp, *p;

	if ((tmp = argv) != NULL) {
		error = verify_area(VERIFY_READ, tmp, sizeof(char *));
		if (error)
			return error;
		while ((p = get_user(tmp++)) != NULL) {
			i++;
			error = verify_area(VERIFY_READ, p, 1);
			if (error)
				return error;
		}
	}
	return i;
}

/*
 * 'copy_string()' copies argument/envelope strings from user
 * memory to free pages in kernel mem. These are in a format ready
 * to be put directly into the top of new user memory.
 *
 * Modified by TYT, 11/24/91 to add the from_kmem argument, which specifies
 * whether the string and the string array are from user or kernel segments:
 * 
 * from_kmem     argv *        argv **
 *    0          user space    user space
 *    1          kernel space  user space
 *    2          kernel space  kernel space
 * 
 * We do this by playing games with the fs segment register.  Since it
 * is expensive to load a segment register, we try to avoid calling
 * set_fs() unless we absolutely have to.
 */
unsigned long copy_strings(int argc,char ** argv,unsigned long *page,
		unsigned long p, int from_kmem)
{
	char *tmp, *tmp1, *pag = NULL;
	int len, offset = 0;
	unsigned long old_fs, new_fs;

	if (!p)
		return 0;	/* bullet-proofing */
	new_fs = get_ds();
	old_fs = get_fs();
	if (from_kmem==2)
		set_fs(new_fs);
	while (argc-- > 0) {
		if (from_kmem == 1)
			set_fs(new_fs);
		if (!(tmp1 = tmp = get_user(argv+argc)))
			panic("VFS: argc is wrong");
		if (from_kmem == 1)
			set_fs(old_fs);
		while (get_user(tmp++));
		len = tmp - tmp1;
		if (p < len) {	/* this shouldn't happen - 128kB */
			set_fs(old_fs);
			return 0;
		}
		while (len) {
			--p; --tmp; --len;
			if (--offset < 0) {
				offset = p % PAGE_SIZE;
				if (from_kmem==2)
					set_fs(old_fs);
				if (!(pag = (char *) page[p/PAGE_SIZE]) &&
				    !(pag = (char *) page[p/PAGE_SIZE] =
				      (unsigned long *) get_free_page(GFP_USER))) 
					return 0;
				if (from_kmem==2)
					set_fs(new_fs);

			}
			if (len == 0 || offset == 0)
			  *(pag + offset) = get_user(tmp);
			else {
			  int bytes_to_copy = (len > offset) ? offset : len;
			  tmp -= bytes_to_copy;
			  p -= bytes_to_copy;
			  offset -= bytes_to_copy;
			  len -= bytes_to_copy;
			  memcpy_fromfs(pag + offset, tmp, bytes_to_copy + 1);
			}
		}
	}
	if (from_kmem==2)
		set_fs(old_fs);
	return p;
}

unsigned long setup_arg_pages(unsigned long p, struct linux_binprm * bprm)
{
	unsigned long stack_base;
	struct vm_area_struct *mpnt;
	int i;

	stack_base = STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE;

	p += stack_base;
	if (bprm->loader)
		bprm->loader += stack_base;
	bprm->exec += stack_base;

	mpnt = (struct vm_area_struct *)kmalloc(sizeof(*mpnt), GFP_KERNEL);
	if (mpnt) {
		mpnt->vm_mm = current->mm;
		mpnt->vm_start = PAGE_MASK & (unsigned long) p;
		mpnt->vm_end = STACK_TOP;
		mpnt->vm_page_prot = PAGE_COPY;
		mpnt->vm_flags = VM_STACK_FLAGS;
		mpnt->vm_ops = NULL;
		mpnt->vm_offset = 0;
		mpnt->vm_inode = NULL;
		mpnt->vm_pte = 0;
		insert_vm_struct(current->mm, mpnt);
		current->mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
	}

	for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
		if (bprm->page[i]) {
			current->mm->rss++;
			put_dirty_page(current,bprm->page[i],stack_base);
		}
		stack_base += PAGE_SIZE;
	}
	return p;
}

/*
 * Read in the complete executable. This is used for "-N" files
 * that aren't on a block boundary, and for files on filesystems
 * without bmap support.
 */
int read_exec(struct inode *inode, unsigned long offset,
	char * addr, unsigned long count, int to_kmem)
{
	struct file file;
	int result = -ENOEXEC;

	if (!inode->i_op || !inode->i_op->default_file_ops)
		goto end_readexec;
	file.f_mode = 1;
	file.f_flags = 0;
	file.f_count = 1;
	file.f_inode = inode;
	file.f_pos = 0;
	file.f_reada = 0;