linux内核之情景分析mmap操作

进程可以通过mmap把一个已打开文件映射到用户空间.
  1. mmap(void*start,size_t length,int prot,int flags,int fd,off_t offset)
start表示用户空间映射的起始地址,offset文件的起始length长度.
  2. asmlinkage long sys_mmap2(unsigned long addr, unsigned long len,
  3. 	unsigned long prot, unsigned long flags,
  4. 	unsigned long fd, unsigned long pgoff)
  5. {
  6. 	return do_mmap2(addr, len, prot, flags, fd, pgoff);
  7. }
其主体是do_mmap2,注意其标志MAP_ANONYMOUS表示匿名映射
  2. /* common code for old and new mmaps */
  3. static inline long do_mmap2(
  4. 	unsigned long addr, unsigned long len,
  5. 	unsigned long prot, unsigned long flags,
  6. 	unsigned long fd, unsigned long pgoff)
  7. {
  8. 	int error = -EBADF;
  9. 	struct file * file = NULL;
  11. 	flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
  12. 	if (!(flags & MAP_ANONYMOUS)) {//map_anonymous表示没有文件,只是在指定位置分配内存
  13. 		file = fget(fd);//上一条表示,没有文件,就跳过if以下,有文件则打开文件
  14. 		if (!file)//如果文件不存在,直接返回错误
  15. 			goto out;
  16. 	}
  18. 	down(&current->mm->mmap_sem);//信号量down操作
  19. 	error = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);//mmap主体操作还是这个
  20. 	up(&current->mm->mmap_sem);//信号量up操作
  22. 	if (file)
  23. 		fput(file);
  24. out:
  25. 	return error;
  26. }
其主体为do_mmap_pgoff
  1. do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
第一个参数为打开文件,第二个地址,第三长度,第四个参数为访问权限,第五个参数为其他控制目的,第6个为偏移量
  2. unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, unsigned long len,
  3. 	unsigned long prot, unsigned long flags, unsigned long pgoff)
  4. {
  5. 	struct mm_struct * mm = current->mm;//获取当前进程的内存描述符
  6. 	struct vm_area_struct * vma;
  7. 	int correct_wcount = 0;
  8. 	int error;
  9.    //file非0表示是文件,其对应一定有相关操作函数.
  10. 	if (file && (!file->f_op || !file->f_op->mmap))
  11. 		return -ENODEV;
  12.   //长度对齐,如果为0,直接返回
  13. 	if ((len = PAGE_ALIGN(len)) == 0)
  14. 		return addr;
  15. //长度大于3g或者addr+len映射区域超过用户空间,返回错误
  16. 	if (len > TASK_SIZE || addr > TASK_SIZE-len)
  17. 		return -EINVAL;
  18. // 偏移量是否超过了长度
  19. 	/* offset overflow? */
  20. 	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
  21. 		return -EINVAL;
  22. //映射次数是否超过了限定
  23. 	/* Too many mappings? */
  24. 	if (mm->map_count > MAX_MAP_COUNT)
  25. 		return -ENOMEM;
  26. //是否加锁?这里不知道了
  27. 	/* mlock MCL_FUTURE? */
  28. 	if (mm->def_flags & VM_LOCKED) {
  29. 		unsigned long locked = mm->locked_vm << PAGE_SHIFT;
  30. 		locked += len;
  31. 		if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
  32. 			return -EAGAIN;
  33. 	}
  35. 	/* Do simple checking here so the lower-level routines won't have
  36. 	 * to. we assume access permissions have been handled by the open
  37. 	 * of the memory object, so we don't do any here.
  38. 	 */
  39. 	if (file != NULL) { //如果文件存在
  40. 		switch (flags & MAP_TYPE) {//映射类型:读写
  41. 		case MAP_SHARED://共享映射
  42. 			if ((prot & PROT_WRITE) && !(file->f_mode & FMODE_WRITE))
  43. 				return -EACCES;
  44.              //确保我们不被允许写在一个只可追加的文件
  45. 			/* Make sure we don't allow writing to an append-only file.. */
  46. 			if (IS_APPEND(file->f_dentry->d_inode) && (file->f_mode & FMODE_WRITE))
  47. 				return -EACCES;
  48.             //确保我们的文件没有锁
  49. 			/* make sure there are no mandatory locks on the file. */
  50. 			if (locks_verify_locked(file->f_dentry->d_inode))
  51. 				return -EAGAIN;
  53. 			/* fall through */
  54. 		case MAP_PRIVATE://私有映射
  55. 			if (!(file->f_mode & FMODE_READ))
  56. 				return -EACCES;
  57. 			break;
  59. 		default:
  60. 			return -EINVAL;
  61. 		}
  62. 	}
  64. 	/* Obtain the address to map to. we verify (or select) it and ensure
  65. 	 * that it represents a valid section of the address space.
  66. 	 */
  67. 	if (flags & MAP_FIXED) {//如果参数flag的标志位map_fixed为0表示,指定映射位置只是一个参考值
  68. 		if (addr & ~PAGE_MASK)
  69. 			return -EINVAL;
  70. 	} else {//不满足由内核从空洞执行分配一个区域
  71. 		addr = get_unmapped_area(addr, len);
  72. 		if (!addr)
  73. 			return -ENOMEM;
  74. 	}
  76. 	/* Determine the object being mapped and call the appropriate
  77. 	 * specific mapper. the address has already been validated, but
  78. 	 * not unmapped, but the maps are removed from the list.
  79. 	 */
  80. 	vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);//从slab获取一个vma结构
  81. 	if (!vma)
  82. 		return -ENOMEM;
  84. 	vma->vm_mm = mm;//指向内存描述符
  85. 	vma->vm_start = addr;//vma的起始地址指向映射的起始地址
  86. 	vma->vm_end = addr + len;//同上
  87. 	vma->vm_flags = vm_flags(prot,flags) | mm->def_flags;//设置vma属性
  89. 	if (file) {//如果file为0,表示匿名映射,仅仅是为了创建虚拟区间,或者仅在于建立从物理空间到虚存空间映射,而非文件映射
  90. 		VM_ClearReadHint(vma);//以下代码设置一堆属性
  91. 		vma->vm_raend = 0;
  93. 		if (file->f_mode & FMODE_READ)
  94. 			vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
  95. 		if (flags & MAP_SHARED) {
  96. 			vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
  98. 			/* This looks strange, but when we don't have the file open
  99. 			 * for writing, we can demote the shared mapping to a simpler
  100. 			 * private mapping. That also takes care of a security hole
  101. 			 * with ptrace() writing to a shared mapping without write
  102. 			 * permissions.
  103. 			 *
  104. 			 * We leave the VM_MAYSHARE bit on, just to get correct output
  105. 			 * from /proc/xxx/maps..
  106. 			 */
  107. 			if (!(file->f_mode & FMODE_WRITE))
  108. 				vma->vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
  109. 		}
  110. 	} else {
  111. 		vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
  112. 		if (flags & MAP_SHARED)
  113. 			vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
  114. 	}
  115. 	vma->vm_page_prot = protection_map[vma->vm_flags & 0x0f];
  116. 	vma->vm_ops = NULL;
  117. 	vma->vm_pgoff = pgoff;//表示所映射内容在文件的起点,此值用于发生缺页异常根据虚存地址计算出相应页面的文件位置
  118. 	vma->vm_file = NULL;
  119. 	vma->vm_private_data = NULL;
  121. 	/* Clear old maps */
  122. 	error = -ENOMEM;
  123. 	if (do_munmap(mm, addr, len))//检查目的地址的vma空间是否已经使用(如果map_fixed设置为1的话)
  124. 		goto free_vma;//已经使用则释放free_vma
  125.     //检查是否超过了限制
  126. 	/* Check against address space limit. */
  127. 	if ((mm->total_vm << PAGE_SHIFT) + len
  128. 	    > current->rlim[RLIMIT_AS].rlim_cur)
  129. 		goto free_vma;
  130.     //检查当前进程专用的可写区间而物理页面不足的情况
  131. 	/* Private writable mapping? Check memory availability.. */
  132. 	if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
  133. 	    !(flags & MAP_NORESERVE)				 &&
  134. 	    !vm_enough_memory(len >> PAGE_SHIFT))
  135. 		goto free_vma;
  137. 	if (file) {//vm_deanwrite职位表示从文件到vma映射,表示不允许同过常规方式读写文件
  138. 		if (vma->vm_flags & VM_DENYWRITE) {
  139. 			error = deny_write_access(file);
  140. 			if (error)
  141. 				goto free_vma;
  142. 			correct_wcount = 1;
  143. 		}
  144. 		vma->vm_file = file;
  145. 		get_file(file);//递增file结构的共享计数
  146. 		error = file->f_op->mmap(file, vma);//一个文件操作必须存在mmap,否则释放vma
  147. 		if (error)
  148. 			goto unmap_and_free_vma;
  149. 	} else if (flags & MAP_SHARED) {//共享映射
  150. 		error = shmem_zero_setup(vma);
  151. 		if (error)
  152. 			goto free_vma;
  153. 	}
  155. 	/* Can addr have changed??
  156. 	 *为了防止flags与addr有变化,再重新设置一遍,
  157. 	 * Answer: Yes, several device drivers can do it in their
  158. 	 *         f_op->mmap method. -DaveM
  159. 	 */
  160. 	flags = vma->vm_flags;
  161. 	addr = vma->vm_start;
  163. 	insert_vm_struct(mm, vma);//插入当前进程的内存描述符
  164. 	if (correct_wcount)
  165. 		atomic_inc(&file->f_dentry->d_inode->i_writecount);
  166. 	
  167. 	mm->total_vm += len >> PAGE_SHIFT;//映射区域+len>>page_shit
  168. 	if (flags & VM_LOCKED) {//需要加锁
  169. 		mm->locked_vm += len >> PAGE_SHIFT;
  170. 		make_pages_present(addr, addr + len);//建立初始映射
  171. 	}
  172. 	return addr;
  174. unmap_and_free_vma:
  175. 	if (correct_wcount)
  176. 		atomic_inc(&file->f_dentry->d_inode->i_writecount);
  177. 	vma->vm_file = NULL;
  178. 	fput(file);
  179. 	/* Undo any partial mapping done by a device driver. */
  180. 	flush_cache_range(mm, vma->vm_start, vma->vm_end);
  181. 	zap_page_range(mm, vma->vm_start, vma->vm_end - vma->vm_start);
  182. 	flush_tlb_range(mm, vma->vm_start, vma->vm_end);
  183. free_vma:
  184. 	kmem_cache_free(vm_area_cachep, vma);
  185. 	return error;
  186. }
以上是文件与虚拟区间之间建立的映射,但具体的映射(从虚拟地址映射到物理地址)还没开始,而是把具体页面的映射推迟到真正需要的时候才进行,具体映射的简历,物理页面的换入和换出分别准备了一些函数,filemap_nopage(),ext2_readpage(),ext2_writepage()
什么时候调用呢
(1)该区间中的一个页面首次收到访问时,会由于页面没映射发生缺页异常,相应的异常处理程序do_no_page(),对于ext2系统,do_no_page()会通过ext2_readpage()分配一个空闲内存页面并从文件读入相应页面,并建立映射.
(2)建立映射后,往页面写使得页面变脏,但页面的内容并不会立即写回文件.而是由内核线程bdflush()周期性的运行时通过page_launder()间接调用ext2_writepage(),将页面的内容写入文件.如果页面很长时间没有收到访问,那就会被try_to_swap_out()解除映射而转入不活跃状态,如果页面是脏的那就也调用ext2_writepage()写入然后再解除映射
(3)解除了映射的页面再次收到访问时又会发生缺页异常,因为页面无映射进入do_no_page()
mmap映射,如果文件映射的一个页面长期得不到访问,将直接把页表项设置为0,如果访问到将重新alloc_page分配一个新页面,然后把文件读取到新页面,再建立映射,对于普通的换入/换出则是发生缺页异常从swap分区查找到换出的页面,然后建立映射













来自为知笔记(Wiz)


posted on 2016-10-27 10:53  笨拙的菜鸟  阅读(2242)  评论(0)    收藏  举报

导航