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.
  1. asmlinkage long sys_mmap2(unsigned long addr, unsigned long len,
  2. unsigned long prot, unsigned long flags,
  3. unsigned long fd, unsigned long pgoff)
  4. {
  5. return do_mmap2(addr, len, prot, flags, fd, pgoff);
  6. }
其主体是do_mmap2,注意其标志MAP_ANONYMOUS表示匿名映射
  1. /* common code for old and new mmaps */
  2. static inline long do_mmap2(
  3. unsigned long addr, unsigned long len,
  4. unsigned long prot, unsigned long flags,
  5. unsigned long fd, unsigned long pgoff)
  6. {
  7. int error = -EBADF;
  8. struct file * file = NULL;
  9. flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
  10. if (!(flags & MAP_ANONYMOUS)) {//map_anonymous表示没有文件,只是在指定位置分配内存
  11. 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
  12. if (!file)//如果文件不存在,直接返回错误
  13. goto out;
  14. }
  15. down(&current->mm->mmap_sem);//信号量down操作
  16. error = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);//mmap主体操作还是这个
  17. up(&current->mm->mmap_sem);//信号量up操作
  18. if (file)
  19. fput(file);
  20. out:
  21. return error;
  22. }
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
  1. unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, unsigned long len,
  2. unsigned long prot, unsigned long flags, unsigned long pgoff)
  3. {
  4. struct mm_struct * mm = current->mm;//获取当前进程的内存描述符
  5. struct vm_area_struct * vma;
  6. int correct_wcount = 0;
  7. int error;
  8. //file非0表示是文件,其对应一定有相关操作函数.
  9. if (file && (!file->f_op || !file->f_op->mmap))
  10. return -ENODEV;
  11. //长度对齐,如果为0,直接返回
  12. if ((len = PAGE_ALIGN(len)) == 0)
  13. return addr;
  14. //长度大于3g或者addr+len映射区域超过用户空间,返回错误
  15. if (len > TASK_SIZE || addr > TASK_SIZE-len)
  16. return -EINVAL;
  17. // 偏移量是否超过了长度
  18. /* offset overflow? */
  19. if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
  20. return -EINVAL;
  21. //映射次数是否超过了限定
  22. /* Too many mappings? */
  23. if (mm->map_count > MAX_MAP_COUNT)
  24. return -ENOMEM;
  25. //是否加锁?这里不知道了
  26. /* mlock MCL_FUTURE? */
  27. if (mm->def_flags & VM_LOCKED) {
  28. unsigned long locked = mm->locked_vm << PAGE_SHIFT;
  29. locked += len;
  30. if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
  31. return -EAGAIN;
  32. }
  33. /* Do simple checking here so the lower-level routines won't have
  34. * to. we assume access permissions have been handled by the open
  35. * of the memory object, so we don't do any here.
  36. */
  37. if (file != NULL) { //如果文件存在
  38. switch (flags & MAP_TYPE) {//映射类型:读写
  39. case MAP_SHARED://共享映射
  40. if ((prot & PROT_WRITE) && !(file->f_mode & FMODE_WRITE))
  41. return -EACCES;
  42. //确保我们不被允许写在一个只可追加的文件
  43. /* Make sure we don't allow writing to an append-only file.. */
  44. if (IS_APPEND(file->f_dentry->d_inode) && (file->f_mode & FMODE_WRITE))
  45. return -EACCES;
  46. //确保我们的文件没有锁
  47. /* make sure there are no mandatory locks on the file. */
  48. if (locks_verify_locked(file->f_dentry->d_inode))
  49. return -EAGAIN;
  50. /* fall through */
  51. case MAP_PRIVATE://私有映射
  52. if (!(file->f_mode & FMODE_READ))
  53. return -EACCES;
  54. break;
  55. default:
  56. return -EINVAL;
  57. }
  58. }
  59. /* Obtain the address to map to. we verify (or select) it and ensure
  60. * that it represents a valid section of the address space.
  61. */
  62. 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
  63. if (addr & ~PAGE_MASK)
  64. return -EINVAL;
  65. } else {//不满足由内核从空洞执行分配一个区域
  66. addr = get_unmapped_area(addr, len);
  67. if (!addr)
  68. return -ENOMEM;
  69. }
  70. /* Determine the object being mapped and call the appropriate
  71. * specific mapper. the address has already been validated, but
  72. * not unmapped, but the maps are removed from the list.
  73. */
  74. vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);//从slab获取一个vma结构
  75. if (!vma)
  76. return -ENOMEM;
  77. vma->vm_mm = mm;//指向内存描述符
  78. vma->vm_start = addr;//vma的起始地址指向映射的起始地址
  79. vma->vm_end = addr + len;//同上
  80. vma->vm_flags = vm_flags(prot,flags) | mm->def_flags;//设置vma属性
  81. 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
  82. VM_ClearReadHint(vma);//以下代码设置一堆属性
  83. vma->vm_raend = 0;
  84. if (file->f_mode & FMODE_READ)
  85. vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
  86. if (flags & MAP_SHARED) {
  87. vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
  88. /* This looks strange, but when we don't have the file open
  89. * for writing, we can demote the shared mapping to a simpler
  90. * private mapping. That also takes care of a security hole
  91. * with ptrace() writing to a shared mapping without write
  92. * permissions.
  93. *
  94. * We leave the VM_MAYSHARE bit on, just to get correct output
  95. * from /proc/xxx/maps..
  96. */
  97. if (!(file->f_mode & FMODE_WRITE))
  98. vma->vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
  99. }
  100. } else {
  101. vma->vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
  102. if (flags & MAP_SHARED)
  103. vma->vm_flags |= VM_SHARED | VM_MAYSHARE;
  104. }
  105. vma->vm_page_prot = protection_map[vma->vm_flags & 0x0f];
  106. vma->vm_ops = NULL;
  107. 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
  108. vma->vm_file = NULL;
  109. vma->vm_private_data = NULL;
  110. /* Clear old maps */
  111. error = -ENOMEM;
  112. 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)
  113. goto free_vma;//已经使用则释放free_vma
  114. //检查是否超过了限制
  115. /* Check against address space limit. */
  116. if ((mm->total_vm << PAGE_SHIFT) + len
  117. > current->rlim[RLIMIT_AS].rlim_cur)
  118. goto free_vma;
  119. //检查当前进程专用的可写区间而物理页面不足的情况
  120. /* Private writable mapping? Check memory availability.. */
  121. if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
  122. !(flags & MAP_NORESERVE) &&
  123. !vm_enough_memory(len >> PAGE_SHIFT))
  124. goto free_vma;
  125. 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
  126. if (vma->vm_flags & VM_DENYWRITE) {
  127. error = deny_write_access(file);
  128. if (error)
  129. goto free_vma;
  130. correct_wcount = 1;
  131. }
  132. vma->vm_file = file;
  133. get_file(file);//递增file结构的共享计数
  134. error = file->f_op->mmap(file, vma);//一个文件操作必须存在mmap,否则释放vma
  135. if (error)
  136. goto unmap_and_free_vma;
  137. } else if (flags & MAP_SHARED) {//共享映射
  138. error = shmem_zero_setup(vma);
  139. if (error)
  140. goto free_vma;
  141. }
  142. /* Can addr have changed??
  143. *为了防止flags与addr有变化,再重新设置一遍,
  144. * Answer: Yes, several device drivers can do it in their
  145. * f_op->mmap method. -DaveM
  146. */
  147. flags = vma->vm_flags;
  148. addr = vma->vm_start;
  149. insert_vm_struct(mm, vma);//插入当前进程的内存描述符
  150. if (correct_wcount)
  151. atomic_inc(&file->f_dentry->d_inode->i_writecount);
  152. mm->total_vm += len >> PAGE_SHIFT;//映射区域+len>>page_shit
  153. if (flags & VM_LOCKED) {//需要加锁
  154. mm->locked_vm += len >> PAGE_SHIFT;
  155. make_pages_present(addr, addr + len);//建立初始映射
  156. }
  157. return addr;
  158. unmap_and_free_vma:
  159. if (correct_wcount)
  160. atomic_inc(&file->f_dentry->d_inode->i_writecount);
  161. vma->vm_file = NULL;
  162. fput(file);
  163. /* Undo any partial mapping done by a device driver. */
  164. flush_cache_range(mm, vma->vm_start, vma->vm_end);
  165. zap_page_range(mm, vma->vm_start, vma->vm_end - vma->vm_start);
  166. flush_tlb_range(mm, vma->vm_start, vma->vm_end);
  167. free_vma:
  168. kmem_cache_free(vm_area_cachep, vma);
  169. return error;
  170. }
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(1991)  评论(0编辑  收藏  举报

导航