Open系统调用

/* 2 */：这一步是需要找到一个没有使用的文件描述符fd

看一下sys_open函数中调用的这个函数get_unused_fd：

738 /*
739  * Find an empty file descriptor entry, and mark it busy.
740  */
741 int get_unused_fd(void)
742 {
743         struct files_struct * files = current->files;  /* (1)获得当前进程的文件打开表，这个前面已经说过了 */
744         int fd, error;
745 
746         error = -EMFILE;
747         write_lock(&files->file_lock);
748 
749 repeat:
750         fd = find_next_zero_bit(files->open_fds,      /* (2)在这个进程文件打开表中寻找还没有使用的fd */
751                                 files->max_fdset, 
752                                 files->next_fd);
753 
754         /*
755          * N.B. For clone tasks sharing a files structure, this test
756          * will limit the total number of files that can be opened.
757          */
758         if (fd >= current->rlim[RLIMIT_NOFILE].rlim_cur)  /* 不能超过限制(考虑到fork进程情况，共享fd情况) */
759                 goto out;
760 
761         /* Do we need to expand the fdset array? */
762         if (fd >= files->max_fdset) {    /* 如果当前进程的fd集合中最大fd比当前申请的fd小，那么需要扩大fdset，需要扩容：具体见http://blog.csdn.net/shanshanpt/article/details/38943731中files_struct结构体描述 */
763                 error = expand_fdset(files, fd);  /* 扩容函数1 */
764                 if (!error) {
765                         error = -EMFILE;
766                         goto repeat;
767                 }
768                 goto out;
769         }
770         
771         /* 
772          * Check whether we need to expand the fd array.
773          */
774         if (fd >= files->max_fds) {    /* 类似于上面，这里是需要扩大fd-array数组，具体见上面链接 */
775                 error = expand_fd_array(files, fd);  /* 扩容函数2 */
776                 if (!error) {
777                         error = -EMFILE;
778                         goto repeat;
779                 }
780                 goto out;
781         }
782 
783         FD_SET(fd, files->open_fds);     /* 将fd加入到打开文件描述符中 */
784         FD_CLR(fd, files->close_on_exec);/* 从close-on-exec中清除 */
785         files->next_fd = fd + 1;         /* 当前描述符是最大的fd，所有next就是fd+1 */
786 #if 1
787         /* Sanity check */
788         if (files->fd[fd] != NULL) {
789                 printk(KERN_WARNING "get_unused_fd: slot %d not NULL!\n", fd);
790                 files->fd[fd] = NULL;
791         }
792 #endif
793         error = fd;
794 
795 out:
796         write_unlock(&files->file_lock);
797         return error;
798 }
799 

ATTENTION：注意上面的max_fdset和max_fds的区别，前者是当前可以容纳的最大的文件描述符的数量大小！后者是当前可以容纳的文件对象的数量大小！对于一个文件对象而言，可以存在多个文件描述符指向这一个文件对象！所以对于申请到的fd，要分别和这两个进行判断！

主要看上面的三个函数，第一个是寻找fd函数find_next_zero_bit：
这个函数的意义就是找到open_fds打开的文件描述符中第一个bit位=0的那一位，作为新的fd返回。(前面也说过，fd的管理是使用位管理的)，反正是一堆位运算，慢慢看吧：

254 static inline unsigned long find_next_zero_bit(void *addr, unsigned long size, unsigned long offset)
255 {
256         unsigned long *p = ((unsigned long *) addr) + (offset >> 5);
257         unsigned long result = offset & ~31UL;
258         unsigned long tmp;
259 
260         if (offset >= size)
261                 return size;
262         size -= result;
263         offset &= 31UL;
264         if (offset) {
265                 tmp = *(p++);
266                 tmp |= ~0UL >> (32-offset);
267                 if (size < 32)
268                         goto found_first;
269                 if (~tmp)
270                         goto found_middle;
271                 size -= 32;
272                 result += 32;
273         }
274         while (size & ~31UL) {
275                 if (~(tmp = *(p++)))
276                         goto found_middle;
277                 result += 32;
278                 size -= 32;
279         }
280         if (!size)
281                 return result;
282         tmp = *p;
283 
284 found_first:
285         tmp |= ~0UL << size;
286         if (tmp == ~0UL)        /* Are any bits zero? */
287                 return result + size; /* Nope. */
288 found_middle:
289         return result + ffz(tmp);
290 }

扩大fdset集合函数：expand_fdset

162 /*
163  * Expand the fdset in the files_struct.  Called with the files spinlock
164  * held for write.
165  */
166 int expand_fdset(struct files_struct *files, int nr)
167 {
168         fd_set *new_openset = 0, *new_execset = 0;
169         int error, nfds = 0;
170 
171         error = -EMFILE;
172         if (files->max_fdset >= NR_OPEN || nr >= NR_OPEN)   /* 不能大于系统最大限制 */
173                 goto out;
174 
175         nfds = files->max_fdset;   /* 当前最大文件描述符 */
176         write_unlock(&files->file_lock);
177 
178         /* Expand to the max in easy steps，下面是一个 简单的扩展过程 */
179         do {
180                 if (nfds < (PAGE_SIZE * 8))
181                         nfds = PAGE_SIZE * 8;
182                 else {
183                         nfds = nfds * 2;
184                         if (nfds > NR_OPEN)
185                                 nfds = NR_OPEN;
186                 }
187         } while (nfds <= nr);
188 
189         error = -ENOMEM;
190         new_openset = alloc_fdset(nfds);  /* 根据新的大小分配新的打开set集合 */
191         new_execset = alloc_fdset(nfds);  /* 根据新的大小分配新的执行时候需要close的set集合 */
192         write_lock(&files->file_lock);
193         if (!new_openset || !new_execset)
194                 goto out;
195 
196         error = 0;
197         
198         /* Copy the existing tables and install the new pointers：将老的数据拷贝到新的内存中来 */
199         if (nfds > files->max_fdset) {
200                 int i = files->max_fdset / (sizeof(unsigned long) * 8);
201                 int count = (nfds - files->max_fdset) / 8;
202                 
203                 /* 
204                  * Don't copy the entire array if the current fdset is
205                  * not yet initialised.  
206                  */
207                 if (i) {
208                         memcpy (new_openset, files->open_fds, files->max_fdset/8);
209                         memcpy (new_execset, files->close_on_exec, files->max_fdset/8);
210                         memset (&new_openset->fds_bits[i], 0, count);
211                         memset (&new_execset->fds_bits[i], 0, count);
212                 }
213                 /* 下面几步骤很重要，将新分配的挂载到files结构体中去 */
214                 nfds = xchg(&files->max_fdset, nfds);
215                 new_openset = xchg(&files->open_fds, new_openset);
216                 new_execset = xchg(&files->close_on_exec, new_execset);
217                 write_unlock(&files->file_lock);
218                 free_fdset (new_openset, nfds);
219                 free_fdset (new_execset, nfds);
220                 write_lock(&files->file_lock);
221                 return 0;
222         } 
223         /* Somebody expanded the array while we slept ... */
224 
225 out:
226         write_unlock(&files->file_lock);
227         if (new_openset)
228                 free_fdset(new_openset, nfds);
229         if (new_execset)
230                 free_fdset(new_execset, nfds);
231         write_lock(&files->file_lock);
232         return error;
233 }

再看一下具体的alloc_fdset函数：

128 /*
129  * Allocate an fdset array, using kmalloc or vmalloc.
130  * Note: the array isn't cleared at allocation time.
131  */
132 fd_set * alloc_fdset(int num)
133 {
134         fd_set *new_fdset;
135         int size = num / 8;
136 
137         if (size <= PAGE_SIZE)
138                 new_fdset = (fd_set *) kmalloc(size, GFP_KERNEL);
139         else
140                 new_fdset = (fd_set *) vmalloc(size);
141         return new_fdset;
142 }

回到上面，看一下扩大fd数组的函数expand_fd_array：

 52 /*
 53  * Expand the fd array in the files_struct.  Called with the files
 54  * spinlock held for write.
 55  */
 56 
 57 int expand_fd_array(struct files_struct *files, int nr)
 58 {
 59         struct file **new_fds;
 60         int error, nfds;
 61 
 62         
 63         error = -EMFILE;
 64         if (files->max_fds >= NR_OPEN || nr >= NR_OPEN)   /* 不能超过最大系统限制 */
 65                 goto out;
 66 
 67         nfds = files->max_fds;     /* 当前进程中最大的fd */
 68         write_unlock(&files->file_lock);
 69 
 70         /* 
 71          * Expand to the max in easy steps, and keep expanding it until
 72          * we have enough for the requested fd array size. 
 73          */
 74         /* 简单的扩展策略 */
 75         do {
 76 #if NR_OPEN_DEFAULT < 256
 77                 if (nfds < 256)
 78                         nfds = 256;
 79                 else 
 80 #endif
 81                 if (nfds < (PAGE_SIZE / sizeof(struct file *)))
 82                         nfds = PAGE_SIZE / sizeof(struct file *);
 83                 else {
 84                         nfds = nfds * 2;
 85                         if (nfds > NR_OPEN)
 86                                 nfds = NR_OPEN;
 87                 }
 88         } while (nfds <= nr);
 89 
 90         error = -ENOMEM;
 91         new_fds = alloc_fd_array(nfds);    /* 分配新的fd_array数组 */
 92         write_lock(&files->file_lock);
 93         if (!new_fds)
 94                 goto out;
 95 
 96         /* Copy the existing array and install the new pointer */
 97 
 98         if (nfds > files->max_fds) {
 99                 struct file **old_fds;
100                 int i;
101                 /* 将当前进行的文件数组指针指向新申请的fd数组！ */
102                 old_fds = xchg(&files->fd, new_fds);
103                 i = xchg(&files->max_fds, nfds);
104 
105                 /* Don't copy/clear the array if we are creating a new
106                    fd array for fork() */
107                 if (i) {   /* 下面将老的数据拷贝过去 */
108                         memcpy(new_fds, old_fds, i * sizeof(struct file *));
109                         /* clear the remainder of the array */
110                         memset(&new_fds[i], 0,
111                                (nfds-i) * sizeof(struct file *)); 
112 
113                         write_unlock(&files->file_lock);
114                         free_fd_array(old_fds, i);
115                         write_lock(&files->file_lock);
116                 }
117         } else {
118                 /* Somebody expanded the array while we slept ... */
119                 write_unlock(&files->file_lock);
120                 free_fd_array(new_fds, nfds);
121                 write_lock(&files->file_lock);
122         }
123         error = 0;
124 out:
125         return error;
126 }