ocfs2_readpage()
ftrace:
3) | ocfs2_readpage() {
3) | ocfs2_inode_lock_with_page() {
3) 0.126 us | down_read();
3) 0.098 us | up_read();
3) + 11.225 us | ocfs2_inode_unlock();
3) ! 11945.43 us | }
3) ! 11946.01 us | }
代码分析：
274 static int ocfs2_readpage(struct file *file, struct page *page)
275 {
276 struct inode *inode = page->mapping->host;
277 struct ocfs2_inode_info *oi = OCFS2_I(inode);
278 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
279 int ret, unlock = 1;
280
281 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
282 (page ? page->index : 0));
283
/*
ocfs2_inode_lock_with_page()是个痛点，为了避免lock inversion又引入了live lock问题，这些
具体问题的细节，暂时还没理解！
调用该函数时，page已经locked, 试探性申请dlm锁，如果dlm资源正在被不可兼容的占用，那么主动放弃申请，
unlock page, 然后再去申请dlm锁;为了当好人，让ocfs2dc获得page lock，如果dlm锁被写进程抢走，那么
在写锁降级的时候，要等待刷盘，很浪费时间的；肯定还暗藏其它问题，要不然怎么会浪费那么多时间！
284 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
285 if (ret != 0) {
286 if (ret == AOP_TRUNCATED_PAGE)
287 unlock = 0;
288 mlog_errno(ret);
289 goto out;
290 }
291
292 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
293 /*
294 * Unlock the page and cycle ip_alloc_sem so that we don't
295 * busyloop waiting for ip_alloc_sem to unlock
296 */
297 ret = AOP_TRUNCATED_PAGE;
298 unlock_page(page);
299 unlock = 0;
300 down_read(&oi->ip_alloc_sem);
301 up_read(&oi->ip_alloc_sem);
302 goto out_inode_unlock;
303 }
割...
305 /*
306 * i_size might have just been updated as we grabed the meta lock. We
307 * might now be discovering a truncate that hit on another node.
308 * block_read_full_page->get_block freaks out if it is asked to read
309 * beyond the end of a file, so we check here. Callers
310 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
311 * and notice that the page they just read isn't needed.
312 *
313 * XXX sys_readahead() seems to get that wrong?
314 */
315 if (start >= i_size_read(inode)) {
316 zero_user(page, 0, PAGE_SIZE);
317 SetPageUptodate(page);
318 ret = 0;
319 goto out_alloc;
320 }
321
322 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
323 ret = ocfs2_readpage_inline(inode, page);
324 else
325 ret = block_read_full_page(page, ocfs2_get_block);
/*
我觉得unlock不应该设置为0， 因为block_read_full_page函数有可能没有unlock page,
所以我改了这一块，只要page锁着，就让unlock等于1，虽然性能上去了，但是读会发生io错误；
可能不像我想得那么简单，因为block_read_full_page中可能会调用submit_bh进行IO， 是异步的，后面可能
需要等待IO完成，才会去unlock page, 难怪读会发生io错误
*/
326 unlock = 0;
327
328 out_alloc:
329 up_read(&OCFS2_I(inode)->ip_alloc_sem);
330 out_inode_unlock:
331 ocfs2_inode_unlock(inode, 0);
332 out:
333 if (unlock)
334 unlock_page(page);
335 return ret;
336 }