真实的I/O调度层处理

1.5.6 真实的I/O调度层处理

现在我们块设备也有了，队列也有了，要提交请求也就可以开始提交了。那就让我们回到generic_make_request来研究一下如何提交请求如何处理请求吧。我们看到，函数最后调用q->make_request_fn(q, bio)。对 make_request_fn 函数的调用可以认为是 IO调度层的入口，该函数用于向请求队列中添加请求。该函数是在创建请求队列时指定的，代码如下（blk_init_queue 函数中）：

q->request_fn = rfn;

blk_queue_make_request(q, __make_request);

 

前面看到函数 blk_queue_make_request 将函数 __make_request 的地址赋予了请求队列 q 的 make_request_fn 成员：

 

void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) {        q->nr_requests = BLKDEV_MAX_RQ;        q->make_request_fn = mfn; ……

 

那么，__make_request函数才是IO调度层的真实入口，来自block/ll_rw_blk.c：

 

2846static int __make_request(request_queue_t *q, struct bio *bio) 2847{ 2848        struct request *req; 2849        int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; 2850        unsigned short prio; 2851        sector_t sector; 2852 2853        sector = bio->bi_sector; 2854        nr_sectors = bio_sectors(bio); 2855        cur_nr_sectors = bio_cur_sectors(bio); 2856        prio = bio_prio(bio); 2857 2858        rw = bio_data_dir(bio); 2859        sync = bio_sync(bio); 2860 2861        /* 2862         * low level driver can indicate that it wants pages above a 2863         * certain limit bounced to low memory (ie for highmem, or even 2864         * ISA dma in theory) 2865         */ 2866        blk_queue_bounce(q, &bio); 2867 2868        spin_lock_prefetch(q->queue_lock); 2869 2870        barrier = bio_barrier(bio); 2871        if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { 2872                err = -EOPNOTSUPP; 2873                goto end_io; 2874        } 2875 2876        spin_lock_irq(q->queue_lock); 2877 2878        if (unlikely(barrier) || elv_queue_empty(q)) 2879                goto get_rq; 2880 2881        el_ret = elv_merge(q, &req, bio); 2882        switch (el_ret) { 2883                case ELEVATOR_BACK_MERGE: 2884                        BUG_ON(!rq_mergeable(req)); 2885 2886                        if (!q->back_merge_fn(q, req, bio)) 2887                                break; 2888 2889                        blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); 2890 2891                        req->biotail->bi_next = bio; 2892                        req->biotail = bio; 2893                        req->nr_sectors = req->hard_nr_sectors += nr_sectors; 2894                        req->ioprio = ioprio_best(req->ioprio, prio); 2895                        drive_stat_acct(req, nr_sectors, 0); 2896                        if (!attempt_back_merge(q, req)) 2897                                elv_merged_request(q, req); 2898                        goto out; 2899 2900                case ELEVATOR_FRONT_MERGE: 2901                        BUG_ON(!rq_mergeable(req)); 2902 2903                        if (!q->front_merge_fn(q, req, bio)) 2904                                break; 2905 2906                        blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); 2907 2908                        bio->bi_next = req->bio; 2909                        req->bio = bio; 2910 2911                        /* 2912                         * may not be valid. if the low level driver said 2913                         * it didn't need a bounce buffer then it better 2914                         * not touch req->buffer either... 2915                         */ 2916                        req->buffer = bio_data(bio); 2917                        req->current_nr_sectors = cur_nr_sectors; 2918                        req->hard_cur_sectors = cur_nr_sectors; 2919                        req->sector = req->hard_sector = sector; 2920                        req->nr_sectors = req->hard_nr_sectors += nr_sectors; 2921                        req->ioprio = ioprio_best(req->ioprio, prio); 2922                        drive_stat_acct(req, nr_sectors, 0); 2923                        if (!attempt_front_merge(q, req)) 2924                                elv_merged_request(q, req); 2925                        goto out; 2926</