提交I/O传输请求

1.4.3 提交I/O传输请求

好了，bio这个数据我们建立好了，随后调用generic_make_request 函数。这个函数是通用块层的入口点，该层只有这一个函数处理请求：

 

3020void generic_make_request(struct bio *bio) 3021{ 3022        request_queue_t *q; 3023        sector_t maxsector; 3024        int ret, nr_sectors = bio_sectors(bio); 3025        dev_t old_dev; 3026 3027        might_sleep(); 3028        /* Test device or partition size, when known. */ 3029        maxsector = bio->bi_bdev->bd_inode->i_size >> 9; 3030        if (maxsector) { 3031                sector_t sector = bio->bi_sector; 3032 3033                if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 3034                        /* 3035                         * This may well happen - the kernel calls bread() 3036                         * without checking the size of the device, e.g., when 3037                         * mounting a device. 3038                         */ 3039                        handle_bad_sector(bio); 3040                        goto end_io; 3041                } 3042        } 3043 3044        /* 3045         * Resolve the mapping until finished. (drivers are 3046         * still free to implement/resolve their own stacking 3047         * by explicitly returning 0) 3048         * 3049         * NOTE: we don't repeat the blk_size check for each new device. 3050         * Stacking drivers are expected to know what they are doing. 3051         */ 3052        maxsector = -1; 3053        old_dev = 0; 3054        do { 3055                char b[BDEVNAME_SIZE]; 3056 3057                q = bdev_get_queue(bio->bi_bdev); 3058                if (!q) { 3059                        printk(KERN_ERR 3060                               "generic_make_request: Trying to access " 3061                                "nonexistent block-device %s (%Lu)/n", 3062                                bdevname(bio->bi_bdev, b), 3063                                (long long) bio->bi_sector); 3064end_io: 3065                        bio_endio(bio, bio->bi_size, -EIO); 3066                        break; 3067                } 3068 3069                if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { 3070                        printk("bio too big device %s (%u > %u)/n", 3071                                bdevname(bio->bi_bdev, b), 3072                                bio_sectors(bio), 3073                                q->max_hw_sectors); 3074                        goto end_io; 3075                } 3076 3077                if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) 3078                        goto end_io; 3079 3080                /* 3081                 * If this device has partitions, remap block n 3082                 * of partition p to block n+start(p) of the disk. 3083                 */ 3084                blk_partition_remap(bio); 3085 3086                if (maxsector != -1) 3087                        blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, 3088                                            maxsector); 3089 3090                blk_add_trace_bio(q, bio, BLK_TA_QUEUE); 3091 3092                maxsector = bio->bi_sector; 3093                old_dev = bio->bi_bdev->bd_dev; 3094 3095                ret = q->make_request_fn(q, bio); 3096        } while (ret); 3097}

 

首先3033行，检查bio->bi_sector没有超过块设备的扇区数。如果超过，则调用handle_bad_sector(bio)函数将bio->bi_flags设置为BIO_EOF标志，然后打印一条内核错误信息。随后跳到end_io标号处执行3065行的bio_endio()函数，并终止。bio_endio()更新bio描述符中的bi_size和bi_sector值，然后调用bio的bi_end_io方法。

 

如果没有超过，那么进入一个循环，通过3057行的bdev_get_queue函数获取与块设备相关的请求队列q；其中地址存放在bio的块设备描述符bi_bdev的bd_disk字段所指向的gendisk结构中，而该块设备描述符则由bio->bi_bdev指向：

 

static inline request_queue_t *bdev_get_queue(struct block_device *bdev) {        return bdev->bd_disk->queue; }

 

其中地址存放在块设备描述符的bd_disk字段所指向的gendisk结构中，而该块设备描述符则由bio->bi_bdev指向。

 

接下来3084行调用blk_partition_remap()函数检查块设备是否指的是一个磁盘分区（bio->bi_bdev不等于bio->bi_dev->bd_contains）。如果是，则从bio->bi_bdev获取分区的hd_struct描述符，从而执行下面的子操作：

 

a) 根据数据传送的方向，更新hd_struct描述符中的read_sectors和reads值或write_sectors和writes值。

b) 调整bio->bi_sector值使得把相对于分区的起始扇区号转变为相对于整有盘的扇区号。

c) 将bio->bi_bdev设置为整个磁盘的块设备描述符（bio->bd_contains）。

 

从现在开始，通用块层、I/O调度程序以及设备驱动程序将忘记磁盘分区的存在，直接作用于整个磁盘。

 

最终，3095行调用q->make_request_fn方法进入块设备的I/O调度层，将bio请求插入请求队列q中。