如果看懂了raid1陣列的run函數，那麼看raid5陣列run就非常輕松了，因為兩者要做的事情都是大同小異。
raid5的run函數很長，但很大一部分跟創建運行是沒有關系的，特別是有一段跟reshape相關的，大多數系統都不關注該功能，因此可以直接跳過。經過刪減之後的run函數如下：
5307 static int run(struct mddev *mddev)
5308 {
5309 struct r5conf *conf;
5310 int working_disks = 0;
5311 int dirty_parity_disks = 0;
5312 struct md_rdev *rdev;
5313 sector_t reshape_offset = 0;
5314 int i;
5315 long long min_offset_diff = 0;
5316 int first = 1;
...
5426 if (mddev->private == NULL)
5427 conf = setup_conf(mddev);
5428 else
5429 conf = mddev->private;
5430
5431 if (IS_ERR(conf))
5432 return PTR_ERR(conf);
5433
5434 conf->min_offset_diff = min_offset_diff;
5435 mddev->thread = conf->thread;
5436 conf->thread = NULL;
5437 mddev->private = conf;
...
5491 /*
5492 * 0 for a fully functional array, 1 or 2 for a degraded array.
5493 */
5494 mddev->degraded = calc_degraded(conf);
...
5503 /* device size must be a multiple of chunk size */
5504 mddev->dev_sectors &= ~(mddev->chunk_sectors - 1);
5505 mddev->resync_max_sectors = mddev->dev_sectors;
...
5556 md_set_array_sectors(mddev, raid5_size(mddev, 0, 0));
5557
5558 if (mddev->queue) {
...
5628 }
5629
5630 return 0;
是不是感覺超級簡單呢，就像有些事情表面上看起來很復雜，但只要認真地去分析之後發現其實是有規律可循的。就像這個run函數，做的事情與raid1的run是相同的，就是建立讀寫的上下文環境。
5427行，創建struct r5conf，跟進函數：
5131 static struct r5conf *setup_conf(struct mddev *mddev)
5132 {
5133 struct r5conf *conf;
5134 int raid_disk, memory, max_disks;
5135 struct md_rdev *rdev;
5136 struct disk_info *disk;
5137 char pers_name[6];
5138
5139 if (mddev->new_level != 5
5140 && mddev->new_level != 4
5141 && mddev->new_level != 6) {
5142 printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n",
5143 mdname(mddev), mddev->new_level);
5144 return ERR_PTR(-EIO);
5145 }
5146 if ((mddev->new_level == 5
5147 && !algorithm_valid_raid5(mddev->new_layout)) ||
5148 (mddev->new_level == 6
5149 && !algorithm_valid_raid6(mddev->new_layout))) {
5150 printk(KERN_ERR "md/raid:%s: layout %d not supported\n",
5151 mdname(mddev), mddev->new_layout);
5152 return ERR_PTR(-EIO);
5153 }
5154 if (mddev->new_level == 6 && mddev->raid_disks < 4) {
5155 printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n",
5156 mdname(mddev), mddev->raid_disks);
5157 return ERR_PTR(-EINVAL);
5158 }
5159
5160 if (!mddev->new_chunk_sectors ||
5161 (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
5162 !is_power_of_2(mddev->new_chunk_sectors)) {
5163 printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n",
5164 mdname(mddev), mddev->new_chunk_sectors << 9);
5165 return ERR_PTR(-EINVAL);
5166 }
5167
5168 conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL);
5169 if (conf == NULL)
5170 goto abort;
5171 spin_lock_init(&conf->device_lock);
5172 init_waitqueue_head(&conf->wait_for_stripe);
5173 init_waitqueue_head(&conf->wait_for_overlap);
5174 INIT_LIST_HEAD(&conf->handle_list);
5175 INIT_LIST_HEAD(&conf->hold_list);
5176 INIT_LIST_HEAD(&conf->delayed_list);
5177 INIT_LIST_HEAD(&conf->bitmap_list);
5178 INIT_LIST_HEAD(&conf->inactive_list);
5179 atomic_set(&conf->active_stripes, 0);
5180 atomic_set(&conf->preread_active_stripes, 0);
5181 atomic_set(&conf->active_aligned_reads, 0);
5182 conf->bypass_threshold = BYPASS_THRESHOLD;
5183 conf->recovery_disabled = mddev->recovery_disabled - 1;
5184
5185 conf->raid_disks = mddev->raid_disks;
5186 if (mddev->reshape_position == MaxSector)
5187 conf->previous_raid_disks = mddev->raid_disks;
5188 else
5189 conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks;
5190 max_disks = max(conf->raid_disks, conf->previous_raid_disks);
5191 conf->scribble_len = scribble_len(max_disks);
5192
5193 conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
5194 GFP_KERNEL);
5195 if (!conf->disks)
5196 goto abort;
5197
5198 conf->mddev = mddev;
5199
5200 if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
5201 goto abort;
5202
5203 conf->level = mddev->new_level;
5204 if (raid5_alloc_percpu(conf) != 0)
5205 goto abort;
5206
5207 pr_debug("raid456: run(%s) called.\n", mdname(mddev));
5208
5209 rdev_for_each(rdev, mddev) {
5210 raid_disk = rdev->raid_disk;
5211 if (raid_disk >= max_disks
5212 || raid_disk < 0)
5213 continue;
5214 disk = conf->disks + raid_disk;
5215
5216 if (test_bit(Replacement, &rdev->flags)) {
5217 if (disk->replacement)
5218 goto abort;
5219 disk->replacement = rdev;
5220 } else {
5221 if (disk->rdev)
5222 goto abort;
5223 disk->rdev = rdev;
5224 }
5225
5226 if (test_bit(In_sync, &rdev->flags)) {
5227 char b[BDEVNAME_SIZE];
5228 printk(KERN_INFO "md/raid:%s: device %s operational as raid"
5229 " disk %d\n",
5230 mdname(mddev), bdevname(rdev->bdev, b), raid_disk);
5231 } else if (rdev->saved_raid_disk != raid_disk)
5232 /* Cannot rely on bitmap to complete recovery */
5233 conf->fullsync = 1;
5234 }
5235
5236 conf->chunk_sectors = mddev->new_chunk_sectors;
5237 conf->level = mddev->new_level;
5238 if (conf->level == 6)
5239 conf->max_degraded = 2;
5240 else
5241 conf->max_degraded = 1;
5242 conf->algorithm = mddev->new_layout;
5243 conf->max_nr_stripes = NR_STRIPES;
5244 conf->reshape_progress = mddev->reshape_position;
5245 if (conf->reshape_progress != MaxSector) {
5246 conf->prev_chunk_sectors = mddev->chunk_sectors;
5247 conf->prev_algo = mddev->layout;
5248 }
5249
5250 memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
5251 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
5252 if (grow_stripes(conf, conf->max_nr_stripes)) {
5253 printk(KERN_ERR
5254 "md/raid:%s: couldn't allocate %dkB for buffers\n",
5255 mdname(mddev), memory);
5256 goto abort;
5257 } else
5258 printk(KERN_INFO "md/raid:%s: allocated %dkB\n",
5259 mdname(mddev), memory);
5260
5261 sprintf(pers_name, "raid%d", mddev->new_level);
5262 conf->thread = md_register_thread(raid5d, mddev, pers_name);
5263 if (!conf->thread) {
5264 printk(KERN_ERR
5265 "md/raid:%s: couldn't allocate thread.\n",
5266 mdname(mddev));
5267 goto abort;
5268 }
5269
5270 return conf;
同樣，這個函數與raid1的setup_conf也很相似。
5139行，檢查陣列級別，支持raid4,5,6。
5147行，檢查raid5的layout是否正確。
5160行，檢查陣列chunk大小，必須為page整數倍並且是2的n次方。
5168行，申請struct r5conf內存空間並初始化。
5185行，設置數據盤數。
5193行，申請struct disk_info，用於保存與磁盤的關聯。
5200行，用於保存struct stripe_head的哈希表，用於快速查找指定扇區的stripe_head。
5209-5234行，最關鍵的是5223行，關聯struct disk_info與struct md_rdev。
5236行，設置條塊大小。
5237行，設置級別。
5241行，設置最大降級磁盤數。
5252行，申請struct stripe_head slab。跟進函數grow_stripes：
1501 static int grow_stripes(struct r5conf *conf, int num)
1502 {
1503 struct kmem_cache *sc;
1504 int devs = max(conf->raid_disks, conf->previous_raid_disks);
1505
1506 if (conf->mddev->gendisk)
1507 sprintf(conf->cache_name[0],
1508 "raid%d-%s", conf->level, mdname(conf->mddev));
1509 else
1510 sprintf(conf->cache_name[0],
1511 "raid%d-%p", conf->level, conf->mddev);
1512 sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]);
1513
1514 conf->active_name = 0;
1515 sc = kmem_cache_create(conf->cache_name[conf->active_name],
1516 sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev),
1517 0, 0, NULL);
1518 if (!sc)
1519 return 1;
1520 conf->slab_cache = sc;
1521 conf->pool_size = devs;
1522 while (num--)
1523 if (!grow_one_stripe(conf))
1524 return 1;
1525 return 0;
1526 }
1504行，計算數據盤數目。
1506行，設置slab名稱。
1515行，創建slab， 空間大小為sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev)是因為struct stripe_head尾部有devs個struct r5dev。
1523行，創建空閒struct stripe_head。然而只是簡單地創建就沒有必要跟進看了，但該函數中隱藏著一個最經常調用的函數release_stripe，所以還是有必要跟進的：
1477 static int grow_one_stripe(struct r5conf *conf)
1478 {
1479 struct stripe_head *sh;
1480 sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
1481 if (!sh)
1482 return 0;
1483
1484 sh->raid_conf = conf;
1485
1486 spin_lock_init(&sh->stripe_lock);
1487
1488 if (grow_buffers(sh)) {
1489 shrink_buffers(sh);
1490 kmem_cache_free(conf->slab_cache, sh);
1491 return 0;
1492 }
1493 /* we just created an active stripe so... */
1494 atomic_set(&sh->count, 1);
1495 atomic_inc(&conf->active_stripes);
1496 INIT_LIST_HEAD(&sh->lru);
1497 release_stripe(sh);
1498 return 1;
1499 }
1480行，新申請一個struct stripe_head。
1484行，關聯struct r5conf。
1488行，grow_buffers，為每個struct r5dev申請一個page頁用於stripe_head頁拷貝和計算校驗。頁指針保存在sh->dev[].page指針中。
1494行，設置struct stripe_head計數器，在1497行release_stripe中會遞減。
1495行，遞增陣列活躍條帶數。
1496行，lru鏈表初始化。
1497行，釋放struct stripe_head，添加到空閒條帶空閒鏈表。release_stripe最終會調用到do_release_stripe，do_release_stripe裡會執行到下面幾行：
228 list_add_tail(&sh->lru, &conf->inactive_list); 229 wake_up(&conf->wait_for_stripe); 230 if (conf->retry_read_aligned) 231 md_wakeup_thread(conf->mddev->thread);
228行，添加struct stripe_head到inactive_list，即條帶空閒鏈表。229行，喚醒等待空閒條帶的請求，因為每個陣列的struct stripe_head資源是有限的，申請不到時就在等待隊列上等候。231行，喚醒條塊讀請求。繼續返回到setup_conf函數中，這裡已經通過grow_stripes為陣列申請了NR_STRIPES個struct stripe_head。5262行，創建raid5主線程。這樣setup_conf函數也結束了，繼續返回到run函數中。5434-5437行，conf和mddev的關聯和賦值。5494-5556行，mddev相關域的賦值。5558行，mddev請求隊列struct queue_limits設置等等初始化。小結一下，raid5的run函數同raid1基本作用是一樣的，都是向上虛擬一個塊設備，向下包裝磁盤，建立讀寫請求的通道。區別在於raid5的讀寫是以struct stripe_head為基礎的，而在raid5的讀寫中也是圍繞著struct stripe_head展開的。下一小節介紹raid10陣列的運行。
出處：http://blog.csdn.net/liumangxiong
