raid10.c

来自「Linux Kernel 2.6.9 for OMAP1710」· C语言 代码 · 共 1,781 行 · 第 1/4 页

	mempool_destroy(conf->r10buf_pool);
	conf->r10buf_pool = NULL;
}

static int raid10_spare_active(mddev_t *mddev)
{
	int i;
	conf_t *conf = mddev->private;
	mirror_info_t *tmp;

	spin_lock_irq(&conf->device_lock);
	/*
	 * Find all non-in_sync disks within the RAID10 configuration
	 * and mark them in_sync
	 */
	for (i = 0; i < conf->raid_disks; i++) {
		tmp = conf->mirrors + i;
		if (tmp->rdev
		    && !tmp->rdev->faulty
		    && !tmp->rdev->in_sync) {
			conf->working_disks++;
			mddev->degraded--;
			tmp->rdev->in_sync = 1;
		}
	}
	spin_unlock_irq(&conf->device_lock);

	print_conf(conf);
	return 0;
}


static int raid10_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
{
	conf_t *conf = mddev->private;
	int found = 0;
	int mirror;
	mirror_info_t *p;

	if (mddev->recovery_cp < MaxSector)
		/* only hot-add to in-sync arrays, as recovery is
		 * very different from resync
		 */
		return 0;
	spin_lock_irq(&conf->device_lock);
	for (mirror=0; mirror < mddev->raid_disks; mirror++)
		if ( !(p=conf->mirrors+mirror)->rdev) {
			p->rdev = rdev;

			blk_queue_stack_limits(mddev->queue,
					       rdev->bdev->bd_disk->queue);
			/* as we don't honour merge_bvec_fn, we must never risk
			 * violating it, so limit ->max_sector to one PAGE, as
			 * a one page request is never in violation.
			 */
			if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
			    mddev->queue->max_sectors > (PAGE_SIZE>>9))
				mddev->queue->max_sectors = (PAGE_SIZE>>9);

			p->head_position = 0;
			rdev->raid_disk = mirror;
			found = 1;
			break;
		}
	spin_unlock_irq(&conf->device_lock);

	print_conf(conf);
	return found;
}

static int raid10_remove_disk(mddev_t *mddev, int number)
{
	conf_t *conf = mddev->private;
	int err = 1;
	mirror_info_t *p = conf->mirrors+ number;

	print_conf(conf);
	spin_lock_irq(&conf->device_lock);
	if (p->rdev) {
		if (p->rdev->in_sync ||
		    atomic_read(&p->rdev->nr_pending)) {
			err = -EBUSY;
			goto abort;
		}
		p->rdev = NULL;
		err = 0;
	}
	if (err)
		MD_BUG();
abort:
	spin_unlock_irq(&conf->device_lock);

	print_conf(conf);
	return err;
}


static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
{
	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private);
	conf_t *conf = mddev_to_conf(r10_bio->mddev);
	int i,d;

	if (bio->bi_size)
		return 1;

	for (i=0; i<conf->copies; i++)
		if (r10_bio->devs[i].bio == bio)
			break;
	if (i == conf->copies)
		BUG();
	update_head_pos(i, r10_bio);
	d = r10_bio->devs[i].devnum;
	if (!uptodate)
		md_error(r10_bio->mddev,
			 conf->mirrors[d].rdev);

	/* for reconstruct, we always reschedule after a read.
	 * for resync, only after all reads
	 */
	if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
	    atomic_dec_and_test(&r10_bio->remaining)) {
		/* we have read all the blocks,
		 * do the comparison in process context in raid10d
		 */
		reschedule_retry(r10_bio);
	}
	rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
	return 0;
}

static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
{
	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private);
	mddev_t *mddev = r10_bio->mddev;
	conf_t *conf = mddev_to_conf(mddev);
	int i,d;

	if (bio->bi_size)
		return 1;

	for (i = 0; i < conf->copies; i++)
		if (r10_bio->devs[i].bio == bio)
			break;
	d = r10_bio->devs[i].devnum;

	if (!uptodate)
		md_error(mddev, conf->mirrors[d].rdev);
	update_head_pos(i, r10_bio);

	while (atomic_dec_and_test(&r10_bio->remaining)) {
		if (r10_bio->master_bio == NULL) {
			/* the primary of several recovery bios */
			md_done_sync(mddev, r10_bio->sectors, 1);
			put_buf(r10_bio);
			break;
		} else {
			r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio;
			put_buf(r10_bio);
			r10_bio = r10_bio2;
		}
	}
	rdev_dec_pending(conf->mirrors[d].rdev, mddev);
	return 0;
}

/*
 * Note: sync and recover and handled very differently for raid10
 * This code is for resync.
 * For resync, we read through virtual addresses and read all blocks.
 * If there is any error, we schedule a write.  The lowest numbered
 * drive is authoritative.
 * However requests come for physical address, so we need to map.
 * For every physical address there are raid_disks/copies virtual addresses,
 * which is always are least one, but is not necessarly an integer.
 * This means that a physical address can span multiple chunks, so we may
 * have to submit multiple io requests for a single sync request.
 */
/*
 * We check if all blocks are in-sync and only write to blocks that
 * aren't in sync
 */
static void sync_request_write(mddev_t *mddev, r10bio_t *r10_bio)
{
	conf_t *conf = mddev_to_conf(mddev);
	int i, first;
	struct bio *tbio, *fbio;

	atomic_set(&r10_bio->remaining, 1);

	/* find the first device with a block */
	for (i=0; i<conf->copies; i++)
		if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags))
			break;

	if (i == conf->copies)
		goto done;

	first = i;
	fbio = r10_bio->devs[i].bio;

	/* now find blocks with errors */
	for (i=first+1 ; i < conf->copies ; i++) {
		int vcnt, j, d;

		if (!test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags))
			continue;
		/* We know that the bi_io_vec layout is the same for
		 * both 'first' and 'i', so we just compare them.
		 * All vec entries are PAGE_SIZE;
		 */
		tbio = r10_bio->devs[i].bio;
		vcnt = r10_bio->sectors >> (PAGE_SHIFT-9);
		for (j = 0; j < vcnt; j++)
			if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
				   page_address(tbio->bi_io_vec[j].bv_page),
				   PAGE_SIZE))
				break;
		if (j == vcnt)
			continue;
		/* Ok, we need to write this bio
		 * First we need to fixup bv_offset, bv_len and
		 * bi_vecs, as the read request might have corrupted these
		 */
		tbio->bi_vcnt = vcnt;
		tbio->bi_size = r10_bio->sectors << 9;
		tbio->bi_idx = 0;
		tbio->bi_phys_segments = 0;
		tbio->bi_hw_segments = 0;
		tbio->bi_hw_front_size = 0;
		tbio->bi_hw_back_size = 0;
		tbio->bi_flags &= ~(BIO_POOL_MASK - 1);
		tbio->bi_flags |= 1 << BIO_UPTODATE;
		tbio->bi_next = NULL;
		tbio->bi_rw = WRITE;
		tbio->bi_private = r10_bio;
		tbio->bi_sector = r10_bio->devs[i].addr;

		for (j=0; j < vcnt ; j++) {
			tbio->bi_io_vec[j].bv_offset = 0;
			tbio->bi_io_vec[j].bv_len = PAGE_SIZE;

			memcpy(page_address(tbio->bi_io_vec[j].bv_page),
			       page_address(fbio->bi_io_vec[j].bv_page),
			       PAGE_SIZE);
		}
		tbio->bi_end_io = end_sync_write;

		d = r10_bio->devs[i].devnum;
		atomic_inc(&conf->mirrors[d].rdev->nr_pending);
		atomic_inc(&r10_bio->remaining);
		md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9);

		generic_make_request(tbio);
	}

done:
	if (atomic_dec_and_test(&r10_bio->remaining)) {
		md_done_sync(mddev, r10_bio->sectors, 1);
		put_buf(r10_bio);
	}
}

/*
 * Now for the recovery code.
 * Recovery happens across physical sectors.
 * We recover all non-is_sync drives by finding the virtual address of
 * each, and then choose a working drive that also has that virt address.
 * There is a separate r10_bio for each non-in_sync drive.
 * Only the first two slots are in use. The first for reading,
 * The second for writing.
 *
 */

static void recovery_request_write(mddev_t *mddev, r10bio_t *r10_bio)
{
	conf_t *conf = mddev_to_conf(mddev);
	int i, d;
	struct bio *bio, *wbio;


	/* move the pages across to the second bio
	 * and submit the write request
	 */
	bio = r10_bio->devs[0].bio;
	wbio = r10_bio->devs[1].bio;
	for (i=0; i < wbio->bi_vcnt; i++) {
		struct page *p = bio->bi_io_vec[i].bv_page;
		bio->bi_io_vec[i].bv_page = wbio->bi_io_vec[i].bv_page;
		wbio->bi_io_vec[i].bv_page = p;
	}
	d = r10_bio->devs[1].devnum;

	atomic_inc(&conf->mirrors[d].rdev->nr_pending);
	md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9);
	generic_make_request(wbio);
}


/*
 * This is a kernel thread which:
 *
 *	1.	Retries failed read operations on working mirrors.
 *	2.	Updates the raid superblock when problems encounter.
 *	3.	Performs writes following reads for array syncronising.
 */

static void raid10d(mddev_t *mddev)
{
	r10bio_t *r10_bio;
	struct bio *bio;
	unsigned long flags;
	conf_t *conf = mddev_to_conf(mddev);
	struct list_head *head = &conf->retry_list;
	int unplug=0;
	mdk_rdev_t *rdev;

	md_check_recovery(mddev);
	md_handle_safemode(mddev);

	for (;;) {
		char b[BDEVNAME_SIZE];
		spin_lock_irqsave(&conf->device_lock, flags);
		if (list_empty(head))
			break;
		r10_bio = list_entry(head->prev, r10bio_t, retry_list);
		list_del(head->prev);
		spin_unlock_irqrestore(&conf->device_lock, flags);

		mddev = r10_bio->mddev;
		conf = mddev_to_conf(mddev);
		if (test_bit(R10BIO_IsSync, &r10_bio->state)) {
			sync_request_write(mddev, r10_bio);
			unplug = 1;
		} else 	if (test_bit(R10BIO_IsRecover, &r10_bio->state)) {
			recovery_request_write(mddev, r10_bio);
			unplug = 1;
		} else {
			int mirror;
			bio = r10_bio->devs[r10_bio->read_slot].bio;
			r10_bio->devs[r10_bio->read_slot].bio = NULL;
			mirror = read_balance(conf, r10_bio);
			r10_bio->devs[r10_bio->read_slot].bio = bio;
			if (mirror == -1) {
				printk(KERN_ALERT "raid10: %s: unrecoverable I/O"
				       " read error for block %llu\n",
				       bdevname(bio->bi_bdev,b),
				       (unsigned long long)r10_bio->sector);
				raid_end_bio_io(r10_bio);
			} else {
				rdev = conf->mirrors[mirror].rdev;
				if (printk_ratelimit())
					printk(KERN_ERR "raid10: %s: redirecting sector %llu to"
					       " another mirror\n",
					       bdevname(rdev->bdev,b),
					       (unsigned long long)r10_bio->sector);
				bio->bi_bdev = rdev->bdev;
				bio->bi_sector = r10_bio->devs[r10_bio->read_slot].addr
					+ rdev->data_offset;
				bio->bi_next = NULL;
				bio->bi_flags &= (1<<BIO_CLONED);
				bio->bi_flags |= 1 << BIO_UPTODATE;
				bio->bi_idx = 0;
				bio->bi_size = r10_bio->sectors << 9;
				bio->bi_rw = READ;
				unplug = 1;
				generic_make_request(bio);
			}
		}
	}
	spin_unlock_irqrestore(&conf->device_lock, flags);
	if (unplug)
		unplug_slaves(mddev);
}


static int init_resync(conf_t *conf)
{
	int buffs;

	buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
	if (conf->r10buf_pool)
		BUG();
	conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf);
	if (!conf->r10buf_pool)
		return -ENOMEM;
	conf->next_resync = 0;
	return 0;
}

/*
 * perform a "sync" on one "block"
 *
 * We need to make sure that no normal I/O request - particularly write
 * requests - conflict with active sync requests.
 *
 * This is achieved by tracking pending requests and a 'barrier' concept
 * that can be installed to exclude normal IO requests.
 *
 * Resync and recovery are handled very differently.
 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
 *
 * For resync, we iterate over virtual addresses, read all copies,
 * and update if there are differences.  If only one copy is live,
 * skip it.
 * For recovery, we iterate over physical addresses, read a good
 * value for each non-in_sync drive, and over-write.
 *
 * So, for recovery we may have several outstanding complex requests for a
 * given address, one for each out-of-sync device.  We model this by allocating
 * a number of r10_bio structures, one for each out-of-sync device.
 * As we setup these structures, we collect all bio's together into a list
 * which we then process collectively to add pages, and then process again
 * to pass to generic_make_request.
 *
 * The r10_bio structures are linked using a borrowed master_bio pointer.
 * This link is counted in ->remaining.  When the r10_bio that points to NULL
 * has its remaining count decremented to 0, the whole complex operation
 * is complete.
 *
 */

static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster)
{
	conf_t *conf = mddev_to_conf(mddev);
	r10bio_t *r10_bio;
	struct bio *biolist = NULL, *bio;
	sector_t max_sector, nr_sectors;
	int disk;
	int i;

	sector_t sectors_skipped = 0;
	int chunks_skipped = 0;

	if (!conf->r10buf_pool)
		if (init_resync(conf))
			return -ENOMEM;

 skipped:
	max_sector = mddev->size << 1;
	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
		max_sector = mddev->resync_max_sectors;
	if (sector_nr >= max_sector) {
		close_sync(conf);