block: allocate struct hd_struct as part of struct bdev_inode

Allocate hd_struct together with struct block_device to pre-load
the lifetime rule changes in preparation of merging the two structures.

Note that part0 was previously embedded into struct gendisk, but is
a separate allocation now, and already points to the block_device instead
of the hd_struct. The lifetime of struct gendisk is still controlled by
the struct device embedded in the part0 hd_struct.

Signed-off-by: Christoph Hellwig <[e-mail]>
Reviewed-by: Jan Kara <[e-mail]>
Reviewed-by: Hannes Reinecke <[e-mail]>
Signed-off-by: Jens Axboe <[e-mail]>

block/blk-core.c

···
 
 static noinline int should_fail_bio(struct bio *bio)
 {
-	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
+	if (should_fail_request(bio->bi_disk->part0->bd_part,
+			bio->bi_iter.bi_size))
 		return -EIO;
 	return 0;
 }
···
 		if (unlikely(blk_partition_remap(bio)))
 			goto end_io;
 	} else {
-		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
+		if (unlikely(bio_check_ro(bio, bio->bi_disk->part0->bd_part)))
 			goto end_io;
 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
 			goto end_io;
···
 		return ret;
 
 	if (rq->rq_disk &&
-	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
+	    should_fail_request(rq->rq_disk->part0->bd_part, blk_rq_bytes(rq)))
 		return BLK_STS_IOERR;
 
 	if (blk_crypto_insert_cloned_request(rq))
···
 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
 	}
 	if (part->partno) {
-		part = &part_to_disk(part)->part0;
+		part = part_to_disk(part)->part0->bd_part;
 		goto again;
 	}
 }
···
 		part_stat_inc(part, ios[sgrp]);
 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
 		part_stat_unlock();
-
-		hd_struct_put(part);
 	}
 }
 
···
 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
 				 unsigned int op)
 {
-	return __part_start_io_acct(&disk->part0, sectors, op);
+	return __part_start_io_acct(disk->part0->bd_part, sectors, op);
 }
 EXPORT_SYMBOL(disk_start_io_acct);
 
···
 		      unsigned long start_time)
 {
 	__part_end_io_acct(part, bio_op(bio), start_time);
-	hd_struct_put(part);
 }
 EXPORT_SYMBOL_GPL(part_end_io_acct);
 
 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
 		      unsigned long start_time)
 {
-	__part_end_io_acct(&disk->part0, op, start_time);
+	__part_end_io_acct(disk->part0->bd_part, op, start_time);
 }
 EXPORT_SYMBOL(disk_end_io_acct);
 

block/blk-flush.c

···
 
 static void blk_account_io_flush(struct request *rq)
 {
-	struct hd_struct *part = &rq->rq_disk->part0;
+	struct hd_struct *part = rq->rq_disk->part0->bd_part;
 
 	part_stat_lock();
 	part_stat_inc(part, ios[STAT_FLUSH]);

block/blk-merge.c

···
 		part_stat_lock();
 		part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
 		part_stat_unlock();
-
-		hd_struct_put(req->part);
 	}
 }
 

block/blk.h

···
 int bdev_resize_partition(struct block_device *bdev, int partno,
 		sector_t start, sector_t length);
 int disk_expand_part_tbl(struct gendisk *disk, int target);
-int hd_ref_init(struct hd_struct *part);
-
-/* no need to get/put refcount of part0 */
-static inline int hd_struct_try_get(struct hd_struct *part)
-{
-	if (part->partno)
-		return percpu_ref_tryget_live(&part->ref);
-	return 1;
-}
-
-static inline void hd_struct_put(struct hd_struct *part)
-{
-	if (part->partno)
-		percpu_ref_put(&part->ref);
-}
-
-static inline void hd_free_part(struct hd_struct *part)
-{
-	bdput(part->bdev);
-	percpu_ref_exit(&part->ref);
-}
 
 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
 		struct page *page, unsigned int len, unsigned int offset,

block/genhd.c

···
 
 void set_capacity(struct gendisk *disk, sector_t sectors)
 {
-	struct block_device *bdev = disk->part0.bdev;
+	struct block_device *bdev = disk->part0;
 
 	spin_lock(&bdev->bd_size_lock);
 	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
···
  * primarily used for stats accounting.
  *
  * CONTEXT:
- * RCU read locked.  The returned partition pointer is always valid
- * because its refcount is grabbed except for part0, which lifetime
- * is same with the disk.
+ * RCU read locked.
  *
  * RETURNS:
  * Found partition on success, part0 is returned if no partition matches
···
 	ptbl = rcu_dereference(disk->part_tbl);
 
 	part = rcu_dereference(ptbl->last_lookup);
-	if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
+	if (part && sector_in_part(part, sector))
 		goto out_unlock;
 
 	for (i = 1; i < ptbl->len; i++) {
 		part = rcu_dereference(ptbl->part[i]);
 
 		if (part && sector_in_part(part, sector)) {
-			/*
-			 * only live partition can be cached for lookup,
-			 * so use-after-free on cached & deleting partition
-			 * can be avoided
-			 */
-			if (!hd_struct_try_get(part))
-				break;
 			rcu_assign_pointer(ptbl->last_lookup, part);
 			goto out_unlock;
 		}
 	}
 
-	part = &disk->part0;
+	part = disk->part0->bd_part;
 out_unlock:
 	rcu_read_unlock();
 	return part;
···
 	 */
 	pm_runtime_set_memalloc_noio(ddev, true);
 
-	disk->part0.bdev->bd_holder_dir =
-			kobject_create_and_add("holders", &ddev->kobj);
+	disk->part0->bd_holder_dir =
+		kobject_create_and_add("holders", &ddev->kobj);
 	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 
 	if (disk->flags & GENHD_FL_HIDDEN) {
···
 
 	disk->flags |= GENHD_FL_UP;
 
-	retval = blk_alloc_devt(&disk->part0, &devt);
+	retval = blk_alloc_devt(disk->part0->bd_part, &devt);
 	if (retval) {
 		WARN_ON(1);
 		return;
···
 		ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
 		WARN_ON(ret);
 		bdi_set_owner(bdi, dev);
-		bdev_add(disk->part0.bdev, devt);
+		bdev_add(disk->part0, devt);
 	}
 	register_disk(parent, disk, groups);
 	if (register_queue)
···
 
 	blk_unregister_queue(disk);
 
-	kobject_put(disk->part0.bdev->bd_holder_dir);
+	kobject_put(disk->part0->bd_holder_dir);
 	kobject_put(disk->slave_dir);
 
-	part_stat_set_all(&disk->part0, 0);
-	disk->part0.bdev->bd_stamp = 0;
+	part_stat_set_all(disk->part0->bd_part, 0);
+	disk->part0->bd_stamp = 0;
 	if (!sysfs_deprecated)
 		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
···
 		 */
 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 		while ((part = disk_part_iter_next(&piter))) {
-			bool is_part0 = part == &disk->part0;
+			bool is_part0 = part == disk->part0->bd_part;
 
 			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
 			       bdevt_str(part_devt(part), devt_buf),
···
 	disk_release_events(disk);
 	kfree(disk->random);
 	disk_replace_part_tbl(disk, NULL);
-	hd_free_part(&disk->part0);
+	bdput(disk->part0);
 	if (disk->queue)
 		blk_put_queue(disk->queue);
 	kfree(disk);
···
 	if (!disk)
 		return NULL;
 
-	disk->part0.bdev = bdev_alloc(disk, 0);
-	if (!disk->part0.bdev)
+	disk->part0 = bdev_alloc(disk, 0);
+	if (!disk->part0)
 		goto out_free_disk;
 
 	disk->node_id = node_id;
···
 		goto out_bdput;
 
 	ptbl = rcu_dereference_protected(disk->part_tbl, 1);
-	rcu_assign_pointer(ptbl->part[0], &disk->part0);
-
-	if (hd_ref_init(&disk->part0))
-		goto out_bdput;
+	rcu_assign_pointer(ptbl->part[0], disk->part0->bd_part);
 
 	disk->minors = minors;
 	rand_initialize_disk(disk);
···
 	return disk;
 
 out_bdput:
-	bdput(disk->part0.bdev);
+	bdput(disk->part0);
 out_free_disk:
 	kfree(disk);
 	return NULL;
···
 	struct disk_part_iter piter;
 	struct hd_struct *part;
 
-	if (disk->part0.bdev->bd_read_only != flag) {
+	if (disk->part0->bd_read_only != flag) {
 		set_disk_ro_uevent(disk, flag);
-		disk->part0.bdev->bd_read_only = flag;
+		disk->part0->bd_read_only = flag;
 	}
 
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);

block/partitions/core.c

···
 static void part_release(struct device *dev)
 {
 	struct hd_struct *p = dev_to_part(dev);
+
 	blk_free_devt(dev->devt);
-	hd_free_part(p);
-	kfree(p);
+	bdput(p->bdev);
 }
 
 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
···
 	.uevent		= part_uevent,
 };
 
-static void hd_struct_free_work(struct work_struct *work)
-{
-	struct hd_struct *part =
-		container_of(to_rcu_work(work), struct hd_struct, rcu_work);
-	struct gendisk *disk = part_to_disk(part);
-
-	/*
-	 * Release the disk reference acquired in delete_partition here.
-	 * We can't release it in hd_struct_free because the final put_device
-	 * needs process context and thus can't be run directly from a
-	 * percpu_ref ->release handler.
-	 */
-	put_device(disk_to_dev(disk));
-
-	part->bdev->bd_start_sect = 0;
-	bdev_set_nr_sectors(part->bdev, 0);
-	part_stat_set_all(part, 0);
-	put_device(part_to_dev(part));
-}
-
-static void hd_struct_free(struct percpu_ref *ref)
-{
-	struct hd_struct *part = container_of(ref, struct hd_struct, ref);
-	struct gendisk *disk = part_to_disk(part);
-	struct disk_part_tbl *ptbl =
-		rcu_dereference_protected(disk->part_tbl, 1);
-
-	rcu_assign_pointer(ptbl->last_lookup, NULL);
-
-	INIT_RCU_WORK(&part->rcu_work, hd_struct_free_work);
-	queue_rcu_work(system_wq, &part->rcu_work);
-}
-
-int hd_ref_init(struct hd_struct *part)
-{
-	if (percpu_ref_init(&part->ref, hd_struct_free, 0, GFP_KERNEL))
-		return -ENOMEM;
-	return 0;
-}
-
 /*
  * Must be called either with bd_mutex held, before a disk can be opened or
  * after all disk users are gone.
···
 	 * ->part_tbl is referenced in this part's release handler, so
 	 *  we have to hold the disk device
 	 */
-	get_device(disk_to_dev(disk));
 	rcu_assign_pointer(ptbl->part[part->partno], NULL);
+	rcu_assign_pointer(ptbl->last_lookup, NULL);
 	kobject_put(part->bdev->bd_holder_dir);
 	device_del(part_to_dev(part));
 
···
 	 */
 	remove_inode_hash(part->bdev->bd_inode);
 
-	percpu_ref_kill(&part->ref);
+	put_device(part_to_dev(part));
 }
 
 static ssize_t whole_disk_show(struct device *dev,
···
 	if (ptbl->part[partno])
 		return ERR_PTR(-EBUSY);
 
-	p = kzalloc(sizeof(*p), GFP_KERNEL);
-	if (!p)
-		return ERR_PTR(-EBUSY);
-
 	bdev = bdev_alloc(disk, partno);
 	if (!bdev)
-		goto out_free;
-	p->bdev = bdev;
+		return ERR_PTR(-ENOMEM);
 
+	p = bdev->bd_part;
 	pdev = part_to_dev(p);
 
 	bdev->bd_start_sect = start;
···
 			goto out_del;
 	}
 
-	err = hd_ref_init(p);
-	if (err) {
-		if (flags & ADDPART_FLAG_WHOLEDISK)
-			goto out_remove_file;
-		goto out_del;
-	}
-
 	/* everything is up and running, commence */
 	bdev_add(bdev, devt);
 	rcu_assign_pointer(ptbl->part[partno], p);
···
 
 out_bdput:
 	bdput(bdev);
-out_free:
-	kfree(p);
 	return ERR_PTR(err);
-out_remove_file:
-	device_remove_file(pdev, &dev_attr_whole_disk);
 out_del:
 	kobject_put(bdev->bd_holder_dir);
 	device_del(pdev);

drivers/block/drbd/drbd_receiver.c

···
 	if (c_min_rate == 0)
 		return false;
 
-	curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
+	curr_events = (int)part_stat_read_accum(disk->part0->bd_part, sectors) -
 			atomic_read(&device->rs_sect_ev);
 
 	if (atomic_read(&device->ap_actlog_cnt)

drivers/block/drbd/drbd_worker.c

···
 	atomic_set(&device->rs_sect_in, 0);
 	atomic_set(&device->rs_sect_ev, 0);
 	device->rs_in_flight = 0;
-	device->rs_last_events = (int)part_stat_read_accum(&disk->part0, sectors);
+	device->rs_last_events =
+		(int)part_stat_read_accum(disk->part0->bd_part, sectors);
 
 	/* Updating the RCU protected object in place is necessary since
 	   this function gets called from atomic context.

drivers/block/zram/zram_drv.c

···
 	zram->disksize = 0;
 
 	set_capacity_and_notify(zram->disk, 0);
-	part_stat_set_all(&zram->disk->part0, 0);
+	part_stat_set_all(zram->disk->part0->bd_part, 0);
 
 	up_write(&zram->init_lock);
 	/* I/O operation under all of CPU are done so let's free */

drivers/md/dm.c

···
 				 * (by eliminating DM's splitting and just using bio_split)
 				 */
 				part_stat_lock();
-				__dm_part_stat_sub(&dm_disk(md)->part0,
+				__dm_part_stat_sub(dm_disk(md)->part0->bd_part,
 						   sectors[op_stat_group(bio_op(bio))], ci.sector_count);
 				part_stat_unlock();
 
···
 static bool md_in_flight_bios(struct mapped_device *md)
 {
 	int cpu;
-	struct hd_struct *part = &dm_disk(md)->part0;
+	struct hd_struct *part = dm_disk(md)->part0->bd_part;
 	long sum = 0;
 
 	for_each_possible_cpu(cpu) {

drivers/md/md.c

···
 	rcu_read_lock();
 	rdev_for_each_rcu(rdev, mddev) {
 		struct gendisk *disk = rdev->bdev->bd_disk;
-		curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
+		curr_events = (int)part_stat_read_accum(disk->part0->bd_part, sectors) -
 			      atomic_read(&disk->sync_io);
 		/* sync IO will cause sync_io to increase before the disk_stats
 		 * as sync_io is counted when a request starts, and

fs/block_dev.c

···
 
 struct bdev_inode {
 	struct block_device bdev;
+	struct hd_struct hd;
 	struct inode vfs_inode;
 };
 
···
 		iput(inode);
 		return NULL;
 	}
+	bdev->bd_part = &BDEV_I(inode)->hd;
+	memset(bdev->bd_part, 0, sizeof(*bdev->bd_part));
+	bdev->bd_part->bdev = bdev;
 	return bdev;
 }
 
···
 static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 {
 	struct gendisk *disk = bdev->bd_disk;
-	int ret;
+	int ret = 0;
 
 	if (!bdev->bd_openers) {
 		if (!bdev_is_partition(bdev)) {
-			ret = -ENXIO;
-			bdev->bd_part = disk_get_part(disk, 0);
-			if (!bdev->bd_part)
-				goto out_clear;
-
 			ret = 0;
 			if (disk->fops->open)
 				ret = disk->fops->open(bdev, mode);
···
 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
 
 			if (ret)
-				goto out_clear;
+				return ret;
 		} else {
 			struct block_device *whole = bdget_disk(disk, 0);
 
···
 			if (ret) {
 				mutex_unlock(&whole->bd_mutex);
 				bdput(whole);
-				goto out_clear;
+				return ret;
 			}
 			whole->bd_part_count++;
 			mutex_unlock(&whole->bd_mutex);
 
-			bdev->bd_part = disk_get_part(disk, bdev->bd_partno);
 			if (!(disk->flags & GENHD_FL_UP) ||
-			    !bdev->bd_part || !bdev_nr_sectors(bdev)) {
+			    !bdev_nr_sectors(bdev)) {
 				__blkdev_put(whole, mode, 1);
 				bdput(whole);
-				ret = -ENXIO;
-				goto out_clear;
+				return -ENXIO;
 			}
 			set_init_blocksize(bdev);
 		}
···
 			bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
 	} else {
 		if (!bdev_is_partition(bdev)) {
-			ret = 0;
 			if (bdev->bd_disk->fops->open)
 				ret = bdev->bd_disk->fops->open(bdev, mode);
 			/* the same as first opener case, read comment there */
···
 	}
 	bdev->bd_openers++;
 	return 0;
-
- out_clear:
-	disk_put_part(bdev->bd_part);
-	bdev->bd_part = NULL;
-	return ret;
 }
 
 struct block_device *blkdev_get_no_open(dev_t dev)
···
 		sync_blockdev(bdev);
 		kill_bdev(bdev);
 		bdev_write_inode(bdev);
-
-		if (!bdev_is_partition(bdev) && disk->fops->release)
-			disk->fops->release(disk, mode);
-
-		disk_put_part(bdev->bd_part);
-		bdev->bd_part = NULL;
 		if (bdev_is_partition(bdev))
 			victim = bdev_whole(bdev);
-	} else {
-		if (!bdev_is_partition(bdev) && disk->fops->release)
-			disk->fops->release(disk, mode);
 	}
+
+	if (!bdev_is_partition(bdev) && disk->fops->release)
+		disk->fops->release(disk, mode);
 	mutex_unlock(&bdev->bd_mutex);
 	if (victim) {
 		__blkdev_put(victim, mode, 1);

include/linux/blk_types.h

···
 } __randomize_layout;
 
 #define bdev_whole(_bdev) \
-	((_bdev)->bd_disk->part0.bdev)
+	((_bdev)->bd_disk->part0)
 
 #define bdev_kobj(_bdev) \
 	(&part_to_dev((_bdev)->bd_part)->kobj)

include/linux/genhd.h

···
 #include <linux/blk_types.h>
 #include <asm/local.h>
 
-#define dev_to_disk(device)	container_of((device), struct gendisk, part0.__dev)
 #define dev_to_part(device)	container_of((device), struct hd_struct, __dev)
-#define disk_to_dev(disk)	(&(disk)->part0.__dev)
 #define part_to_dev(part)	(&((part)->__dev))
+
+#define dev_to_disk(device)	(dev_to_part(device)->bdev->bd_disk)
+#define disk_to_dev(disk)	(part_to_dev((disk)->part0->bd_part))
 
 extern const struct device_type disk_type;
 extern struct device_type part_type;
···
 };
 
 struct hd_struct {
-	struct percpu_ref ref;
-
 	struct block_device *bdev;
 	struct device __dev;
 	int partno;
-	struct rcu_work rcu_work;
 };
 
 /**
···
 	 * helpers.
 	 */
 	struct disk_part_tbl __rcu *part_tbl;
-	struct hd_struct part0;
+	struct block_device *part0;
 
 	const struct block_device_operations *fops;
 	struct request_queue *queue;
···
 
 static inline int get_disk_ro(struct gendisk *disk)
 {
-	return disk->part0.bdev->bd_read_only;
+	return disk->part0->bd_read_only;
 }
 
 extern void disk_block_events(struct gendisk *disk);
···
 
 static inline sector_t get_capacity(struct gendisk *disk)
 {
-	return bdev_nr_sectors(disk->part0.bdev);
+	return bdev_nr_sectors(disk->part0);
 }
 
 int bdev_disk_changed(struct block_device *bdev, bool invalidate);

include/linux/part_stat.h

···
 #define part_stat_add(part, field, addnd)	do {			\
 	__part_stat_add((part), field, addnd);				\
 	if ((part)->partno)						\
-		__part_stat_add(&part_to_disk((part))->part0,		\
-				field, addnd);				\
+		__part_stat_add(part_to_disk((part))->part0->bd_part,	\
+			field, addnd); \
 } while (0)
 
 #define part_stat_dec(part, field)					\