md: allow configuring logical block size

Previously, raid array used the maximum logical block size (LBS)
of all member disks. Adding a larger LBS disk at runtime could
unexpectedly increase RAID's LBS, risking corruption of existing
partitions. This can be reproduced by:

```
  # LBS of sd[de] is 512 bytes, sdf is 4096 bytes.
  mdadm -CRq /dev/md0 -l1 -n3 /dev/sd[de] missing --assume-clean

  # LBS is 512
  cat /sys/block/md0/queue/logical_block_size

  # create partition md0p1
  parted -s /dev/md0 mklabel gpt mkpart primary 1MiB 100%
  lsblk | grep md0p1

  # LBS becomes 4096 after adding sdf
  mdadm --add -q /dev/md0 /dev/sdf
  cat /sys/block/md0/queue/logical_block_size

  # partition lost
  partprobe /dev/md0
  lsblk | grep md0p1
```

Simply restricting larger-LBS disks is inflexible. In some scenarios,
only disks with 512 bytes LBS are available currently, but later, disks
with 4KB LBS may be added to the array.

Making LBS configurable is the best way to solve this scenario.
After this patch, the raid will:
  - store LBS in disk metadata
  - add a read-write sysfs 'mdX/logical_block_size'

Future mdadm should support setting LBS via metadata field during RAID
creation and the new sysfs. Though the kernel allows runtime LBS changes,
users should avoid modifying it after creating partitions or filesystems
to prevent compatibility issues.

Only 1.x metadata supports configurable LBS. 0.90 metadata inits all
fields to default values at auto-detect. Supporting 0.90 would require
more extensive changes and no such use case has been observed.

Note that many RAID paths rely on PAGE_SIZE alignment, including for
metadata I/O. A larger LBS than PAGE_SIZE will result in metadata
read/write failures. So this config should be prevented.

Link: https://lore.kernel.org/linux-raid/20251103125757.1405796-6-linan666@huaweicloud.com
Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Yu Kuai <yukuai@fnnas.com>

9 files changed, 95 insertions, 1 deletions

diff --git a/Documentation/admin-guide/md.rst b/Documentation/admin-guide/md.rst
index deed823eab01..dc7eab191caa 100644
--- a/Documentation/admin-guide/md.rst
+++ b/Documentation/admin-guide/md.rst
@@ -238,6 +238,16 @@ All md devices contain:
      the number of devices in a raid4/5/6, or to support external
      metadata formats which mandate such clipping.
 
+  logical_block_size
+     Configure the array's logical block size in bytes. This attribute
+     is only supported for 1.x meta. Write the value before starting
+     array. The final array LBS uses the maximum between this
+     configuration and LBS of all combined devices. Note that
+     LBS cannot exceed PAGE_SIZE before RAID supports folio.
+     WARNING: Arrays created on new kernel cannot be assembled at old
+     kernel due to padding check, Set module parameter 'check_new_feature'
+     to false to bypass, but data loss may occur.
+
   reshape_position
      This is either ``none`` or a sector number within the devices of
      the array where ``reshape`` is up to.  If this is set, the three
diff --git a/drivers/md/md-linear.c b/drivers/md/md-linear.c
index 25a6ddedea65..8d7b82c4a723 100644
--- a/drivers/md/md-linear.c
+++ b/drivers/md/md-linear.c
@@ -72,6 +72,7 @@ static int linear_set_limits(struct mddev *mddev)
 
 	md_init_stacking_limits(&lim);
 	lim.max_hw_sectors = mddev->chunk_sectors;
+	lim.logical_block_size = mddev->logical_block_size;
 	lim.max_write_zeroes_sectors = mddev->chunk_sectors;
 	lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors;
 	lim.io_min = mddev->chunk_sectors << 9;
diff --git a/drivers/md/md.c b/drivers/md/md.c
index 9676e2477df6..7b5c5967568f 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -1999,6 +1999,7 @@ static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struc
 		mddev->layout = le32_to_cpu(sb->layout);
 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 		mddev->dev_sectors = le64_to_cpu(sb->size);
+		mddev->logical_block_size = le32_to_cpu(sb->logical_block_size);
 		mddev->events = ev1;
 		mddev->bitmap_info.offset = 0;
 		mddev->bitmap_info.space = 0;
@@ -2208,6 +2209,7 @@ static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
 	sb->level = cpu_to_le32(mddev->level);
 	sb->layout = cpu_to_le32(mddev->layout);
+	sb->logical_block_size = cpu_to_le32(mddev->logical_block_size);
 	if (test_bit(FailFast, &rdev->flags))
 		sb->devflags |= FailFast1;
 	else
@@ -5936,6 +5938,68 @@ static struct md_sysfs_entry md_serialize_policy =
 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
        serialize_policy_store);
 
+static int mddev_set_logical_block_size(struct mddev *mddev,
+				unsigned int lbs)
+{
+	int err = 0;
+	struct queue_limits lim;
+
+	if (queue_logical_block_size(mddev->gendisk->queue) >= lbs) {
+		pr_err("%s: Cannot set LBS smaller than mddev LBS %u\n",
+		       mdname(mddev), lbs);
+		return -EINVAL;
+	}
+
+	lim = queue_limits_start_update(mddev->gendisk->queue);
+	lim.logical_block_size = lbs;
+	pr_info("%s: logical_block_size is changed, data may be lost\n",
+		mdname(mddev));
+	err = queue_limits_commit_update(mddev->gendisk->queue, &lim);
+	if (err)
+		return err;
+
+	mddev->logical_block_size = lbs;
+	/* New lbs will be written to superblock after array is running */
+	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
+	return 0;
+}
+
+static ssize_t
+lbs_show(struct mddev *mddev, char *page)
+{
+	return sprintf(page, "%u\n", mddev->logical_block_size);
+}
+
+static ssize_t
+lbs_store(struct mddev *mddev, const char *buf, size_t len)
+{
+	unsigned int lbs;
+	int err = -EBUSY;
+
+	/* Only 1.x meta supports configurable LBS */
+	if (mddev->major_version == 0)
+		return -EINVAL;
+
+	if (mddev->pers)
+		return -EBUSY;
+
+	err = kstrtouint(buf, 10, &lbs);
+	if (err < 0)
+		return -EINVAL;
+
+	err = mddev_lock(mddev);
+	if (err)
+		goto unlock;
+
+	err = mddev_set_logical_block_size(mddev, lbs);
+
+unlock:
+	mddev_unlock(mddev);
+	return err ?: len;
+}
+
+static struct md_sysfs_entry md_logical_block_size =
+__ATTR(logical_block_size, 0644, lbs_show, lbs_store);
 
 static struct attribute *md_default_attrs[] = {
 	&md_level.attr,
@@ -5958,6 +6022,7 @@ static struct attribute *md_default_attrs[] = {
 	&md_consistency_policy.attr,
 	&md_fail_last_dev.attr,
 	&md_serialize_policy.attr,
+	&md_logical_block_size.attr,
 	NULL,
 };
 
@@ -6088,6 +6153,17 @@ int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
 			return -EINVAL;
 	}
 
+	/*
+	 * Before RAID adding folio support, the logical_block_size
+	 * should be smaller than the page size.
+	 */
+	if (lim->logical_block_size > PAGE_SIZE) {
+		pr_err("%s: logical_block_size must not larger than PAGE_SIZE\n",
+			mdname(mddev));
+		return -EINVAL;
+	}
+	mddev->logical_block_size = lim->logical_block_size;
+
 	return 0;
 }
 EXPORT_SYMBOL_GPL(mddev_stack_rdev_limits);
@@ -6699,6 +6775,7 @@ static void md_clean(struct mddev *mddev)
 	mddev->chunk_sectors = 0;
 	mddev->ctime = mddev->utime = 0;
 	mddev->layout = 0;
+	mddev->logical_block_size = 0;
 	mddev->max_disks = 0;
 	mddev->events = 0;
 	mddev->can_decrease_events = 0;
diff --git a/drivers/md/md.h b/drivers/md/md.h
index fd6e001c1d38..6985f2829bbd 100644
--- a/drivers/md/md.h
+++ b/drivers/md/md.h
@@ -433,6 +433,7 @@ struct mddev {
 	sector_t			array_sectors; /* exported array size */
 	int				external_size; /* size managed
 							* externally */
+	unsigned int			logical_block_size;
 	__u64				events;
 	/* If the last 'event' was simply a clean->dirty transition, and
 	 * we didn't write it to the spares, then it is safe and simple
diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c
index fbf763401521..47aee1b1d4d1 100644
--- a/drivers/md/raid0.c
+++ b/drivers/md/raid0.c
@@ -380,6 +380,7 @@ static int raid0_set_limits(struct mddev *mddev)
 	lim.max_hw_sectors = mddev->chunk_sectors;
 	lim.max_write_zeroes_sectors = mddev->chunk_sectors;
 	lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors;
+	lim.logical_block_size = mddev->logical_block_size;
 	lim.io_min = mddev->chunk_sectors << 9;
 	lim.io_opt = lim.io_min * mddev->raid_disks;
 	lim.chunk_sectors = mddev->chunk_sectors;
diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index 592a40233004..57d50465eed1 100644
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c
@@ -3213,6 +3213,7 @@ static int raid1_set_limits(struct mddev *mddev)
 	md_init_stacking_limits(&lim);
 	lim.max_write_zeroes_sectors = 0;
 	lim.max_hw_wzeroes_unmap_sectors = 0;
+	lim.logical_block_size = mddev->logical_block_size;
 	lim.features |= BLK_FEAT_ATOMIC_WRITES;
 	err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
 	if (err)
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index 14dcd5142eb4..84be4cc7e873 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -4000,6 +4000,7 @@ static int raid10_set_queue_limits(struct mddev *mddev)
 	md_init_stacking_limits(&lim);
 	lim.max_write_zeroes_sectors = 0;
 	lim.max_hw_wzeroes_unmap_sectors = 0;
+	lim.logical_block_size = mddev->logical_block_size;
 	lim.io_min = mddev->chunk_sectors << 9;
 	lim.chunk_sectors = mddev->chunk_sectors;
 	lim.io_opt = lim.io_min * raid10_nr_stripes(conf);
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 24b32a0c95b4..cdbc7eba5c54 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -7745,6 +7745,7 @@ static int raid5_set_limits(struct mddev *mddev)
 	stripe = roundup_pow_of_two(data_disks * (mddev->chunk_sectors << 9));
 
 	md_init_stacking_limits(&lim);
+	lim.logical_block_size = mddev->logical_block_size;
 	lim.io_min = mddev->chunk_sectors << 9;
 	lim.io_opt = lim.io_min * (conf->raid_disks - conf->max_degraded);
 	lim.features |= BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE;
diff --git a/include/uapi/linux/raid/md_p.h b/include/uapi/linux/raid/md_p.h
index ac74133a4768..310068bb2a1d 100644
--- a/include/uapi/linux/raid/md_p.h
+++ b/include/uapi/linux/raid/md_p.h
@@ -291,7 +291,8 @@ struct mdp_superblock_1 {
 	__le64	resync_offset;	/* data before this offset (from data_offset) known to be in sync */
 	__le32	sb_csum;	/* checksum up to devs[max_dev] */
 	__le32	max_dev;	/* size of devs[] array to consider */
-	__u8	pad3[64-32];	/* set to 0 when writing */
+	__le32  logical_block_size;	/* same as q->limits->logical_block_size */
+	__u8	pad3[64-36];	/* set to 0 when writing */
 
 	/* device state information. Indexed by dev_number.
 	 * 2 bytes per device