Hi all This are the test steps: [root@storageqe-54 ~]# mdadm -CR /dev/md0 -l1 -n2 /dev/loop0 /dev/loop1 -e 1.0 --size=1G [root@storageqe-54 ~]# mdadm --wait /dev/md0 [root@storageqe-54 ~]# mdadm -G /dev/md0 --size=5G mdadm: component size of /dev/md0 has been set to 5242880K [root@storageqe-54 ~]# mdadm --wait /dev/md0 [root@storageqe-54 ~]# mdadm -G /dev/md0 --size=6G mdadm: Cannot set device size for /dev/md0: No space left on device loop0 7:0 0 7.8G 0 loop └─md0 9:0 0 5G 0 raid1 loop1 7:1 0 7.8G 0 loop └─md0 9:0 0 5G 0 raid1 The reason is that it calcluates the max usable space in a wrong way for super1.0 array. It uses rdev->sectors to calcuate the max usable space rather than the whole disk size. At first it grow size and it's successful. Because rdev->sectors is set to the whole disk size when creating raid device. super_1_load sets rdev->sectors to sb->data_size. In mdadm it sets sb->data_size to the whole disk size rathan than the raid device size. I think it's a wrong place too. Because of this error, it can grow size successfully at the first time. It re-sets rdev->sectors to real raid device after reshaping to the new size. So it can't grow size anymore. Xiao Ni (2): super1.0 calculates max sectors in a wrong way Don't need to reset superblock start address for super1.0 drivers/md/md.c | 27 +++++++++++++++++++++++---- 1 file changed, 23 insertions(+), 4 deletions(-) -- 2.7.5
