In two device configs of RAID1/RAID5 where one device can be missing in the degraded mount, or in the configs such as four devices RAID6 where two devices can be missing, in these type of configs it can form two separate set of devices where each of the set can be mounted without the other set. And if user does that and writes the data, one set would have to loose the data, as of now we just loose the data without the user consent depending on the SB generation number which user won't have any aware.
This patch introduces a new SB flag BTRFS_SUPER_FLAG_VOL_MOVED_ON which will be set when RAID group is mounted with missing device, so when RAID is mounted with no missing device and if all the devices contains this flag then we know that each of these set are mounted without the other set. In this scenario this patch will fail the mount so that user can decide which set to keep the data. Now the procedure to reassemble the disks would be to continue to mount the good set first without the device set on which new data can be ignored, and later on run btrfs device scan to bring-in the missing device and complete the RAID group which then shall reset the flag BTRFS_SUPER_FLAG_VOL_MOVED_ON. Signed-off-by: Anand Jain <anand.j...@oracle.com> --- fs/btrfs/disk-io.c | 53 ++++++++++++++++++++++++++++++++++++++++- fs/btrfs/volumes.c | 14 +++++++++-- include/uapi/linux/btrfs_tree.h | 1 + 3 files changed, 65 insertions(+), 3 deletions(-) diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index dfdab849037b..62b80226d157 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -59,7 +59,8 @@ BTRFS_HEADER_FLAG_RELOC |\ BTRFS_SUPER_FLAG_ERROR |\ BTRFS_SUPER_FLAG_SEEDING |\ - BTRFS_SUPER_FLAG_METADUMP) + BTRFS_SUPER_FLAG_METADUMP|\ + BTRFS_SUPER_FLAG_VOL_MOVED_ON) static const struct extent_io_ops btree_extent_io_ops; static void end_workqueue_fn(struct btrfs_work *work); @@ -2552,6 +2553,43 @@ static int btrfs_read_roots(struct btrfs_fs_info *fs_info) return 0; } +bool volume_has_split_brain(struct btrfs_fs_info *fs_info) +{ + unsigned long devs_moved_on = 0; + struct btrfs_fs_devices *fs_devs = fs_info->fs_devices; + struct list_head *head = &fs_devs->devices; + struct btrfs_device *dev; + +again: + list_for_each_entry(dev, head, dev_list) { + struct buffer_head *bh; + struct btrfs_super_block *sb; + + if (!dev->devid) + continue; + + bh = btrfs_read_dev_super(dev->bdev); + if (IS_ERR(bh)) + continue; + + sb = (struct btrfs_super_block *)bh->b_data; + if (btrfs_super_flags(sb) & BTRFS_SUPER_FLAG_VOL_MOVED_ON) + devs_moved_on++; + brelse(bh); + } + + fs_devs = fs_devs->seed; + if (fs_devs) { + head = &fs_devs->devices; + goto again; + } + + if (devs_moved_on == fs_info->fs_devices->total_devices) + return true; + else + return false; +} + int open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, char *options) @@ -3044,6 +3082,19 @@ int open_ctree(struct super_block *sb, goto fail_sysfs; } + if (fs_info->fs_devices->missing_devices) { + btrfs_set_super_flags(fs_info->super_copy, + fs_info->super_copy->flags | BTRFS_SUPER_FLAG_VOL_MOVED_ON); + } else if (fs_info->super_copy->flags & BTRFS_SUPER_FLAG_VOL_MOVED_ON) { + if (volume_has_split_brain(fs_info)) { + btrfs_err(fs_info, + "Detected 'moved_on' flag on all device"); + goto fail_sysfs; + } + btrfs_set_super_flags(fs_info->super_copy, + fs_info->super_copy->flags & ~BTRFS_SUPER_FLAG_VOL_MOVED_ON); + } + fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, "btrfs-cleaner"); if (IS_ERR(fs_info->cleaner_kthread)) diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index e7dd996831f2..ba32e2d73fbe 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -1988,8 +1988,13 @@ int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, device->fs_devices->num_devices--; device->fs_devices->total_devices--; - if (device->missing) + if (device->missing) { device->fs_devices->missing_devices--; + if (!device->missing) + btrfs_set_super_flags(fs_info->super_copy, + fs_info->super_copy->flags & + ~BTRFS_SUPER_FLAG_VOL_MOVED_ON); + } btrfs_assign_next_active_device(fs_info, device, NULL); @@ -2062,8 +2067,13 @@ void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, list_del_rcu(&srcdev->dev_list); list_del_rcu(&srcdev->dev_alloc_list); fs_devices->num_devices--; - if (srcdev->missing) + if (srcdev->missing) { fs_devices->missing_devices--; + if (!srcdev->missing) + btrfs_set_super_flags(fs_info->super_copy, + fs_info->super_copy->flags & + ~BTRFS_SUPER_FLAG_VOL_MOVED_ON); + } if (srcdev->writeable) fs_devices->rw_devices--; diff --git a/include/uapi/linux/btrfs_tree.h b/include/uapi/linux/btrfs_tree.h index 10689e1fdf11..b53731b5b5a0 100644 --- a/include/uapi/linux/btrfs_tree.h +++ b/include/uapi/linux/btrfs_tree.h @@ -455,6 +455,7 @@ struct btrfs_free_space_header { #define BTRFS_SUPER_FLAG_SEEDING (1ULL << 32) #define BTRFS_SUPER_FLAG_METADUMP (1ULL << 33) +#define BTRFS_SUPER_FLAG_VOL_MOVED_ON (1ULL << 36) /* -- 2.13.1 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html