bcachefs version upgrade isn't written to disk immediately (umount is needed) [1166d22b67675611902976b36dbf38d45848826f]

[jpsollie]

when performing a version upgrade at mount, apparently the upgraded version isn't written to disk immediately,
but gets written after a umount, eg my multimedia fs gets mounted at boot:

#dmesg | grep bcachefs
[   40.053674] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): starting version 1.12: rebalance_work_acct_fix opts=metadata_replicas=2,data_checksum=xxhash,metadata_target=SSDM,foreground_target=SSDC,background_target=HDD,promote_target=SSDC
[   40.053684] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): recovering from clean shutdown, journal seq 88863359
[   40.053688] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): Doing compatible version upgrade from 1.12: rebalance_work_acct_fix to 1.13: inode_has_child_snapshots
[   40.148281] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): accounting_read...
[   42.840116] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): alloc_read... done
[   42.854930] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): stripes_read... done
[   42.854934] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): snapshots_read... done
[   43.003431] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): going read-write
[   43.219441] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): journal_replay... done
[   43.219444] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): check_inodes...
[   73.048135] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): resume_logged_ops... done
[   73.052540] [ T4446] bcachefs (ac7b718e-2f5b-4910-af1b-0255fa366f19): delete_dead_inodes... done

#bcachefs show-super /dev/md3
Device:                                     (unknown device)
External UUID:                             ac7b718e-2f5b-4910-af1b-0255fa366f19
Internal UUID:                             2a4933d1-83cb-4aa5-8635-fed38050445b
Magic number:                              c68573f6-66ce-90a9-d96a-60cf803df7ef
Device index:                              0
Label:                                     multimedia
Version:                                   1.12: rebalance_work_acct_fix
Version upgrade complete:                  1.12: rebalance_work_acct_fix
Oldest version on disk:                    1.12: rebalance_work_acct_fix
Created:                                   Sat Dec 16 22:09:09 2023
Sequence number:                           649
Time of last write:                        Thu Oct 10 10:12:25 2024
Superblock size:                           7.40 KiB/1.00 MiB
Clean:                                     1
Devices:                                   14
Sections:                                  members_v1,replicas_v0,disk_groups,clean,journal_seq_blacklist,journal_v2,counters,members_v2,errors,ext,downgrade
Features:                                  journal_seq_blacklist_v3,reflink,new_siphash,inline_data,new_extent_overwrite,btree_ptr_v2,reflink_inline_data,new_varint,journal_no_flush,alloc_v2,extents_across_btree_nodes
Compat features:                           alloc_info,alloc_metadata,extents_above_btree_updates_done,bformat_overflow_done
# umount /data/multimedia

# bcachefs show-super /dev/md3
Device:                                     (unknown device)
External UUID:                             ac7b718e-2f5b-4910-af1b-0255fa366f19
Internal UUID:                             2a4933d1-83cb-4aa5-8635-fed38050445b
Magic number:                              c68573f6-66ce-90a9-d96a-60cf803df7ef
Device index:                              0
Label:                                     multimedia
Version:                                   1.13: inode_has_child_snapshots
Version upgrade complete:                  1.13: inode_has_child_snapshots
Oldest version on disk:                    1.12: rebalance_work_acct_fix
Created:                                   Sat Dec 16 22:09:09 2023
Sequence number:                           654
Time of last write:                        Thu Oct 10 10:39:26 2024
Superblock size:                           7.41 KiB/1.00 MiB
Clean:                                     1
Devices:                                   14
Sections:                                  members_v1,replicas_v0,disk_groups,clean,journal_seq_blacklist,journal_v2,counters,members_v2,errors,ext,downgrade
Features:                                  journal_seq_blacklist_v3,reflink,new_siphash,inline_data,new_extent_overwrite,btree_ptr_v2,reflink_inline_data,new_varint,journal_no_flush,alloc_v2,extents_across_btree_nodes
Compat features:                           alloc_info,alloc_metadata,extents_above_btree_updates_done,bformat_overflow_done

this may result in a warning "incomplete upgrade" when the system uncleanly unmounts while it did a version upgrade at boot

[jpsollie]

something like this is needed (BUT THIS ONE WOULD COULD DEADLOCK, SO DO NOT IMPLEMENT):

--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -620,7 +620,7 @@ static bool check_version_upgrade(struct bch_fs *c)
                bch_info(c, "%s", buf.buf);
 
                bch2_sb_upgrade(c, new_version);
-
+               c->vfs_sb->s_op->sync_fs(c->vfs_sb, 0);
                printbuf_exit(&buf);
                return true;
        }

... but I can't imagine a vfs sync() can be called before the upgrade is completed, so this patch won't work