Initialize freespace_initialized bits to 0 in member's flags and update
member's cached version for each device in bch2_fs_initialize.

It's possible for the bits to be set to 1 before fs is initialized and if
call to bch2_trans_mark_dev_sbs (just before bch2_fs_freespace_init) fails
bits remain to be 1 which can later indirectly trigger BUG condition in
bch2_bucket_alloc_freelist during shutdown.

Reported-by: syzbot+2b6a17991a6af64f9489@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=2b6a17991a6af64f9489
Fixes: bbe682c76789 ("bcachefs: Ensure devices are always correctly initialized")
Suggested-by: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Piotr Zalewski <pZ010001011111@proton.me>
---

Notes:
    changes in v2:
        - unconditionally set freespace initialized bits to false at
          the top of bch2_fs_initialized instead of only if
          bch2_trans_mark_dev_sbs fails

    Link to v1: https://lore.kernel.org/linux-bcachefs/20241020170708.67044-2-pZ010001011111@proton.me/

 fs/bcachefs/recovery.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 67bba156cce9..ed3dbe5802b5 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -1030,6 +1030,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	struct bch_inode_unpacked root_inode, lostfound_inode;
 	struct bkey_inode_buf packed_inode;
 	struct qstr lostfound = QSTR("lost+found");
+	struct bch_member *m;
 	int ret;
 
 	bch_notice(c, "initializing new filesystem");
@@ -1046,6 +1047,13 @@ int bch2_fs_initialize(struct bch_fs *c)
 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 		bch2_write_super(c);
 	}
+
+	for_each_member_device(c, ca) {
+		m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
+		SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, false);
+		ca->mi = bch2_mi_to_cpu(m);
+	}
+
 	mutex_unlock(&c->sb_lock);
 
 	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
-- 
2.47.0

Hi all,

Any comments on this? I saw the most recent master and I think this might 
be not applying anymore so I can send v3 if needed. 

Best regards, Piotr Zalewski

On Monday, October 21st, 2024 at 8:10 PM, Piotr Zalewski <pZ010001011111@proton.me> wrote:

> Initialize freespace_initialized bits to 0 in member's flags and update
> member's cached version for each device in bch2_fs_initialize.
> 
> It's possible for the bits to be set to 1 before fs is initialized and if
> call to bch2_trans_mark_dev_sbs (just before bch2_fs_freespace_init) fails
> bits remain to be 1 which can later indirectly trigger BUG condition in
> bch2_bucket_alloc_freelist during shutdown.
> 
> Reported-by: syzbot+2b6a17991a6af64f9489@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=2b6a17991a6af64f9489
> Fixes: bbe682c76789 ("bcachefs: Ensure devices are always correctly initialized")
> Suggested-by: Kent Overstreet kent.overstreet@linux.dev
> 
> Signed-off-by: Piotr Zalewski pZ010001011111@proton.me
> 
> ---
> 
> Notes:
> changes in v2:
> - unconditionally set freespace initialized bits to false at
> the top of bch2_fs_initialized instead of only if
> bch2_trans_mark_dev_sbs fails
> 
> Link to v1: https://lore.kernel.org/linux-bcachefs/20241020170708.67044-2-pZ010001011111@proton.me/
> 
> fs/bcachefs/recovery.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
> 
> diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
> index 67bba156cce9..ed3dbe5802b5 100644
> --- a/fs/bcachefs/recovery.c
> +++ b/fs/bcachefs/recovery.c
> @@ -1030,6 +1030,7 @@ int bch2_fs_initialize(struct bch_fs *c)
> struct bch_inode_unpacked root_inode, lostfound_inode;
> struct bkey_inode_buf packed_inode;
> struct qstr lostfound = QSTR("lost+found");
> + struct bch_member *m;
> int ret;
> 
> bch_notice(c, "initializing new filesystem");
> @@ -1046,6 +1047,13 @@ int bch2_fs_initialize(struct bch_fs *c)
> SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
> 
> bch2_write_super(c);
> }
> +
> + for_each_member_device(c, ca) {
> + m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
> 
> + SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, false);
> + ca->mi = bch2_mi_to_cpu(m);
> 
> + }
> +
> mutex_unlock(&c->sb_lock);
> 
> 
> c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
> 
> --
> 2.47.0
>

On Fri, Oct 25, 2024 at 11:42:38PM +0000, Piotr Zalewski wrote:
> Hi all,
> 
> Any comments on this? I saw the most recent master and I think this might 
> be not applying anymore so I can send v3 if needed. 

I just missed it before, send me the rebased version and I'll apply