From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 8 Jun 2025 00:18:16 +0000 (-0400)
Subject: bcachefs: Make sure opts.read_only gets propagated back to VFS
X-Git-Url: https://git.kernel.dk/?a=commitdiff_plain;h=f946ce0be45e75801583a5371fccf7466961d9d0;p=linux-block.git

bcachefs: Make sure opts.read_only gets propagated back to VFS

If we think we're read-only but the VFS doesn't, fun will ensue.

And now that we know we have to be able to do this safely, just make
nochanges imply ro.

Reported-by: syzbot+a7d6ceaba099cc21dee4@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---

diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 85d13f800165..3063a8ddc2df 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -2490,6 +2490,14 @@ static int bch2_fs_get_tree(struct fs_context *fc)
 	if (ret)
 		goto err_stop_fs;
 
+	/*
+	 * We might be doing a RO mount because other options required it, or we
+	 * have no alloc info and it's a small image with no room to regenerate
+	 * it
+	 */
+	if (c->opts.read_only)
+		fc->sb_flags |= SB_RDONLY;
+
 	sb = sget(fc->fs_type, NULL, bch2_set_super, fc->sb_flags|SB_NOSEC, c);
 	ret = PTR_ERR_OR_ZERO(sb);
 	if (ret)
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 06d0ba0fbffb..520eb72c4971 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -752,9 +752,11 @@ int bch2_fs_recovery(struct bch_fs *c)
 			? min(c->opts.recovery_pass_last, BCH_RECOVERY_PASS_snapshots_read)
 			: BCH_RECOVERY_PASS_snapshots_read;
 		c->opts.nochanges = true;
-		c->opts.read_only = true;
 	}
 
+	if (c->opts.nochanges)
+		c->opts.read_only = true;
+
 	mutex_lock(&c->sb_lock);
 	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 	bool write_sb = false;