Commit 0c0cbfdb84725e9933a24ecf47c61bdeeda06ba2 dropped the ctx->pos
update before the call to dir_emit. This breaks the userspace
implementation, causing the directory reads to be stuck in an infinite
loop. This doesn't happen in the kernel because the vfs handles the
updates to ctx->pos, but in the fuse implementation nobody updates
it.

Signed-off-by: Ariel Miculas <[email protected]>
---
V1 -> V2:
- added a comment explaining why the update to ctx->pos is necessary

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

diff --git a/fs/bcachefs/dirent.c b/fs/bcachefs/dirent.c
index c67460d8205d..d743da89308e 100644
--- a/fs/bcachefs/dirent.c
+++ b/fs/bcachefs/dirent.c
@@ -534,6 +534,14 @@ int bch2_empty_dir_trans(struct btree_trans *trans, 
subvol_inum dir)
 static int bch2_dir_emit(struct dir_context *ctx, struct bkey_s_c_dirent d, 
subvol_inum target)
 {
        struct qstr name = bch2_dirent_get_name(d);
+       /*
+        * Although not required by the kernel code, updating ctx->pos is needed
+        * for the bcachefs FUSE driver. Without this update, the FUSE
+        * implementation will be stuck in an infinite loop when reading
+        * directories (via the bcachefs_fuse_readdir callback).
+        * In kernel space, ctx->pos is updated by the VFS code.
+        */
+       ctx->pos = d.k->p.offset;
        bool ret = dir_emit(ctx, name.name,
                      name.len,
                      target.inum,
-- 
2.34.1


Reply via email to