Ceph » rgw

I reproduced it with some debug printing added and see that when `RadosBucket::check_quota` [1] is called, the `owner` is set to the user that makes the request, not to the user that owns the bucket. Though the `info.owner` is properly set to the bucket owner. So this simple patch makes it work as expected:

diff --git a/src/rgw/driver/rados/rgw_sal_rados.cc b/src/rgw/driver/rados/rgw_sal_rados.cc
index cd0e4e231b6..7ac20b0b19c 100644
--- a/src/rgw/driver/rados/rgw_sal_rados.cc
+++ b/src/rgw/driver/rados/rgw_sal_rados.cc
@@ -785,7 +785,7 @@ int RadosBucket::check_empty(const DoutPrefixProvider* dpp, optional_yield y)
 int RadosBucket::check_quota(const DoutPrefixProvider *dpp, RGWQuota& quota, uint64_t obj_size,
                                optional_yield y, bool check_size_only)
 {
-    return store->getRados()->check_quota(dpp, owner->get_id(), get_key(),
+    return store->getRados()->check_quota(dpp, info.owner, get_key(),
                                          quota, obj_size, y, check_size_only);
 }

It leaves the question though if it is expected that the `owner` is set to the user that makes the request. Isn't the bug actually that RadosBucket's `owner` member is wrongly initialized?

[1] https://github.com/ceph/ceph/blob/7e5b2542ab3379cf227774e05eef5f8494ef9da9/src/rgw/driver/rados/rgw_sal_rados.cc#L788

I would expect owner and info.owner to be the same, so something is definitely wrong here. We'd talked about removing the SAL User type entirely, but that's a bigger fix.

Mykola Golub wrote:

It leaves the question though if it is expected that the `owner` is set to the user that makes the request. Isn't the bug actually that RadosBucket's `owner` member is wrongly initialized?

yes, it definitely is a bug. most/all of the ops initialize rgw::sal::Bucket with the authorized user s->user instead of the bucket's actual owner. i think rgw::sal::Bucket should only expose a rgw_user get_owner() { return info.owner; } instead of holding a pointer to rgw::sal::User