Merge pull request ceph#58213 from benhanokh/ref_count

rgw/rgw_rados: fix server side-copy orphans tail-objects

Reviewed-by: Casey Bodley <[email protected]>

Author: cbodley

qa/workunits/rgw/test_rgw_orphan_list.sh

@@ -6,7 +6,8 @@ set -e
 # if defined, debug messages will be displayed and prepended with the string
 # debug="DEBUG"
 
-huge_size=5100 # in megabytes
+#huge_size=5100 # in megabytes
+huge_size=51 # in megabytes
 big_size=7 # in megabytes
 
 huge_obj=/tmp/huge_obj.temp.$$
@@ -376,6 +377,95 @@ done
 
 mys3cmd rb --recursive s3://$o_bkt
 
+############################################################
+# copy multipart objects and delete destination
+
+o_bkt="orig-mp-bkt-5"
+d_bkt="copy-mp-bkt-5"
+
+mys3cmd mb s3://$o_bkt
+
+for f in $(seq 2) ;do
+    dest_obj="orig-multipart-obj-$f"
+    mys3cmd put -q $huge_obj s3://${o_bkt}/$dest_obj
+done
+
+mys3cmd mb s3://$d_bkt
+
+mys3cmd cp s3://${o_bkt}/orig-multipart-obj-1 \
+	s3://${d_bkt}/copied-multipart-obj-1
+
+for f in $(seq 5 5) ;do
+    dest_obj="orig-multipart-obj-$f"
+    mys3cmd put -q $huge_obj s3://${d_bkt}/$dest_obj
+done
+
+mys3cmd rb --recursive s3://$d_bkt
+
+#####################################################################
+# FORCE GARBAGE COLLECTION
+sleep 6 # since for testing age at which gc can happen is 5 secs
+radosgw-admin gc process --include-all
+#####################################################################
+
+############################################################
+# copy multipart objects and delete original then destination
+
+o_bkt="orig-mp-bkt-6"
+d_bkt="copy-mp-bkt-6"
+
+mys3cmd mb s3://$o_bkt
+
+for f in $(seq 2) ;do
+    dest_obj="orig-multipart-obj-$f"
+    mys3cmd put -q $huge_obj s3://${o_bkt}/$dest_obj
+done
+
+mys3cmd mb s3://$d_bkt
+
+mys3cmd cp s3://${o_bkt}/orig-multipart-obj-1 \
+	s3://${d_bkt}/copied-multipart-obj-1
+
+for f in $(seq 5 5) ;do
+    dest_obj="orig-multipart-obj-$f"
+    mys3cmd put -q $huge_obj s3://${d_bkt}/$dest_obj
+done
+
+mys3cmd rb --recursive s3://$o_bkt
+mys3cmd rb --recursive s3://$d_bkt
+
+############################################################
+# copy multipart objects and delete destination then original
+
+o_bkt="orig-mp-bkt-7"
+d_bkt="copy-mp-bkt-7"
+
+mys3cmd mb s3://$o_bkt
+
+for f in $(seq 2) ;do
+    dest_obj="orig-multipart-obj-$f"
+    mys3cmd put -q $huge_obj s3://${o_bkt}/$dest_obj
+done
+
+mys3cmd mb s3://$d_bkt
+
+mys3cmd cp s3://${o_bkt}/orig-multipart-obj-1 \
+	s3://${d_bkt}/copied-multipart-obj-1
+
+for f in $(seq 5 5) ;do
+    dest_obj="orig-multipart-obj-$f"
+    mys3cmd put -q $huge_obj s3://${d_bkt}/$dest_obj
+done
+
+mys3cmd rb --recursive s3://$d_bkt
+mys3cmd rb --recursive s3://$o_bkt
+
+#####################################################################
+# FORCE GARBAGE COLLECTION
+sleep 6 # since for testing age at which gc can happen is 5 secs
+radosgw-admin gc process --include-all
+#####################################################################
+
 ########################################################################
 # SWIFT TESTS
 

src/rgw/driver/rados/rgw_rados.cc

@@ -2915,6 +2915,7 @@ int RGWRados::swift_versioning_copy(RGWObjectCtx& obj_ctx,
   jspan_context no_trace{false, false};
 
   r = copy_obj(obj_ctx,
+               obj_ctx,  /* src and dest share an obj_ctx */
                owner,
                remote_user,
                NULL, /* req_info *info */
@@ -3014,6 +3015,7 @@ int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx,
     jspan_context no_trace{false, false};
 
     int ret = copy_obj(obj_ctx,
+                       obj_ctx,  /* src and dest share an obj_ctx */
                        owner,
                        remote_user,
                        nullptr,       /* req_info *info */
@@ -4154,7 +4156,7 @@ int RGWFetchObjFilter_Default::filter(CephContext *cct,
   return 0;
 }
 
-int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx,
+int RGWRados::fetch_remote_obj(RGWObjectCtx& dest_obj_ctx,
                const rgw_user& user_id,
                req_info *info,
                const rgw_zone_id& source_zone,
@@ -4206,7 +4208,7 @@ int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx,
   using namespace rgw::putobj;
   jspan_context no_trace{false, false};
   AtomicObjectProcessor processor(&aio, this, dest_bucket_info, nullptr,
-                                  owner, obj_ctx, dest_obj, olh_epoch,
+                                  owner, dest_obj_ctx, dest_obj, olh_epoch,
 				  tag, rctx.dpp, rctx.y, no_trace);
   RGWRESTConn *conn;
   auto& zone_conn_map = svc.zone->get_zone_conn_map();
@@ -4290,7 +4292,7 @@ int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx,
 
   if (copy_if_newer) {
     /* need to get mtime for destination */
-    ret = get_obj_state(rctx.dpp, &obj_ctx, dest_bucket_info, stat_dest_obj, &dest_state, &manifest, stat_follow_olh, rctx.y);
+    ret = get_obj_state(rctx.dpp, &dest_obj_ctx, dest_bucket_info, stat_dest_obj, &dest_state, &manifest, stat_follow_olh, rctx.y);
     if (ret < 0)
       goto set_err_state;
 
@@ -4499,8 +4501,8 @@ int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx,
 
     if (copy_if_newer && canceled) {
       ldpp_dout(rctx.dpp, 20) << "raced with another write of obj: " << dest_obj << dendl;
-      obj_ctx.invalidate(dest_obj); /* object was overwritten */
-      ret = get_obj_state(rctx.dpp, &obj_ctx, dest_bucket_info, stat_dest_obj, &dest_state, &manifest, stat_follow_olh, rctx.y);
+      dest_obj_ctx.invalidate(dest_obj); /* object was overwritten */
+      ret = get_obj_state(rctx.dpp, &dest_obj_ctx, dest_bucket_info, stat_dest_obj, &dest_state, &manifest, stat_follow_olh, rctx.y);
       if (ret < 0) {
         ldpp_dout(rctx.dpp, 0) << "ERROR: " << __func__ << ": get_err_state() returned ret=" << ret << dendl;
         goto set_err_state;
@@ -4534,7 +4536,7 @@ int RGWRados::fetch_remote_obj(RGWObjectCtx& obj_ctx,
     // for OP_LINK_OLH to call set_olh() with a real olh_epoch
     if (olh_epoch && *olh_epoch > 0) {
       constexpr bool log_data_change = true;
-      ret = set_olh(rctx.dpp, obj_ctx, dest_bucket_info, dest_obj, false, nullptr,
+      ret = set_olh(rctx.dpp, dest_obj_ctx, dest_bucket_info, dest_obj, false, nullptr,
                     *olh_epoch, real_time(), false, rctx.y, zones_trace, log_data_change);
     } else {
       // we already have the latest copy
@@ -4609,7 +4611,8 @@ int RGWRados::copy_obj_to_remote_dest(const DoutPrefixProvider *dpp,
  * err: stores any errors resulting from the get of the original object
  * Returns: 0 on success, -ERR# otherwise.
  */
-int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
+int RGWRados::copy_obj(RGWObjectCtx& src_obj_ctx,
+	       RGWObjectCtx& dest_obj_ctx,
                const ACLOwner& owner,
                const rgw_user& remote_user,
                req_info *info,
@@ -4670,7 +4673,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
   if (remote_src || !source_zone.empty()) {
     rgw_zone_set_entry source_trace_entry{source_zone.id, std::nullopt};
     const req_context rctx{dpp, y, nullptr};
-    return fetch_remote_obj(obj_ctx, remote_user, info, source_zone,
+    return fetch_remote_obj(dest_obj_ctx, remote_user, info, source_zone,
                dest_obj, src_obj, dest_bucket_info, &src_bucket_info,
                dest_placement, src_mtime, mtime, mod_ptr,
                unmod_ptr, high_precision_time,
@@ -4680,7 +4683,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
   }
 
   map<string, bufferlist> src_attrs;
-  RGWRados::Object src_op_target(this, src_bucket_info, obj_ctx, src_obj);
+  RGWRados::Object src_op_target(this, src_bucket_info, src_obj_ctx, src_obj);
   RGWRados::Object::Read read_op(&src_op_target);
 
   read_op.conds.mod_ptr = mod_ptr;
@@ -4741,7 +4744,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
   RGWObjManifest *amanifest = nullptr;
 
   constexpr bool follow_olh = true;
-  ret = get_obj_state(dpp, &obj_ctx, src_bucket_info, src_obj,
+  ret = get_obj_state(dpp, &src_obj_ctx, src_bucket_info, src_obj,
                       &astate, &amanifest, follow_olh, y);
   if (ret < 0) {
     return ret;
@@ -4818,7 +4821,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
 
   if (copy_data) { /* refcounting tail wouldn't work here, just copy the data */
     attrs.erase(RGW_ATTR_TAIL_TAG);
-    return copy_obj_data(obj_ctx, owner, dest_bucket_info, dest_placement, read_op, obj_size - 1, dest_obj,
+    return copy_obj_data(dest_obj_ctx, owner, dest_bucket_info, dest_placement, read_op, obj_size - 1, dest_obj,
                          mtime, real_time(), attrs, olh_epoch, delete_at, petag, dpp, y);
   }
 
@@ -4835,7 +4838,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
   RGWObjManifest *pmanifest; 
   ldpp_dout(dpp, 20) << "dest_obj=" << dest_obj << " src_obj=" << src_obj << " copy_itself=" << (int)copy_itself << dendl;
 
-  RGWRados::Object dest_op_target(this, dest_bucket_info, obj_ctx, dest_obj);
+  RGWRados::Object dest_op_target(this, dest_bucket_info, dest_obj_ctx, dest_obj);
   RGWRados::Object::Write write_op(&dest_op_target);
 
   string tag;

src/rgw/driver/rados/rgw_rados.h

@@ -1133,7 +1133,7 @@ class RGWRados
                std::string *ptag,
                std::string *petag, optional_yield y);
 
-  int fetch_remote_obj(RGWObjectCtx& obj_ctx,
+  int fetch_remote_obj(RGWObjectCtx& dest_obj_ctx,
                        const rgw_user& user_id,
                        req_info *info,
                        const rgw_zone_id& source_zone,
@@ -1180,7 +1180,8 @@ class RGWRados
    *                             are overwritten by values contained in attrs parameter.
    * Returns: 0 on success, -ERR# otherwise.
    */
-  int copy_obj(RGWObjectCtx& obj_ctx,
+  int copy_obj(RGWObjectCtx& src_obj_ctx,
+               RGWObjectCtx& dest_obj_ctx,
                const ACLOwner& owner, // owner of destination object
                const rgw_user& remote_user, // uid for fetch_remote_obj() auth
                req_info *info,

src/rgw/driver/rados/rgw_sal_rados.cc

@@ -2903,6 +2903,7 @@ int RadosObject::copy_object(const ACLOwner& owner,
 				optional_yield y)
 {
   return store->getRados()->copy_obj(*rados_ctx,
+				     *static_cast<RadosObject*>(dest_object)->rados_ctx,
 				     owner,
 				     remote_user,
 				     info,

src/tools/rados/rados.cc

@@ -59,6 +59,8 @@
 #include "RadosImport.h"
 
 #include "osd/ECUtil.h"
+#include "objclass/objclass.h"
+#include "cls/refcount/cls_refcount_ops.h"
 
 using namespace std::chrono_literals;
 using namespace librados;
@@ -2750,8 +2752,30 @@ static int rados_tool_common(const std::map < std::string, std::string > &opts,
     }
     else
       ret = 0;
-    string s(bl.c_str(), bl.length());
-    cout << s;
+
+    if (attr_name == "refcount")  {
+      obj_refcount oref;
+      auto p = bl.cbegin();
+      decode(oref, p);
+      for (auto itr = oref.refs.begin(); itr != oref.refs.end(); itr++) {
+	if (!itr->first.empty()) {
+	  cout << itr->first << "::" << itr->second << std::endl;
+	}
+	else {
+	  cout << "wildcard reference::" << itr->second << std::endl;
+	}
+      }
+      if (!oref.retired_refs.empty()) {
+	cout << "--------------------------------------" << std::endl;
+	for (const auto & ref : oref.retired_refs) {
+	  cout << "retired_refs::" << ref << std::endl;
+	}
+      }
+    }
+    else {
+      string s(bl.c_str(), bl.length());
+      cout << s << std::endl;
+    }
   } else if (strcmp(nargs[0], "rmxattr") == 0) {
     if (!pool_name || nargs.size() < (obj_name ? 2 : 3)) {
       usage(cerr);