rgw: add tail tag to track tail instance

Use tail_tag instead of idtag. The tail_tag is modified when either
creating a new tail, or adding a refcount to the tail (e.g., when
copying object). When just modifying the object's head, tail_tag
should remain the same. The gc will then use the tail_tag (if
exists, idtag if not) as the refcount id.

Fixes: http://tracker.ceph.com/issues/20234
Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
(cherry picked from commit f71a3e525d9a69fe537428f25930fb207a602afe)

diff --git a/src/rgw/rgw_common.h b/src/rgw/rgw_common.h
index 4fb476859e70b3baec11ce99fc87358247e63882..8d299cf39d475c4818491ec3c6c3fee0ef785d20 100644 (file)
--- a/src/rgw/rgw_common.h
+++ b/src/rgw/rgw_common.h
@@ -67,6 +67,7 @@ using ceph::crypto::MD5;
 #define RGW_ATTR_EXPIRES       RGW_ATTR_PREFIX "expires"
 #define RGW_ATTR_DELETE_AT     RGW_ATTR_PREFIX "delete_at"
 #define RGW_ATTR_ID_TAG        RGW_ATTR_PREFIX "idtag"
+#define RGW_ATTR_TAIL_TAG      RGW_ATTR_PREFIX "tail_tag"
 #define RGW_ATTR_SHADOW_OBJ            RGW_ATTR_PREFIX "shadow_name"
 #define RGW_ATTR_MANIFEST      RGW_ATTR_PREFIX "manifest"
 #define RGW_ATTR_USER_MANIFEST  RGW_ATTR_PREFIX "user_manifest"

diff --git a/src/rgw/rgw_op.cc b/src/rgw/rgw_op.cc
index c59c5376653234a84876bdc2978299aacb897aed..ac6f7b70081aece857f51238abbd19cf778f217e 100644 (file)
--- a/src/rgw/rgw_op.cc
+++ b/src/rgw/rgw_op.cc
@@ -3017,6 +3017,7 @@ int RGWPutObjProcessor_Multipart::do_complete(size_t accounted_size,
   head_obj_op.meta.owner = s->owner.get_id();
   head_obj_op.meta.delete_at = delete_at;
   head_obj_op.meta.zones_trace = zones_trace;
+  head_obj_op.meta.modify_tail = true;
 
   int r = head_obj_op.write_meta(obj_len, accounted_size, attrs);
   if (r < 0)
@@ -5450,6 +5451,7 @@ void RGWCompleteMultipart::execute()
   obj_op.meta.ptag = &s->req_id; /* use req_id as operation tag */
   obj_op.meta.owner = s->owner.get_id();
   obj_op.meta.flags = PUT_OBJ_CREATE;
+  obj_op.meta.modify_tail = true;
   op_ret = obj_op.write_meta(ofs, accounted_size, attrs);
   if (op_ret < 0)
     return;

diff --git a/src/rgw/rgw_rados.cc b/src/rgw/rgw_rados.cc
index 7a1fce857151b99077ef251e0afaa9ba02cfd1cd..7ca3184aec2756dd820d515f3db1233a9f027197 100644 (file)
--- a/src/rgw/rgw_rados.cc
+++ b/src/rgw/rgw_rados.cc
@@ -2780,6 +2780,7 @@ int RGWPutObjProcessor_Atomic::do_complete(size_t accounted_size, const string&
   obj_op.meta.delete_at = delete_at;
   obj_op.meta.user_data = user_data;
   obj_op.meta.zones_trace = zones_trace;
+  obj_op.meta.modify_tail = true;
 
   r = obj_op.write_meta(obj_len, accounted_size, attrs);
   if (r < 0) {
@@ -6823,7 +6824,8 @@ int RGWRados::swift_versioning_restore(RGWObjectCtx& obj_ctx,
  * Returns: 0 on success, -ERR# otherwise.
  */
 int RGWRados::Object::Write::_do_write_meta(uint64_t size, uint64_t accounted_size,
-                                           map<string, bufferlist>& attrs, bool assume_noent,
+                                           map<string, bufferlist>& attrs,
+                                           bool assume_noent, bool modify_tail,
                                            void *_index_op)
 {
   RGWRados::Bucket::UpdateIndex *index_op = static_cast<RGWRados::Bucket::UpdateIndex *>(_index_op);
@@ -6856,7 +6858,7 @@ int RGWRados::Object::Write::_do_write_meta(uint64_t size, uint64_t accounted_si
   if (!ptag && !index_op->get_optag()->empty()) {
     ptag = index_op->get_optag();
   }
-  r = target->prepare_atomic_modification(op, reset_obj, ptag, meta.if_match, meta.if_nomatch, false);
+  r = target->prepare_atomic_modification(op, reset_obj, ptag, meta.if_match, meta.if_nomatch, false, modify_tail);
   if (r < 0)
     return r;
 
@@ -7072,13 +7074,13 @@ int RGWRados::Object::Write::write_meta(uint64_t size, uint64_t accounted_size,
   bool assume_noent = (meta.if_match == NULL && meta.if_nomatch == NULL);
   int r;
   if (assume_noent) {
-    r = _do_write_meta(size, accounted_size, attrs, assume_noent, (void *)&index_op);
+    r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op);
     if (r == -EEXIST) {
       assume_noent = false;
     }
   }
   if (!assume_noent) {
-    r = _do_write_meta(size, accounted_size, attrs, assume_noent, (void *)&index_op);
+    r = _do_write_meta(size, accounted_size, attrs, assume_noent, meta.modify_tail, (void *)&index_op);
   }
   return r;
 }
@@ -7391,6 +7393,12 @@ static void set_copy_attrs(map<string, bufferlist>& src_attrs,
     if (!attrs[RGW_ATTR_ETAG].length()) {
       attrs[RGW_ATTR_ETAG] = src_attrs[RGW_ATTR_ETAG];
     }
+    if (!attrs[RGW_ATTR_TAIL_TAG].length()) {
+      auto ttiter = src_attrs.find(RGW_ATTR_TAIL_TAG);
+      if (ttiter != src_attrs.end()) {
+        attrs[RGW_ATTR_TAIL_TAG] = src_attrs[RGW_ATTR_TAIL_TAG];
+      }
+    }
     break;
   case RGWRados::ATTRSMOD_MERGE:
     for (map<string, bufferlist>::iterator it = src_attrs.begin(); it != src_attrs.end(); ++it) {
@@ -7422,6 +7430,7 @@ int RGWRados::rewrite_obj(RGWBucketInfo& dest_bucket_info, rgw_obj& obj)
     return ret;
 
   attrset.erase(RGW_ATTR_ID_TAG);
+  attrset.erase(RGW_ATTR_TAIL_TAG);
 
   uint64_t max_chunk_size;
 
@@ -8117,6 +8126,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
   }
 
   if (!copy_itself) {
+    attrs.erase(RGW_ATTR_TAIL_TAG);
     manifest = astate->manifest;
     const rgw_bucket_placement& tail_placement = manifest.get_tail_placement();
     if (tail_placement.bucket.name.empty()) {
@@ -8163,6 +8173,7 @@ int RGWRados::copy_obj(RGWObjectCtx& obj_ctx,
   write_op.meta.category = category;
   write_op.meta.olh_epoch = olh_epoch;
   write_op.meta.delete_at = delete_at;
+  write_op.meta.modify_tail = !copy_itself;
 
   ret = write_op.write_meta(obj_size, astate->accounted_size, attrs);
   if (ret < 0) {
@@ -8475,7 +8486,7 @@ int RGWRados::Object::complete_atomic_modification()
     return 0;
   }
 
-  string tag = state->obj_tag.to_str();
+  string tag = (state->tail_tag.length() > 0 ? state->tail_tag.to_str() : state->obj_tag.to_str());
   return store->gc->send_chain(chain, tag, false);  // do it async
 }
 
@@ -8692,13 +8703,17 @@ int RGWRados::defer_gc(void *ctx, const RGWBucketInfo& bucket_info, const rgw_ob
     return -EINVAL;
   }
 
-  if (state->obj_tag.length() == 0) {// check for backward compatibility
+  string tag;
+
+  if (state->tail_tag.length() > 0) {
+    tag = state->tail_tag.c_str();
+  } else if (state->obj_tag.length() > 0) {
+    tag = state->obj_tag.c_str();
+  } else {
     ldout(cct, 20) << "state->obj_tag is empty, not deferring gc operation" << dendl;
     return -EINVAL;
   }
 
-  string tag = state->obj_tag.c_str();
-
   ldout(cct, 0) << "defer chain tag=" << tag << dendl;
 
   return gc->defer_chain(tag, false);
@@ -8861,7 +8876,7 @@ int RGWRados::Object::Delete::delete_obj()
     return -ENOENT;
   }
 
-  r = target->prepare_atomic_modification(op, false, NULL, NULL, NULL, true);
+  r = target->prepare_atomic_modification(op, false, NULL, NULL, NULL, true, false);
   if (r < 0)
     return r;
 
@@ -9195,6 +9210,10 @@ int RGWRados::get_obj_state_impl(RGWObjectCtx *rctx, const RGWBucketInfo& bucket
     s->shadow_obj[bl.length()] = '\0';
   }
   s->obj_tag = s->attrset[RGW_ATTR_ID_TAG];
+  auto ttiter = s->attrset.find(RGW_ATTR_TAIL_TAG);
+  if (ttiter != s->attrset.end()) {
+    s->tail_tag = s->attrset[RGW_ATTR_TAIL_TAG];
+  }
 
   bufferlist manifest_bl = s->attrset[RGW_ATTR_MANIFEST];
   if (manifest_bl.length()) {
@@ -9467,7 +9486,8 @@ void RGWRados::SystemObject::invalidate_state()
 }
 
 int RGWRados::Object::prepare_atomic_modification(ObjectWriteOperation& op, bool reset_obj, const string *ptag,
-                                                  const char *if_match, const char *if_nomatch, bool removal_op)
+                                                  const char *if_match, const char *if_nomatch, bool removal_op,
+                                                  bool modify_tail)
 {
   int r = get_state(&state, false);
   if (r < 0)
@@ -9551,6 +9571,9 @@ int RGWRados::Object::prepare_atomic_modification(ObjectWriteOperation& op, bool
   ldout(store->ctx(), 10) << "setting object write_tag=" << state->write_tag << dendl;
 
   op.setxattr(RGW_ATTR_ID_TAG, bl);
+  if (modify_tail) {
+    op.setxattr(RGW_ATTR_TAIL_TAG, bl);
+  }
 
   return 0;
 }

diff --git a/src/rgw/rgw_rados.h b/src/rgw/rgw_rados.h
index 23caafd6314f48a58ba33ee624ec8ca836b58c65..b911601be19bbc6e309da42abb9bcceae6a44148 100644 (file)
--- a/src/rgw/rgw_rados.h
+++ b/src/rgw/rgw_rados.h
@@ -874,6 +874,7 @@ struct RGWObjState {
   ceph::real_time mtime;
   uint64_t epoch;
   bufferlist obj_tag;
+  bufferlist tail_tag;
   string write_tag;
   bool fake_tag;
   RGWObjManifest manifest;
@@ -908,6 +909,9 @@ struct RGWObjState {
     if (rhs.obj_tag.length()) {
       obj_tag = rhs.obj_tag;
     }
+    if (rhs.tail_tag.length()) {
+      tail_tag = rhs.tail_tag;
+    }
     write_tag = rhs.write_tag;
     fake_tag = rhs.fake_tag;
     if (rhs.has_manifest) {
@@ -2702,7 +2706,7 @@ public:
     void invalidate_state();
 
     int prepare_atomic_modification(librados::ObjectWriteOperation& op, bool reset_obj, const string *ptag,
-                                    const char *ifmatch, const char *ifnomatch, bool removal_op);
+                                    const char *ifmatch, const char *ifnomatch, bool removal_op, bool modify_tail);
     int complete_atomic_modification();
 
   public:
@@ -2798,17 +2802,19 @@ public:
         bool canceled;
         const string *user_data;
         rgw_zone_set *zones_trace;
+        bool modify_tail;
 
         MetaParams() : mtime(NULL), rmattrs(NULL), data(NULL), manifest(NULL), ptag(NULL),
                  remove_objs(NULL), category(RGW_OBJ_CATEGORY_MAIN), flags(0),
-                 if_match(NULL), if_nomatch(NULL), olh_epoch(0), canceled(false), user_data(nullptr), zones_trace(nullptr) {}
+                 if_match(NULL), if_nomatch(NULL), olh_epoch(0), canceled(false), user_data(nullptr), zones_trace(nullptr),
+                 modify_tail(false) {}
       } meta;
 
       explicit Write(RGWRados::Object *_target) : target(_target) {}
 
       int _do_write_meta(uint64_t size, uint64_t accounted_size,
                      map<std::string, bufferlist>& attrs,
-                     bool assume_noent,
+                     bool modify_tail, bool assume_noent,
                      void *index_op);
       int write_meta(uint64_t size, uint64_t accounted_size,
                      map<std::string, bufferlist>& attrs);