osd: return ENOENT if pool information is invalid during tier-flush

During tier-flush, OSD sends reference increase message to target OSD.
At this point, sending message with invalid pool information (e.g., deleted pool)
causes unexpected behavior.

Therefore, this commit return ENOENT early before sending the message

fixes: https://tracker.ceph.com/issues/53294

Signed-off-by: Myoungwon Oh <myoungwon.oh@samsung.com>
(cherry picked from commit 3de27b25cb6104d810f7af81809a0813df6c46a4)

diff --git a/src/osd/PrimaryLogPG.cc b/src/osd/PrimaryLogPG.cc
index f7b15a5f8e1fdacc9a02c6447724030fa1da3642..c86f40bb392df19e066b8e78967eb08c3a6261d7 100644 (file)
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -7114,7 +7114,11 @@ int PrimaryLogPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops)
          goto fail;
        }
        pg_t raw_pg;
-       get_osdmap()->object_locator_to_pg(target_name, target_oloc, raw_pg);
+       result = get_osdmap()->object_locator_to_pg(target_name, target_oloc, raw_pg);
+       if (result < 0) {
+         dout(5) << " pool information is invalid: " << result << dendl;
+         break;
+       }
        hobject_t target(target_name, target_oloc.key, target_snapid,
                raw_pg.ps(), raw_pg.pool(),
                target_oloc.nspace);
@@ -7263,7 +7267,11 @@ int PrimaryLogPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops)
 
        pg_t raw_pg;
        chunk_info_t chunk_info;
-       get_osdmap()->object_locator_to_pg(tgt_name, tgt_oloc, raw_pg);
+       result = get_osdmap()->object_locator_to_pg(tgt_name, tgt_oloc, raw_pg);
+       if (result < 0) {
+         dout(5) << " pool information is invalid: " << result << dendl;
+         break;
+       }
        hobject_t target(tgt_name, tgt_oloc.key, snapid_t(),
                         raw_pg.ps(), raw_pg.pool(),
                         tgt_oloc.nspace);
@@ -10589,7 +10597,10 @@ int PrimaryLogPG::do_cdc(const object_info_t& oi,
   for (auto p : cdc_chunks) {
     bufferlist chunk;
     chunk.substr_of(bl, p.first, p.second);
-    hobject_t target = get_fpoid_from_chunk(oi.soid, chunk);
+    auto [ret, target] = get_fpoid_from_chunk(oi.soid, chunk);
+    if (ret < 0) {
+      return ret;
+    }
     chunks[p.first] = std::move(chunk);
     chunk_map[p.first] = chunk_info_t(0, p.second, target);
     total_length += p.second;
@@ -10597,11 +10608,12 @@ int PrimaryLogPG::do_cdc(const object_info_t& oi,
   return total_length;
 }
 
-hobject_t PrimaryLogPG::get_fpoid_from_chunk(const hobject_t soid, bufferlist& chunk)
+std::pair<int, hobject_t> PrimaryLogPG::get_fpoid_from_chunk(
+  const hobject_t soid, bufferlist& chunk)
 {
   pg_pool_t::fingerprint_t fp_algo = pool.info.get_fingerprint_type();
   if (fp_algo == pg_pool_t::TYPE_FINGERPRINT_NONE) {
-    return hobject_t();
+    return make_pair(-EINVAL, hobject_t());
   }
   object_t fp_oid = [&fp_algo, &chunk]() -> string {
     switch (fp_algo) {
@@ -10622,11 +10634,14 @@ hobject_t PrimaryLogPG::get_fpoid_from_chunk(const hobject_t soid, bufferlist& c
   oloc.pool = pool.info.get_dedup_tier();
   // check if dedup_tier isn't set
   ceph_assert(oloc.pool > 0);
-  get_osdmap()->object_locator_to_pg(fp_oid, oloc, raw_pg);
+  int ret = get_osdmap()->object_locator_to_pg(fp_oid, oloc, raw_pg);
+  if (ret < 0) {
+    return make_pair(ret, hobject_t());
+  }
   hobject_t target(fp_oid, oloc.key, snapid_t(),
                    raw_pg.ps(), raw_pg.pool(),
                    oloc.nspace);
-  return target;
+  return make_pair(0, target);
 }
 
 int PrimaryLogPG::finish_set_dedup(hobject_t oid, int r, ceph_tid_t tid, uint64_t offset)

diff --git a/src/osd/PrimaryLogPG.h b/src/osd/PrimaryLogPG.h
index b92c46bf4a9265d11af87de36426a5d84fa1b1ab..4a54e6ebe2cd7de884dedb1a6842675b692df270 100644 (file)
--- a/src/osd/PrimaryLogPG.h
+++ b/src/osd/PrimaryLogPG.h
@@ -1458,7 +1458,7 @@ protected:
   int do_cdc(const object_info_t& oi, std::map<uint64_t, chunk_info_t>& chunk_map,
             std::map<uint64_t, bufferlist>& chunks);
   int start_dedup(OpRequestRef op, ObjectContextRef obc);
-  hobject_t get_fpoid_from_chunk(const hobject_t soid, bufferlist& chunk);
+  std::pair<int, hobject_t> get_fpoid_from_chunk(const hobject_t soid, bufferlist& chunk);
   int finish_set_dedup(hobject_t oid, int r, ceph_tid_t tid, uint64_t offset);
   int finish_set_manifest_refcount(hobject_t oid, int r, ceph_tid_t tid, uint64_t offset);
 

diff --git a/src/test/librados/tier_cxx.cc b/src/test/librados/tier_cxx.cc
index f34accf8f1340612db5e5ca7c176356c5ae509e9..e8f345bd5adc93587c933cd67823fded55304ad8 100644 (file)
--- a/src/test/librados/tier_cxx.cc
+++ b/src/test/librados/tier_cxx.cc
@@ -5678,6 +5678,9 @@ TEST_F(LibRadosTwoPoolsPP, TierFlushDuringFlush) {
   // delete temp pool, so flushing chunk will fail
   ASSERT_EQ(0, s_cluster.pool_delete(temp_pool_name.c_str()));
 
+  // wait for maps to settle
+  cluster.wait_for_latest_osdmap();
+
   // flush to check if proper error is returned
   {
     ObjectReadOperation op;