SERVER-41943 Validate the given namespace for refineCollectionShardKey

3 files changed, 57 insertions, 5 deletions

diff --git a/src/mongo/db/s/config/configsvr_refine_collection_shard_key_command.cpp b/src/mongo/db/s/config/configsvr_refine_collection_shard_key_command.cpp
index e31af7aa0de..37b8d03b4e7 100644
--- a/src/mongo/db/s/config/configsvr_refine_collection_shard_key_command.cpp
+++ b/src/mongo/db/s/config/configsvr_refine_collection_shard_key_command.cpp
@@ -33,6 +33,8 @@
 
 #include "mongo/db/auth/authorization_session.h"
 #include "mongo/db/commands.h"
+#include "mongo/s/catalog/dist_lock_manager.h"
+#include "mongo/s/grid.h"
 #include "mongo/s/request_types/refine_collection_shard_key_gen.h"
 #include "mongo/util/log.h"
 
@@ -49,18 +51,50 @@ public:
         using InvocationBase::InvocationBase;
 
         void typedRun(OperationContext* opCtx) {
+            const NamespaceString& nss = ns();
+
             uassert(ErrorCodes::IllegalOperation,
                     "_configsvrRefineCollectionShardKey can only be run on config servers",
                     serverGlobalParams.clusterRole == ClusterRole::ConfigServer);
             uassert(ErrorCodes::InvalidOptions,
-                    str::stream()
-                        << "refineCollectionShardKey must be called with majority writeConcern",
+                    "refineCollectionShardKey must be called with majority writeConcern",
                     opCtx->getWriteConcern().wMode == WriteConcernOptions::kMajority);
 
             // Set the operation context read concern level to local for reads into the config
             // database.
             repl::ReadConcernArgs::get(opCtx) =
                 repl::ReadConcernArgs(repl::ReadConcernLevel::kLocalReadConcern);
+
+            const auto catalogClient = Grid::get(opCtx)->catalogClient();
+
+            // Acquire distlocks on the namespace's database and collection.
+            DistLockManager::ScopedDistLock dbDistLock(uassertStatusOK(
+                catalogClient->getDistLockManager()->lock(opCtx,
+                                                          nss.db(),
+                                                          "refineCollectionShardKey",
+                                                          DistLockManager::kDefaultLockTimeout)));
+            DistLockManager::ScopedDistLock collDistLock(uassertStatusOK(
+                catalogClient->getDistLockManager()->lock(opCtx,
+                                                          nss.ns(),
+                                                          "refineCollectionShardKey",
+                                                          DistLockManager::kDefaultLockTimeout)));
+
+            const auto collStatus =
+                catalogClient->getCollection(opCtx, nss, repl::ReadConcernLevel::kLocalReadConcern);
+
+            uassert(ErrorCodes::NamespaceNotSharded,
+                    str::stream() << "refineCollectionShardKey namespace " << nss.toString()
+                                  << " is not sharded",
+                    collStatus != ErrorCodes::NamespaceNotFound);
+
+            const auto collType = uassertStatusOK(collStatus).value;
+
+            uassert(ErrorCodes::StaleEpoch,
+                    str::stream()
+                        << "refineCollectionShardKey namespace "
+                        << nss.toString()
+                        << " has a different epoch than mongos had in its routing table cache",
+                    request().getEpoch() == collType.getEpoch());
         }
 
     private:
diff --git a/src/mongo/s/commands/cluster_refine_collection_shard_key_cmd.cpp b/src/mongo/s/commands/cluster_refine_collection_shard_key_cmd.cpp
index fa96a95b10f..23d8d1a3314 100644
--- a/src/mongo/s/commands/cluster_refine_collection_shard_key_cmd.cpp
+++ b/src/mongo/s/commands/cluster_refine_collection_shard_key_cmd.cpp
@@ -35,11 +35,14 @@
 #include "mongo/db/commands.h"
 #include "mongo/s/grid.h"
 #include "mongo/s/request_types/refine_collection_shard_key_gen.h"
+#include "mongo/util/fail_point_service.h"
 #include "mongo/util/log.h"
 
 namespace mongo {
 namespace {
 
+MONGO_FAIL_POINT_DEFINE(hangRefineCollectionShardKeyAfterRefresh);
+
 class RefineCollectionShardKeyCommand final : public TypedCommand<RefineCollectionShardKeyCommand> {
 public:
     using Request = RefineCollectionShardKey;
@@ -49,13 +52,23 @@ public:
         using InvocationBase::InvocationBase;
 
         void typedRun(OperationContext* opCtx) {
-            const NamespaceString nss = ns();
+            const NamespaceString& nss = ns();
+
+            const auto routingInfo = uassertStatusOK(
+                Grid::get(opCtx)->catalogCache()->getShardedCollectionRoutingInfoWithRefresh(opCtx,
+                                                                                             nss));
 
-            ConfigsvrRefineCollectionShardKey configsvrRefineCollShardKey(nss, request().getKey());
+            if (MONGO_FAIL_POINT(hangRefineCollectionShardKeyAfterRefresh)) {
+                log() << "Hit hangRefineCollectionShardKeyAfterRefresh failpoint";
+                MONGO_FAIL_POINT_PAUSE_WHILE_SET_OR_INTERRUPTED(
+                    opCtx, hangRefineCollectionShardKeyAfterRefresh);
+            }
+
+            ConfigsvrRefineCollectionShardKey configsvrRefineCollShardKey(
+                nss, request().getKey(), routingInfo.cm()->getVersion().epoch());
             configsvrRefineCollShardKey.setDbName(request().getDbName());
 
             auto configShard = Grid::get(opCtx)->shardRegistry()->getConfigShard();
-
             auto cmdResponse = uassertStatusOK(configShard->runCommandWithFixedRetryAttempts(
                 opCtx,
                 ReadPreferenceSetting(ReadPreference::PrimaryOnly),
diff --git a/src/mongo/s/request_types/refine_collection_shard_key.idl b/src/mongo/s/request_types/refine_collection_shard_key.idl
index bd6284b2709..5b03128620f 100644
--- a/src/mongo/s/request_types/refine_collection_shard_key.idl
+++ b/src/mongo/s/request_types/refine_collection_shard_key.idl
@@ -57,3 +57,8 @@ commands:
                 type: object
                 description: "The index specification document to use as the new shard key."
                 optional: false
+            epoch:
+                type: objectid
+                description: "The expected epoch of the namespace provided to
+                              refineCollectionShardKey."
+                optional: false