grpc 1.15.0 → 1.16.0.pre1

data/src/core/lib/iomgr/tcp_posix.h

@@ -31,7 +31,10 @@
 
 #include <grpc/support/port_platform.h>
 
+#include "src/core/lib/iomgr/port.h"
+
 #include "src/core/lib/debug/trace.h"
+#include "src/core/lib/iomgr/buffer_list.h"
 #include "src/core/lib/iomgr/endpoint.h"
 #include "src/core/lib/iomgr/ev_posix.h"
 

data/src/core/lib/iomgr/tcp_server_posix.cc

@@ -226,7 +226,7 @@ static void on_read(void* arg, grpc_error* err) {
       gpr_log(GPR_INFO, "SERVER_CONNECT: incoming connection: %s", addr_str);
     }
 
-    grpc_fd* fdobj = grpc_fd_create(fd, name, false);
+    grpc_fd* fdobj = grpc_fd_create(fd, name, true);
 
     read_notifier_pollset =
         sp->server->pollsets[static_cast<size_t>(gpr_atm_no_barrier_fetch_add(
@@ -362,7 +362,7 @@ static grpc_error* clone_port(grpc_tcp_listener* listener, unsigned count) {
     listener->sibling = sp;
     sp->server = listener->server;
     sp->fd = fd;
-    sp->emfd = grpc_fd_create(fd, name, false);
+    sp->emfd = grpc_fd_create(fd, name, true);
     memcpy(&sp->addr, &listener->addr, sizeof(grpc_resolved_address));
     sp->port = port;
     sp->port_index = listener->port_index;

data/src/core/lib/iomgr/tcp_server_utils_posix_common.cc

@@ -105,7 +105,7 @@ static grpc_error* add_socket_to_server(grpc_tcp_server* s, int fd,
     s->tail = sp;
     sp->server = s;
     sp->fd = fd;
-    sp->emfd = grpc_fd_create(fd, name, false);
+    sp->emfd = grpc_fd_create(fd, name, true);
     memcpy(&sp->addr, addr, sizeof(grpc_resolved_address));
     sp->port = port;
     sp->port_index = port_index;
@@ -166,6 +166,9 @@ grpc_error* grpc_tcp_server_prepare_socket(grpc_tcp_server* s, int fd,
     if (err != GRPC_ERROR_NONE) goto error;
     err = grpc_set_socket_reuse_addr(fd, 1);
     if (err != GRPC_ERROR_NONE) goto error;
+    err = grpc_set_socket_tcp_user_timeout(fd, s->channel_args,
+                                           false /* is_client */);
+    if (err != GRPC_ERROR_NONE) goto error;
   }
   err = grpc_set_socket_no_sigpipe_if_possible(fd);
   if (err != GRPC_ERROR_NONE) goto error;

data/src/core/lib/iomgr/tcp_windows.cc

@@ -296,7 +296,7 @@ static void on_write(void* tcpp, grpc_error* error) {
 
 /* Initiates a write. */
 static void win_write(grpc_endpoint* ep, grpc_slice_buffer* slices,
-                      grpc_closure* cb) {
+                      grpc_closure* cb, void* arg) {
   grpc_tcp* tcp = (grpc_tcp*)ep;
   grpc_winsocket* socket = tcp->socket;
   grpc_winsocket_callback_info* info = &socket->write_info;

data/src/core/lib/iomgr/timer_generic.cc

@@ -48,22 +48,22 @@ grpc_core::TraceFlag grpc_timer_trace(false, "timer");
 grpc_core::TraceFlag grpc_timer_check_trace(false, "timer_check");
 
 /* A "timer shard". Contains a 'heap' and a 'list' of timers. All timers with
- * deadlines earlier than 'queue_deadline" cap are maintained in the heap and
+ * deadlines earlier than 'queue_deadline_cap' are maintained in the heap and
  * others are maintained in the list (unordered). This helps to keep the number
  * of elements in the heap low.
  *
  * The 'queue_deadline_cap' gets recomputed periodically based on the timer
  * stats maintained in 'stats' and the relevant timers are then moved from the
- * 'list' to 'heap'
+ * 'list' to 'heap'.
  */
 typedef struct {
   gpr_mu mu;
   grpc_time_averaged_stats stats;
-  /* All and only timers with deadlines <= this will be in the heap. */
+  /* All and only timers with deadlines < this will be in the heap. */
   grpc_millis queue_deadline_cap;
-  /* The deadline of the next timer due in this shard */
+  /* The deadline of the next timer due in this shard. */
   grpc_millis min_deadline;
-  /* Index of this timer_shard in the g_shard_queue */
+  /* Index of this timer_shard in the g_shard_queue. */
   uint32_t shard_queue_index;
   /* This holds all timers with deadlines < queue_deadline_cap. Timers in this
      list have the top bit of their deadline set to 0. */
@@ -85,7 +85,7 @@ static timer_shard** g_shard_queue;
 
 #ifndef NDEBUG
 
-/* == Hash table for duplicate timer detection == */
+/* == DEBUG ONLY: hash table for duplicate timer detection == */
 
 #define NUM_HASH_BUCKETS 1009 /* Prime number close to 1000 */
 
@@ -177,7 +177,7 @@ static void remove_from_ht(grpc_timer* t) {
   t->hash_table_next = nullptr;
 }
 
-/* If a timer is added to a timer shard (either heap or a list), it cannot
+/* If a timer is added to a timer shard (either heap or a list), it must
  * be pending. A timer is added to hash table only-if it is added to the
  * timer shard.
  * Therefore, if timer->pending is false, it cannot be in hash table */
@@ -256,7 +256,7 @@ static grpc_millis compute_min_deadline(timer_shard* shard) {
 static void timer_list_init() {
   uint32_t i;
 
-  g_num_shards = GPR_MIN(1, 2 * gpr_cpu_num_cores());
+  g_num_shards = GPR_CLAMP(2 * gpr_cpu_num_cores(), 1, 32);
   g_shards =
       static_cast<timer_shard*>(gpr_zalloc(g_num_shards * sizeof(*g_shards)));
   g_shard_queue = static_cast<timer_shard**>(
@@ -291,7 +291,7 @@ static void timer_list_init() {
 static void timer_list_shutdown() {
   size_t i;
   run_some_expired_timers(
-      GPR_ATM_MAX, nullptr,
+      GRPC_MILLIS_INF_FUTURE, nullptr,
       GRPC_ERROR_CREATE_FROM_STATIC_STRING("Timer list shutdown"));
   for (i = 0; i < g_num_shards; i++) {
     timer_shard* shard = &g_shards[i];
@@ -489,7 +489,7 @@ static void timer_cancel(grpc_timer* timer) {
    'queue_deadline_cap') into into shard->heap.
    Returns 'true' if shard->heap has atleast ONE element
    REQUIRES: shard->mu locked */
-static int refill_heap(timer_shard* shard, grpc_millis now) {
+static bool refill_heap(timer_shard* shard, grpc_millis now) {
   /* Compute the new queue window width and bound by the limits: */
   double computed_deadline_delta =
       grpc_time_averaged_stats_update_average(&shard->stats) *
@@ -714,9 +714,10 @@ static grpc_timer_check_result timer_check(grpc_millis* next) {
 #if GPR_ARCH_64
     gpr_log(GPR_INFO,
             "TIMER CHECK BEGIN: now=%" PRId64 " next=%s tls_min=%" PRId64
-            " glob_min=%" PRIdPTR,
+            " glob_min=%" PRId64,
             now, next_str, min_timer,
-            gpr_atm_no_barrier_load((gpr_atm*)(&g_shared_mutables.min_timer)));
+            static_cast<grpc_millis>(gpr_atm_no_barrier_load(
+                (gpr_atm*)(&g_shared_mutables.min_timer))));
 #else
     gpr_log(GPR_INFO, "TIMER CHECK BEGIN: now=%" PRId64 " next=%s min=%" PRId64,
             now, next_str, min_timer);

data/src/core/lib/iomgr/timer_heap.cc

@@ -95,7 +95,7 @@ void grpc_timer_heap_init(grpc_timer_heap* heap) {
 
 void grpc_timer_heap_destroy(grpc_timer_heap* heap) { gpr_free(heap->timers); }
 
-int grpc_timer_heap_add(grpc_timer_heap* heap, grpc_timer* timer) {
+bool grpc_timer_heap_add(grpc_timer_heap* heap, grpc_timer* timer) {
   if (heap->timer_count == heap->timer_capacity) {
     heap->timer_capacity =
         GPR_MAX(heap->timer_capacity + 1, heap->timer_capacity * 3 / 2);
@@ -122,7 +122,7 @@ void grpc_timer_heap_remove(grpc_timer_heap* heap, grpc_timer* timer) {
   note_changed_priority(heap, heap->timers[i]);
 }
 
-int grpc_timer_heap_is_empty(grpc_timer_heap* heap) {
+bool grpc_timer_heap_is_empty(grpc_timer_heap* heap) {
   return heap->timer_count == 0;
 }
 

data/src/core/lib/iomgr/timer_heap.h

@@ -29,8 +29,8 @@ typedef struct {
   uint32_t timer_capacity;
 } grpc_timer_heap;
 
-/* return 1 if the new timer is the first timer in the heap */
-int grpc_timer_heap_add(grpc_timer_heap* heap, grpc_timer* timer);
+/* return true if the new timer is the first timer in the heap */
+bool grpc_timer_heap_add(grpc_timer_heap* heap, grpc_timer* timer);
 
 void grpc_timer_heap_init(grpc_timer_heap* heap);
 void grpc_timer_heap_destroy(grpc_timer_heap* heap);
@@ -39,6 +39,6 @@ void grpc_timer_heap_remove(grpc_timer_heap* heap, grpc_timer* timer);
 grpc_timer* grpc_timer_heap_top(grpc_timer_heap* heap);
 void grpc_timer_heap_pop(grpc_timer_heap* heap);
 
-int grpc_timer_heap_is_empty(grpc_timer_heap* heap);
+bool grpc_timer_heap_is_empty(grpc_timer_heap* heap);
 
 #endif /* GRPC_CORE_LIB_IOMGR_TIMER_HEAP_H */

data/src/core/lib/iomgr/timer_manager.cc

@@ -61,6 +61,14 @@ static uint64_t g_timed_waiter_generation;
 
 static void timer_thread(void* completed_thread_ptr);
 
+// For debug of the timer manager crash only.
+// TODO (mxyan): remove after bug is fixed.
+#ifdef GRPC_DEBUG_TIMER_MANAGER
+extern int64_t g_timer_manager_init_count;
+extern int64_t g_timer_manager_shutdown_count;
+extern int64_t g_fork_count;
+#endif  // GRPC_DEBUG_TIMER_MANAGER
+
 static void gc_completed_threads(void) {
   if (g_completed_threads != nullptr) {
     completed_thread* to_gc = g_completed_threads;
@@ -92,8 +100,7 @@ static void start_timer_thread_and_unlock(void) {
 
 void grpc_timer_manager_tick() {
   grpc_core::ExecCtx exec_ctx;
-  grpc_millis next = GRPC_MILLIS_INF_FUTURE;
-  grpc_timer_check(&next);
+  grpc_timer_check(nullptr);
 }
 
 static void run_some_timers() {
@@ -102,9 +109,12 @@ static void run_some_timers() {
   // remove a waiter from the pool, and start another thread if necessary
   --g_waiter_count;
   if (g_waiter_count == 0 && g_threaded) {
+    // The number of timer threads is always increasing until all the threads
+    // are stopped. In rare cases, if a large number of timers fire
+    // simultaneously, we may end up using a large number of threads.
     start_timer_thread_and_unlock();
   } else {
-    // if there's no thread waiting with a timeout, kick an existing
+    // if there's no thread waiting with a timeout, kick an existing untimed
     // waiter so that the next deadline is not missed
     if (!g_has_timed_waiter) {
       if (grpc_timer_check_trace.enabled()) {
@@ -284,6 +294,11 @@ static void start_threads(void) {
 void grpc_timer_manager_init(void) {
   gpr_mu_init(&g_mu);
   gpr_cv_init(&g_cv_wait);
+#ifdef GRPC_DEBUG_TIMER_MANAGER
+  // For debug of the timer manager crash only.
+  // TODO (mxyan): remove after bug is fixed.
+  g_timer_manager_init_count++;
+#endif
   gpr_cv_init(&g_cv_shutdown);
   g_threaded = false;
   g_thread_count = 0;
@@ -319,6 +334,11 @@ static void stop_threads(void) {
 }
 
 void grpc_timer_manager_shutdown(void) {
+#ifdef GRPC_DEBUG_TIMER_MANAGER
+  // For debug of the timer manager crash only.
+  // TODO (mxyan): remove after bug is fixed.
+  g_timer_manager_shutdown_count++;
+#endif
   stop_threads();
 
   gpr_mu_destroy(&g_mu);
@@ -327,6 +347,11 @@ void grpc_timer_manager_shutdown(void) {
 }
 
 void grpc_timer_manager_set_threading(bool threaded) {
+#ifdef GRPC_DEBUG_TIMER_MANAGER
+  // For debug of the timer manager crash only.
+  // TODO (mxyan): remove after bug is fixed.
+  g_fork_count++;
+#endif
   if (threaded) {
     start_threads();
   } else {

data/src/core/lib/iomgr/timer_manager.h

@@ -23,8 +23,8 @@
 
 #include <stdbool.h>
 
-/* Timer Manager tries to keep one thread waiting for the next timeout at all
-   times */
+/* Timer Manager tries to keep only one thread waiting for the next timeout at
+   all times, and thus effectively preventing the thundering herd problem. */
 
 void grpc_timer_manager_init(void);
 void grpc_timer_manager_shutdown(void);

data/src/core/lib/iomgr/udp_server.cc

@@ -152,7 +152,7 @@ GrpcUdpListener::GrpcUdpListener(grpc_udp_server* server, int fd,
   grpc_sockaddr_to_string(&addr_str, addr, 1);
   gpr_asprintf(&name, "udp-server-listener:%s", addr_str);
   gpr_free(addr_str);
-  emfd_ = grpc_fd_create(fd, name, false);
+  emfd_ = grpc_fd_create(fd, name, true);
   memcpy(&addr_, addr, sizeof(grpc_resolved_address));
   GPR_ASSERT(emfd_);
   gpr_free(name);

data/src/core/lib/security/credentials/alts/grpc_alts_credentials_client_options.cc

@@ -72,7 +72,8 @@ static void target_service_account_destroy(
 static const grpc_alts_credentials_options_vtable vtable = {
     alts_client_options_copy, alts_client_options_destroy};
 
-grpc_alts_credentials_options* grpc_alts_credentials_client_options_create() {
+grpc_alts_credentials_options* grpc_alts_credentials_client_options_create(
+    void) {
   auto client_options = static_cast<grpc_alts_credentials_client_options*>(
       gpr_zalloc(sizeof(grpc_alts_credentials_client_options)));
   client_options->base.vtable = &vtable;

data/src/core/lib/security/credentials/alts/grpc_alts_credentials_server_options.cc

@@ -36,7 +36,8 @@ static void alts_server_options_destroy(
 static const grpc_alts_credentials_options_vtable vtable = {
     alts_server_options_copy, alts_server_options_destroy};
 
-grpc_alts_credentials_options* grpc_alts_credentials_server_options_create() {
+grpc_alts_credentials_options* grpc_alts_credentials_server_options_create(
+    void) {
   grpc_alts_credentials_server_options* server_options =
       static_cast<grpc_alts_credentials_server_options*>(
           gpr_zalloc(sizeof(*server_options)));

data/src/core/lib/security/security_connector/security_connector.cc

@@ -59,8 +59,8 @@ static const char* installed_roots_path =
 
 /** Environment variable used as a flag to enable/disable loading system root
     certificates from the OS trust store. */
-#ifndef GRPC_USE_SYSTEM_SSL_ROOTS_ENV_VAR
-#define GRPC_USE_SYSTEM_SSL_ROOTS_ENV_VAR "GRPC_USE_SYSTEM_SSL_ROOTS"
+#ifndef GRPC_NOT_USE_SYSTEM_SSL_ROOTS_ENV_VAR
+#define GRPC_NOT_USE_SYSTEM_SSL_ROOTS_ENV_VAR "GRPC_NOT_USE_SYSTEM_SSL_ROOTS"
 #endif
 
 #ifndef TSI_OPENSSL_ALPN_SUPPORT
@@ -1192,10 +1192,10 @@ const char* DefaultSslRootStore::GetPemRootCerts() {
 
 grpc_slice DefaultSslRootStore::ComputePemRootCerts() {
   grpc_slice result = grpc_empty_slice();
-  char* use_system_roots_env_value =
-      gpr_getenv(GRPC_USE_SYSTEM_SSL_ROOTS_ENV_VAR);
-  const bool use_system_roots = gpr_is_true(use_system_roots_env_value);
-  gpr_free(use_system_roots_env_value);
+  char* not_use_system_roots_env_value =
+      gpr_getenv(GRPC_NOT_USE_SYSTEM_SSL_ROOTS_ENV_VAR);
+  const bool not_use_system_roots = gpr_is_true(not_use_system_roots_env_value);
+  gpr_free(not_use_system_roots_env_value);
   // First try to load the roots from the environment.
   char* default_root_certs_path =
       gpr_getenv(GRPC_DEFAULT_SSL_ROOTS_FILE_PATH_ENV_VAR);
@@ -1218,7 +1218,7 @@ grpc_slice DefaultSslRootStore::ComputePemRootCerts() {
     gpr_free(pem_root_certs);
   }
   // Try loading roots from OS trust store if flag is enabled.
-  if (GRPC_SLICE_IS_EMPTY(result) && use_system_roots) {
+  if (GRPC_SLICE_IS_EMPTY(result) && !not_use_system_roots) {
     result = LoadSystemRootCerts();
   }
   // Fallback to roots manually shipped with gRPC.

data/src/core/lib/security/transport/secure_endpoint.cc

@@ -254,7 +254,7 @@ static void flush_write_staging_buffer(secure_endpoint* ep, uint8_t** cur,
 }
 
 static void endpoint_write(grpc_endpoint* secure_ep, grpc_slice_buffer* slices,
-                           grpc_closure* cb) {
+                           grpc_closure* cb, void* arg) {
   GPR_TIMER_SCOPE("secure_endpoint.endpoint_write", 0);
 
   unsigned i;
@@ -342,7 +342,7 @@ static void endpoint_write(grpc_endpoint* secure_ep, grpc_slice_buffer* slices,
     return;
   }
 
-  grpc_endpoint_write(ep->wrapped_ep, &ep->output_buffer, cb);
+  grpc_endpoint_write(ep->wrapped_ep, &ep->output_buffer, cb, arg);
 }
 
 static void endpoint_shutdown(grpc_endpoint* secure_ep, grpc_error* why) {

data/src/core/lib/security/transport/security_handshaker.cc

@@ -259,7 +259,7 @@ static grpc_error* on_handshake_next_done_locked(
     grpc_slice_buffer_reset_and_unref_internal(&h->outgoing);
     grpc_slice_buffer_add(&h->outgoing, to_send);
     grpc_endpoint_write(h->args->endpoint, &h->outgoing,
-                        &h->on_handshake_data_sent_to_peer);
+                        &h->on_handshake_data_sent_to_peer, nullptr);
   } else if (handshaker_result == nullptr) {
     // There is nothing to send, but need to read from peer.
     grpc_endpoint_read(h->args->endpoint, h->args->read_buffer,

data/src/core/lib/security/transport/server_auth_filter.cc

@@ -41,6 +41,11 @@ struct call_data {
   grpc_transport_stream_op_batch* recv_initial_metadata_batch;
   grpc_closure* original_recv_initial_metadata_ready;
   grpc_closure recv_initial_metadata_ready;
+  grpc_error* recv_initial_metadata_error;
+  grpc_closure recv_trailing_metadata_ready;
+  grpc_closure* original_recv_trailing_metadata_ready;
+  grpc_error* recv_trailing_metadata_error;
+  bool seen_recv_trailing_metadata_ready;
   grpc_metadata_array md;
   const grpc_metadata* consumed_md;
   size_t num_consumed_md;
@@ -111,7 +116,16 @@ static void on_md_processing_done_inner(grpc_call_element* elem,
         batch->payload->recv_initial_metadata.recv_initial_metadata,
         remove_consumed_md, elem, "Response metadata filtering error");
   }
-  GRPC_CLOSURE_SCHED(calld->original_recv_initial_metadata_ready, error);
+  calld->recv_initial_metadata_error = GRPC_ERROR_REF(error);
+  grpc_closure* closure = calld->original_recv_initial_metadata_ready;
+  calld->original_recv_initial_metadata_ready = nullptr;
+  if (calld->seen_recv_trailing_metadata_ready) {
+    GRPC_CALL_COMBINER_START(calld->call_combiner,
+                             &calld->recv_trailing_metadata_ready,
+                             calld->recv_trailing_metadata_error,
+                             "continue recv_trailing_metadata_ready");
+  }
+  GRPC_CLOSURE_SCHED(closure, error);
 }
 
 // Called from application code.
@@ -180,8 +194,31 @@ static void recv_initial_metadata_ready(void* arg, grpc_error* error) {
       return;
     }
   }
-  GRPC_CLOSURE_RUN(calld->original_recv_initial_metadata_ready,
-                   GRPC_ERROR_REF(error));
+  grpc_closure* closure = calld->original_recv_initial_metadata_ready;
+  calld->original_recv_initial_metadata_ready = nullptr;
+  if (calld->seen_recv_trailing_metadata_ready) {
+    GRPC_CALL_COMBINER_START(calld->call_combiner,
+                             &calld->recv_trailing_metadata_ready,
+                             calld->recv_trailing_metadata_error,
+                             "continue recv_trailing_metadata_ready");
+  }
+  GRPC_CLOSURE_RUN(closure, GRPC_ERROR_REF(error));
+}
+
+static void recv_trailing_metadata_ready(void* user_data, grpc_error* err) {
+  grpc_call_element* elem = static_cast<grpc_call_element*>(user_data);
+  call_data* calld = static_cast<call_data*>(elem->call_data);
+  if (calld->original_recv_initial_metadata_ready != nullptr) {
+    calld->recv_trailing_metadata_error = GRPC_ERROR_REF(err);
+    calld->seen_recv_trailing_metadata_ready = true;
+    GRPC_CALL_COMBINER_STOP(calld->call_combiner,
+                            "deferring recv_trailing_metadata_ready until "
+                            "after recv_initial_metadata_ready");
+    return;
+  }
+  err = grpc_error_add_child(
+      GRPC_ERROR_REF(err), GRPC_ERROR_REF(calld->recv_initial_metadata_error));
+  GRPC_CLOSURE_RUN(calld->original_recv_trailing_metadata_ready, err);
 }
 
 static void auth_start_transport_stream_op_batch(
@@ -195,6 +232,12 @@ static void auth_start_transport_stream_op_batch(
     batch->payload->recv_initial_metadata.recv_initial_metadata_ready =
         &calld->recv_initial_metadata_ready;
   }
+  if (batch->recv_trailing_metadata) {
+    calld->original_recv_trailing_metadata_ready =
+        batch->payload->recv_trailing_metadata.recv_trailing_metadata_ready;
+    batch->payload->recv_trailing_metadata.recv_trailing_metadata_ready =
+        &calld->recv_trailing_metadata_ready;
+  }
   grpc_call_next_op(elem, batch);
 }
 
@@ -208,6 +251,9 @@ static grpc_error* init_call_elem(grpc_call_element* elem,
   GRPC_CLOSURE_INIT(&calld->recv_initial_metadata_ready,
                     recv_initial_metadata_ready, elem,
                     grpc_schedule_on_exec_ctx);
+  GRPC_CLOSURE_INIT(&calld->recv_trailing_metadata_ready,
+                    recv_trailing_metadata_ready, elem,
+                    grpc_schedule_on_exec_ctx);
   // Create server security context.  Set its auth context from channel
   // data and save it in the call context.
   grpc_server_security_context* server_ctx =
@@ -227,7 +273,10 @@ static grpc_error* init_call_elem(grpc_call_element* elem,
 /* Destructor for call_data */
 static void destroy_call_elem(grpc_call_element* elem,
                               const grpc_call_final_info* final_info,
-                              grpc_closure* ignored) {}
+                              grpc_closure* ignored) {
+  call_data* calld = static_cast<call_data*>(elem->call_data);
+  GRPC_ERROR_UNREF(calld->recv_initial_metadata_error);
+}
 
 /* Constructor for channel_data */
 static grpc_error* init_channel_elem(grpc_channel_element* elem,