From 32dbf764fd5b7f6941c49750b49dbba253bd3234 Mon Sep 17 00:00:00 2001
From: zhongtao <[email protected]>
Date: Mon, 13 Feb 2023 15:36:58 +0800
Subject: [PATCH 16/53] isulad-shim support execSync with timeout

Signed-off-by: zhongtao <[email protected]>
---
 .../executor/container_cb/execution_stream.c  |  2 +-
 .../modules/runtime/isula/isula_rt_ops.c      | 37 +++++++++++++++---
 src/utils/cutils/utils.c                      | 39 +++++++++++++++++++
 src/utils/cutils/utils.h                      |  5 +++
 4 files changed, 77 insertions(+), 6 deletions(-)

diff --git a/src/daemon/executor/container_cb/execution_stream.c b/src/daemon/executor/container_cb/execution_stream.c
index fde0335e..1a7353b5 100644
--- a/src/daemon/executor/container_cb/execution_stream.c
+++ b/src/daemon/executor/container_cb/execution_stream.c
@@ -161,7 +161,7 @@ static int container_exec_cb(const container_exec_request *request, container_ex
 
     if (exec_container(cont, request, *response, stdinfd, stdout_handler, stderr_handler) != 0) {
         ret = -1;
-        goto out;
+        goto pack_err_response;
     }
 
     goto out;
diff --git a/src/daemon/modules/runtime/isula/isula_rt_ops.c b/src/daemon/modules/runtime/isula/isula_rt_ops.c
index 5463bb1b..6f2b4f7d 100644
--- a/src/daemon/modules/runtime/isula/isula_rt_ops.c
+++ b/src/daemon/modules/runtime/isula/isula_rt_ops.c
@@ -205,6 +205,10 @@ static void show_shim_runtime_errlog(const char *workdir)
     char buf1[SHIM_LOG_SIZE] = { 0 };
     char buf2[SHIM_LOG_SIZE] = { 0 };
 
+    if (g_isulad_errmsg != NULL) {
+        return;
+    }
+
     get_err_message(buf1, sizeof(buf1), workdir, "shim-log.json");
     get_err_message(buf2, sizeof(buf2), workdir, "log.json");
     ERROR("shim-log: %s", buf1);
@@ -686,8 +690,29 @@ static int status_to_exit_code(int status)
     return exit_code;
 }
 
+static int try_wait_pid(pid_t pid)
+{
+    if (waitpid(pid, NULL, WNOHANG) == pid) {
+        return 0;
+    }
+
+    return 1;
+}
+
+static void kill_and_show_err(pid_t pid)
+{
+    int nret = 0;
+    kill(pid, SIGKILL);
+    // wait atmost 0.5 seconds
+    DO_RETRY_CALL(5, 100000, nret, try_wait_pid, pid);
+    if (nret != 0) {
+        WARN("Fail to wait isulad-shim");
+    }
+    isulad_set_error_message("Exec container error;exec timeout");
+}
+
 static int shim_create(bool fg, const char *id, const char *workdir, const char *bundle, const char *runtime_cmd,
-                       int *exit_code)
+                       int *exit_code, const int64_t timeout)
 {
     pid_t pid = 0;
     int exec_fd[2] = { -1, -1 };
@@ -778,7 +803,7 @@ realexec:
         goto out;
     }
 
-    status = util_wait_for_pid_status(pid);
+    status = util_waitpid_with_timeout(pid, timeout, kill_and_show_err);
     if (status < 0) {
         ERROR("failed wait shim-parent %d exit %s", pid, strerror(errno));
         ret = -1;
@@ -792,7 +817,9 @@ realexec:
 out:
     if (ret != 0) {
         show_shim_runtime_errlog(workdir);
-        kill(pid, SIGKILL); /* can kill other process? */
+        if (timeout <= 0) {
+            kill(pid, SIGKILL); /* can kill other process? */
+        }
     }
 
     return ret;
@@ -901,7 +928,7 @@ int rt_isula_create(const char *id, const char *runtime, const rt_create_params_
     }
 
     get_runtime_cmd(runtime, &cmd);
-    ret = shim_create(false, id, workdir, params->bundle, cmd, NULL);
+    ret = shim_create(false, id, workdir, params->bundle, cmd, NULL, -1);
     if (ret != 0) {
         runtime_call_delete_force(workdir, runtime, id);
         ERROR("%s: failed create shim process", id);
@@ -1173,7 +1200,7 @@ int rt_isula_exec(const char *id, const char *runtime, const rt_exec_params_t *p
     }
 
     get_runtime_cmd(runtime, &cmd);
-    ret = shim_create(fg_exec(params), id, workdir, bundle, cmd, exit_code);
+    ret = shim_create(fg_exec(params), id, workdir, bundle, cmd, exit_code, params->timeout);
     if (ret != 0) {
         ERROR("%s: failed create shim process for exec %s", id, exec_id);
         goto errlog_out;
diff --git a/src/utils/cutils/utils.c b/src/utils/cutils/utils.c
index f99b28e4..2c4c01e4 100644
--- a/src/utils/cutils/utils.c
+++ b/src/utils/cutils/utils.c
@@ -311,6 +311,45 @@ rep:
     return 0;
 }
 
+/*
+ * If timeout <= 0, blocking wait pid.
+ * If timeout > 0, non-blocking wait pid with timeout.
+ * When waitpid timeout, calling handle_timeout_callback_t.
+ */
+int util_waitpid_with_timeout(pid_t pid, const int64_t timeout, handle_timeout_callback_t cb)
+{
+    int nret = 0;
+    time_t start_time = time(NULL);
+    time_t end_time;
+    double interval;
+
+    if (timeout <= 0) {
+        return util_wait_for_pid_status(pid);
+    }
+
+    for (;;) {
+        nret = waitpid(pid, NULL, WNOHANG);
+        if (nret == pid) {
+            break;
+        }
+        if (nret == -1 && errno != EINTR) {
+            return -1;
+        }
+        end_time = time(NULL);
+        interval = difftime(end_time, start_time);
+        if (nret == 0 && interval >= timeout) {
+            INFO("Wait %d timeout", pid);
+            if (cb != NULL) {
+                cb(pid);
+            }
+            return -1;
+        }
+        // sleep some time instead to avoid cpu full running and then retry.
+        sleep(0.1);
+    }
+    return 0;
+}
+
 int util_wait_for_pid_status(pid_t pid)
 {
     int st;
diff --git a/src/utils/cutils/utils.h b/src/utils/cutils/utils.h
index 6261dc05..01107605 100644
--- a/src/utils/cutils/utils.h
+++ b/src/utils/cutils/utils.h
@@ -302,6 +302,9 @@ typedef struct _proc_t {
         processor; /* current (or most recent?) CPU */
 } proc_t;
 
+// handle waitpid timeout.
+typedef void(*handle_timeout_callback_t)(pid_t pid);
+
 struct signame {
     int num;
     const char *name;
@@ -329,6 +332,8 @@ char *util_strdup_s(const char *src);
 
 int util_wait_for_pid(pid_t pid);
 
+int util_waitpid_with_timeout(pid_t pid, const int64_t timeout, handle_timeout_callback_t cb);
+
 void util_contain_errmsg(const char *errmsg, int *exit_code);
 
 char *util_short_digest(const char *digest);
-- 
2.25.1