iSulad/0184-do-check-process-alive-read-pid-ppid-info-in-runtime.patch

From 02b51963a37da893cc52a35562dd32f772e9e497 Mon Sep 17 00:00:00 2001
From: jikai <jikai11@huawei.com>
Date: Fri, 24 Nov 2023 17:36:50 +0800
Subject: [PATCH 184/198] do check process alive read pid ppid info in runtime

Signed-off-by: jikai <jikai11@huawei.com>
---
 src/daemon/modules/api/runtime_api.h          | 17 ++++++++
 .../container/container_gc/containers_gc.c    | 19 +++++++--
 .../modules/container/restore/restore.c       | 29 ++++++++++---
 .../modules/container/supervisor/supervisor.c | 17 +++++++-
 .../modules/runtime/engines/lcr/lcr_rt_ops.c  | 11 +++++
 .../modules/runtime/engines/lcr/lcr_rt_ops.h  |  4 ++
 .../modules/runtime/isula/isula_rt_ops.c      | 13 +++++-
 .../modules/runtime/isula/isula_rt_ops.h      |  3 ++
 src/daemon/modules/runtime/runtime.c          | 42 +++++++++++++++++++
 src/daemon/modules/runtime/shim/shim_rt_ops.c | 22 ++++++++++
 src/daemon/modules/runtime/shim/shim_rt_ops.h |  3 ++
 .../modules/service/service_container.c       |  1 +
 12 files changed, 169 insertions(+), 12 deletions(-)

diff --git a/src/daemon/modules/api/runtime_api.h b/src/daemon/modules/api/runtime_api.h
index 1f23efe3..b0d70493 100644
--- a/src/daemon/modules/api/runtime_api.h
+++ b/src/daemon/modules/api/runtime_api.h
@@ -208,6 +208,15 @@ typedef struct _rt_runtime_rebuild_config_params_t {
     const char *rootpath;
 } rt_rebuild_config_params_t;
 
+typedef struct _rt_runtime_read_pid_ppid_info_params_t {
+    int pid;
+} rt_read_pid_ppid_info_params_t;
+
+typedef struct _rt_runtime_detect_process_params_t {
+    int pid;
+    uint64_t start_time;
+} rt_detect_process_params_t;
+
 struct rt_ops {
     /* detect whether runtime is of this runtime type */
     bool (*detect)(const char *runtime);
@@ -245,6 +254,10 @@ struct rt_ops {
     int (*rt_resize)(const char *name, const char *runtime, const rt_resize_params_t *params);
     int (*rt_exec_resize)(const char *name, const char *runtime, const rt_exec_resize_params_t *params);
     int (*rt_rebuild_config)(const char *name, const char *runtime, const rt_rebuild_config_params_t *params);
+
+    int (*rt_read_pid_ppid_info)(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                                 pid_ppid_info_t *pid_info);
+    int (*rt_detect_process)(const char *name, const char *runtime, const rt_detect_process_params_t *params);
 };
 
 int runtime_create(const char *name, const char *runtime, const rt_create_params_t *params);
@@ -269,6 +282,10 @@ int runtime_rebuild_config(const char *name, const char *runtime, const rt_rebui
 void free_rt_listpids_out_t(rt_listpids_out_t *out);
 int runtime_resize(const char *name, const char *runtime, const rt_resize_params_t *params);
 int runtime_exec_resize(const char *name, const char *runtime, const rt_exec_resize_params_t *params);
+
+int runtime_read_pid_ppid_info(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                               pid_ppid_info_t *pid_info);
+int runtime_detect_process(const char *name, const char *runtime, const rt_detect_process_params_t *params);
 bool is_default_runtime(const char *name);
 
 int runtime_init(void);
diff --git a/src/daemon/modules/container/container_gc/containers_gc.c b/src/daemon/modules/container/container_gc/containers_gc.c
index 2d16dee8..3f2473d5 100644
--- a/src/daemon/modules/container/container_gc/containers_gc.c
+++ b/src/daemon/modules/container/container_gc/containers_gc.c
@@ -386,6 +386,7 @@ static void gc_monitor_process(const char *id, pid_t pid, unsigned long long sta
 {
     INFO("Received garbage collector monitor of %s with pid %d", id, pid);
 
+    // for shim-v2, here is a ppid, which is always 0
     if (util_process_alive(pid, start_time)) {
         int ret = kill(pid, SIGKILL);
         if (ret < 0 && errno != ESRCH) {
@@ -461,14 +462,16 @@ static void gc_container_process(struct linked_list *it)
     char *runtime = NULL;
     char *id = NULL;
     container_garbage_config_gc_containers_element *gc_cont = NULL;
+    rt_detect_process_params_t detect_params = {
+        .pid = gc_cont->pid,
+        .start_time = gc_cont->start_time,
+    };
 
     gc_cont = (container_garbage_config_gc_containers_element *)it->elem;
     id = gc_cont->id;
     runtime = gc_cont->runtime;
-    pid = gc_cont->pid;
-    start_time = gc_cont->start_time;
 
-    if (util_process_alive(pid, start_time) == false) {
+    if (runtime_detect_process(id, runtime, &detect_params) < 0) {
         ret = clean_container_resource(id, runtime, pid);
         if (ret != 0) {
             WARN("Failed to clean resources of container %s", id);
@@ -495,7 +498,15 @@ static void gc_container_process(struct linked_list *it)
         free(it);
     } else {
         try_to_resume_container(id, runtime);
-        ret = kill(pid, SIGKILL);
+
+        rt_kill_params_t kill_params = {
+            .signal = SIGKILL,
+            .stop_signal = SIGKILL,
+            .pid = pid,
+            .start_time = start_time,
+        };
+
+        ret = runtime_kill(id, runtime, &kill_params);
         if (ret < 0 && errno != ESRCH) {
             ERROR("Can not kill process (pid=%d) with SIGKILL for container %s", pid, id);
         }
diff --git a/src/daemon/modules/container/restore/restore.c b/src/daemon/modules/container/restore/restore.c
index 44ed14df..48a3ae4b 100644
--- a/src/daemon/modules/container/restore/restore.c
+++ b/src/daemon/modules/container/restore/restore.c
@@ -58,7 +58,12 @@ static int restore_supervisor(const container_t *cont)
     char *exit_fifo = NULL;
     char *id = cont->common_config->id;
     char *statepath = cont->state_path;
+    char *runtime = cont->runtime;
     pid_ppid_info_t pid_info = { 0 };
+    rt_detect_process_params_t params = {
+        .pid = cont->state->state->pid,
+        .start_time = cont->state->state->start_time,
+    };
 
     nret = snprintf(container_state, sizeof(container_state), "%s/%s", statepath, id);
     if (nret < 0 || (size_t)nret >= sizeof(container_state)) {
@@ -81,7 +86,7 @@ static int restore_supervisor(const container_t *cont)
         goto out;
     }
 
-    if (!util_process_alive(cont->state->state->pid, cont->state->state->start_time)) {
+    if (runtime_detect_process(id, runtime, &params) != 0) {
         ERROR("Container %s pid %d already dead, skip add supervisor", id, cont->state->state->pid);
         close(exit_fifo_fd);
         ret = -1;
@@ -112,8 +117,10 @@ static int post_stopped_container_to_gc(const char *id, const char *runtime, con
 {
     int ret = 0;
     pid_ppid_info_t pid_info = { 0 };
+    rt_read_pid_ppid_info_params_t params = { 0 };
+    params.pid = old_pid_info->pid;
 
-    (void)util_read_pid_ppid_info(old_pid_info->pid, &pid_info);
+    (void)runtime_read_pid_ppid_info(id, runtime, &params, &pid_info);
     if (pid_info.ppid == 0) {
         pid_info.ppid = old_pid_info->ppid;
         pid_info.pstart_time = old_pid_info->pstart_time;
@@ -180,9 +187,15 @@ static void restore_stopped_container(Container_Status status, const container_t
     pid_ppid_info_t pid_info = { 0 };
 
     if (status != CONTAINER_STATUS_STOPPED && status != CONTAINER_STATUS_CREATED) {
-        if (util_process_alive(cont->state->state->pid, cont->state->state->start_time)) {
+        rt_detect_process_params_t params = {
+            .pid = cont->state->state->pid,
+            .start_time = cont->state->state->start_time,
+        };
+        if (runtime_detect_process(id, cont->runtime, &params) == 0) {
             pid_info.pid = cont->state->state->pid;
+            pid_info.start_time = cont->state->state->start_time;
         }
+
         if (util_process_alive(cont->state->state->p_pid, cont->state->state->p_start_time)) {
             pid_info.ppid = cont->state->state->p_pid;
             pid_info.pstart_time = cont->state->state->p_start_time;
@@ -204,8 +217,11 @@ static void restore_running_container(Container_Status status, container_t *cont
     int nret = 0;
     const char *id = cont->common_config->id;
     pid_ppid_info_t pid_info = { 0 };
+    rt_read_pid_ppid_info_params_t params = {
+        .pid = info->pid,
+    };
 
-    nret = util_read_pid_ppid_info(info->pid, &pid_info);
+    nret = runtime_read_pid_ppid_info(id, cont->runtime, &params, &pid_info);
     if (nret == 0) {
         try_to_set_container_running(status, cont, &pid_info);
         container_state_reset_has_been_manual_stopped(cont->state);
@@ -234,10 +250,13 @@ static void restore_paused_container(Container_Status status, container_t *cont,
     int nret = 0;
     const char *id = cont->common_config->id;
     pid_ppid_info_t pid_info = { 0 };
+    rt_read_pid_ppid_info_params_t params = {
+        .pid = info->pid,
+    };
 
     container_state_set_paused(cont->state);
 
-    nret = util_read_pid_ppid_info(info->pid, &pid_info);
+    nret = runtime_read_pid_ppid_info(id, cont->runtime, &params, &pid_info);
     if (nret == 0) {
         try_to_set_paused_container_pid(status, cont, &pid_info);
         container_state_reset_has_been_manual_stopped(cont->state);
diff --git a/src/daemon/modules/container/supervisor/supervisor.c b/src/daemon/modules/container/supervisor/supervisor.c
index f77f58d7..b5ff6166 100644
--- a/src/daemon/modules/container/supervisor/supervisor.c
+++ b/src/daemon/modules/container/supervisor/supervisor.c
@@ -46,6 +46,7 @@
 #include "cgroup.h"
 #include "specs_api.h"
 #endif
+#include "runtime_api.h"
 
 pthread_mutex_t g_supervisor_lock = PTHREAD_MUTEX_INITIALIZER;
 struct epoll_descr g_supervisor_descr;
@@ -177,6 +178,17 @@ static void *clean_resources_thread(void *arg)
     char *runtime = data->runtime;
     unsigned long long start_time = data->pid_info.start_time;
     pid_t pid = data->pid_info.pid;
+    rt_detect_process_params_t detect_params = {
+        .pid = pid,
+        .start_time = start_time,
+    };
+    rt_kill_params_t kill_params = {
+        .signal = SIGKILL,
+        .stop_signal = SIGKILL,
+        .pid = pid,
+        .start_time = start_time
+    };
+
     int retry_count = 0;
     int max_retry = 10;
 #ifdef ENABLE_CRI_API_V1
@@ -192,7 +204,7 @@ static void *clean_resources_thread(void *arg)
     prctl(PR_SET_NAME, "Clean resource");
 
 retry:
-    if (false == util_process_alive(pid, start_time)) {
+    if (runtime_detect_process(name, runtime, &detect_params) < 0) {
         ret = clean_container_resource(name, runtime, pid);
         // clean_container_resource failed, do not log error message,
         // just add to gc to retry clean resource.
@@ -200,7 +212,7 @@ retry:
             ERROR("Failed to clean resources of container %s", name);
         }
     } else {
-        ret = kill(pid, SIGKILL);
+        ret = runtime_kill(name, runtime, &kill_params);
         if (ret < 0 && errno != ESRCH) {
             ERROR("Can not kill process (pid=%d) with SIGKILL for container %s", pid, name);
         }
@@ -212,6 +224,7 @@ retry:
         }
 
         // get info of init process in container for debug problem of container
+        // but for shim-v2, this might be a misleading debug info
         proc_t *c_proc = util_get_process_proc_info(pid);
         if (c_proc != NULL) {
             ERROR("Container %s into GC with process state: {cmd: %s, state: %c, pid: %d}", name, c_proc->cmd, c_proc->state,
diff --git a/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.c b/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.c
index 64a8adbc..eb3afb94 100644
--- a/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.c
+++ b/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.c
@@ -941,3 +941,14 @@ out:
     free_oci_runtime_spec(oci_spec);
     return ret;
 }
+
+int rt_lcr_read_pid_ppid_info(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                              pid_ppid_info_t *pid_info)
+{
+    return util_read_pid_ppid_info(params->pid, pid_info);
+}
+
+int rt_lcr_detect_process(const char *name, const char *runtime, const rt_detect_process_params_t *params)
+{
+    return util_process_alive(params->pid, params->start_time) ? 0 : -1;
+}
diff --git a/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.h b/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.h
index 7403544d..85ebe6f7 100644
--- a/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.h
+++ b/src/daemon/modules/runtime/engines/lcr/lcr_rt_ops.h
@@ -48,6 +48,10 @@ int rt_lcr_resize(const char *id, const char *runtime, const rt_resize_params_t
 int rt_lcr_exec_resize(const char *id, const char *runtime, const rt_exec_resize_params_t *params);
 int rt_lcr_kill(const char *id, const char *runtime, const rt_kill_params_t *params);
 int rt_lcr_rebuild_config(const char *name, const char *runtime, const rt_rebuild_config_params_t *params);
+
+int rt_lcr_read_pid_ppid_info(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                              pid_ppid_info_t *pid_info);
+int rt_lcr_detect_process(const char *name, const char *runtime, const rt_detect_process_params_t *params);
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/daemon/modules/runtime/isula/isula_rt_ops.c b/src/daemon/modules/runtime/isula/isula_rt_ops.c
index 0adb3858..35c09921 100644
--- a/src/daemon/modules/runtime/isula/isula_rt_ops.c
+++ b/src/daemon/modules/runtime/isula/isula_rt_ops.c
@@ -2173,4 +2173,15 @@ int rt_isula_kill(const char *id, const char *runtime, const rt_kill_params_t *p
 int rt_isula_rebuild_config(const char *name, const char *runtime, const rt_rebuild_config_params_t *params)
 {
     return 0;
-}
\ No newline at end of file
+}
+
+int rt_isula_read_pid_ppid_info(const char *id, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                                pid_ppid_info_t *pid_info)
+{
+    return util_read_pid_ppid_info(params->pid, pid_info);
+}
+
+int rt_isula_detect_process(const char *id, const char *runtime, const rt_detect_process_params_t *params)
+{
+    return util_process_alive(params->pid, params->start_time) ? 0 : -1;
+}
diff --git a/src/daemon/modules/runtime/isula/isula_rt_ops.h b/src/daemon/modules/runtime/isula/isula_rt_ops.h
index 1e5e049a..88236a1e 100644
--- a/src/daemon/modules/runtime/isula/isula_rt_ops.h
+++ b/src/daemon/modules/runtime/isula/isula_rt_ops.h
@@ -48,6 +48,9 @@ int rt_isula_exec_resize(const char *id, const char *runtime, const rt_exec_resi
 int rt_isula_kill(const char *id, const char *runtime, const rt_kill_params_t *params);
 int rt_isula_rebuild_config(const char *name, const char *runtime, const rt_rebuild_config_params_t *params);
 
+int rt_isula_read_pid_ppid_info(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                                pid_ppid_info_t *pid_info);
+int rt_isula_detect_process(const char *name, const char *runtime, const rt_detect_process_params_t *params);
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/daemon/modules/runtime/runtime.c b/src/daemon/modules/runtime/runtime.c
index 43b78bca..7be140dc 100644
--- a/src/daemon/modules/runtime/runtime.c
+++ b/src/daemon/modules/runtime/runtime.c
@@ -46,6 +46,8 @@ static const struct rt_ops g_lcr_rt_ops = {
     .rt_exec_resize = rt_lcr_exec_resize,
     .rt_kill = rt_lcr_kill,
     .rt_rebuild_config = rt_lcr_rebuild_config,
+    .rt_read_pid_ppid_info = rt_lcr_read_pid_ppid_info,
+    .rt_detect_process = rt_lcr_detect_process,
 };
 
 static const struct rt_ops g_isula_rt_ops = {
@@ -67,6 +69,8 @@ static const struct rt_ops g_isula_rt_ops = {
     .rt_exec_resize = rt_isula_exec_resize,
     .rt_kill = rt_isula_kill,
     .rt_rebuild_config = rt_isula_rebuild_config,
+    .rt_read_pid_ppid_info = rt_isula_read_pid_ppid_info,
+    .rt_detect_process = rt_isula_detect_process,
 };
 
 #ifdef ENABLE_SHIM_V2
@@ -89,6 +93,8 @@ static const struct rt_ops g_shim_rt_ops = {
     .rt_exec_resize = rt_shim_exec_resize,
     .rt_kill = rt_shim_kill,
     .rt_rebuild_config = rt_shim_rebuild_config,
+    .rt_read_pid_ppid_info = rt_shim_read_pid_ppid_info,
+    .rt_detect_process = rt_shim_detect_process,
 };
 #endif
 
@@ -534,6 +540,42 @@ out:
     return ret;
 }
 
+int runtime_read_pid_ppid_info(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                               pid_ppid_info_t *pid_info)
+{
+    const struct rt_ops *ops = NULL;
+
+    if (name == NULL || runtime == NULL || params == NULL) {
+        ERROR("Invalid arguments for runtime exec resize");
+        return -1;
+    }
+
+    ops = rt_ops_query(runtime);
+    if (ops == NULL) {
+        ERROR("Failed to get runtime ops");
+        return -1;
+    }
+
+    return ops->rt_read_pid_ppid_info(name, runtime, params, pid_info);
+}
+
+int runtime_detect_process(const char *name, const char *runtime, const rt_detect_process_params_t *params)
+{
+    const struct rt_ops *ops = NULL;
+
+    if (name == NULL || runtime == NULL || params == NULL) {
+        ERROR("Invalid arguments for runtime process alive");
+        return -1;
+    }
+
+    ops = rt_ops_query(runtime);
+    if (ops == NULL) {
+        return -1;
+    }
+
+    return ops->rt_detect_process(name, runtime, params);
+}
+
 bool is_default_runtime(const char *name)
 {
     const char *runtimes[] = { "lcr", "runc", "kata-runtime" };
diff --git a/src/daemon/modules/runtime/shim/shim_rt_ops.c b/src/daemon/modules/runtime/shim/shim_rt_ops.c
index 30b5e442..268d66d9 100644
--- a/src/daemon/modules/runtime/shim/shim_rt_ops.c
+++ b/src/daemon/modules/runtime/shim/shim_rt_ops.c
@@ -847,3 +847,25 @@ int rt_shim_rebuild_config(const char *name, const char *runtime, const rt_rebui
 {
     return 0;
 }
+
+int rt_shim_read_pid_ppid_info(const char *name, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                               pid_ppid_info_t *pid_info)
+{
+    if (pid_info == NULL) {
+        ERROR("Invalid input params");
+        return -1;
+    }
+
+    pid_info->pid = params->pid;
+    return 0;
+}
+
+int rt_shim_detect_process(const char *name, const char *runtime, const rt_detect_process_params_t *params)
+{
+    if (shim_v2_kill(name, NULL, 0, false) != 0) {
+        ERROR("%s: detect process failed", name);
+        return -1;
+    }
+
+    return 0;
+}
diff --git a/src/daemon/modules/runtime/shim/shim_rt_ops.h b/src/daemon/modules/runtime/shim/shim_rt_ops.h
index 2df34f4c..a3968cf5 100644
--- a/src/daemon/modules/runtime/shim/shim_rt_ops.h
+++ b/src/daemon/modules/runtime/shim/shim_rt_ops.h
@@ -64,6 +64,9 @@ bool is_valid_v2_runtime(const char* name);
 
 int rt_shim_rebuild_config(const char *name, const char *runtime, const rt_rebuild_config_params_t *params);
 
+int rt_shim_read_pid_ppid_info(const char *id, const char *runtime, const rt_read_pid_ppid_info_params_t *params,
+                               pid_ppid_info_t *pid_info);
+int rt_shim_detect_process(const char *id, const char *runtime, const rt_detect_process_params_t *params);
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/daemon/modules/service/service_container.c b/src/daemon/modules/service/service_container.c
index 250e8299..dbf56776 100644
--- a/src/daemon/modules/service/service_container.c
+++ b/src/daemon/modules/service/service_container.c
@@ -1513,6 +1513,7 @@ out:
 
 static int send_signal_to_process(pid_t pid, unsigned long long start_time, uint32_t stop_signal, uint32_t signal)
 {
+    // for shim-v2, here is a ppid, which is always 0
     if (util_process_alive(pid, start_time) == false) {
         if (signal == stop_signal || signal == SIGKILL) {
             WARN("Process %d is not alive", pid);
-- 
2.34.1