LcoreMgr: add clean_pid_auto_check and clean_lcore support

./build/app/LcoreMgr --clean_pid_auto_check ./build/app/LcoreMgr --clean_lcore 30 Signed-off-by: Frank Du <[email protected]>
OpenVisualCloud · Nov 13, 2023 · a718d2d · a718d2d
1 parent f462f54
commit a718d2d
Show file tree

Hide file tree

Showing 6 changed files with 235 additions and 37 deletions.
diff --git a/app/tools/lcore_shmem_mgr.c b/app/tools/lcore_shmem_mgr.c
@@ -2,14 +2,94 @@
  * Copyright(c) 2023 Intel Corporation
  */
 
+#include <errno.h>
+#include <getopt.h>
 #include <mtl/mtl_lcore_shm_api.h>
+#include <stdlib.h>
 
 #include "log.h"
 
+enum lsm_args_cmd {
+  LSM_ARG_UNKNOWN = 0,
+  LSM_ARG_HELP = 0x100, /* start from end of ascii */
+  LSM_ARG_INFO,
+  LSM_ARG_CLEAN_PID_AUTO_CHECK,
+  LSM_ARG_CLEAN_LCORE,
+  LSM_ARG_MAX,
+};
+
+/*
+struct option {
+   const char *name;
+   int has_arg;
+   int *flag;
+   int val;
+};
+*/
+static struct option lsm_args_options[] = {
+    {"help", no_argument, 0, LSM_ARG_HELP},
+    {"info", no_argument, 0, LSM_ARG_INFO},
+    {"clean_pid_auto_check", no_argument, 0, LSM_ARG_CLEAN_PID_AUTO_CHECK},
+    {"clean_lcore", required_argument, 0, LSM_ARG_CLEAN_LCORE},
+    {0, 0, 0, 0},
+};
+
+static void lsm_print_help() {
+  printf("\n");
+  printf("##### Usage: #####\n\n");
+
+  printf("Params:\n");
+  printf(" --help: Print the help information\n");
+  printf(" --info: Print lcore shared manager detail info\n");
+  printf(" --info: Print lcore shared manager detail info\n");
+  printf(" --clean_pid_auto_check: Clean the dead entries if PID is not active\n");
+  printf(" --clean_lcore <lcore id>: Clean the entry by lcore ID\n");
+
+  printf("\n");
+}
+
 int main(int argc, char** argv) {
-  MTL_MAY_UNUSED(argc);
-  MTL_MAY_UNUSED(argv);
+  int cmd = -1, opt_idx = 0;
+  int ret;
+
+  while (1) {
+    cmd = getopt_long_only(argc, argv, "hv", lsm_args_options, &opt_idx);
+    if (cmd == -1) break;
+
+    switch (cmd) {
+      case LSM_ARG_INFO:
+        mtl_lcore_shm_print();
+        break;
+      case LSM_ARG_CLEAN_PID_AUTO_CHECK:
+        ret = mtl_lcore_shm_clean(MTL_LCORE_CLEAN_PID_AUTO_CHECK, NULL, 0);
+        if (ret > 0)
+          info("Total %d dead lcores detected and deleted\n", ret);
+        else if (ret == 0)
+          info("No dead lcores detected\n");
+        else
+          err("Fail %d to clean shm by auto PID check\n", ret);
+        break;
+      case LSM_ARG_CLEAN_LCORE:
+        int lcore = atoi(optarg);
+        if (lcore < 0) {
+          err("lcore %d is not valid\n", lcore);
+          return -EIO;
+        }
+        struct mtl_lcore_clean_pid_info pid;
+        pid.lcore = lcore;
+        ret = mtl_lcore_shm_clean(MTL_LCORE_CLEAN_LCORE, &pid, sizeof(pid));
+        if (ret >= 0)
+          info("Succ to delete lcore %d\n", lcore);
+        else
+          err("Fail %d to delete lcore %d\n", ret, lcore);
+        break;
+        break;
+      case LSM_ARG_HELP:
+      default:
+        lsm_print_help();
+        return -1;
+    }
+  }
 
-  mtl_lcore_shm_print();
   return 0;
 }
diff --git a/include/mtl_lcore_shm_api.h b/include/mtl_lcore_shm_api.h
@@ -18,6 +18,22 @@
 extern "C" {
 #endif
 
+struct mtl_lcore_clean_pid_info {
+  uint32_t lcore;
+};
+
+/** lcore clean action */
+enum mtl_lcore_clean_action {
+  /** auto, no args.
+   * Remove lcore usage if the PID is inactive under the same hostname and user.
+   */
+  MTL_LCORE_CLEAN_PID_AUTO_CHECK = 0,
+  /** clean as PID info, args to struct mtl_lcore_clean_pid_info */
+  MTL_LCORE_CLEAN_LCORE,
+  /** max value of this enum */
+  MTL_LCORE_CLEAN_MAX,
+};
+
 /**
  * Print out the legacy lcore manager(shared memory) status.
  *
@@ -27,6 +43,21 @@ extern "C" {
  */
 int mtl_lcore_shm_print(void);
 
+/**
+ * Clean the unused lcore from the legacy lcore manager(shared memory).
+ * @param action
+ *   The action type.
+ * @param args
+ *   The args to the action type.
+ * @param args_sz
+ *   The size of the args.
+ *
+ * @return
+ *   - 0 if successful.
+ *   - <0: Error code if fail.
+ */
+int mtl_lcore_shm_clean(enum mtl_lcore_clean_action action, void* args, size_t args_sz);
+
 #if defined(__cplusplus)
 }
 #endif

diff --git a/lib/src/mt_main.c b/lib/src/mt_main.c
@@ -4,10 +4,6 @@
 
 #include "mt_main.h"
 
-#ifndef WINDOWSENV
-#include <pwd.h>
-#endif
-
 #include "datapath/mt_queue.h"
 #include "dev/mt_dev.h"
 #include "mt_admin.h"
@@ -402,31 +398,6 @@ static int _mt_stop(struct mtl_main_impl* impl) {
   return 0;
 }
 
-static int mt_user_info_init(struct mtl_main_impl* impl) {
-  int ret = -EIO;
-  struct mt_user_info* info = &impl->u_info;
-
-#ifdef WINDOWSENV /* todo */
-  MTL_MAY_UNUSED(ret);
-  snprintf(info->hostname, sizeof(info->hostname), "%s", "unknow");
-  snprintf(info->user, sizeof(info->user), "%s", "unknow");
-#else
-  ret = gethostname(info->hostname, sizeof(info->hostname));
-  if (ret < 0) {
-    warn("%s, gethostname fail %d\n", __func__, ret);
-    snprintf(info->hostname, sizeof(info->hostname), "%s", "unknow");
-  }
-  uid_t uid = getuid();
-  struct passwd* user_info = getpwuid(uid);
-  snprintf(info->user, sizeof(info->user), "%s",
-           user_info ? user_info->pw_name : "unknow");
-#endif
-
-  info->pid = getpid();
-
-  return 0;
-}
-
 mtl_handle mtl_init(struct mtl_init_params* p) {
   struct mtl_main_impl* impl = NULL;
   int socket[MTL_PORT_MAX], ret;
@@ -493,7 +464,7 @@ mtl_handle mtl_init(struct mtl_init_params* p) {
   impl = mt_rte_zmalloc_socket(sizeof(*impl), socket[MTL_PORT_P]);
   if (!impl) goto err_exit;
 
-  mt_user_info_init(impl);
+  mt_user_info_init(&impl->u_info);
 
 #ifndef WINDOWSENV
   if (geteuid() == 0)

diff --git a/lib/src/mt_sch.c b/lib/src/mt_sch.c
@@ -4,8 +4,11 @@
 
 #include "mt_sch.h"
 
+#include <signal.h>
+
 #include "mt_log.h"
 #include "mt_stat.h"
+#include "mtl_lcore_shm_api.h"
 #include "st2110/st_rx_ancillary_session.h"
 #include "st2110/st_rx_audio_session.h"
 #include "st2110/st_rx_video_session.h"
@@ -1038,12 +1041,95 @@ int mtl_lcore_shm_print(void) {
 
   for (int i = 0; i < RTE_MAX_LCORE; i++) {
     shm_entry = &lcore_shm->lcores_info[i];
-    if (shm_entry->active) {
-      info("%s, lcore %d active on %s@%s with pid: %d\n", __func__, i, shm_entry->user,
-           shm_entry->hostname, (int)shm_entry->pid);
-    }
+
+    if (!shm_entry->active) continue;
+    info("%s, lcore %d active on %s@%s with pid: %d\n", __func__, i, shm_entry->user,
+         shm_entry->hostname, (int)shm_entry->pid);
   }
 
   sch_lcore_shm_uinit(&lcore_mgr);
   return 0;
 }
+
+static int lcore_shm_clean_auto_pid(struct mt_lcore_mgr* lcore_mgr) {
+  struct mt_user_info u_info;
+  int clean = 0;
+
+  memset(&u_info, 0, sizeof(u_info));
+  mt_user_info_init(&u_info);
+
+  struct mt_lcore_shm* lcore_shm = lcore_mgr->lcore_shm;
+  struct mt_lcore_shm_entry* shm_entry;
+  for (int i = 0; i < RTE_MAX_LCORE; i++) {
+    shm_entry = &lcore_shm->lcores_info[i];
+
+    if (!shm_entry->active) continue;
+    if (0 != strncmp(shm_entry->hostname, u_info.hostname, sizeof(shm_entry->hostname)))
+      continue;
+    if (0 != strncmp(shm_entry->user, u_info.user, sizeof(shm_entry->user))) continue;
+    /* now check if PID is active with zero signal */
+    int result = kill(shm_entry->pid, 0);
+    if (0 == result) continue;
+    clean++;
+    notice("%s, delete dead lcore %d from the shared mem, PID %d\n", __func__, i,
+           (int)shm_entry->pid);
+  }
+
+  return clean;
+}
+
+static int lcore_shm_clean_id(struct mt_lcore_mgr* lcore_mgr, void* args,
+                              size_t args_sz) {
+  struct mtl_lcore_clean_pid_info* info = args;
+
+  if (!args) {
+    err("%s, NULL args\n", __func__);
+    return -EINVAL;
+  }
+  if (args_sz != sizeof(*info)) {
+    err("%s, error args_sz %" PRIu64 "\n", __func__, args_sz);
+    return -EINVAL;
+  }
+  uint32_t lcore = info->lcore;
+  if (lcore >= RTE_MAX_LCORE) {
+    err("%s, invalid lcore %u\n", __func__, lcore);
+    return -EINVAL;
+  }
+
+  struct mt_lcore_shm* lcore_shm = lcore_mgr->lcore_shm;
+  struct mt_lcore_shm_entry* shm_entry = &lcore_shm->lcores_info[lcore];
+  if (!shm_entry->active) {
+    err("%s, lcore %u is inactive\n", __func__, lcore);
+    return -EINVAL;
+  }
+
+  shm_entry->active = false;
+  notice("%s, delete lcore %u from the shared mem, PID %d\n", __func__, lcore,
+         (int)shm_entry->pid);
+  return 0;
+}
+
+int mtl_lcore_shm_clean(enum mtl_lcore_clean_action action, void* args, size_t args_sz) {
+  struct mt_lcore_mgr lcore_mgr;
+
+  int ret = sch_lcore_shm_init(&lcore_mgr, false);
+  if (ret < 0) return ret;
+
+  MTL_MAY_UNUSED(args);
+  MTL_MAY_UNUSED(args_sz);
+  switch (action) {
+    case MTL_LCORE_CLEAN_PID_AUTO_CHECK:
+      ret = lcore_shm_clean_auto_pid(&lcore_mgr);
+      break;
+    case MTL_LCORE_CLEAN_LCORE:
+      ret = lcore_shm_clean_id(&lcore_mgr, args, args_sz);
+      break;
+    default:
+      err("%s, unknown action %d\n", __func__, action);
+      ret = -EINVAL;
+      break;
+  }
+
+  sch_lcore_shm_uinit(&lcore_mgr);
+  return ret;
+}
diff --git a/lib/src/mt_util.c b/lib/src/mt_util.c
@@ -4,6 +4,10 @@
 
 #include "mt_util.h"
 
+#ifndef WINDOWSENV
+#include <pwd.h>
+#endif
+
 #include "datapath/mt_queue.h"
 #include "mt_log.h"
 #include "mt_main.h"
@@ -879,3 +883,27 @@ const char* mt_native_afxdp_port2if(const char* port) {
   }
   return port + strlen(native_afxdp_port_prefix);
 }
+
+int mt_user_info_init(struct mt_user_info* info) {
+  int ret = -EIO;
+
+#ifdef WINDOWSENV /* todo */
+  MTL_MAY_UNUSED(ret);
+  snprintf(info->hostname, sizeof(info->hostname), "%s", "unknow");
+  snprintf(info->user, sizeof(info->user), "%s", "unknow");
+#else
+  ret = gethostname(info->hostname, sizeof(info->hostname));
+  if (ret < 0) {
+    warn("%s, gethostname fail %d\n", __func__, ret);
+    snprintf(info->hostname, sizeof(info->hostname), "%s", "unknow");
+  }
+  uid_t uid = getuid();
+  struct passwd* user_info = getpwuid(uid);
+  snprintf(info->user, sizeof(info->user), "%s",
+           user_info ? user_info->pw_name : "unknow");
+#endif
+
+  info->pid = getpid();
+
+  return 0;
+}
diff --git a/lib/src/mt_util.h b/lib/src/mt_util.h
@@ -251,4 +251,6 @@ const char* mt_dpdk_afpkt_port2if(const char* port);
 const char* mt_kernel_port2if(const char* port);
 const char* mt_native_afxdp_port2if(const char* port);
 
+int mt_user_info_init(struct mt_user_info* info);
+
 #endif