in_podman_metrics: fix multiple cgroup v2 issues

plugins/in_podman_metrics/podman_metrics.c

@@ -113,10 +113,15 @@ static int collect_container_data(struct flb_in_metrics *ctx)
                 metadata_token_start = strstr(metadata, JSON_SUBFIELD_IMAGE_NAME);
                 if (metadata_token_start) {
                     metadata_token_stop = strstr(metadata_token_start + JSON_SUBFIELD_SIZE_IMAGE_NAME+1, "\\\"");
-                    metadata_token_size = metadata_token_stop - metadata_token_start - JSON_SUBFIELD_SIZE_IMAGE_NAME;
-
-                    strncpy(image_name, metadata_token_start+JSON_SUBFIELD_SIZE_IMAGE_NAME, metadata_token_size);
-                    image_name[metadata_token_size] = '\0';
+                    if (metadata_token_stop) {
+                        metadata_token_size = metadata_token_stop - metadata_token_start - JSON_SUBFIELD_SIZE_IMAGE_NAME;
+                        strncpy(image_name, metadata_token_start+JSON_SUBFIELD_SIZE_IMAGE_NAME, metadata_token_size);
+                        image_name[metadata_token_size] = '\0';
+                    }
+                    else {
+                        strncpy(image_name, "unknown", IMAGE_NAME_SIZE - 1);
+                        image_name[sizeof("unknown") - 1] = '\0';
+                    }
 
                     flb_plg_trace(ctx->ins, "Found image name %s", image_name);
                     add_container_to_list(ctx, id, name, image_name);
@@ -229,10 +234,19 @@ static int create_counter(struct flb_in_metrics *ctx, struct cmt_counter **count
         return -1;
     }
 
-    if (strcmp(metric_name, COUNTER_CPU) == 0 || strcmp(metric_name, COUNTER_CPU_USER) == 0) {
-        fvalue = fvalue / 1000000000;
-        flb_plg_trace(ctx->ins, "Converting %s from nanoseconds to seconds (%lu -> %lu)", metric_name, value, fvalue);
-
+    if (strcmp(metric_name, COUNTER_CPU) == 0 ||
+        strcmp(metric_name, COUNTER_CPU_USER) == 0) {
+        if (ctx->cgroup_version == CGROUP_V2) {
+            /* cgroup v2 cpu.stat reports in microseconds */
+            fvalue = fvalue / 1000000;
+        }
+        else {
+            /* cgroup v1 cpuacct reports in nanoseconds */
+            fvalue = fvalue / 1000000000;
+        }
+        flb_plg_trace(ctx->ins,
+                      "Converting %s to seconds (%lu -> %lu)",
+                      metric_name, value, fvalue);
     }
 
     labels = (char *[]){id, name, image_name, interface};

plugins/in_podman_metrics/podman_metrics_config.h

@@ -83,6 +83,7 @@
 
 /* Key names in .stat files */
 #define STAT_KEY_RSS                    "rss"
+#define V2_STAT_KEY_RSS                 "anon"
 #define STAT_KEY_CPU                    "usage_usec"
 #define STAT_KEY_CPU_USER               "user_usec"
 
@@ -112,7 +113,8 @@
 #define V2_SYSFS_FILE_MEMORY_LIMIT     "memory.max"
 #define V2_SYSFS_FILE_CPU_STAT         "cpu.stat"
 #define V2_SYSFS_FILE_PIDS             "cgroup.procs"
-#define V2_SYSFS_FILE_PIDS_ALT         "containers/cgroup.procs"
+#define V2_SYSFS_FILE_PIDS_ALT         "container/cgroup.procs"
+#define V2_SYSFS_FILE_PIDS_ALT_LEGACY  "containers/cgroup.procs"
 #define V2_SYSFS_FILE_IO_STAT          "io.stat"
 
 /* Values used to construct counters/gauges names and descriptions */

plugins/in_podman_metrics/podman_metrics_data.c

@@ -409,6 +409,59 @@ int fill_counters_with_sysfs_data_v1(struct flb_in_metrics *ctx)
     return 0;
 }
 
+/*
+ * Read uint64_t value from sysfs file, with special handling for the "max"
+ * keyword used by cgroup v2 to indicate an unlimited resource.
+ * Returns 0 when file contains "max".
+ */
+static uint64_t read_from_sysfs_or_max(struct flb_in_metrics *ctx,
+                                        flb_sds_t dir,
+                                        flb_sds_t name)
+{
+    char path[SYSFS_FILE_PATH_SIZE];
+    char buf[32];
+    uint64_t value = UINT64_MAX;
+    FILE *fp;
+    int c;
+
+    if (dir == NULL) {
+        return value;
+    }
+
+    snprintf(path, sizeof(path), "%s/%s", dir, name);
+
+    fp = fopen(path, "r");
+    if (!fp) {
+        flb_plg_warn(ctx->ins, "Failed to read %s", path);
+        return value;
+    }
+
+    if (fgets(buf, sizeof(buf), fp) == NULL) {
+        flb_plg_warn(ctx->ins, "Failed to read a value from %s", path);
+        fclose(fp);
+        return UINT64_MAX;
+    }
+
+    /* cgroup v2 uses "max" to indicate unlimited */
+    if (strncmp(buf, "max", 3) == 0) {
+        flb_plg_debug(ctx->ins, "%s: max (unlimited)", path);
+        fclose(fp);
+        return 0;
+    }
+
+    c = sscanf(buf, "%lu", &value);
+    if (c != 1) {
+        flb_plg_warn(ctx->ins,
+                     "Failed to read a number from %s", path);
+        fclose(fp);
+        return UINT64_MAX;
+    }
+
+    fclose(fp);
+    flb_plg_debug(ctx->ins, "%s: %lu", path, value);
+    return value;
+}
+
 /*
  * Iterate over previously created container list. For each entry, generate its
  * path in sysfs system directory. From this path, grab data about container metrics
@@ -430,15 +483,22 @@ int fill_counters_with_sysfs_data_v2(struct flb_in_metrics *ctx)
 
         cnt->memory_usage = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_MEMORY, NULL);
         cnt->memory_max_usage = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_MAX_MEMORY, NULL);
-        cnt->rss = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_MEMORY_STAT, STAT_KEY_RSS);
-        cnt->memory_limit = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_MEMORY_LIMIT, NULL);
+        cnt->rss = get_data_from_sysfs(ctx, path,
+                                       V2_SYSFS_FILE_MEMORY_STAT,
+                                       V2_STAT_KEY_RSS);
+        cnt->memory_limit = read_from_sysfs_or_max(ctx, path,
+                                                    V2_SYSFS_FILE_MEMORY_LIMIT);
         cnt->cpu_user = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_CPU_STAT, STAT_KEY_CPU_USER);
         cnt->cpu = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_CPU_STAT, STAT_KEY_CPU);
         read_io_stat(ctx, path, V2_SYSFS_FILE_IO_STAT, cnt);
         pid = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_PIDS, NULL);
         if (!pid || pid == UINT64_MAX) {
             pid = get_data_from_sysfs(ctx, path, V2_SYSFS_FILE_PIDS_ALT, NULL);
         }
+        if (!pid || pid == UINT64_MAX) {
+            pid = get_data_from_sysfs(ctx, path,
+                                      V2_SYSFS_FILE_PIDS_ALT_LEGACY, NULL);
+        }
         if (pid && pid != UINT64_MAX) {
             get_net_data_from_proc(ctx, cnt, pid);
         }