DLPX-88427 Fix garbage stat names in estat metaslab-alloc BPF program

PR URL: https://www.github.com/delphix/performance-diagnostics/pull/121

Author: dbshah12

bpf/estat/metaslab-alloc.c

@@ -4,6 +4,15 @@
  * SPDX-License-Identifier: GPL-2.0-or-later
  */
 
+/*
+ * On some kernel versions (e.g. 6.14) linux/bpf.h is pulled in transitively
+ * via metaslab.h -> spa.h -> zfs_context.h -> vfs.h -> security.h and
+ * contains sizeof(struct bpf_wq) where bpf_wq is only forward-declared,
+ * causing a compile error (DLPX-96701).  Blocking security.h (which this BPF
+ * program does not need) stops the chain without touching linux/bpf.h itself.
+ */
+#define __LINUX_SECURITY_H
+
 #include <sys/metaslab.h>
 #include <sys/metaslab_impl.h>
 #include <sys/vdev_impl.h>
@@ -103,6 +112,19 @@ metaslab_group_alloc_exit(struct pt_regs *ctx)
 		axis = success;
 	}
 
+	/*
+	 * On some engine versions a kernel bug (DLPX-88427) causes vd_name to
+	 * contain raw memory bytes instead of a vdev path or type string.  The
+	 * first byte of any valid name is either '/' (device path) or an ASCII
+	 * letter (vdev type like "mirror").  If it falls outside printable ASCII
+	 * (0x20–0x7e) the name is garbage; overwrite it with a safe fallback so
+	 * we never emit unparseable metric names.
+	 */
+	if (data->vd_name[0] < 0x20 || data->vd_name[0] > 0x7e) {
+		char unknown[] = "unknown";
+		__builtin_memcpy(data->vd_name, unknown, sizeof(unknown));
+	}
+
 	AGGREGATE_DATA(data->vd_name, axis,
 		bpf_ktime_get_ns() - data->ts, data->asize);
 

telegraf/metaslab-alloc-stats.sh

@@ -91,10 +91,9 @@
   ]
   json_string_fields = ["iops(/s)", "avg latency(us)", "stddev(us)", "throughput(k/s)", "microseconds"]
 
-# Collect output from "estat metaslab-alloc -jm 10" via wrapper script.
-# The wrapper filters out metrics with garbage "name" tags (DLPX-88427).
+# Collect output from "estat metaslab-alloc -jm 10"
 [[inputs.execd]]
-  command = ["/etc/telegraf/metaslab-alloc-stats.sh"]
+  command = ["estat", "metaslab-alloc", "-jm", "10"]
   name_override = "estat_metaslab-alloc"
   signal = "none"
   restart_delay = "30s"