From 5c431cdadf7d8c81afa46915b85f6a63aaff9871 Mon Sep 17 00:00:00 2001
From: Aleksandr Nogikh <nogikh@google.com>
Date: Tue, 27 Jan 2026 14:29:05 +0000
Subject: pkg/instance: extract crash dumps

If the crash dumps are enabled, collect one crash dump per each crash
when fuzzing locally.

Optionally also collect crash dumps in pkg/instance's Test().

Inspired by Chenxi Huang's downstream changes to syzkaller.
---
 pkg/instance/dump.go | 86 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 86 insertions(+)
 create mode 100644 pkg/instance/dump.go

(limited to 'pkg/instance')

diff --git a/pkg/instance/dump.go b/pkg/instance/dump.go
new file mode 100644
index 000000000..1544f3f93
--- /dev/null
+++ b/pkg/instance/dump.go
@@ -0,0 +1,86 @@
+// Copyright 2026 syzkaller project authors. All rights reserved.
+// Use of this source code is governed by Apache 2 LICENSE that can be found in the LICENSE file.
+
+package instance
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"time"
+
+	"github.com/google/syzkaller/pkg/log"
+	"github.com/google/syzkaller/sys/targets"
+	"github.com/google/syzkaller/vm"
+	"github.com/google/syzkaller/vm/vmimpl"
+)
+
+func ExtractMemoryDump(inst *vm.Instance, target *targets.Target, path string) error {
+	// TODO: if the instance has not yet panicked, we could cause the panic
+	// by writing to /proc/sys/kernel/sysrq. But the problem is that we don't
+	// want to enable CONFIG_MAGIC_SYSRQ during fuzzing.
+	const (
+		// To extract the crash dump, we need to SSH into the crash kernel.
+		// As it takes time for the new kernel to fully load and we don't
+		// know it exactly, we do several attempts.
+		maxRetries = 100
+		retrySleep = 3 * time.Second
+		// Using more restrictive masks somhow causes the crash utility to fail.
+		cmd = "/usr/sbin/makedumpfile -F -c -d 0 /proc/vmcore"
+	)
+	if target.OS != targets.Linux {
+		return fmt.Errorf("memory dump collection is only supported on linux")
+	}
+	var lastErr error
+	for i := 0; i < maxRetries; i++ {
+		err := extractKdumpInner(inst, path, cmd)
+		if err == nil {
+			return nil
+		}
+		lastErr = err
+		log.Logf(2, "[instance #%d] failed to extract memory dump: %v",
+			inst.Index(), err)
+		time.Sleep(retrySleep)
+	}
+	return fmt.Errorf("failed to extract memory dump after %v attempts: %w", maxRetries, lastErr)
+}
+
+func extractKdumpInner(inst *vm.Instance, path, cmd string) error {
+	// We need a long timeout for dump extraction, it can be hundreds of megabytes.
+	// 1 hour timeout should be enough for the typical scenarios.
+	ctx, cancel := context.WithTimeout(context.Background(), time.Hour)
+	defer cancel()
+
+	outc, errc, err := inst.RunStream(ctx, cmd)
+	if err != nil {
+		return fmt.Errorf("failed to start command: %w", err)
+	}
+
+	f, err := os.Create(path)
+	if err != nil {
+		return fmt.Errorf("failed to create dump file: %w", err)
+	}
+	defer f.Close()
+
+	for {
+		select {
+		case chunk, ok := <-outc:
+			if !ok {
+				outc = nil
+				continue
+			}
+			if chunk.Type != vmimpl.OutputStdout {
+				// Filter out console and stderr.
+				continue
+			}
+			if _, err := f.Write(chunk.Data); err != nil {
+				return fmt.Errorf("failed to write to dump file: %w", err)
+			}
+		case err := <-errc:
+			return err
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
+	return nil
+}
-- 
cgit mrf-deployment