When Network Frames Attack: Fixing a Heap Buffer Overflow in lwIP's Hosted Driver

Introduction

Embedded networking stacks sit at a fascinating — and dangerous — intersection: they must be lean enough to run on microcontrollers with kilobytes of RAM, yet robust enough to handle untrusted input arriving over the air from any device within wireless range. When a validation step is missing at that boundary, the consequences can be severe.

This post examines a critical heap buffer overflow found in drivers/lwip_hosted/lwip_hosted.c, the glue layer between the ESP-hosted wireless interface and the lwIP TCP/IP stack. The root cause is deceptively simple: a frame length value arriving from the network was never checked against the maximum size that the packet buffer allocator could safely handle. One crafted packet from a nearby attacker was enough to corrupt heap memory — and potentially achieve arbitrary code execution on the target device.

Who should read this? Embedded developers working with lwIP, ESP-IDF, or any network driver that bridges raw frames into a managed buffer system. The pattern is common; the lesson is universal.

The Vulnerability Explained

Architecture Background

The lwip_hosted driver acts as a network interface (netif) driver for lwIP when the wireless connectivity is provided by a co-processor running the ESP-hosted firmware. The data path looks like this:

Wireless co-processor
        │
        │  (raw frame + esp_hosted_frame_info_t metadata)
        ▼
lwip_hosted_rx_input()          ← vulnerability lives here
        │
        │  pbuf_alloc() + memcpy()
        ▼
lwIP network stack

When a frame arrives, lwip_hosted_rx_input is called with:
- info — a pointer to esp_hosted_frame_info_t containing frame metadata
- payload — a pointer to the raw frame bytes
- len — the reported length of the frame (a size_t, i.e., 32- or 64-bit)

The Dangerous Code Path

Before the fix, the function validated only the obvious error conditions:

// BEFORE — vulnerable code
if (!s_lwip.initialized || info == NULL || payload == NULL || len == 0U)
{
    return;
}

After passing these checks, the code called pbuf_alloc to allocate a packet buffer and then copied len bytes into it. Here is the critical problem:

pbuf_alloc accepts a u16_t (unsigned 16-bit integer) for the size parameter. If len is larger than 65535 (the maximum value of a u16_t), the value is silently truncated when passed to pbuf_alloc. The allocator happily creates a small buffer — say, a buffer for the truncated size — while the subsequent memcpy still uses the original, full-size len value.

The result: more bytes are written into the buffer than the buffer can hold, overflowing into adjacent heap memory.

Integer Truncation: A Visual Breakdown

Attacker sends frame with len = 0x00010010 (65,552 bytes)
                                    │
                                    ▼
pbuf_alloc((u16_t)len)  →  pbuf_alloc(0x0010)  →  allocates 16-byte buffer
                                    │
memcpy(pbuf->payload, frame, len)   │
       └──────────────────────────► copies 65,552 bytes into 16-byte buffer
                                    │
                                    ▼
                           ████ HEAP CORRUPTION ████

This is a classic CWE-122: Heap-based Buffer Overflow triggered via CWE-190: Integer Overflow or Wraparound (specifically, truncation on implicit narrowing conversion).

Real-World Exploitability

This vulnerability is remotely exploitable by any attacker on the same wireless network segment — no authentication required. An attacker needs only to:

1. Connect to (or be present on) the same Wi-Fi network or wireless segment as the target device.
2. Craft a raw network frame where the length field in esp_hosted_frame_info_t is set to a value greater than 65535.
3. Transmit the frame.

The heap corruption that follows can be leveraged for:

• Denial of Service — crashing the device by corrupting heap metadata or critical data structures.
• Information Disclosure — overwriting memory in ways that cause sensitive data to be returned or logged.
• Remote Code Execution — with sufficient control over heap layout, an attacker may overwrite function pointers or return addresses, redirecting execution flow.

On resource-constrained embedded targets, heap layout is often predictable, making exploitation more reliable than on desktop operating systems with ASLR and heap hardening.

The Fix

What Changed

The fix adds a single, targeted bounds check immediately after the existing input validation:

// AFTER — patched code
if (!s_lwip.initialized || info == NULL || payload == NULL || len == 0U ||
    len > UINT16_MAX) /* pbuf_alloc takes u16_t; prevent silent truncation */
{
    return;
}

Why This Works

By rejecting any len value that exceeds UINT16_MAX (65,535), the fix guarantees that the subsequent cast from size_t to u16_t is lossless. The allocated pbuf will always be large enough to hold exactly len bytes, eliminating the overflow condition entirely.

The fix is minimal and surgical — it does not change the happy path for valid frames, does not introduce new data structures, and adds negligible runtime overhead (a single integer comparison).

Before vs. After

The Comment Matters

Notice that the patch includes an inline comment:

len > UINT16_MAX) /* pbuf_alloc takes u16_t; prevent silent truncation */

This is not cosmetic. It explains why the check exists — a future maintainer who doesn't know the pbuf_alloc signature might otherwise consider this check redundant and remove it. Good security fixes document their reasoning.

Conclusion

This vulnerability is a textbook example of how a missing bounds check at a type boundary can turn routine network processing into a critical security flaw. The attack surface is wide — any device on the wireless network — and the potential impact reaches all the way to remote code execution.

The fix is elegant in its simplicity: one comparison, one comment, one early return. But finding it requires understanding the full data flow from the wireless co-processor through the driver into the lwIP allocator, and recognizing that size_t and u16_t are not interchangeable.

Key Takeaways

• 🔴 Silent integer truncation is dangerous — always validate before narrowing a type at security-sensitive boundaries.
• 🟡 Network input is untrusted input — apply the same rigor to incoming frames as you would to user-supplied form data.
• 🟢 One line can fix a critical vulnerability — but only if you understand why it's needed. Document your intent.
• 🔵 Enable -Wconversion in your builds — let the compiler be your first line of defense.

Embedded systems are increasingly connected, and increasingly targeted. Building the habit of validating every external input — especially at the network boundary — is one of the highest-leverage security investments a firmware team can make.

This vulnerability was identified and fixed by automated security scanning. Automated tools can catch entire classes of vulnerabilities before they ship — consider integrating security scanning into your CI/CD pipeline.