[Bug] legacy USB host trusts device-controlled bNumInterfaces and can trigger out-of-bounds access in fixed interface array
維護者通常 1 天內回覆
評估
- 難度
- 4/5
- 預估耗時
- 3-5 天
- 新手友好度
- 48/100
- Issue 類型
- 缺陷
- 描述清晰度
- 描述清楚
- 活躍度
- 冷清
- 技術堆疊
- c
研究方向
從 components/legacy/usb/usbhost/core/usbhost_core.c 中的 USB_MAX_INTERFACE 和 struct uinstance 開始,接著檢查所引用位置的 attach 和 detach 迴圈,以及 hub.c:452 中的 hub 路徑。確認 bNumInterfaces 如何在列舉和清理過程中傳遞。完成的標準是兩條路徑都能安全處理超過固定陣列容量的數量,包括部分 attach 失敗的情況。
由索引模型根據 Issue 內容生成。
描述
RT-Thread Version
master, verified on current worktree at commit 25295501c0cc7181d6a541a867fdf7214879ddf8
Hardware Type/Architectures
Any BSP enabling the legacy USB host stack and connecting to external USB devices
Develop Toolchain
GCC
Describe the bug
Summary
A memory-safety issue exists in RT-Thread's legacy USB host stack because it trusts the device-controlled bNumInterfaces field from the USB configuration descriptor, but stores interface state in a fixed-size internal array of only 8 entries.
USB_MAX_INTERFACE is fixed at 8, and struct uinstance contains only:
struct uhintf *intf[USB_MAX_INTERFACE];
However, the USB device's configuration descriptor field bNumInterfaces is attacker-controlled external input, and both attach and detach paths iterate directly up to device->cfg_desc->bNumInterfaces without comparing it against USB_MAX_INTERFACE.
This makes out-of-bounds access reachable from a malicious USB device.
Relevant Code
Internal capacity limit:
components/drivers/include/drivers/usb_host.h:22components/drivers/include/drivers/usb_host.h:66
#define USB_MAX_INTERFACE 0x08
struct uinstance
{
...
struct uhintf *intf[USB_MAX_INTERFACE];
...
};
Untrusted external field:
components/drivers/include/drivers/usb_common.h:294
rt_uint8_t bNumInterfaces;
Attach path:
components/legacy/usb/usbhost/core/usbhost_core.c:190components/legacy/usb/usbhost/core/usbhost_core.c:229
for (i = 0; i < device->cfg_desc->bNumInterfaces; i++)
{
...
device->intf[i] = (struct uhintf *)rt_malloc(sizeof(struct uhintf));
...
}
Detach path:
components/legacy/usb/usbhost/core/usbhost_core.c:276
for (i = 0; i < device->cfg_desc->bNumInterfaces; i++)
{
if (device->intf[i] == RT_NULL) continue;
...
}
Hub path that makes follow-on detach realistic even if attach fails partway:
components/legacy/usb/usbhost/core/hub.c:452
The attach return value is not handled in a way that prevents later detach-driven access from the same untrusted device lifecycle.
Correct Vulnerability Characterization
The issue is real, but the trigger conditions need to be stated precisely.
1. Out-of-bounds read is the easiest effect to reach
Because the detach path also trusts bNumInterfaces, a malicious USB device can cause the host to read past the 8-entry intf[] array during later disconnect / reconnect / cleanup flows.
2. Out-of-bounds write is also reachable
It is not enough to set only bNumInterfaces = 9. To actually reach the write at device->intf[i] = ... for i >= 8, the malicious USB device must also provide interface descriptors with indices beyond 7, and those descriptors must match an enabled host class driver so that the allocation path is taken.
If those conditions are met, device->intf[i] becomes a real out-of-bounds write.
3. Static storage makes corruption more dangerous
struct uinstance objects come from a static array:
components/legacy/usb/usbhost/core/usbhost_core.c:18
static struct uinstance dev[USB_MAX_DEVICE];
So writing beyond intf[7] is likely to corrupt adjacent fields of the same instance or neighboring device instances, not harmless padding.
Why This Is a Security Issue
This bug is reachable from an untrusted USB device during normal host-side enumeration.
A malicious USB peripheral can provide a crafted configuration descriptor with an oversized bNumInterfaces value and additional interface descriptors designed to drive the host into out-of-bounds access.
This is not just a robustness issue. It is a real memory-safety problem in the legacy USB host stack, with at least:
- Out-of-bounds read during detach / cleanup
- Out-of-bounds write during attach if enough matching interfaces are supplied
Depending on target layout and allocator behavior, the practical impact may include:
- Crash
- Denial of service
- Corruption of adjacent host USB state
- Undefined behavior during later device handling
Steps to Reproduce
A practical PoC should use a programmable USB device/emulator (for example a USB gadget setup, Facedancer-like platform, or custom firmware) and present a crafted configuration descriptor.
Minimal Out-of-Bounds Read PoC
- Enable RT-Thread legacy USB host support.
- Connect a malicious USB device.
- Provide a configuration descriptor with:
bNumInterfaces > 8- Enough malformed or unsupported interface layout so the attach path does not fully populate all
device->intf[]entries
- Trigger disconnect / reconnect or any path that causes cleanup / detach.
- Observe that the detach logic iterates up to the untrusted
bNumInterfacesvalue and readsdevice->intf[i]past the 8-entry array.
Stronger Out-of-Bounds Write PoC
- Provide a crafted configuration descriptor with:
bNumInterfaces > 8- Actual interface descriptors for indices
0..8or higher - Interface class/subclass values that match enabled host class drivers
- Let the host enumerate the device normally.
- When
i >= 8, the attach logic writes todevice->intf[i]even though the fixed array ends atintf[7].
Expected Behavior
RT-Thread should treat bNumInterfaces as untrusted input and reject or clamp any device that reports more interfaces than the internal host representation can store.
At minimum:
- Reject
bNumInterfaces > USB_MAX_INTERFACE - Avoid indexing
device->intf[i]unlessi < USB_MAX_INTERFACE - Apply the same validation consistently in both attach and detach paths
Actual Behavior
RT-Thread directly trusts bNumInterfaces from the device configuration descriptor and uses it to drive loops that index a fixed-size 8-entry internal array.
This allows a malicious USB device to cause out-of-bounds access in the legacy USB host stack.
Suggested Fix
The fix should be applied in all paths that use bNumInterfaces to index device->intf[], including both enumeration and detach/cleanup.
At minimum:
- Reject or clamp
device->cfg_desc->bNumInterfacesif it exceedsUSB_MAX_INTERFACE - Do not allocate or access
device->intf[i]unlessi < USB_MAX_INTERFACE - Ensure failure during attach cannot leave a later detach path iterating on the untrusted original interface count
Kindly let me know if you intend to request a CVE ID upon confirmation of the vulnerability.
Other additional context
No response
- 主要語言
- C
- 星號
- 12.3k
- 分支
- 5.5k
- 平均合併
- 3 天 8 小時
- 30 天內合併 PR
- 27
環境準備
在瀏覽器裡用你自己的 GitHub 帳號啟動這個專案的開發容器。
- 沒有 Dockerfile 或 Docker Compose 檔案
- 有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue,再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫,在一個分支上完成修改。
- 送出 Pull Request,並在描述裡引用這個 Issue 編號。
RT-Thread/rt-thread 的其他 Issue
-
[Bug] [netdev] ping crashes the shell with a division by zero when the target is unreachable (received == 0)可能已有人在做 @r3wretrhy 於 4 天前認領。 未關閉bug Component component: net
難度 1/5 1 小時以內 新手友好度 90/100
RT-Thread/rt-thread#11852 · 1 則留言 ·
維護者通常 1 天內回覆
-
[bsp][stm32][bluepill] README「快速上手」缺少重新生成 MDK 工程这一步,按文档操作无法编译通过可能已有人在做 @moment-NEW 於 6 天前認領。 未關閉in progress
難度 2/5 1-3 小時 新手友好度 75/100
RT-Thread/rt-thread#11818 · 4 則留言 · 已指派 1 人 ·
維護者通常 1 天內回覆
-
BSP BSP: Loongson bug RT-Smart
難度 2/5 1-3 小時 新手友好度 72/100
RT-Thread/rt-thread#11717 · 2 則留言 ·
維護者通常 1 天內回覆
-
Arch: RISC-V BSP BSP: HPMicro bug
難度 2/5 1-3 小時 新手友好度 72/100
RT-Thread/rt-thread#11687 · 3 則留言 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 68/100
RT-Thread/rt-thread#11472 · 1 則留言 ·
維護者通常 1 天內回覆
查看 RT-Thread/rt-thread 的全部 Issue
相似的 Issue
-
難度 2/5 1-3 小時 新手友好度 66/100
維護者通常 5 天內回覆
-
area:backend bug launch:p2 scope:core
難度 2/5 1-3 小時 新手友好度 62/100
Chelis-Lang/chelis#3494 ·
維護者通常 1 天內回覆
-
IO.get_env on Node truncates names at embedded NUL可能已有人在做 @Yi-111-a 今天認領。 未關閉
難度 2/5 1-3 小時 新手友好度 82/100
HigherOrderCO/Bend#1449 · 1 則留言 ·
-
難度 1/5 1 小時以內 新手友好度 88/100
FujiNetWIFI/fujinet-firmware#1872 ·
維護者通常 1 天內回覆
-
bug C/C++ code
難度 2/5 1-3 小時 新手友好度 70/100
webarkit/WebARKitLib#84 ·
維護者通常 1 天內回覆