I previously attempted to apply a fix within the i2cdev_ioctl_smbus function. 
While this approach was successful in preventing the warning, I found that the 
required changes were quite extensive. The WARN is triggered by the 
trace_smbus_write tracepoint, which performs a memcpy(__entry->buf, 
data->block, len) for write operations on three specific block protocols: 
I2C_SMBUS_BLOCK_DATA, I2C_SMBUS_I2C_BLOCK_DATA, and I2C_SMBUS_BLOCK_PROC_CALL. 
To fix this in i2cdev_ioctl_smbus, it would be necessary to add checks for all 
three of these cases, which makes the solution rather complex.

--xiaomeng
-----邮件原件-----
发件人: Steven Rostedt <[email protected]> 
发送时间: 2025年9月3日 3:50
收件人: zhangxiaomeng (A) <[email protected]>
抄送: [email protected]; [email protected]; [email protected]; 
[email protected]; [email protected]
主题: Re: [PATCH] i2c: Fix OOB access in trace_event_raw_event_smbus_write

On Thu, 21 Aug 2025 01:23:12 +0000
Xiaomeng Zhang <[email protected]> wrote:

> The smbus_write tracepoint copies __entry->len bytes into a fixed 
> I2C_SMBUS_BLOCK_MAX + 2 buffer. Oversized lengths (e.g., 46) exceed 
> the destination and over-read the source buffer, triggering OOB 
> warning:
> 
> memcpy: detected field-spanning write (size 48) of single field 
> "entry->buf" at include/trace/events/smbus.h:60 (size 34)
> 
> Clamp the copy size to I2C_SMBUS_BLOCK_MAX + 2 before memcpy().
> This only affects tracing and does not change I2C transfer behavior.
> 
> Fixes: 8a325997d95d ("i2c: Add message transfer tracepoints for SMBUS 
> [ver #2]")
> Signed-off-by: Xiaomeng Zhang <[email protected]>
> ---
>  include/trace/events/smbus.h | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/include/trace/events/smbus.h 
> b/include/trace/events/smbus.h index 71a87edfc46d..e306d8b928c3 100644
> --- a/include/trace/events/smbus.h
> +++ b/include/trace/events/smbus.h
> @@ -57,6 +57,8 @@ TRACE_EVENT_CONDITION(smbus_write,
>               case I2C_SMBUS_I2C_BLOCK_DATA:
>                       __entry->len = data->block[0] + 1;
>               copy:
> +                     if (__entry->len > I2C_SMBUS_BLOCK_MAX + 2)
> +                             __entry->len = I2C_SMBUS_BLOCK_MAX + 2;
>                       memcpy(__entry->buf, data->block, __entry->len);
>                       break;
>               case I2C_SMBUS_QUICK:

The code has:

                switch (protocol) {
                case I2C_SMBUS_BYTE_DATA:
                        __entry->len = 1;
                        goto copy;
                case I2C_SMBUS_WORD_DATA:
                case I2C_SMBUS_PROC_CALL:
                        __entry->len = 2;
                        goto copy;
                case I2C_SMBUS_BLOCK_DATA:
                case I2C_SMBUS_BLOCK_PROC_CALL:
                case I2C_SMBUS_I2C_BLOCK_DATA:
                        __entry->len = data->block[0] + 1;
                copy:   
                        memcpy(__entry->buf, data->block, __entry->len);
                        break;
                case I2C_SMBUS_QUICK:
                case I2C_SMBUS_BYTE:
                case I2C_SMBUS_I2C_BLOCK_BROKEN:
                default:
                        __entry->len = 0;
                }

I only see two calls to the copy where one is len = 1 and the other is len = 2. 
Why not put the check before the copy label?

-- Steve

Reply via email to