[PATCH] i3c: fix potential NULL dereference in send_ccc_cmd_locked

[PATCH] i3c: fix potential NULL dereference in send_ccc_cmd_locked

[Hongling Zeng]

The retry loop added in commit 02cc832191dd ("i3c: master: Validate GET
 CCC payload length and retry Direct GET once") resets payload actual_len
before each attempt. However, cmd->dests is only validated for DIRECT
CCC commands (I3C_CCC_DIRECT bit set). For broadcast commands, dests
can be NULL while ndests > 0, causing a NULL pointer dereference at
cmd->dests[i].payload.actual_len.

Fix by restricting the actual_len reset to DIRECT commands.

Fixes: 02cc832191dd ("i3c: master: Validate GET CCC payload length and retry Direct GET once")
Signed-off-by: Hongling Zeng <zenghongling@kylinos.cn>
---
 drivers/i3c/master.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/i3c/master.c b/drivers/i3c/master.c
index fd3e79d10c84..21212c34c31e 100644
--- a/drivers/i3c/master.c
+++ b/drivers/i3c/master.c
@@ -1051,7 +1051,7 @@ static int i3c_master_send_ccc_cmd_locked(struct i3c_master_controller *master,
 	for (attempt = 0; attempt < max_attempts; attempt++) {
 		unsigned int i;
 
-		if (cmd->rnw)
+		if (cmd->rnw && (cmd->id & I3C_CCC_DIRECT))
 			for (i = 0; i < cmd->ndests; i++)
 				cmd->dests[i].payload.actual_len = 0;
 
-- 
2.25.1

Re: [PATCH] i3c: fix potential NULL dereference in send_ccc_cmd_locked

[Mukesh Savaliya]

On 7/28/2026 12:00 PM, Hongling Zeng wrote:
> The retry loop added in commit 02cc832191dd ("i3c: master: Validate GET
>   CCC payload length and retry Direct GET once") resets payload actual_len
> before each attempt. However, cmd->dests is only validated for DIRECT
> CCC commands (I3C_CCC_DIRECT bit set). For broadcast commands, dests
> can be NULL while ndests > 0, causing a NULL pointer dereference at
> cmd->dests[i].payload.actual_len.
> 
> Fix by restricting the actual_len reset to DIRECT commands.
> 
> Fixes: 02cc832191dd ("i3c: master: Validate GET CCC payload length and retry Direct GET once")
> Signed-off-by: Hongling Zeng <zenghongling@kylinos.cn>
> ---
>   drivers/i3c/master.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/i3c/master.c b/drivers/i3c/master.c
> index fd3e79d10c84..21212c34c31e 100644
> --- a/drivers/i3c/master.c
> +++ b/drivers/i3c/master.c
> @@ -1051,7 +1051,7 @@ static int i3c_master_send_ccc_cmd_locked(struct i3c_master_controller *master,
>   	for (attempt = 0; attempt < max_attempts; attempt++) {
>   		unsigned int i;
>   
> -		if (cmd->rnw)
> +		if (cmd->rnw && (cmd->id & I3C_CCC_DIRECT))

Please review this recent fix @
https://lore.kernel.org/all/amJau9LgehYLnR20@lizhi-Precision-Tower-5810/

>   			for (i = 0; i < cmd->ndests; i++)
>   				cmd->dests[i].payload.actual_len = 0;
>   

Re: [PATCH] i3c: fix potential NULL dereference in send_ccc_cmd_locked

[Hongling Zeng]

在 2026年07月28日 18:27, Mukesh Savaliya 写道:
>
>
> On 7/28/2026 12:00 PM, Hongling Zeng wrote:
>> The retry loop added in commit 02cc832191dd ("i3c: master: Validate GET
>>   CCC payload length and retry Direct GET once") resets payload 
>> actual_len
>> before each attempt. However, cmd->dests is only validated for DIRECT
>> CCC commands (I3C_CCC_DIRECT bit set). For broadcast commands, dests
>> can be NULL while ndests > 0, causing a NULL pointer dereference at
>> cmd->dests[i].payload.actual_len.
>>
>> Fix by restricting the actual_len reset to DIRECT commands.
>>
>> Fixes: 02cc832191dd ("i3c: master: Validate GET CCC payload length 
>> and retry Direct GET once")
>> Signed-off-by: Hongling Zeng <zenghongling@kylinos.cn>
>> ---
>>   drivers/i3c/master.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/i3c/master.c b/drivers/i3c/master.c
>> index fd3e79d10c84..21212c34c31e 100644
>> --- a/drivers/i3c/master.c
>> +++ b/drivers/i3c/master.c
>> @@ -1051,7 +1051,7 @@ static int 
>> i3c_master_send_ccc_cmd_locked(struct i3c_master_controller *master,
>>       for (attempt = 0; attempt < max_attempts; attempt++) {
>>           unsigned int i;
>>   -        if (cmd->rnw)
>> +        if (cmd->rnw && (cmd->id & I3C_CCC_DIRECT))
>
> Please review this recent fix @
> https://lore.kernel.org/all/amJau9LgehYLnR20@lizhi-Precision-Tower-5810/
>
|

|Thanks, Adrian's fix makes more sense by validating the input at the
entry point rather than fixing only the retry loop use site.|

|
>>               for (i = 0; i < cmd->ndests; i++)
>>                   cmd->dests[i].payload.actual_len = 0;