[PATCH v2 1/1] hisi_sas: Fix possible buffer overflows in prep_ssp_v3_hw

[PATCH v2 1/1] hisi_sas: Fix possible buffer overflows in prep_ssp_v3_hw

[Xiaohui Zhang]

From: Zhang Xiaohui <ruc_zhangxiaohui@163.com>

prep_ssp_v3_hw() calls memcpy() without checking the destination
size may trigger a buffer overflower.

buf_cmd should be a ssp_tmf_iu struct through the analysis of
the command below:
buf_cmd = hisi_sas_cmd_hdr_addr_mem(slot) +
        sizeof(struct ssp_frame_hdr);

Then buf_cmd + 12 should point to tag, so the length parameter
of memcpy() should not exceed sizeof(__be16)+sizeof(u8)*14):
struct ssp_tmf_iu {
    u8     lun[8];
    u16    _r_a;
    u8     tmf;
    u8     _r_b;
    __be16 tag;
    u8     _r_c[14];
} __attribute__ ((packed));

Signed-off-by: Zhang Xiaohui <ruc_zhangxiaohui@163.com>
---
 drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
index 7133ca859..d02831c17 100644
--- a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
+++ b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
@@ -1267,7 +1267,9 @@ static void prep_ssp_v3_hw(struct hisi_hba *hisi_hba,
 	memcpy(buf_cmd, &task->ssp_task.LUN, 8);
 	if (!tmf) {
 		buf_cmd[9] = ssp_task->task_attr | (ssp_task->task_prio << 3);
-		memcpy(buf_cmd + 12, scsi_cmnd->cmnd, scsi_cmnd->cmd_len);
+		memcpy(buf_cmd + 12, scsi_cmnd->cmnd,
+		       min_t(unsigned short, scsi_cmnd->cmd_len,
+			     sizeof(__be16) + sizeof(u8) * 14));
 	} else {
 		buf_cmd[10] = tmf->tmf;
 		switch (tmf->tmf) {
-- 
2.17.1

Re: [PATCH v2 1/1] hisi_sas: Fix possible buffer overflows in prep_ssp_v3_hw

[John Garry]

On 09/12/2020 12:48, Xiaohui Zhang wrote:
> From: Zhang Xiaohui <ruc_zhangxiaohui@163.com>
> 
> prep_ssp_v3_hw() calls memcpy() without checking the destination
> size may trigger a buffer overflower.
> 
> buf_cmd should be a ssp_tmf_iu struct through the analysis of

hmmm... but you change the !tmf path.

> the command below:
> buf_cmd = hisi_sas_cmd_hdr_addr_mem(slot) +
>          sizeof(struct ssp_frame_hdr);
> 
> Then buf_cmd + 12 should point to tag, so the length parameter
> of memcpy() should not exceed sizeof(__be16)+sizeof(u8)*14):
> struct ssp_tmf_iu {
>      u8     lun[8];
>      u16    _r_a;
>      u8     tmf;
>      u8     _r_b;
>      __be16 tag;
>      u8     _r_c[14];
> } __attribute__ ((packed));
> 
> Signed-off-by: Zhang Xiaohui <ruc_zhangxiaohui@163.com>
> ---
>   drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
> index 7133ca859..d02831c17 100644
> --- a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
> +++ b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
> @@ -1267,7 +1267,9 @@ static void prep_ssp_v3_hw(struct hisi_hba *hisi_hba,
>   	memcpy(buf_cmd, &task->ssp_task.LUN, 8);
>   	if (!tmf) {
>   		buf_cmd[9] = ssp_task->task_attr | (ssp_task->task_prio << 3);
> -		memcpy(buf_cmd + 12, scsi_cmnd->cmnd, scsi_cmnd->cmd_len);
> +		memcpy(buf_cmd + 12, scsi_cmnd->cmnd,
> +		       min_t(unsigned short, scsi_cmnd->cmd_len,
> +			     sizeof(__be16) + sizeof(u8) * 14));

Again, this is not the right thing to do, and I don't think that this 
code needs fixing at all.

There should be a contract already that the driver is not sent cdb len > 
16, and this is sizeof(ssp_command_iu.cdb).

And, if we were sent too much data, then we would/should error.

Thanks,
John