Re: [PATCH v2 4/7] scsi: qla4xxx: replace deprecated strncpy with strscpy

[Kees Cook <keescook@chromium.org>]

On Wed, Feb 28, 2024 at 10:59:04PM +0000, Justin Stitt wrote:
> Replace 3 instances of strncpy in ql4_mbx.c
> 
> No bugs exist in the current implementation as some care was taken to
> ensure the write length was decreased by one to leave some space for a
> NUL-byte. However, instead of using strncpy(dest, src, LEN-1) we can opt
> for strscpy(dest, src, sizeof(dest)) which will result in
> NUL-termination as well. It should be noted that the entire chap_table
> is zero-allocated so the NUL-padding provided by strncpy is not needed.
> 
> While here, I noticed that MIN_CHAP_SECRET_LEN was not used anywhere.
> Since strscpy gives us the number of bytes copied into the destination
> buffer (or an -E2BIG) we can check both for an error during copying and
> also for a non-length compliant secret. Add a new jump label so we can
> properly clean up our chap_table should we have to abort due to bad
> secret.
> 
> The third instance in this file involves some more peculiar handling of
> strings:
> |	uint32_t mbox_cmd[MBOX_REG_COUNT];
> |	...
> |	memset(&mbox_cmd, 0, sizeof(mbox_cmd));
> |	...
> |	mbox_cmd[0] = MBOX_CMD_SET_PARAM;
> |	if (param == SET_DRVR_VERSION) {
> |		mbox_cmd[1] = SET_DRVR_VERSION;
> |		strncpy((char *)&mbox_cmd[2], QLA4XXX_DRIVER_VERSION,
> |			MAX_DRVR_VER_LEN - 1);
> 
> mbox_cmd has a size of 8:
> |	#define MBOX_REG_COUNT 8
> ... and its type width is 4 bytes. Hence, we have 32 bytes to work with
> here. The first 4 bytes are used as a flag for the MBOX_CMD_SET_PARAM.
> The next 4 bytes are used for SET_DRVR_VERSION. We now have 32-8=24
> bytes remaining -- which thankfully is what MAX_DRVR_VER_LEN is equal to
> |	#define MAX_DRVR_VER_LEN                    24
> 
> ... and the thing we're copying into this pseudo-string buffer is
> |	#define QLA4XXX_DRIVER_VERSION        "5.04.00-k6"
> 
> ... which is great because its less than 24 bytes (therefore we aren't
> truncating the source).
> 
> All to say, there's no bug in the existing implementation (yay!) but we
> can clean the code up a bit by using strscpy().
> 
> In ql4_os.c, there aren't any strncpy() uses to replace but there are
> some existing strscpy() calls that could be made more idiomatic. Where
> possible, use strscpy(dest, src, sizeof(dest)). Note that
> chap_rec->password has a size of ISCSI_CHAP_AUTH_SECRET_MAX_LEN
> |	#define ISCSI_CHAP_AUTH_SECRET_MAX_LEN	256
> ... while the current strscpy usage uses QL4_CHAP_MAX_SECRET_LEN
> |	#define QL4_CHAP_MAX_SECRET_LEN 100
> ... however since chap_table->secret was set and bounded properly in its
> string assignment its probably safe here to switch over to sizeof().
> 
> |	struct iscsi_chap_rec {
> 	...
> |		char username[ISCSI_CHAP_AUTH_NAME_MAX_LEN];
> |		uint8_t password[ISCSI_CHAP_AUTH_SECRET_MAX_LEN];
> 	...
> |	};
> 
> |	strscpy(chap_rec->password, chap_table->secret,
> |		QL4_CHAP_MAX_SECRET_LEN);
> 
> Signed-off-by: Justin Stitt <justinstitt@google.com>
> ---
>  drivers/scsi/qla4xxx/ql4_mbx.c | 17 ++++++++++++-----
>  drivers/scsi/qla4xxx/ql4_os.c  | 14 +++++++-------
>  2 files changed, 19 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/scsi/qla4xxx/ql4_mbx.c b/drivers/scsi/qla4xxx/ql4_mbx.c
> index 249f1d7021d4..75125d2021f5 100644
> --- a/drivers/scsi/qla4xxx/ql4_mbx.c
> +++ b/drivers/scsi/qla4xxx/ql4_mbx.c
> @@ -1641,6 +1641,7 @@ int qla4xxx_set_chap(struct scsi_qla_host *ha, char *username, char *password,
>  	struct ql4_chap_table *chap_table;
>  	uint32_t chap_size = 0;
>  	dma_addr_t chap_dma;
> +	ssize_t secret_len;
>  
>  	chap_table = dma_pool_zalloc(ha->chap_dma_pool, GFP_KERNEL, &chap_dma);
>  	if (chap_table == NULL) {
> @@ -1652,9 +1653,13 @@ int qla4xxx_set_chap(struct scsi_qla_host *ha, char *username, char *password,
>  		chap_table->flags |= BIT_6; /* peer */
>  	else
>  		chap_table->flags |= BIT_7; /* local */
> -	chap_table->secret_len = strlen(password);
> -	strncpy(chap_table->secret, password, MAX_CHAP_SECRET_LEN - 1);
> -	strncpy(chap_table->name, username, MAX_CHAP_NAME_LEN - 1);
> +
> +	secret_len = strscpy(chap_table->secret, password,
> +			     sizeof(chap_table->secret));
> +	if (secret_len < MIN_CHAP_SECRET_LEN)
> +		goto cleanup_chap_table;
> +	chap_table->secret_len = (uint8_t)secret_len;
> +	strscpy(chap_table->name, username, sizeof(chap_table->name));
>  	chap_table->cookie = cpu_to_le16(CHAP_VALID_COOKIE);
>  
>  	if (is_qla40XX(ha)) {
> @@ -1679,6 +1684,8 @@ int qla4xxx_set_chap(struct scsi_qla_host *ha, char *username, char *password,
>  		memcpy((struct ql4_chap_table *)ha->chap_list + idx,
>  		       chap_table, sizeof(struct ql4_chap_table));
>  	}
> +
> +cleanup_chap_table:
>  	dma_pool_free(ha->chap_dma_pool, chap_table, chap_dma);
>  	if (rval != QLA_SUCCESS)
>  		ret =  -EINVAL;
> @@ -2281,8 +2288,8 @@ int qla4_8xxx_set_param(struct scsi_qla_host *ha, int param)
>  	mbox_cmd[0] = MBOX_CMD_SET_PARAM;
>  	if (param == SET_DRVR_VERSION) {
>  		mbox_cmd[1] = SET_DRVR_VERSION;
> -		strncpy((char *)&mbox_cmd[2], QLA4XXX_DRIVER_VERSION,
> -			MAX_DRVR_VER_LEN - 1);
> +		strscpy((char *)&mbox_cmd[2], QLA4XXX_DRIVER_VERSION,
> +			MAX_DRVR_VER_LEN);
>  	} else {
>  		ql4_printk(KERN_ERR, ha, "%s: invalid parameter 0x%x\n",
>  			   __func__, param);
> diff --git a/drivers/scsi/qla4xxx/ql4_os.c b/drivers/scsi/qla4xxx/ql4_os.c
> index 675332e49a7b..17cccd14765f 100644
> --- a/drivers/scsi/qla4xxx/ql4_os.c
> +++ b/drivers/scsi/qla4xxx/ql4_os.c
> @@ -799,10 +799,10 @@ static int qla4xxx_get_chap_list(struct Scsi_Host *shost, uint16_t chap_tbl_idx,
>  
>  		chap_rec->chap_tbl_idx = i;
>  		strscpy(chap_rec->username, chap_table->name,
> -			ISCSI_CHAP_AUTH_NAME_MAX_LEN);
> -		strscpy(chap_rec->password, chap_table->secret,
> -			QL4_CHAP_MAX_SECRET_LEN);
> -		chap_rec->password_length = chap_table->secret_len;
> +			sizeof(chap_rec->username));
> +		chap_rec->password_length = strscpy(chap_rec->password,
> +						    chap_table->secret,
> +						    sizeof(chap_rec->password));

This hunk took me a bit to convince myself it's safe, but yes, I agree.
It all boils down to sizeof(chap_table->secret) being less than
sizeof(chap_rec->password), so there's no behavioral change here.

Reviewed-by: Kees Cook <keescook@chromium.org>

-- 
Kees Cook