Skip to contentSkip to conversation

Improve zcrypt reply message verification checks

PATCH Superseded Harald Freudenberger 3 participants 5 messages Jul 10 – Aug 4

Subsystems (2)

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

Harald Freudenberger

author maintainer
#0 lore raw

v5 cover Improve zcrypt reply message verification checks

Add or improve checks related to buffer sizes and reply sizes to the handling of replies from the crypto cards for CCA and EP11 (AP message type 6) messages. The verification code related to reply field length was not designed well and thus firmware deficiencies could lead to unexpected behavior in the zcrypt device driver. Thus improve the code to more closely inspect especially length fields at message replies.

The 3 parts of this patch deal with CCA, EP11 and (CCA) RNG replies and improve the checking for reply buffer size by using size_t instead of int. RNG replies an additional check makes sure the hard coded limit of the data buffer is not exceeded. Also there was a condition with additional data for an CCA reply where some of the field values where unchecked used to invoke memcpy into user
space. zcrypt_msgtype6_receive() now checks all the relevant fields before convert_type86_xcrb() uses them.

Improve zcrypt reply message verification checks

Add or improve checks related to buffer sizes and reply sizes to the handling of replies from the crypto cards for CCA and EP11 (AP message type 6) messages. The verification code related to reply field length was not designed well and thus firmware deficiencies could lead to unexpected behavior in the zcrypt device driver. Thus improve the code to more closely inspect especially length fields at message replies.

Changelog:
v1 - initial patch
v2 - fixed typo in header check_for_overflow -> check_add_overflow. v3 - rephrased and smoothed subject and text of the patch. It is now
"s390/zcrypt: Improve zcrypt reply message verification checks" and the text does not talk about malicious cards any more. Updated Reviewed-by tags
v4 - As sashiko clearly states the addition of two 32 bit values can
mathematically never overflow a 64 bit value and thus the check_add_overflow() was total overkill - removed.
v5 - Sashiko had a by-catch related to the very same fields. Under
some circumstances the fields count1 and offset1 of the CPRB where not checked but used for a memcpy to userspace. So again a rework of the check of these x86 header fields in the receiving function before the CPRB is processed to be copied in parts to userspace. Removed all reviewed-by as I want to have another developer look onto this patch again.

Harald Freudenberger (1):
s390/zcrypt: Improve zcrypt reply message verification checks

 drivers/s390/crypto/zcrypt_msgtype6.c | 42 ++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 8 deletions(-)

--
2.43.0

author maintainer
#1 lore raw

PATCH v5 1/1 s390/zcrypt: Improve zcrypt reply message verification checks

Add or improve checks related to buffer sizes and reply sizes to the handling of replies from the crypto cards for CCA and EP11 (AP message type 6) messages. The verification code related to reply field length was not designed well and thus firmware deficiencies could lead to unexpected behavior in the zcrypt device driver. Thus improve the code to more closely inspect especially length fields at message replies.

The 3 hunks of this patch deal with CCA, EP11 and (CCA) RNG replies and improve the checking for reply buffer size by using size_t instead of int. RNG replies an additional check makes sure the hard coded limit of the data buffer is not exceeded. Also there was a condition with additional data for an CCA reply where some of the field values where unchecked used to invoke memcpy into user
space. zcrypt_msgtype6_receive() now checks all the relevant fields before convert_type86_xcrb() uses them.

---
 drivers/s390/crypto/zcrypt_msgtype6.c | 42 ++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 8 deletions(-)
drivers/s390/crypto/zcrypt_msgtype6.c +34 -8
index 40f72cdf284d..8252fd185663 100644
@@ -691,6 +691,13 @@ static int convert_type86_rng(struct zcrypt_queue *zq,
 
 	if (msg->cprbx.ccp_rtcode != 0 || msg->cprbx.ccp_rscode != 0)
 		return -EINVAL;
+	/*
+	 * Note that offset2 and count2 have already been checked in
+	 * zcrypt_msgtype6_receive(). So only check for not exceeding
+	 * the hard coded rng buffer size.
+	 */
+	if (msg->fmt2.count2 > ZCRYPT_RNG_BUFFER_SIZE)
+		return -EMSGSIZE;
 	memcpy(buffer, data + msg->fmt2.offset2, msg->fmt2.count2);
 	return msg->fmt2.count2;
 }
@@ -853,7 +860,7 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
 	};
 	struct ap_response_type *resp_type = &msg->response;
 	struct type86x_reply *t86r;
-	int len;
+	size_t len, len1, len2 = 0;
 
 	/* Copy the reply message to the request message buffer. */
 	if (!reply)
@@ -863,7 +870,8 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
 	    t86r->cprbx.cprb_ver_id == 0x02) {
 		switch (resp_type->type) {
 		case CEXXC_RESPONSE_TYPE_ICA:
-			len = sizeof(struct type86x_reply) + t86r->length;
+			len = (size_t)sizeof(struct type86x_reply) +
+				(size_t)t86r->length;
 			if (len > reply->bufsize || len > msg->bufsize ||
 			    len != reply->len) {
 				pr_debug("len mismatch => EMSGSIZE\n");
@@ -874,10 +882,27 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
 			msg->len = len;
 			break;
 		case CEXXC_RESPONSE_TYPE_XCRB:
-			if (t86r->fmt2.count2)
-				len = t86r->fmt2.offset2 + t86r->fmt2.count2;
-			else
-				len = t86r->fmt2.offset1 + t86r->fmt2.count1;
+			len1 = (size_t)t86r->fmt2.offset1 +
+				(size_t)t86r->fmt2.count1;
+			if (t86r->fmt2.offset1 > reply->len ||
+			    t86r->fmt2.count1 > reply->len ||
+			    len1 > reply->len) {
+				pr_debug("len mismatch => EMSGSIZE\n");
+				msg->rc = -EMSGSIZE;
+				goto out;
+			}
+			if (t86r->fmt2.count2) {
+				len2 = (size_t)t86r->fmt2.offset2 +
+					(size_t)t86r->fmt2.count2;
+				if (t86r->fmt2.offset2 > reply->len ||
+				    t86r->fmt2.count2 > reply->len ||
+				    len2 > reply->len) {
+					pr_debug("len mismatch => EMSGSIZE\n");
+					msg->rc = -EMSGSIZE;
+					goto out;
+				}
+			}
+			len = max_t(size_t, len1, len2);
 			if (len > reply->bufsize || len > msg->bufsize ||
 			    len != reply->len) {
 				pr_debug("len mismatch => EMSGSIZE\n");
@@ -917,7 +942,7 @@ static void zcrypt_msgtype6_receive_ep11(struct ap_queue *aq,
 	};
 	struct ap_response_type *resp_type = &msg->response;
 	struct type86_ep11_reply *t86r;
-	int len;
+	size_t len;
 
 	/* Copy the reply message to the request message buffer. */
 	if (!reply)
@@ -927,7 +952,8 @@ static void zcrypt_msgtype6_receive_ep11(struct ap_queue *aq,
 	    t86r->cprbx.cprb_ver_id == 0x04) {
 		switch (resp_type->type) {
 		case CEXXC_RESPONSE_TYPE_EP11:
-			len = t86r->fmt2.offset1 + t86r->fmt2.count1;
+			len = (size_t)t86r->fmt2.offset1 +
+				(size_t)t86r->fmt2.count1;
 			if (len > reply->bufsize || len > msg->bufsize ||
 			    len != reply->len) {
 				pr_debug("len mismatch => EMSGSIZE\n");
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH v5 1/1] s390/zcrypt: Improve zcrypt reply message verification checks

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Medium] Uninitialized memory usage and buffer reuse leading to kernel heap information leak --

--- Patch [1]: [PATCH v5 1/1] s390/zcrypt: Improve zcrypt reply message verification checks --- commit 52b1c279b5cba2ecec495171bd1e75b2e86fc109
Author: Harald Freudenberger <[email protected]>

s390/zcrypt: Improve zcrypt reply message verification checks

This commit improves reply message verification checks for CCA and EP11 messages in the zcrypt device driver. It updates length field checking to use size_t and ensures bounds checking before using values to invoke memcpy.

Quoted 14 lines
drivers/s390/crypto/zcrypt_msgtype6.c +1 -1
index 40f72cdf284d3..8252fd1856638 100644
@@ -853,7 +860,7 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
 	};
 	struct ap_response_type *resp_type = &msg->response;
 	struct type86x_reply *t86r;
-	int len;
+	size_t len, len1, len2 = 0;
 
 	/* Copy the reply message to the request message buffer. */
 	if (!reply)

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct ap_queue *aq, · line 870

Quoted 1 line
drivers/s390/crypto/zcrypt_msgtype6.c +0 -0
@@ -863,7 +870,8 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
 	    t86r->cprbx.cprb_ver_id == 0x02) {

[Severity: Medium]
This is a pre-existing issue, but is it possible for a short reply to cause an out-of-bounds read here? The cprb_ver_id field is accessed before verifying that reply->len is large enough to contain the CPRBX header.

Additionally, if the hardware returns a response that hits the else branch in zcrypt_msgtype6_receive():

drivers/s390/crypto/zcrypt_msgtype6.c:zcrypt_msgtype6_receive() {
	...
	} else {
		memcpy(msg->msg, reply->msg, sizeof(error_reply));
		msg->len = sizeof(error_reply);
	}
	...
}

Only 8 bytes (sizeof(error_reply)) are copied, and the rest of msg->msg remains user-controlled from the original request.

Could a malicious user craft a request via ZSECSENDCPRB embedding specific values for cprb_ver_id and fmt2.offset2/count2, and if a short response forces the else branch, bypass verification?

If convert_response_xcrb() reads the mixed buffer, sees TYPE86_RSP_CODE, and validates cprb_ver_id == 0x02 from the user's data, it would call convert_type86_xcrb(), which then copies arbitrary kernel heap memory to user space using the user-controlled offsets:

drivers/s390/crypto/zcrypt_msgtype6.c:convert_type86_xcrb() {
	...
		if (z_copy_to_user(userspace, xcrb->reply_data_addr,
				   data + msg->fmt2.offset2, msg->fmt2.count2))
			return -EFAULT;
	...
}
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
maintainer
#3 lore raw

Re: [PATCH v5 1/1] s390/zcrypt: Improve zcrypt reply message verification checks

In reply to Harald Freudenberger
Quoted 16 lines · Harald Freudenberger

Add or improve checks related to buffer sizes and reply sizes to the handling of replies from the crypto cards for CCA and EP11 (AP message type 6) messages. The verification code related to reply field length was not designed well and thus firmware deficiencies could lead to unexpected behavior in the zcrypt device driver. Thus improve the code to more closely inspect especially length fields at message replies.

The 3 hunks of this patch deal with CCA, EP11 and (CCA) RNG replies and improve the checking for reply buffer size by using size_t instead of int. RNG replies an additional check makes sure the hard coded limit of the data buffer is not exceeded. Also there was a condition with additional data for an CCA reply where some of the field values where unchecked used to invoke memcpy into user
space. zcrypt_msgtype6_receive() now checks all the relevant fields before convert_type86_xcrb() uses them.

Quoted 2 lines

See my comments below.

Quoted 8 lines
---
  drivers/s390/crypto/zcrypt_msgtype6.c | 42 ++++++++++++++++++++++-----
  1 file changed, 34 insertions(+), 8 deletions(-)
drivers/s390/crypto/zcrypt_msgtype6.c +0 -0
index 40f72cdf284d..8252fd185663 100644

[...]

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct ap_queue *aq, · line 874

Quoted 7 lines
drivers/s390/crypto/zcrypt_msgtype6.c +2 -1
@@ -863,7 +870,8 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
  	    t86r->cprbx.cprb_ver_id == 0x02) {
  		switch (resp_type->type) {
  		case CEXXC_RESPONSE_TYPE_ICA:
-			len = sizeof(struct type86x_reply) + t86r->length;
+			len = (size_t)sizeof(struct type86x_reply) +
+				(size_t)t86r->length;

Is the explicit cast for sizeof() really necessary. I would assume, that the following should be sufficient:

	len = sizeof(struct type86x_reply) +
		(size_t)t86r->length;
Quoted 6 lines
drivers/s390/crypto/zcrypt_msgtype6.c +0 -0
  			if (len > reply->bufsize || len > msg->bufsize ||
  			    len != reply->len) {
  				pr_debug("len mismatch => EMSGSIZE\n");
@@ -874,10 +882,27 @@ static void zcrypt_msgtype6_receive(struct ap_queue *aq,
  			msg->len = len;
  			break;

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct ap_queue *aq, · line 889

Quoted 10 lines
drivers/s390/crypto/zcrypt_msgtype6.c +5 -4
  		case CEXXC_RESPONSE_TYPE_XCRB:
-			if (t86r->fmt2.count2)
-				len = t86r->fmt2.offset2 + t86r->fmt2.count2;
-			else
-				len = t86r->fmt2.offset1 + t86r->fmt2.count1;
+			len1 = (size_t)t86r->fmt2.offset1 +
+				(size_t)t86r->fmt2.count1;
+			if (t86r->fmt2.offset1 > reply->len ||
+			    t86r->fmt2.count1 > reply->len ||
+			    len1 > reply->len) {

Wouldn't it be sufficient to check only (len1 > reply->len)? If (t86r->fmt2.offset1 > reply->len) is true, than also (len1 > reply->len) will be true (and the same for count1).

Or did I miss something?

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct ap_queue *aq, · line 899

Quoted 10 lines
drivers/s390/crypto/zcrypt_msgtype6.c +10 -0
+				pr_debug("len mismatch => EMSGSIZE\n");
+				msg->rc = -EMSGSIZE;
+				goto out;
+			}
+			if (t86r->fmt2.count2) {
+				len2 = (size_t)t86r->fmt2.offset2 +
+					(size_t)t86r->fmt2.count2;
+				if (t86r->fmt2.offset2 > reply->len ||
+				    t86r->fmt2.count2 > reply->len ||
+				    len2 > reply->len) {

Same here.

[...]

Signature · 2 lines
Mit freundlichen Grüßen / Kind regards
Holger Dengler
author maintainer
#4 lore raw

Re: [PATCH v5 1/1] s390/zcrypt: Improve zcrypt reply message verification checks

Quoted 44 lines · Holger Dengler
Quoted 16 lines · Harald Freudenberger

Add or improve checks related to buffer sizes and reply sizes to the handling of replies from the crypto cards for CCA and EP11 (AP message type 6) messages. The verification code related to reply field length was not designed well and thus firmware deficiencies could lead to unexpected behavior in the zcrypt device driver. Thus improve the code to more closely inspect especially length fields at message replies.

The 3 hunks of this patch deal with CCA, EP11 and (CCA) RNG replies and improve the checking for reply buffer size by using size_t instead of int. RNG replies an additional check makes sure the hard coded limit of the data buffer is not exceeded. Also there was a condition with additional data for an CCA reply where some of the field values where unchecked used to invoke memcpy into user
space. zcrypt_msgtype6_receive() now checks all the relevant fields before convert_type86_xcrb() uses them.

Quoted 2 lines

See my comments below.

Quoted 9 lines
---
  drivers/s390/crypto/zcrypt_msgtype6.c | 42 
++++++++++++++++++++++-----
  1 file changed, 34 insertions(+), 8 deletions(-)
drivers/s390/crypto/zcrypt_msgtype6.c +0 -0
index 40f72cdf284d..8252fd185663 100644

[...]

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct · line 875

Quoted 8 lines
drivers/s390/crypto/zcrypt_msgtype6.c +2 -1
@@ -863,7 +870,8 @@ static void zcrypt_msgtype6_receive(struct 
 ap_queue *aq,
  	    t86r->cprbx.cprb_ver_id == 0x02) {
  		switch (resp_type->type) {
  		case CEXXC_RESPONSE_TYPE_ICA:
-			len = sizeof(struct type86x_reply) + t86r->length;
+			len = (size_t)sizeof(struct type86x_reply) +
+				(size_t)t86r->length;

Is the explicit cast for sizeof() really necessary. I would assume, that the following should be sufficient:

Quoted 3 lines
	len = sizeof(struct type86x_reply) +
		(size_t)t86r->length;

Yes - removed.

Quoted 22 lines
Quoted 7 lines
drivers/s390/crypto/zcrypt_msgtype6.c +0 -0
  			if (len > reply->bufsize || len > msg->bufsize ||
  			    len != reply->len) {
  				pr_debug("len mismatch => EMSGSIZE\n");
drivers/s390/crypto/zcrypt_msgtype6.c +0 -0
@@ -874,10 +882,27 @@ static void zcrypt_msgtype6_receive(struct 
 ap_queue *aq,
  			msg->len = len;
  			break;

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct · line 890

Quoted 10 lines
drivers/s390/crypto/zcrypt_msgtype6.c +5 -4
  		case CEXXC_RESPONSE_TYPE_XCRB:
-			if (t86r->fmt2.count2)
-				len = t86r->fmt2.offset2 + t86r->fmt2.count2;
-			else
-				len = t86r->fmt2.offset1 + t86r->fmt2.count1;
+			len1 = (size_t)t86r->fmt2.offset1 +
+				(size_t)t86r->fmt2.count1;
+			if (t86r->fmt2.offset1 > reply->len ||
+			    t86r->fmt2.count1 > reply->len ||
+			    len1 > reply->len) {

Wouldn't it be sufficient to check only (len1 > reply->len)? If (t86r->fmt2.offset1 > reply->len) is true, than also (len1 > reply->len) will be true (and the same for count1).

Quoted 1 line

Or did I miss something?

Well this calculation is tricky. So let me summarize what I think should be checked:
1) offset1 should lie in the buffer
==> offset1 < reply->len should be true
2) the block should fit into the buffer
==> count1 <= reply->len should be true
with that it is clear and no need to check that the end (offset1 + count1) is also covered:
==> offset1 + count1 <= reply->len should then be implicitly true

Maybe have a look at v6 of this code. I reworked this again and now it clearly distinguishes between validations of the message fields (count, offset) and checks about length of messages and buffer sizes.

Quoted 15 lines

drivers/s390/crypto/zcrypt_msgtype6.c static void zcrypt_msgtype6_receive(struct ap_queue *aq, · line 899

Quoted 11 lines
drivers/s390/crypto/zcrypt_msgtype6.c +10 -0
+				pr_debug("len mismatch => EMSGSIZE\n");
+				msg->rc = -EMSGSIZE;
+				goto out;
+			}
+			if (t86r->fmt2.count2) {
+				len2 = (size_t)t86r->fmt2.offset2 +
+					(size_t)t86r->fmt2.count2;
+				if (t86r->fmt2.offset2 > reply->len ||
+				    t86r->fmt2.count2 > reply->len ||
+				    len2 > reply->len) {

Same here.

[...]