Skip to contentSkip to conversation

s390/qeth: reject oversized SNMP requests

PATCH Applied Julian Wiedmann 2 participants 3 messages Aug 23, 2019 – Aug 24, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 s390/qeth: reject oversized SNMP requests

Commit d4c08afafa04 ("s390/qeth: streamline SNMP cmd code") removed the bounds checking for req_len, under the assumption that the check in qeth_alloc_cmd() would suffice.

But that code path isn't sufficiently robust to handle a user-provided data_length, which could overflow (when adding the cmd header overhead) before being checked against QETH_BUFSIZE. We end up allocating just a tiny iob, and the subsequent copy_from_user() writes past the end of that iob.

Special-case this path and add a coarse bounds check, to protect against maliciuous requests. This let's the subsequent code flow do its normal job and precise checking, without risk of overflow.

Fixes d4c08afafa04 ("s390/qeth: streamline SNMP cmd code")
---
 drivers/s390/net/qeth_core_main.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/s390/net/qeth_core_main.c +4 -0
index 9c3310c4d61d..6502b148541e 100644
@@ -4374,6 +4374,10 @@ static int qeth_snmp_command(struct qeth_card *card, char __user *udata)
 	    get_user(req_len, &ureq->hdr.req_len))
 		return -EFAULT;
 
+	/* Sanitize user input, to avoid overflows in iob size calculation: */
+	if (req_len > QETH_BUFSIZE)
+		return -EINVAL;
+
 	iob = qeth_get_adapter_cmd(card, IPA_SETADP_SET_SNMP_CONTROL, req_len);
 	if (!iob)
 		return -ENOMEM;
Signature · 1 line
2.17.1
#1 lore raw

+cc Dan

Quoted 38 lines · Julian Wiedmann

Commit d4c08afafa04 ("s390/qeth: streamline SNMP cmd code") removed the bounds checking for req_len, under the assumption that the check in qeth_alloc_cmd() would suffice.

But that code path isn't sufficiently robust to handle a user-provided data_length, which could overflow (when adding the cmd header overhead) before being checked against QETH_BUFSIZE. We end up allocating just a tiny iob, and the subsequent copy_from_user() writes past the end of that iob.

Special-case this path and add a coarse bounds check, to protect against maliciuous requests. This let's the subsequent code flow do its normal job and precise checking, without risk of overflow.

Fixes d4c08afafa04 ("s390/qeth: streamline SNMP cmd code")
---
 drivers/s390/net/qeth_core_main.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/s390/net/qeth_core_main.c +4 -0
index 9c3310c4d61d..6502b148541e 100644
@@ -4374,6 +4374,10 @@ static int qeth_snmp_command(struct qeth_card *card, char __user *udata)
 	    get_user(req_len, &ureq->hdr.req_len))
 		return -EFAULT;
 
+	/* Sanitize user input, to avoid overflows in iob size calculation: */
+	if (req_len > QETH_BUFSIZE)
+		return -EINVAL;
+
 	iob = qeth_get_adapter_cmd(card, IPA_SETADP_SET_SNMP_CONTROL, req_len);
 	if (!iob)
 		return -ENOMEM;
maintainer Applied
#2 lore raw
In reply to Julian Wiedmann

From: Julian Wiedmann <[email protected]>
Date: Fri, 23 Aug 2019 11:29:23 +0200

Quoted 10 lines

Commit d4c08afafa04 ("s390/qeth: streamline SNMP cmd code") removed the bounds checking for req_len, under the assumption that the check in qeth_alloc_cmd() would suffice.

But that code path isn't sufficiently robust to handle a user-provided data_length, which could overflow (when adding the cmd header overhead) before being checked against QETH_BUFSIZE. We end up allocating just a tiny iob, and the subsequent copy_from_user() writes past the end of that iob.

Quoted 8 lines

Special-case this path and add a coarse bounds check, to protect against maliciuous requests. This let's the subsequent code flow do its normal job and precise checking, without risk of overflow.

Fixes d4c08afafa04 ("s390/qeth: streamline SNMP cmd code")

Applied.