laforge submitted this change.

View Change

Approvals: Jenkins Builder: Verified laforge: Looks good to me, but someone else must approve pespin: Looks good to me, approved
gb: ns2: clamp embedded NS PDU to msgb tailroom

Per TS 48.016 9.2.7.2 an NS-STATUS PDU embeds the PDU that caused the
error. reject_status_msg() copies it in without any length check, and
ns2_tx_status() clamps it against nsvc->bind->mtu only, not against
the space left in the reply msgb.

For an NS-over-IP bind that MTU is 65527 (gprs_ns2_udp.c:344), while
the STATUS msgb from ns2_msgb_alloc() offers 3052 usable bytes
(NS_ALLOC_SIZE 3072 minus NS_ALLOC_HEADROOM 20), of which 3048 remain
after the NS header and the Cause IE. A received PDU of 3046 bytes or
more therefore makes msgb_tvlv_put() -> msgb_put() hit MSGB_ABORT,
which aborts the process. No NS-VC, handshake or credentials are
needed to reach either site.

Clamp against the msgb tailroom in both places, and use a saturating
clamp instead of a subtraction on an unsigned (the old form could
underflow if the MTU were smaller than the message built so far).

bssgp_tx_status() in the same tree already does this correctly via
OSMO_MIN(..., msgb_tailroom(msg) - 3).

Fixes: 6a161497cfd8cacfef10c2e81cff85a45447e107
Change-Id: Iac1c5c15875f30f34d0f7e9fc74264530a02b6ae
---
M src/gb/gprs_ns2.c
M src/gb/gprs_ns2_message.c
2 files changed, 32 insertions(+), 9 deletions(-)

diff --git a/src/gb/gprs_ns2.c b/src/gb/gprs_ns2.c
index 4e496c1..8d77e9e 100644
--- a/src/gb/gprs_ns2.c
+++ b/src/gb/gprs_ns2.c
@@ -751,10 +751,21 @@
case NS_CAUSE_PDU_INCOMP_PSTATE:
case NS_CAUSE_PROTO_ERR_UNSPEC:
case NS_CAUSE_INVAL_ESSENT_IE:
- case NS_CAUSE_MISSING_ESSENT_IE:
- msgb_tvlv_put(msg, NS_IE_PDU, msgb_l2len(orig_msg),
- orig_msg->l2h);
+ case NS_CAUSE_MISSING_ESSENT_IE: {
+ /* The original PDU can be larger than what is left in our own
+ * msgb (NS_ALLOC_SIZE), so clamp it. TvLV header is 3 bytes. */
+ const unsigned int avail = msgb_tailroom(msg);
+ unsigned int embed_len = msgb_l2len(orig_msg);
+
+ if (avail < 3)
+ embed_len = 0;
+ else if (avail < 3 + embed_len)
+ embed_len = avail - 3;
+
+ if (embed_len > 0)
+ msgb_tvlv_put(msg, NS_IE_PDU, embed_len, orig_msg->l2h);
break;
+ }
default:
break;
}
diff --git a/src/gb/gprs_ns2_message.c b/src/gb/gprs_ns2_message.c
index de63b7a..0fa2e26 100644
--- a/src/gb/gprs_ns2_message.c
+++ b/src/gb/gprs_ns2_message.c
@@ -426,7 +426,7 @@
struct msgb *msg = ns2_msgb_alloc();
struct gprs_ns_hdr *nsh;
uint16_t encoded_nsvci;
- unsigned int orig_len, max_orig_len;
+ unsigned int orig_len, avail;

log_set_context(LOG_CTX_GB_NSE, nsvc->nse);
log_set_context(LOG_CTX_GB_NSVC, nsvc);
@@ -458,12 +458,24 @@
case NS_CAUSE_INVAL_ESSENT_IE:
case NS_CAUSE_MISSING_ESSENT_IE:
/* Section 9.2.7.2: Static conditions for NS PDU */
- /* ensure the PDU doesn't exceed the MTU */
+ /* Trim available space to the MTU. bind->mtu can be far larger
+ * than NS_ALLOC_SIZE (65527 for NS-over-IP), so the msgb
+ * tailroom is the other bound we need. */
+ avail = msgb_tailroom(msg);
+ if (msgb_length(msg) > nsvc->bind->mtu)
+ avail = 0;
+ else if (nsvc->bind->mtu - msgb_length(msg) < avail)
+ avail = nsvc->bind->mtu - msgb_length(msg);
+
+ /* Figure out how many origin PDU bytes fit; TvLV header is 3 bytes: */
orig_len = msgb_l2len(orig_msg);
- max_orig_len = msgb_length(msg) + TVLV_GROSS_LEN(orig_len);
- if (max_orig_len > nsvc->bind->mtu)
- orig_len -= max_orig_len - nsvc->bind->mtu;
- msgb_tvlv_put(msg, NS_IE_PDU, orig_len, orig_msg->l2h);
+ if (avail < 3)
+ orig_len = 0;
+ else if (avail < 3 + orig_len)
+ orig_len = avail - 3;
+
+ if (orig_len > 0)
+ msgb_tvlv_put(msg, NS_IE_PDU, orig_len, orig_msg->l2h);
break;
case NS_CAUSE_BVCI_UNKNOWN:
/* Section 9.2.7.3: Static conditions for BVCI */

To view, visit change 44138. To unsubscribe, or for help writing mail filters, visit settings.

Gerrit-MessageType: merged
Gerrit-Project: libosmocore
Gerrit-Branch: master
Gerrit-Change-Id: Iac1c5c15875f30f34d0f7e9fc74264530a02b6ae
Gerrit-Change-Number: 44138
Gerrit-PatchSet: 3
Gerrit-Owner: ws1004 <ws1004@kakao.com>
Gerrit-Reviewer: Jenkins Builder
Gerrit-Reviewer: daniel <daniel@totalueberwachung.de>
Gerrit-Reviewer: fixeria <vyanitskiy@sysmocom.de>
Gerrit-Reviewer: laforge <laforge@osmocom.org>
Gerrit-Reviewer: lynxis <lynxis@fe80.eu>
Gerrit-Reviewer: pespin <pespin@sysmocom.de>