mirrors/meta-openembedded
1
0 Fork 0
mirror of https://github.com/openembedded/meta-openembedded.git synced 2026-10-04 14:56:41 +00:00
Code Issues Projects Releases Wiki Activity

wolfssl: patch CVE-2026-6092

Browse Source 
Details:
https://nvd.nist.gov/vuln/detail/cve-2026-6092

Signed-off-by: Ankur Tyagi <ankur.tyagi85@gmail.com>
Signed-off-by: Anuj Mittal <anuj.mittal@oss.qualcomm.com>
This commit is contained in:
Ankur Tyagi authored and Anuj Mittal committed 2026-09-15 10:13:39 +05:30
1 parent c897678f87
commit 4b124297a2
2 files changed
+164
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

meta-networking/recipes-connectivity/wolfssl/files/CVE-2026-6092.patch
+163
View File
@@ -0,0 +1,163 @@
From 5e2fa43aed59a2524c33a443c93445882519677b Mon Sep 17 00:00:00 2001
From: Eric Blankenhorn <eric@wolfssl.com>
Date: Wed, 8 Apr 2026 15:06:11 -0500
Subject: [PATCH] Fix ETM on resumption
(cherry picked from commit af5369636a938faf4a79d1f550d3b206c51fafec)
CVE: CVE-2026-6092
Upstream-Status: Backport [https://github.com/wolfSSL/wolfssl/commit/af5369636a938faf4a79d1f550d3b206c51fafec]
Signed-off-by: Ankur Tyagi <ankur.tyagi85@gmail.com>
---
src/internal.c | 20 +++++++----
tests/api/test_tls.c | 83 ++++++++++++++++++++++++++++++++++++++++++++
tests/api/test_tls.h | 2 ++
3 files changed, 99 insertions(+), 6 deletions(-)
diff --git a/src/internal.c b/src/internal.c
index ad1587e0f..267c75a5d 100644
--- a/src/internal.c
+++ b/src/internal.c
@@ -38436,13 +38436,21 @@ static int AddPSKtoPreMasterSecret(WOLFSSL* ssl)
#if defined(HAVE_TLS_EXTENSIONS) && defined(HAVE_ENCRYPT_THEN_MAC) && \
!defined(WOLFSSL_AEAD_ONLY)
- if (ssl->options.encThenMac && ssl->specs.cipher_type == block) {
- ret = TLSX_EncryptThenMac_Respond(ssl);
- if (ret != 0)
- goto out;
+ /* Only respond to ETM here when resumption actually succeeded;
+ * HandleTlsResumption populates ssl->specs via SetCipherSpecs only
+ * on the success path. If resumption failed, resuming has been
+ * cleared and ssl->specs.cipher_type is still zero-initialized,
+ * so we must defer the ETM decision until after MatchSuite. */
+ if (ssl->options.resuming) {
+ if (ssl->options.encThenMac &&
+ ssl->specs.cipher_type == block) {
+ ret = TLSX_EncryptThenMac_Respond(ssl);
+ if (ret != 0)
+ goto out;
+ }
+ else
+ ssl->options.encThenMac = 0;
}
- else
- ssl->options.encThenMac = 0;
#endif
if (ssl->options.clientState == CLIENT_KEYEXCHANGE_COMPLETE) {
WOLFSSL_LEAVE("DoClientHello", ret);
diff --git a/tests/api/test_tls.c b/tests/api/test_tls.c
index 565006171..f6cbb6d38 100644
--- a/tests/api/test_tls.c
+++ b/tests/api/test_tls.c
@@ -730,6 +730,89 @@ int test_tls12_no_null_compression(void)
* uppercase names like "SECP384R1" do not match the lowercase "secp384r1"
* entry; they fall through to the wolfCrypt ECC look-up which uses
* XSTRCASECMP. */
+/* Regression test for the encrypt-then-MAC silent-disable bug.
+ *
+ * Before the fix, when a client sent a 32-byte session ID in its ClientHello
+ * (so the server set ssl->options.resuming = 1) but the server's session
+ * cache did not contain that session, DoClientHello would run an
+ * encrypt_then_mac decision *before* MatchSuite/SetCipherSpecs had populated
+ * ssl->specs.cipher_type. Because cipher_type was zero-initialized
+ * (== stream, not block), the ETM block cleared encThenMac to 0, and the
+ * post-MatchSuite block could not re-enable it. The connection then
+ * silently negotiated MAC-then-encrypt instead of encrypt-then-MAC.
+ *
+ * This test forces a stale-resumption ClientHello against a server with an
+ * empty session cache, using a CBC-mode cipher suite, and asserts that the
+ * server still negotiates encrypt-then-MAC. */
+int test_tls12_etm_failed_resumption(void)
+{
+ EXPECT_DECLS;
+#if defined(HAVE_MANUAL_MEMIO_TESTS_DEPENDENCIES) && \
+ !defined(WOLFSSL_NO_TLS12) && defined(HAVE_ENCRYPT_THEN_MAC) && \
+ !defined(WOLFSSL_AEAD_ONLY) && !defined(NO_RSA) && !defined(NO_AES) && \
+ defined(HAVE_AES_CBC) && !defined(NO_SHA256) && \
+ defined(HAVE_SESSION_TICKET) && defined(HAVE_ECC)
+ /* TLS_ECDHE_RSA_WITH_AES_128_CBC_SHA256 — a CBC suite, where ETM applies. */
+ const char* cbcSuite = "ECDHE-RSA-AES128-SHA256";
+ WOLFSSL_CTX *ctx_c = NULL, *ctx_s = NULL;
+ WOLFSSL *ssl_c = NULL, *ssl_s = NULL;
+ WOLFSSL_SESSION *sess = NULL;
+ struct test_memio_ctx test_ctx;
+
+ /* First handshake: establish a session-ID-based session on the client.
+ * Disable TLS 1.2 session tickets on both sides so resumption uses the
+ * session ID path (not tickets), which is the path the bug lives on. */
+ XMEMSET(&test_ctx, 0, sizeof(test_ctx));
+ ExpectIntEQ(test_memio_setup(&test_ctx, &ctx_c, &ctx_s, &ssl_c, &ssl_s,
+ wolfTLSv1_2_client_method, wolfTLSv1_2_server_method), 0);
+ ExpectIntEQ(wolfSSL_NoTicketTLSv12(ssl_c), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_NoTicketTLSv12(ssl_s), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_set_cipher_list(ssl_c, cbcSuite), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_set_cipher_list(ssl_s, cbcSuite), WOLFSSL_SUCCESS);
+ ExpectIntEQ(test_memio_do_handshake(ssl_c, ssl_s, 10, NULL), 0);
+ /* Sanity: the first handshake itself must use ETM. */
+ ExpectIntEQ(ssl_s->options.encThenMac, 1);
+ ExpectNotNull(sess = wolfSSL_get1_session(ssl_c));
+
+ wolfSSL_free(ssl_c); ssl_c = NULL;
+ wolfSSL_free(ssl_s); ssl_s = NULL;
+ wolfSSL_CTX_free(ctx_c); ctx_c = NULL;
+ wolfSSL_CTX_free(ctx_s); ctx_s = NULL;
+
+ /* Second handshake against a *fresh* server context (empty cache). The
+ * client offers the saved session, so the server's ClientHello parser
+ * sets options.resuming = 1, but HandleTlsResumption then fails to find
+ * the session and clears resuming. Pre-fix, ETM was silently dropped
+ * here. */
+ XMEMSET(&test_ctx, 0, sizeof(test_ctx));
+ ExpectIntEQ(test_memio_setup(&test_ctx, &ctx_c, &ctx_s, &ssl_c, &ssl_s,
+ wolfTLSv1_2_client_method, wolfTLSv1_2_server_method), 0);
+ /* The internal session cache is process-global, so the saved session is
+ * still findable via the cache. Disable lookups on this server SSL
+ * directly so that HandleTlsResumption hits its "session lookup failed"
+ * path — exactly the scenario the bug fix targets. */
+ ssl_s->options.sessionCacheOff = 1;
+ ExpectIntEQ(wolfSSL_NoTicketTLSv12(ssl_c), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_NoTicketTLSv12(ssl_s), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_set_cipher_list(ssl_c, cbcSuite), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_set_cipher_list(ssl_s, cbcSuite), WOLFSSL_SUCCESS);
+ ExpectIntEQ(wolfSSL_set_session(ssl_c, sess), WOLFSSL_SUCCESS);
+ ExpectIntEQ(test_memio_do_handshake(ssl_c, ssl_s, 10, NULL), 0);
+ /* The server should NOT have actually resumed (fresh ctx, empty cache). */
+ ExpectIntEQ(ssl_s->options.resuming, 0);
+ /* And — the regression check — encrypt-then-MAC must still be active. */
+ ExpectIntEQ(ssl_s->options.encThenMac, 1);
+ ExpectIntEQ(ssl_c->options.encThenMac, 1);
+
+ wolfSSL_SESSION_free(sess);
+ wolfSSL_free(ssl_c);
+ wolfSSL_free(ssl_s);
+ wolfSSL_CTX_free(ctx_c);
+ wolfSSL_CTX_free(ctx_s);
+#endif
+ return EXPECT_RESULT();
+}
+
int test_tls_set_curves_list_ecc_fallback(void)
{
EXPECT_DECLS;
diff --git a/tests/api/test_tls.h b/tests/api/test_tls.h
index 46d540434..9d28a599a 100644
--- a/tests/api/test_tls.h
+++ b/tests/api/test_tls.h
@@ -30,6 +30,7 @@ int test_tls13_curve_intersection(void);
int test_tls_certreq_order(void);
int test_tls12_bad_cv_sig_alg(void);
int test_tls12_no_null_compression(void);
+int test_tls12_etm_failed_resumption(void);
int test_tls_set_curves_list_ecc_fallback(void);
#define TEST_TLS_DECLS \
@@ -41,6 +42,7 @@ int test_tls_set_curves_list_ecc_fallback(void);
TEST_DECL_GROUP("tls", test_tls_certreq_order), \
TEST_DECL_GROUP("tls", test_tls12_bad_cv_sig_alg), \
TEST_DECL_GROUP("tls", test_tls12_no_null_compression), \
+ TEST_DECL_GROUP("tls", test_tls12_etm_failed_resumption), \
TEST_DECL_GROUP("tls", test_tls_set_curves_list_ecc_fallback)
#endif /* TESTS_API_TEST_TLS_H */
meta-networking/recipes-connectivity/wolfssl/wolfssl_5.9.1.bb
+1
View File
@@ -22,6 +22,7 @@ SRC_URI = " \
file://CVE-2026-6091-1.patch \
file://CVE-2026-6091-2.patch \
file://CVE-2026-6091-3.patch \
file://CVE-2026-6092.patch \
"
SRCREV = "1d363f3adceba9d1478230ede476a37b0dcdef24"