Integer underflow in wolfSSL packet sniffer <= 5.9.0 allows an attacker to cause a program crash in the AEAD decryption path by injecting a TLS record shorter than the explicit IV plus authentication tag into traffic inspected by ssl_DecodePacket. The underflow wraps a 16-bit length to a large value that is passed to AEAD decryption routines, causing a large out-of-bounds read and crash. An unauthenticated attacker can trigger this remotely via malformed TLS Application Data records.

A 1-byte stack buffer over-read was identified in the MatchDomainName function (src/internal.c) during wildcard hostname validation when the LEFT_MOST_WILDCARD_ONLY flag is active. If a wildcard * exhausts the entire hostname string, the function reads one byte past the buffer without a bounds check, which could cause a crash.

When restoring a session from cache, a pointer from the serialized session data is used in a free operation without validation. An attacker who can poison the session cache could trigger an arbitrary free. Exploitation requires the ability to inject a crafted session into the cache and for the application to call specific session restore APIs.

A padding oracle exists in wolfSSL's PKCS7 CBC decryption that could allow an attacker to recover plaintext through repeated decryption queries with modified ciphertext. In previous versions of wolfSSL the interior padding bytes are not validated.

In TLSX_EchChangeSNI, the ctx->extensions branch set extensions unconditionally even when TLSX_Find returned NULL. This caused TLSX_UseSNI to attach the attacker-controlled publicName to the shared WOLFSSL_CTX when no inner SNI was configured. TLSX_EchRestoreSNI then failed to clean it up because its removal was gated on serverNameX != NULL. The inner ClientHello was sized before the pollution but written after it, causing TLSX_SNI_Write to memcpy 255 bytes past the allocation boundary.

wolfSSL_X509_verify_cert in the OpenSSL compatibility layer accepts a certificate chain in which the leaf's signature is not checked, if the attacker supplies an untrusted intermediate with Basic Constraints `CA:FALSE` that is legitimately signed by a trusted root. An attacker who obtains any leaf certificate from a trusted CA (e.g. a free DV cert from Let's Encrypt) can forge a certificate for any subject name with any public key and arbitrary signature bytes, and the function returns `WOLFSSL_SUCCESS` / `X509_V_OK`. The native wolfSSL TLS handshake path (`ProcessPeerCerts`) is not susceptible and the issue is limited to applications using the OpenSSL compatibility API directly, which would include integrations of wolfSSL into nginx and haproxy.

wolfSSL's wc_PKCS7_DecodeAuthEnvelopedData() does not properly sanitize the AES-GCM authentication tag length received and has no lower bounds check. A man-in-the-middle can therefore truncate the mac field from 16 bytes to 1 byte, reducing the tag check from 2⁻¹²⁸ to 2⁻⁸.

In wolfSSL's EVP layer, the ChaCha20-Poly1305 AEAD decryption path in wolfSSL_EVP_CipherFinal (and related EVP cipher finalization functions) fails to verify the authentication tag before returning plaintext to the caller. When an application uses the EVP API to perform ChaCha20-Poly1305 decryption, the implementation computes or accepts the tag but does not compare it against the expected value.

An integer overflow existed in the wolfCrypt CMAC implementation, that could be exploited to forge CMAC tags. The function wc_CmacUpdate used the guard `if (cmac->totalSz != 0)` to skip XOR-chaining on the first block (where digest is all-zeros and the XOR is a no-op). However, totalSz is word32 and wraps to zero after 2^28 block flushes (4 GiB), causing the guard to erroneously discard the live CBC-MAC chain state. Any two messages sharing a common suffix beyond the 4 GiB mark then produce identical CMAC tags, enabling a zero-work prefix-substitution forgery. The fix removes the guard, making the XOR unconditional; the no-op property on the first block is preserved because digest is zero-initialized by wc_InitCmac_ex.

wolfSSL's ECCSI signature verifier `wc_VerifyEccsiHash` decodes the `r` and `s` scalars from the signature blob via `mp_read_unsigned_bin` with no check that they lie in `[1, q-1]`. A crafted forged signature could verify against any message for any identity, using only publicly-known constants.

A heap use-after-free exists in wolfSSL's TLS 1.3 post-quantum cryptography (PQC) hybrid KeyShare processing. In the error handling path of TLSX_KeyShare_ProcessPqcHybridClient() in src/tls.c, the inner function TLSX_KeyShare_ProcessPqcClient_ex() frees a KyberKey object upon encountering an error. The caller then invokes TLSX_KeyShare_FreeAll(), which attempts to call ForceZero() on the already-freed KyberKey, resulting in writes of zero bytes over freed heap memory.

X.509 date buffer overflow in wolfSSL_X509_notAfter / wolfSSL_X509_notBefore. A buffer overflow may occur when parsing date fields from a crafted X.509 certificate via the compatibility layer API. This is only triggered when calling these two APIs directly from an application, and does not affect TLS or certificate verify operations in wolfSSL.

Heap buffer overflow in CertFromX509 via AuthorityKeyIdentifier size confusion. A heap buffer overflow occurs when converting an X.509 certificate internally due to incorrect size handling of the AuthorityKeyIdentifier extension.

Dual-Algorithm CertificateVerify out-of-bounds read. When processing a dual-algorithm CertificateVerify message, an out-of-bounds read can occur on crafted input. This can only occur when --enable-experimental and --enable-dual-alg-certs is used when building wolfSSL.

Heap out-of-bounds read in PKCS7 parsing. A crafted PKCS7 message can trigger an OOB read on the heap. The missing bounds check is in the indefinite-length end-of-content verification loop in PKCS7_VerifySignedData().

A stack buffer overflow exists in wolfSSL's PKCS7 implementation in the wc_PKCS7_DecryptOri() function in wolfcrypt/src/pkcs7.c. When processing a CMS EnvelopedData message containing an OtherRecipientInfo (ORI) recipient, the function copies an ASN.1-parsed OID into a fixed 32-byte stack buffer (oriOID[MAX_OID_SZ]) via XMEMCPY without first validating that the parsed OID length does not exceed MAX_OID_SZ. A crafted CMS EnvelopedData message with an ORI recipient containing an OID longer than 32 bytes triggers a stack buffer overflow. Exploitation requires the library to be built with --enable-pkcs7 (disabled by default) and the application to have registered an ORI decrypt callback via wc_PKCS7_SetOriDecryptCb().

Heap buffer overflow in DTLS 1.3 ACK message processing. A remote attacker can send a crafted DTLS 1.3 ACK message that triggers a heap buffer overflow.

URI nameConstraints from constrained intermediate CAs are parsed but not enforced during certificate chain verification in wolfcrypt/src/asn.c. A compromised or malicious sub-CA could issue leaf certificates with URI SAN entries that violate the nameConstraints of the issuing CA, and wolfSSL would accept them as valid.

Missing hash/digest size and OID checks allow digests smaller than allowed when verifying ECDSA certificates, or smaller than is appropriate for the relevant key type, to be accepted by signature verification functions. This could lead to reduced security of ECDSA certificate-based authentication if the public CA key used is also known. This affects ECDSA/ECC verification when EdDSA or ML-DSA is also enabled.

An integer underflow issue exists in wolfSSL when parsing the Subject Alternative Name (SAN) extension of X.509 certificates. A malformed certificate can specify an entry length larger than the enclosing sequence, causing the internal length counter to wrap during parsing. This results in incorrect handling of certificate data. The issue is limited to configurations using the original ASN.1 parsing implementation which is off by default.

Two potential heap out-of-bounds write locations existed in DecodeObjectId() in wolfcrypt/src/asn.c. First, a bounds check only validates one available slot before writing two OID arc values (out[0] and out[1]), enabling a 2-byte out-of-bounds write when outSz equals 1. Second, multiple callers pass sizeof(decOid) (64 bytes on 64-bit platforms) instead of the element count MAX_OID_SZ (32), causing the function to accept crafted OIDs with 33 or more arcs that write past the end of the allocated buffer.

Heap-based buffer overflow in the KCAPI ECC code path of wc_ecc_import_x963_ex() in wolfSSL wolfcrypt allows a remote attacker to write attacker-controlled data past the bounds of the pubkey_raw buffer via a crafted oversized EC public key point. The WOLFSSL_KCAPI_ECC code path copies the input to key->pubkey_raw (132 bytes) using XMEMCPY without a bounds check, unlike the ATECC code path which includes a length validation. This can be triggered during TLS key exchange when a malicious peer sends a crafted ECPoint in ServerKeyExchange.

1-byte OOB heap read in wc_PKCS7_DecodeEnvelopedData via zero-length encrypted content. A vulnerability existed in wolfSSL 5.8.4 and earlier, where a 1-byte out-of-bounds heap read in wc_PKCS7_DecodeEnvelopedData could be triggered by a crafted CMS EnvelopedData message with zero-length encrypted content. Note that PKCS7 support is disabled by default.

Heap Overflow in TLS 1.3 ECH parsing. An integer underflow existed in ECH extension parsing logic when calculating a buffer length, which resulted in writing beyond the bounds of an allocated buffer. Note that in wolfSSL, ECH is off by default, and the ECH standard is still evolving.

Two buffer overflow vulnerabilities existed in the wolfSSL CRL parser when parsing CRL numbers: a heap-based buffer overflow could occur when improperly storing the CRL number as a hexadecimal string, and a stack-based overflow for sufficiently sized CRL numbers. With appropriately crafted CRLs, either of these out of bound writes could be triggered. Note this only affects builds that specifically enable CRL support, and the user would need to load a CRL from an untrusted source.

Out-of-bounds read in ALPN parsing due to incomplete validation. wolfSSL 5.8.4 and earlier contained an out-of-bounds read in ALPN handling when built with ALPN enabled (HAVE_ALPN / --enable-alpn). A crafted ALPN protocol list could trigger an out-of-bounds read, leading to a potential process crash (denial of service). Note that ALPN is disabled by default, but is enabled for these 3rd party compatibility features: enable-apachehttpd, enable-bind, enable-curl, enable-haproxy, enable-hitch, enable-lighty, enable-jni, enable-nginx, enable-quic.

Missing required cryptographic step in the TLS 1.3 client HelloRetryRequest handshake logic in wolfSSL could lead to a compromise in the confidentiality of TLS-protected communications via a crafted HelloRetryRequest followed by a ServerHello message that omits the required key_share extension, resulting in derivation of predictable traffic secrets from (EC)DHE shared secret. This issue does not affect the client's authentication of the server during TLS handshakes.

An integer overflow vulnerability existed in the static function wolfssl_add_to_chain, that caused heap corruption when certificate data was written out of bounds of an insufficiently sized certificate buffer. wolfssl_add_to_chain is called by these API: wolfSSL_CTX_add_extra_chain_cert, wolfSSL_CTX_add1_chain_cert, wolfSSL_add0_chain_cert. These API are enabled for 3rd party compatibility features: enable-opensslall, enable-opensslextra, enable-lighty, enable-stunnel, enable-nginx, enable-haproxy. This issue is not remotely exploitable, and would require that the application context loading certificates is compromised.

A heap-buffer-overflow vulnerability exists in wolfSSL's wolfSSL_d2i_SSL_SESSION() function. When deserializing session data with SESSION_CERTS enabled, certificate and session id lengths are read from an untrusted input without bounds validation, allowing an attacker to overflow fixed-size buffers and corrupt heap memory. A maliciously crafted session would need to be loaded from an external source to trigger this vulnerability. Internal sessions were not vulnerable.

In wolfSSL 5.8.2 and earlier, a logic flaw existed in the TLS 1.2 server state machine implementation. The server could incorrectly accept the CertificateVerify message before the ClientKeyExchange message had been received. This issue affects wolfSSL before 5.8.4 (wolfSSL 5.8.2 and earlier is vulnerable, 5.8.4 is not vulnerable). In 5.8.4 wolfSSL would detect the issue later in the handshake. 5.9.0 was further hardened to catch the issue earlier in the handshake.

Integer underflow in wolfSSL packet sniffer <= 5.8.4 allows an attacker to cause a buffer overflow in the AEAD decryption path by injecting a TLS record shorter than the explicit IV plus authentication tag into traffic inspected by ssl_DecodePacket. The underflow wraps a 16-bit length to a large value that is passed to AEAD decryption routines, causing heap buffer overflow and a crash. An unauthenticated attacker can trigger this remotely via malformed TLS Application Data records.

In the OpenSSL compatibility layer implementation, the function RAND_poll() was not behaving as expected and leading to the potential for predictable values returned from RAND_bytes() after fork() is called. This can lead to weak or predictable random numbers generated in applications that are both using RAND_bytes() and doing fork() operations. This only affects applications explicitly calling RAND_bytes() after fork() and does not affect any internal TLS operations. Although RAND_bytes() documentation in OpenSSL calls out not being safe for use with fork() without first calling RAND_poll(), an additional code change was also made in wolfSSL to make RAND_bytes() behave similar to OpenSSL after a fork() call without calling RAND_poll(). Now the Hash-DRBG used gets reseeded after detecting running in a new process. If making use of RAND_bytes() and calling fork() we recommend updating to the latest version of wolfSSL. Thanks to Per Allansson from Appgate for the report.

In function MatchDomainName(), input param str is treated as a NULL terminated string despite being user provided and unchecked. Specifically, the function X509_check_host() takes in a pointer and length to check against, with no requirements that it be NULL terminated. If a caller was attempting to do a name check on a non-NULL terminated buffer, the code would read beyond the bounds of the input array until it found a NULL terminator.This issue affects wolfSSL: through 5.7.0.

A malicious TLS1.2 server can force a TLS1.3 client with downgrade capability to use a ciphersuite that it did not agree to and achieve a successful connection. This is because, aside from the extensions, the client was skipping fully parsing the server hello. https://doi.org/10.46586/tches.v2024.i1.457-500

An issue was discovered in wolfSSL before 5.7.0. A safe-error attack via Rowhammer, namely FAULT+PROBE, leads to ECDSA key disclosure. When WOLFSSL_CHECK_SIG_FAULTS is used in signing operations with private ECC keys, such as in server-side TLS connections, the connection is halted if any fault occurs. The success rate in a certain amount of connection requests can be processed via an advanced technique for ECDSA key recovery.

Generating the ECDSA nonce k samples a random number r and then truncates this randomness with a modular reduction mod n where n is the order of the elliptic curve. Meaning k = r mod n. The division used during the reduction estimates a factor q_e by dividing the upper two digits (a digit having e.g. a size of 8 byte) of r by the upper digit of n and then decrements q_e in a loop until it has the correct size. Observing the number of times q_e is decremented through a control-flow revealing side-channel reveals a bias in the most significant bits of k. Depending on the curve this is either a negligible bias or a significant bias large enough to reconstruct k with lattice reduction methods. For SECP160R1, e.g., we find a bias of 15 bits.

The side-channel protected T-Table implementation in wolfSSL up to version 5.6.5 protects against a side-channel attacker with cache-line resolution. In a controlled environment such as Intel SGX, an attacker can gain a per instruction sub-cache-line resolution allowing them to break the cache-line-level protection. For details on the attack refer to: https://doi.org/10.46586/tches.v2024.i1.457-500

Remotely executed SEGV and out of bounds read allows malicious packet sender to crash or cause an out of bounds read via sending a malformed packet with the correct length.

wolfSSL prior to 5.6.6 did not check that messages in one (D)TLS record do not span key boundaries. As a result, it was possible to combine (D)TLS messages using different keys into one (D)TLS record. The most extreme edge case is that, in (D)TLS 1.3, it was possible that an unencrypted (D)TLS 1.3 record from the server containing first a ServerHello message and then the rest of the first server flight would be accepted by a wolfSSL client. In (D)TLS 1.3 the handshake is encrypted after the ServerHello but a wolfSSL client would accept an unencrypted flight from the server. This does not compromise key negotiation and authentication so it is assigned a low severity rating.

In wolfSSL prior to 5.6.6, if callback functions are enabled (via the WOLFSSL_CALLBACKS flag), then a malicious TLS client or network attacker can trigger a buffer over-read on the heap of 5 bytes (WOLFSSL_CALLBACKS is only intended for debugging).

wolfSSL SP Math All RSA implementation is vulnerable to the Marvin Attack, new variation of a timing Bleichenbacher style attack, when built with the following options to configure: --enable-all CFLAGS="-DWOLFSSL_STATIC_RSA" The define “WOLFSSL_STATIC_RSA” enables static RSA cipher suites, which is not recommended, and has been disabled by default since wolfSSL 3.6.6.  Therefore the default build since 3.6.6, even with "--enable-all", is not vulnerable to the Marvin Attack. The vulnerability is specific to static RSA cipher suites, and expected to be padding-independent. The vulnerability allows an attacker to decrypt ciphertexts and forge signatures after probing with a large number of test observations. However the server’s private key is not exposed.

If a TLS 1.3 client gets neither a PSK (pre shared key) extension nor a KSE (key share extension) when connecting to a malicious server, a default predictable buffer gets used for the IKM (Input Keying Material) value when generating the session master secret. Using a potentially known IKM value when generating the session master secret key compromises the key generated, allowing an eavesdropper to reconstruct it and potentially allowing access to or meddling with message contents in the session. This issue does not affect client validation of connected servers, nor expose private key information, but could result in an insecure TLS 1.3 session when not controlling both sides of the connection. wolfSSL recommends that TLS 1.3 client side users update the version of wolfSSL used. 

An issue was discovered in wolfSSL before 5.5.0. A fault injection attack on RAM via Rowhammer leads to ECDSA key disclosure. Users performing signing operations with private ECC keys, such as in server-side TLS connections, might leak faulty ECC signatures. These signatures can be processed via an advanced technique for ECDSA key recovery. (In 5.5.0 and later, WOLFSSL_CHECK_SIG_FAULTS can be used to address the vulnerability.)

In wolfSSL before 5.5.2, if callback functions are enabled (via the WOLFSSL_CALLBACKS flag), then a malicious TLS 1.3 client or network attacker can trigger a buffer over-read on the heap of 5 bytes. (WOLFSSL_CALLBACKS is only intended for debugging.)

In wolfSSL before 5.5.1, malicious clients can cause a buffer overflow during a TLS 1.3 handshake. This occurs when an attacker supposedly resumes a previous TLS session. During the resumption Client Hello a Hello Retry Request must be triggered. Both Client Hellos are required to contain a list of duplicate cipher suites to trigger the buffer overflow. In total, two Client Hellos have to be sent: one in the resumed session, and a second one as a response to a Hello Retry Request message.

An issue was discovered in wolfSSL before 5.5.0. When a TLS 1.3 client connects to a wolfSSL server and SSL_clear is called on its session, the server crashes with a segmentation fault. This occurs in the second session, which is created through TLS session resumption and reuses the initial struct WOLFSSL. If the server reuses the previous session structure (struct WOLFSSL) by calling wolfSSL_clear(WOLFSSL* ssl) on it, the next received Client Hello (that resumes the previous session) crashes the server. Note that this bug is only triggered when resuming sessions using TLS session resumption. Only servers that use wolfSSL_clear instead of the recommended SSL_free; SSL_new sequence are affected. Furthermore, wolfSSL_clear is part of wolfSSL's compatibility layer and is not enabled by default. It is not part of wolfSSL's native API.

wolfSSL before 5.4.0 allows remote attackers to cause a denial of service via DTLS because a check for return-routability can be skipped.