src/internal.c in wolfSSL wolfSSH through 1.5.0 admits the server-to-client Diffie-Hellman group exchange messages SSH_MSG_KEX_DH_GEX_GROUP (31) and SSH_MSG_KEX_DH_GEX_REPLY (33) when a server receives them from an unauthenticated client. IsMessageAllowedServer() applies no direction check to the key exchange message range: when the peer is keying and no particular message is expected, which is the state a server is in for the whole window after it processes the client's KEXINIT because nothing sets handshake->expectMsgId there, the function falls out of its expectation branch without a verdict and reaches a numeric bound that admits every message id from 30 through 34. A client that negotiates diffie-hellman-group-exchange-sha256 and then sends message 31 makes the server run the client-side handler DoKexDhGexGroup(), which validates the attacker-supplied group with two 8-round Miller-Rabin primality tests, one on p and one on (p-1)/2, on a value of up to 8192 bits. The handler then returns success: the server stores the attacker's prime and generator, generates a Diffie-Hellman key pair in the attacker's group, and sends the client-role message SSH_MSG_KEX_DH_GEX_INIT (32) back to the attacker. Published RFC 3526 safe primes are the worst-case input and cost the attacker nothing to obtain. The primality validation was added in 1.5.0; versions from 1.2.0 through 1.4.22 admit the same message and enter the same client-role path without the primality cost. Message 33 is admitted as well, but on a server it is rejected before any cryptography because no public key check callback is registered, so it carries no comparable cost. Builds that define WOLFSSH_NO_DH_GEX_SHA256, which is implied by WOLFSSH_NO_DH or NO_SHA256, are unaffected.
Unsigned integer underflow in wstrncat() in src/port.c in wolfSSL wolfSSH from v1.4.11 through v1.5.0 on non-Windows platforms allows an authenticated remote attacker to write one out-of-bounds null byte past the end of a stack buffer by sending a crafted SFTP path. wolfSSH_RealPath() in src/ssh.c appends each path component with a remaining-size bound (outSz - curSz) rather than the full destination size, so once the accumulated path reaches half the output buffer the size_t computation n - strlen(s1) - 1 wraps to near SIZE_MAX. The strncat() call is then effectively unbounded and copies the whole component; when that component exactly fills the remainder of the buffer, its terminating null is written one byte past the end. The caller's own length check keeps the copied data inside the buffer, so the overflow is limited to that single null byte, which may corrupt an adjacent stack value and crash the process. Applications that call the public wolfSSH_RealPath() with an output buffer smaller than the input path are additionally exposed to an unbounded copy, because the word32 expression outSz - segSz in that length check also wraps.
In wolfSSH through 1.5.0 built with --enable-fwd, DoChannelOpen() in src/internal.c gates only direct-tcpip channel opens with the forwarding policy callback. forwarded-tcpip opens are admitted without an authorization check and are not capped in number, allowing a malicious SSH peer to make an endpoint allocate unbounded per-channel buffers for forwarding channels the application never authorized. A client also does not check a forwarded-tcpip open against the forwards it registered with a tcpip-forward request, as RFC 4254 section 7.2 requires, so a malicious server can open forwarding channels for addresses and ports the client never asked it to forward.
wolfSSH does not validate that the ECDSA curve identifier in a KEXDH_REPLY host key blob matches the algorithm negotiated during key exchange. In ParseECCPubKey() (src/internal.c), the blob's algorithm string is used to derive the curve via NameToId/wcPrimeForId without checking against the negotiated ssh->handshake->pubKeyId, and the RFC 5656 curve identifier string is discarded via GetSkip() rather than compared. An active network man-in-the-middle attacker can substitute a host key blob containing a different ECDSA curve, causing the client to import the key on the wrong curve. Because the attacker controls the private key for the substituted curve, signature verification passes. Exploitation requires an active MitM position and a lax public key check callback (e.g., TOFU, algorithm-name-only check, or fingerprint match against the parsed key).
CVE-2026-81341 is a vulnerability in wolfSSL Inc.'s wolfEngine before version 1.4.1 where the AES-CCM nonce for TLS 1.2 and DTLS 1.2 records is incorrectly sourced from a constant value rather than the TLS sequence number. This causes reuse of the same key and nonce pair across multiple records, weakening confidentiality and integrity. The issue affects only wolfEngine with AES-CCM cipher suites, which are not enabled by default and must be explicitly selected. TLS 1.3 and non-TLS uses are not affected.
wolfEngine versions up to 1.4.0 have a vulnerability in their AES-GCM implementation for TLS 1.2 and DTLS 1.2. The explicit 8-byte AES-GCM nonce is generated once per connection and not incremented per record, causing reuse of the key and nonce pair. This nonce reuse leaks the keystream and the GHASH authentication key, enabling attackers to forge authentication tags. AES-CCM, TLS 1.3, and non-TLS cipher uses are not affected. The vulnerability has a high severity with a CVSS score of 7.4.
wolfProvider versions up to 1.2.1 have a vulnerability where the 8-byte explicit AES-GCM nonce is generated once per connection and never incremented per record. This causes every TLS 1.2 and DTLS 1.2 AES-GCM record within a connection to be encrypted with the same key and nonce pair, leading to keystream disclosure and authentication tag forgery. AES-CCM, TLS 1.3, and non-TLS cipher uses are not affected.
Stack Buffer Overflow in wc_HpkeLabeledExtract via Oversized ECH Config. A vulnerability existed in wolfSSL 5.8.4 ECH (Encrypted Client Hello) support, where a maliciously crafted ECH config could cause a stack buffer overflow on the client side, leading to potential remote execution and client program crash. This could be exploited by a malicious TLS server supporting ECH. Note that ECH is off by default, and is only enabled with enable-ech.
CVE-2026-3503 is a medium severity vulnerability in wolfSSL's wolfCrypt library version 5.8.2 affecting ARM Cortex-M microcontrollers. The flaw involves incorrect usage of seeds in the pseudo-random number generator (PRNG) within post-quantum cryptographic implementations (ML-KEM and ML-DSA). Physical attackers can induce transient faults to corrupt or redirect seed or pointer values during the Keccak-based expansion process, potentially compromising key material or cryptographic outcomes. Exploitation requires physical access and high attack complexity, with no user interaction needed. This vulnerability impacts confidentiality due to possible key leakage but has limited scope and availability impact. No known exploits are reported in the wild. Organizations using affected wolfSSL versions on ARM Cortex-M devices should prioritize patching or apply hardware-level protections against fault injection attacks.
Exporting a TPM based RSA key larger than 2048 bits from the TPM could overrun a stack buffer if the default `MAX_RSA_KEY_BITS=2048` is used. If your TPM 2.0 module supports RSA key sizes larger than 2048 bit and your applications supports creating or importing an RSA private or public key larger than 2048 bits and your application calls `wolfTPM2_RsaKey_TpmToWolf` on that key, then a stack buffer could be overrun. If the `MAX_RSA_KEY_BITS` build-time macro is set correctly (RSA bits match what TPM hardware is capable of) for the hardware target, then a stack overrun is not possible.