tor before 0.4.9.9 was prone to an out-of-bounds write when parsing a consensus or detached signature with unexpected signature digest type. Impact is minor for most Tor roles, but potentially major for directory authorities. This is TROVE-2026-019.

Tor before 0.4.9.7 has an out-of-bounds read when an END, a TRUNCATE, or a TRUNCATED cell lacks a reason in its payload, aka TROVE-2026-011.

Tor before 0.4.9.7 has an out-of-bounds read by one byte via a malformed BEGIN cell, aka TROVE-2026-007.

tor before 0.4.9.9 was prone to a NULL write after free when sending a CONFLUXSWITCH cell fails. The return value of relaysendcommandfromedge() was ignored, so a send failure (which calls circuitmarkforclose() and removes the leg via cfxdelleg()) would go undetected, causing the caller to write to the now-freed current leg and resulting in a crash. This is TROVE-2026-017.

Tor before 0.4.9.7 has a NULL pointer dereference when a CERT cell is received out of order, aka TROVE-2026-006.

Tor before 0.4.9.7, when circuit queue memory pressure exists, can experience a client crash because of a double close of a circuit, aka TROVE-2026-009.

A use-after-free vulnerability in SVG Animation has been discovered. An exploit built on this vulnerability has been discovered in the wild targeting Firefox and Tor Browser users on Windows.

Tor before 0.3.5.16, 0.4.5.10, and 0.4.6.7 mishandles the relationship between batch-signature verification and single-signature verification, leading to a remote assertion failure, aka TROVE-2021-007.

A Denial of Service issue was fixed in tor 0.4.7.8 related to congestion control. RTT estimation can become wedged, seriously degrading congestion control performance on all circuits. This impacts clients, onion services, and relays, and can be triggered remotely by a malicious endpoint.

References: https://gitlab.torproject.org/tpo/core/tor/-/issues/40626 https://lists.torproject.org/pipermail/tor-announce/2022-June/000242.html https://github.com/torproject/tor/commit/5a25374209689466e10906a77e66ad717a615a02 https://github.com/torproject/tor/commit/b0496d40197dd5b4fb7b694c1410082d4e34dda6

Tor before 0.4.9.7 mishandles accounting of the conflux out-of-order queue during the clearing of a queue, aka TROVE-2026-010.

Tor before 0.4.9.7 can attempt or accept BEGINDIR via conflux legs, aka TROVE-2026-008.

tor before 0.4.9.9 was prone to an infinite loop when decompressing a truncated zlib/gzip stream with done=1. A truncated stream never reaches ZSTREAMEND, causing zlib to return ZBUFERROR with no input remaining, which bufaddcompress() mistook for a full output buffer and retried forever. Fixed by returning TORCOMPRESSERROR in that case so the caller can abort cleanly. This is TROVE-2026-021.

Tor before 0.4.9.9 was prone to a compression bomb bypass where an attacker could concatenate many gzip or zlib sub-streams, each just under the per-stream detection threshold, to avoid the compression bomb check entirely. This is TROVE-2026-022.

DISPUTED The daemon in Tor through 0.4.1.8 and 0.4.2.x through 0.4.2.6 does not verify that a rendezvous node is known before attempting to connect to it, which might make it easier for remote attackers to discover circuit information. NOTE: The network team of Tor claims this is an intended behavior and not a vulnerability.

The comparetoraddrtoaddrpolicy function in or/policies.c in Tor before 0.2.2.39, and 0.2.3.x before 0.2.3.21-rc, allows remote attackers to cause a denial of service (assertion failure and daemon exit) via a zero-valued port field that is not properly handled during policy comparison.

The tortimegm function in common/util.c in Tor before 0.2.2.39, and 0.2.3.x before 0.2.3.22-rc, does not properly validate time values, which allows remote attackers to cause a denial of service (assertion failure and daemon exit) via a malformed directory object, a different vulnerability than CVE-2012-4419.

Tor before 0.2.4.20, when OpenSSL 1.x is used in conjunction with a certain HardwareAccel setting on Intel Sandy Bridge and Ivy Bridge platforms, does not properly generate random numbers for (1) relay identity keys and (2) hidden-service identity keys, which might make it easier for remote attackers to bypass cryptographic protection mechanisms via unspecified vectors.