mirror of https://gitlab.torproject.org/tpo/core/arti http://git.dilluti0n.com/mirrors/arti.git/atom?h=arti-v2.3.0 2026-04-28T16:27:56Z 2026-04-28T16:27:56Z Clara Engler cve@cve.cx 2026-04-28T08:09:13Z urn:sha1:67c4e02d595fed7ce3afc06e43f2577244b67ff4 If extra-infos is set to NULL, which might be the case for micro descriptors or even router descriptors because the field is optional there, this SQL query fails because it cannot LEFT JOIN server.extra_unsigned_sha1 when this field is NULL. To fix this, we simply add an additional clause to the WHERE statement that filters such rows out. 2026-04-28T16:27:44Z Clara Engler cve@cve.cx 2026-04-23T15:30:35Z urn:sha1:86f129e1ae01cac52912d3fbb1a0fef320d52ce7 This commit changes the schema to make the fingerprint optional. Reason for this is, that microdescriptors are also stored in this table and microdescriptors only contain an OPTIONAL onion-key, meaning it may not always be possible to determine this value from a microdescriptor, thereby making it silly to require it here. 2026-04-27T15:46:58Z Clara Engler cve@cve.cx 2026-04-14T09:01:32Z urn:sha1:6b26ebb430da64239c1698798b2ee1543129cba7 This commit makes use of the presence of only one descriptor hash inside consensus_router_descriptor_member by adding a respective AND clause to the select statements of missing descriptors to only select the rows with a non-null SHA-1 (or SHA-2 for the sake of microdescs) in the consensus_router_descriptor_member table. This is defensive programming and should not be required in practice because the docid as well as the "XOR" CHECK should already ensure that this value is always non-NULL. However, if it still happens and slips through, the statement would return NULL rows, which is not what we want and was previously prevented by ensuring this field was never NULL in the first place. 2026-04-27T15:46:58Z Clara Engler cve@cve.cx 2026-04-14T08:49:56Z urn:sha1:0b07bd3ee6a19af1025c02c52971a9071b937620 This commit modifies the consensus_router_descriptor_member table in the database schema, removing the NOT NULL constraint on unsigned_sha1 and unsigned_sha2 by replacing it with a new CHECK constraint that checks that either one of them is set but not both. The reason for this is as follows: We are going to use this table to compute the queue of missing descriptors, which means that we can only populate this table with the data we know from the consensus. The consensus however tells us only one of those hashes, namely sha1 in the case of a consensus-ns and sha2 in the case of a consensus-md. In other words: This commit can also be seen as an effort to change the design of the operation in such a way that the queue is obtained directly from the database and not computed at the start during state transition. 2026-04-21T14:52:33Z Nick Mathewson nickm@torproject.org 2026-04-19T20:06:44Z urn:sha1:80460f968db2c90d3d1dbc6a1592bff5e68f8de9 Rust 1.95 warns us that this is an experimental API that could be stabilized in the future. 2026-03-31T12:41:59Z Ian Jackson ijackson@chiark.greenend.org.uk 2026-03-30T10:13:28Z urn:sha1:c7a6c82f4511e002ee7b937c3881ff3f286a1120 This method verifies all the signatures, and checks that the signing authority is in the provided list. Anyway, authcerts aren't really self-signed: they're a signature by KS_auth_id_rsa on KP_auth_sign_rsa. Note that there is also a `verify_selfcert` method which does only some of the checks, and has some code duplication. That will be cleaned up later. 2026-03-24T18:49:16Z David Goulet dgoulet@torproject.org 2026-03-24T18:49:16Z urn:sha1:2346ec5dc6de3aba42a4cfd008fda0b6e878c2bf Force use of standard hasher with weak_tables. Closes #2418 See merge request tpo/core/arti!3801 2026-03-24T16:04:08Z Tobias Stoeckmann tobias@stoeckmann.org 2026-03-24T16:04:08Z urn:sha1:1fb2b8dd4aee7c943ce90d6631be473f2a296fe6 Typos found with codespell 2026-03-24T14:22:29Z Nick Mathewson nickm@torproject.org 2026-03-24T14:22:29Z urn:sha1:0a4cd08f3bd0a0cf78cffc31e05251628b86f243 Closes #2418. Fixes TROVE-2026-005, where we would use a less cryptographically secure (and probably less DoS resistant) hash function for these tables if: - We are built alongside another crate that uses `weak-table` - That crate enables the `weak-table/ahash` feature. - We are running on a system without hardware AES. Severity: Low 2026-03-19T10:37:47Z Ian Jackson ijackson@chiark.greenend.org.uk 2026-03-11T13:26:11Z urn:sha1:8667d8562ac40df206c5cd70d74ba58911d78671 We want to stop deriving NetdocParseable directly for body structs. Doing so reveals a call site here in tor-dirmirror where a consensus is parsed and the body data used, but without verifying the signatures. Do this explicitly with the hoop-jumping which is going to become deliberately unavoidable. Add a TODO comment because I'm not sure we have decided explicitloy that this is OK.