Client operation

Every Tor that is not a directory server (that is, those that do not have a DirPort set) implements this section.

Downloading network-status documents

Each client maintains a list of directory authorities. Insofar as possible, clients SHOULD all use the same list.

  [Newer versions of Tor ( and later):
   Each client also maintains a list of default fallback directory mirrors
   (fallbacks). Each released version of Tor MAY have a different list,
   depending on the mirrors that satisfy the fallback directory criteria at
   release time.]

Clients try to have a live consensus network-status document at all times. A network-status document is "live" if the time in its valid-after field has passed, and the time in its valid-until field has not passed.

When a client has no consensus network-status document, it downloads it from a randomly chosen fallback directory mirror or authority. Clients prefer fallbacks to authorities, trying them earlier and more frequently. In all other cases, the client downloads from caches randomly chosen from among those believed to be V3 directory servers. (This information comes from the network-status documents.)

After receiving any response client MUST discard any network-status documents that it did not request.

On failure, the client waits briefly, then tries that network-status document again from another cache. The client does not build circuits until it has a live network-status consensus document, and it has descriptors for a significant proportion of the routers that it believes are running (this is configurable using torrc options and consensus parameters).

  [Newer versions of Tor ( and later):
   If the consensus contains Exits (the typical case), Tor will build both
   exit and internal circuits. When bootstrap completes, Tor will be ready
   to handle an application requesting an exit circuit to services like the
   World Wide Web.

If the consensus does not contain Exits, Tor will only build internal circuits. In this case, earlier statuses will have included "internal" as indicated above. When bootstrap completes, Tor will be ready to handle an application requesting an internal circuit to hidden services at ".onion" addresses.

If a future consensus contains Exits, exit circuits may become available.]

(Note: clients can and should pick caches based on the network-status information they have: once they have first fetched network-status info from an authority or fallback, they should not need to go to the authority directly again, and should only choose the fallback at random, based on its consensus weight in the current consensus.)

To avoid swarming the caches whenever a consensus expires, the clients download new consensuses at a randomly chosen time after the caches are expected to have a fresh consensus, but before their consensus will expire. (This time is chosen uniformly at random from the interval between the time 3/4 into the first interval after the consensus is no longer fresh, and 7/8 of the time remaining after that before the consensus is invalid.)

   [For example, if a client has a consensus that became valid at 1:00,
    and is fresh until 2:00, and expires at 4:00, that client will fetch
    a new consensus at a random time between 2:45 and 3:50, since 3/4
    of the one-hour interval is 45 minutes, and 7/8 of the remaining 75
    minutes is 65 minutes.]

Clients may choose to download the microdescriptor consensus instead of the general network status consensus. In that case they should use the same update strategy as for the normal consensus. They should not download more than one consensus flavor.

When a client does not have a live consensus, it will generally use the most recent consensus it has if that consensus is "reasonably live". A "reasonably live" consensus is one that expired less than 24 hours ago.

Downloading server descriptors or microdescriptors

Clients try to have the best descriptor for each router. A descriptor is "best" if:

  • It is listed in the consensus network-status document.

Periodically (currently every 10 seconds) clients check whether there are any "downloadable" descriptors. A descriptor is downloadable if:

      - It is the "best" descriptor for some router.
      - The descriptor was published at least 10 minutes in the past.
        (This prevents clients from trying to fetch descriptors that the
        mirrors have probably not yet retrieved and cached.)
      - The client does not currently have it.
      - The client is not currently trying to download it.
      - The client would not discard it immediately upon receiving it.
      - The client thinks it is running and valid (see section 5.4.1 below).

If at least 16 known routers have downloadable descriptors, or if enough time (currently 10 minutes) has passed since the last time the client tried to download descriptors, it launches requests for all downloadable descriptors.

When downloading multiple server descriptors, the client chooses multiple mirrors so that:

     - At least 3 different mirrors are used, except when this would result
       in more than one request for under 4 descriptors.
     - No more than 128 descriptors are requested from a single mirror.
     - Otherwise, as few mirrors as possible are used.
   After choosing mirrors, the client divides the descriptors among them

After receiving any response the client MUST discard any descriptors that it did not request.

When a descriptor download fails, the client notes it, and does not consider the descriptor downloadable again until a certain amount of time has passed. (Currently 0 seconds for the first failure, 60 seconds for the second, 5 minutes for the third, 10 minutes for the fourth, and 1 day thereafter.) Periodically (currently once an hour) clients reset the failure count.

Clients retain the most recent descriptor they have downloaded for each router so long as it is listed in the consensus. If it is not listed, they keep it so long as it is not too old (currently, ROUTER_MAX_AGE=48 hours) and no better router descriptor has been downloaded for the same relay. Caches retain descriptors until they are at least OLD_ROUTER_DESC_MAX_AGE=5 days old.

Clients which chose to download the microdescriptor consensus instead of the general consensus must download the referenced microdescriptors instead of server descriptors. Clients fetch and cache microdescriptors preemptively from dir mirrors when starting up, like they currently fetch descriptors. After bootstrapping, clients only need to fetch the microdescriptors that have changed.

When a client gets a new microdescriptor consensus, it looks to see if there are any microdescriptors it needs to learn, and launches a request for them.

Clients maintain a cache of microdescriptors along with metadata like when it was last referenced by a consensus, and which identity key it corresponds to. They keep a microdescriptor until it hasn't been mentioned in any consensus for a week. Future clients might cache them for longer or shorter times.

Downloading extra-info documents

Any client that uses extra-info documents should implement this section.

Note that generally, clients don't need extra-info documents.

Periodically, the Tor instance checks whether it is missing any extra-info documents: in other words, if it has any server descriptors with an extra-info-digest field that does not match any of the extra-info documents currently held. If so, it downloads whatever extra-info documents are missing. Clients try to download from caches. We follow the same splitting and back-off rules as in section 5.2.

Retrying failed downloads

This section applies to caches as well as to clients.

When a client fails to download a resource (a consensus, a router descriptor, a microdescriptor, etc) it waits for a certain amount of time before retrying the download. To determine the amount of time to wait, clients use a randomized exponential backoff algorithm. (Specifically, they use a variation of the "decorrelated jitter" algorithm from .)

The specific formula used to compute the 'i+1'th delay is:

        Delay_{i+1} = MIN(cap, random_between(lower_bound, upper_bound)))
          where upper_bound = MAX(lower_bound+1, Delay_i * 3)
                lower_bound = MAX(1, base_delay).

The value of 'cap' is set to INT_MAX; the value of 'base_delay' depends on what is being downloaded, whether the client is fully bootstrapped, how the client is configured, and where it is downloading from. Current base_delay values are:

   Consensus objects, as a non-bridge cache:
         0 (TestingServerConsensusDownloadInitialDelay)

   Consensus objects, as a client or bridge that has bootstrapped:
         0 (TestingClientConsensusDownloadInitialDelay)

   Consensus objects, as a client or bridge that is bootstrapping,
   when connecting to an authority because no "fallback" caches are
         0 (ClientBootstrapConsensusAuthorityOnlyDownloadInitialDelay)

   Consensus objects, as a client or bridge that is bootstrapping,
   when "fallback" caches are known but connecting to an authority
         6 (ClientBootstrapConsensusAuthorityDownloadInitialDelay)

   Consensus objects, as a client or bridge that is bootstrapping,
   when downloading from a "fallback" cache.
         0 (ClientBootstrapConsensusFallbackDownloadInitialDelay)

   Bridge descriptors, as a bridge-using client when at least one bridge
   is usable:
         10800 (TestingBridgeDownloadInitialDelay)

   Bridge descriptors, otherwise:
         0 (TestingBridgeBootstrapDownloadInitialDelay)

   Other objects, as cache or authority:
         0 (TestingServerDownloadInitialDelay)

   Other objects, as client:
         0 (TestingClientDownloadInitialDelay)