Remove some cargo-cult gcc hacks around tor_assert and predict_unlikely; instead...

[tor/rransom.git] / ReleaseNotes

This document summarizes new features and bugfixes in each stable release
of Tor. If you want to see more detailed descriptions of the changes in
each development snapshot, see the ChangeLog file.

Changes in version 0.2.0.32 - 2008-11-20
  Tor 0.2.0.32 fixes a major security problem in Debian and Ubuntu
  packages (and maybe other packages) noticed by Theo de Raadt, fixes
  a smaller security flaw that might allow an attacker to access local
  services, further improves hidden service performance, and fixes a
  variety of other issues.

  o Security fixes:
    - The "User" and "Group" config options did not clear the
      supplementary group entries for the Tor process. The "User" option
      is now more robust, and we now set the groups to the specified
      user's primary group. The "Group" option is now ignored. For more
      detailed logging on credential switching, set CREDENTIAL_LOG_LEVEL
      in common/compat.c to LOG_NOTICE or higher. Patch by Jacob Appelbaum
      and Steven Murdoch. Bugfix on 0.0.2pre14. Fixes bug 848 and 857.
    - The "ClientDNSRejectInternalAddresses" config option wasn't being
      consistently obeyed: if an exit relay refuses a stream because its
      exit policy doesn't allow it, we would remember what IP address
      the relay said the destination address resolves to, even if it's
      an internal IP address. Bugfix on 0.2.0.7-alpha; patch by rovv.

  o Major bugfixes:
    - Fix a DOS opportunity during the voting signature collection process
      at directory authorities. Spotted by rovv. Bugfix on 0.2.0.x.

  o Major bugfixes (hidden services):
    - When fetching v0 and v2 rendezvous service descriptors in parallel,
      we were failing the whole hidden service request when the v0
      descriptor fetch fails, even if the v2 fetch is still pending and
      might succeed. Similarly, if the last v2 fetch fails, we were
      failing the whole hidden service request even if a v0 fetch is
      still pending. Fixes bug 814. Bugfix on 0.2.0.10-alpha.
    - When extending a circuit to a hidden service directory to upload a
      rendezvous descriptor using a BEGIN_DIR cell, almost 1/6 of all
      requests failed, because the router descriptor has not been
      downloaded yet. In these cases, do not attempt to upload the
      rendezvous descriptor, but wait until the router descriptor is
      downloaded and retry. Likewise, do not attempt to fetch a rendezvous
      descriptor from a hidden service directory for which the router
      descriptor has not yet been downloaded. Fixes bug 767. Bugfix
      on 0.2.0.10-alpha.

  o Minor bugfixes:
    - Fix several infrequent memory leaks spotted by Coverity.
    - When testing for libevent functions, set the LDFLAGS variable
      correctly. Found by Riastradh.
    - Avoid a bug where the FastFirstHopPK 0 option would keep Tor from
      bootstrapping with tunneled directory connections. Bugfix on
      0.1.2.5-alpha. Fixes bug 797. Found by Erwin Lam.
    - When asked to connect to A.B.exit:80, if we don't know the IP for A
      and we know that server B rejects most-but-not all connections to
      port 80, we would previously reject the connection. Now, we assume
      the user knows what they were asking for. Fixes bug 752. Bugfix
      on 0.0.9rc5. Diagnosed by BarkerJr.
    - If we overrun our per-second write limits a little, count this as
      having used up our write allocation for the second, and choke
      outgoing directory writes. Previously, we had only counted this when
      we had met our limits precisely. Fixes bug 824. Patch from by rovv.
      Bugfix on 0.2.0.x (??).
    - Remove the old v2 directory authority 'lefkada' from the default
      list. It has been gone for many months.
    - Stop doing unaligned memory access that generated bus errors on
      sparc64. Bugfix on 0.2.0.10-alpha. Fixes bug 862.
    - Make USR2 log-level switch take effect immediately. Bugfix on
      0.1.2.8-beta.

  o Minor bugfixes (controller):
    - Make DNS resolved events into "CLOSED", not "FAILED". Bugfix on
      0.1.2.5-alpha. Fix by Robert Hogan. Resolves bug 807.


Changes in version 0.2.0.31 - 2008-09-03
  Tor 0.2.0.31 addresses two potential anonymity issues, starts to fix
  a big bug we're seeing where in rare cases traffic from one Tor stream
  gets mixed into another stream, and fixes a variety of smaller issues.

  o Major bugfixes:
    - Make sure that two circuits can never exist on the same connection
      with the same circuit ID, even if one is marked for close. This
      is conceivably a bugfix for bug 779. Bugfix on 0.1.0.4-rc.
    - Relays now reject risky extend cells: if the extend cell includes
      a digest of all zeroes, or asks to extend back to the relay that
      sent the extend cell, tear down the circuit. Ideas suggested
      by rovv.
    - If not enough of our entry guards are available so we add a new
      one, we might use the new one even if it overlapped with the
      current circuit's exit relay (or its family). Anonymity bugfix
      pointed out by rovv.

  o Minor bugfixes:
    - Recover 3-7 bytes that were wasted per memory chunk. Fixes bug
      794; bug spotted by rovv. Bugfix on 0.2.0.1-alpha.
    - Correctly detect the presence of the linux/netfilter_ipv4.h header
      when building against recent kernels. Bugfix on 0.1.2.1-alpha.
    - Pick size of default geoip filename string correctly on windows.
      Fixes bug 806. Bugfix on 0.2.0.30.
    - Make the autoconf script accept the obsolete --with-ssl-dir
      option as an alias for the actually-working --with-openssl-dir
      option. Fix the help documentation to recommend --with-openssl-dir.
      Based on a patch by "Dave". Bugfix on 0.2.0.1-alpha.
    - Disallow session resumption attempts during the renegotiation
      stage of the v2 handshake protocol. Clients should never be trying
      session resumption at this point, but apparently some did, in
      ways that caused the handshake to fail. Bug found by Geoff Goodell.
      Bugfix on 0.2.0.20-rc.
    - When using the TransPort option on OpenBSD, and using the User
      option to change UID and drop privileges, make sure to open
      /dev/pf before dropping privileges. Fixes bug 782. Patch from
      Christopher Davis. Bugfix on 0.1.2.1-alpha.
    - Try to attach connections immediately upon receiving a RENDEZVOUS2
      or RENDEZVOUS_ESTABLISHED cell. This can save a second or two
      on the client side when connecting to a hidden service. Bugfix
      on 0.0.6pre1. Found and fixed by Christian Wilms; resolves bug 743.
    - When closing an application-side connection because its circuit is
      getting torn down, generate the stream event correctly. Bugfix on
      0.1.2.x. Anonymous patch.


Changes in version 0.2.0.30 - 2008-07-15
  This new stable release switches to a more efficient directory
  distribution design, adds features to make connections to the Tor
  network harder to block, allows Tor to act as a DNS proxy, adds separate
  rate limiting for relayed traffic to make it easier for clients to
  become relays, fixes a variety of potential anonymity problems, and
  includes the usual huge pile of other features and bug fixes.

  o New v3 directory design:
    - Tor now uses a new way to learn about and distribute information
      about the network: the directory authorities vote on a common
      network status document rather than each publishing their own
      opinion. Now clients and caches download only one networkstatus
      document to bootstrap, rather than downloading one for each
      authority. Clients only download router descriptors listed in
      the consensus. Implements proposal 101; see doc/spec/dir-spec.txt
      for details.
    - Set up moria1, tor26, and dizum as v3 directory authorities
      in addition to being v2 authorities. Also add three new ones:
      ides (run by Mike Perry), gabelmoo (run by Karsten Loesing), and
      dannenberg (run by CCC).
    - Switch to multi-level keys for directory authorities: now their
      long-term identity key can be kept offline, and they periodically
      generate a new signing key. Clients fetch the "key certificates"
      to keep up to date on the right keys. Add a standalone tool
      "tor-gencert" to generate key certificates. Implements proposal 103.
    - Add a new V3AuthUseLegacyKey config option to make it easier for
      v3 authorities to change their identity keys if another bug like
      Debian's OpenSSL RNG flaw appears.
    - Authorities and caches fetch the v2 networkstatus documents
      less often, now that v3 is recommended.

  o Make Tor connections stand out less on the wire:
    - Use an improved TLS handshake designed by Steven Murdoch in proposal
      124, as revised in proposal 130. The new handshake is meant to
      be harder for censors to fingerprint, and it adds the ability
      to detect certain kinds of man-in-the-middle traffic analysis
      attacks. The new handshake format includes version negotiation for
      OR connections as described in proposal 105, which will allow us
      to improve Tor's link protocol more safely in the future.
    - Enable encrypted directory connections by default for non-relays,
      so censor tools that block Tor directory connections based on their
      plaintext patterns will no longer work. This means Tor works in
      certain censored countries by default again.
    - Stop including recognizeable strings in the commonname part of
      Tor's x509 certificates.

  o Implement bridge relays:
    - Bridge relays (or "bridges" for short) are Tor relays that aren't
      listed in the main Tor directory. Since there is no complete public
      list of them, even an ISP that is filtering connections to all the
      known Tor relays probably won't be able to block all the bridges.
      See doc/design-paper/blocking.pdf and proposal 125 for details.
    - New config option BridgeRelay that specifies you want to be a
      bridge relay rather than a normal relay. When BridgeRelay is set
      to 1, then a) you cache dir info even if your DirPort ins't on,
      and b) the default for PublishServerDescriptor is now "bridge"
      rather than "v2,v3".
    - New config option "UseBridges 1" for clients that want to use bridge
      relays instead of ordinary entry guards. Clients then specify
      bridge relays by adding "Bridge" lines to their config file. Users
      can learn about a bridge relay either manually through word of
      mouth, or by one of our rate-limited mechanisms for giving out
      bridge addresses without letting an attacker easily enumerate them
      all. See https://www.torproject.org/bridges for details.
    - Bridge relays behave like clients with respect to time intervals
      for downloading new v3 consensus documents -- otherwise they
      stand out. Bridge users now wait until the end of the interval,
      so their bridge relay will be sure to have a new consensus document.

  o Implement bridge directory authorities:
    - Bridge authorities are like normal directory authorities, except
      they don't serve a list of known bridges. Therefore users that know
      a bridge's fingerprint can fetch a relay descriptor for that bridge,
      including fetching updates e.g. if the bridge changes IP address,
      yet an attacker can't just fetch a list of all the bridges.
    - Set up Tonga as the default bridge directory authority.
    - Bridge authorities refuse to serve bridge descriptors or other
      bridge information over unencrypted connections (that is, when
      responding to direct DirPort requests rather than begin_dir cells.)
    - Bridge directory authorities do reachability testing on the
      bridges they know. They provide router status summaries to the
      controller via "getinfo ns/purpose/bridge", and also dump summaries
      to a file periodically, so we can keep internal stats about which
      bridges are functioning.
    - If bridge users set the UpdateBridgesFromAuthority config option,
      but the digest they ask for is a 404 on the bridge authority,
      they fall back to contacting the bridge directly.
    - Bridges always use begin_dir to publish their server descriptor to
      the bridge authority using an anonymous encrypted tunnel.
    - Early work on a "bridge community" design: if bridge authorities set
      the BridgePassword config option, they will serve a snapshot of
      known bridge routerstatuses from their DirPort to anybody who
      knows that password. Unset by default.
    - Tor now includes an IP-to-country GeoIP file, so bridge relays can
      report sanitized aggregated summaries in their extra-info documents
      privately to the bridge authority, listing which countries are
      able to reach them. We hope this mechanism will let us learn when
      certain countries start trying to block bridges.
    - Bridge authorities write bridge descriptors to disk, so they can
      reload them after a reboot. They can also export the descriptors
      to other programs, so we can distribute them to blocked users via
      the BridgeDB interface, e.g. via https://bridges.torproject.org/
      and bridges@torproject.org.

  o Tor can be a DNS proxy:
    - The new client-side DNS proxy feature replaces the need for
      dns-proxy-tor: Just set "DNSPort 9999", and Tor will now listen
      for DNS requests on port 9999, use the Tor network to resolve them
      anonymously, and send the reply back like a regular DNS server.
      The code still only implements a subset of DNS.
    - Add a new AutomapHostsOnResolve option: when it is enabled, any
      resolve request for hosts matching a given pattern causes Tor to
      generate an internal virtual address mapping for that host. This
      allows DNSPort to work sensibly with hidden service users. By
      default, .exit and .onion addresses are remapped; the list of
      patterns can be reconfigured with AutomapHostsSuffixes.
    - Add an "-F" option to tor-resolve to force a resolve for a .onion
      address. Thanks to the AutomapHostsOnResolve option, this is no
      longer a completely silly thing to do.

  o Major features (relay usability):
    - New config options RelayBandwidthRate and RelayBandwidthBurst:
      a separate set of token buckets for relayed traffic. Right now
      relayed traffic is defined as answers to directory requests, and
      OR connections that don't have any local circuits on them. See
      proposal 111 for details.
    - Create listener connections before we setuid to the configured
      User and Group. Now non-Windows users can choose port values
      under 1024, start Tor as root, and have Tor bind those ports
      before it changes to another UID. (Windows users could already
      pick these ports.)
    - Added a new ConstrainedSockets config option to set SO_SNDBUF and
      SO_RCVBUF on TCP sockets. Hopefully useful for Tor servers running
      on "vserver" accounts. Patch from coderman.

  o Major features (directory authorities):
    - Directory authorities track weighted fractional uptime and weighted
      mean-time-between failures for relays. WFU is suitable for deciding
      whether a node is "usually up", while MTBF is suitable for deciding
      whether a node is "likely to stay up." We need both, because
      "usually up" is a good requirement for guards, while "likely to
      stay up" is a good requirement for long-lived connections.
    - Directory authorities use a new formula for selecting which relays
      to advertise as Guards: they must be in the top 7/8 in terms of
      how long we have known about them, and above the median of those
      nodes in terms of weighted fractional uptime.
    - Directory authorities use a new formula for selecting which relays
      to advertise as Stable: when we have 4 or more days of data, use
      median measured MTBF rather than median declared uptime. Implements
      proposal 108.
    - Directory authorities accept and serve "extra info" documents for
      routers. Routers now publish their bandwidth-history lines in the
      extra-info docs rather than the main descriptor. This step saves
      60% (!) on compressed router descriptor downloads. Servers upload
      extra-info docs to any authority that accepts them; directory
      authorities now allow multiple router descriptors and/or extra
      info documents to be uploaded in a single go. Authorities, and
      caches that have been configured to download extra-info documents,
      download them as needed. Implements proposal 104.
    - Authorities now list relays who have the same nickname as
      a different named relay, but list them with a new flag:
      "Unnamed". Now we can make use of relays that happen to pick the
      same nickname as a server that registered two years ago and then
      disappeared. Implements proposal 122.
    - Store routers in a file called cached-descriptors instead of in
      cached-routers. Initialize cached-descriptors from cached-routers
      if the old format is around. The new format allows us to store
      annotations along with descriptors, to record the time we received
      each descriptor, its source, and its purpose: currently one of
      general, controller, or bridge.

  o Major features (other):
    - New config options WarnPlaintextPorts and RejectPlaintextPorts so
      Tor can warn and/or refuse connections to ports commonly used with
      vulnerable-plaintext protocols. Currently we warn on ports 23,
      109, 110, and 143, but we don't reject any. Based on proposal 129
      by Kevin Bauer and Damon McCoy.
    - Integrate Karsten Loesing's Google Summer of Code project to publish
      hidden service descriptors on a set of redundant relays that are a
      function of the hidden service address. Now we don't have to rely
      on three central hidden service authorities for publishing and
      fetching every hidden service descriptor. Implements proposal 114.
    - Allow tunnelled directory connections to ask for an encrypted
      "begin_dir" connection or an anonymized "uses a full Tor circuit"
      connection independently. Now we can make anonymized begin_dir
      connections for (e.g.) more secure hidden service posting and
      fetching.

  o Major bugfixes (crashes and assert failures):
    - Stop imposing an arbitrary maximum on the number of file descriptors
      used for busy servers. Bug reported by Olaf Selke; patch from
      Sebastian Hahn.
    - Avoid possible failures when generating a directory with routers
      with over-long versions strings, or too many flags set.
    - Fix a rare assert error when we're closing one of our threads:
      use a mutex to protect the list of logs, so we never write to the
      list as it's being freed. Fixes the very rare bug 575, which is
      kind of the revenge of bug 222.
    - Avoid segfault in the case where a badly behaved v2 versioning
      directory sends a signed networkstatus with missing client-versions.
    - When we hit an EOF on a log (probably because we're shutting down),
      don't try to remove the log from the list: just mark it as
      unusable. (Bulletproofs against bug 222.)

  o Major bugfixes (code security fixes):
    - Detect size overflow in zlib code. Reported by Justin Ferguson and
      Dan Kaminsky.
    - Rewrite directory tokenization code to never run off the end of
      a string. Fixes bug 455. Patch from croup.
    - Be more paranoid about overwriting sensitive memory on free(),
      as a defensive programming tactic to ensure forward secrecy.

  o Major bugfixes (anonymity fixes):
    - Reject requests for reverse-dns lookup of names that are in
      a private address space. Patch from lodger.
    - Never report that we've used more bandwidth than we're willing to
      relay: it leaks how much non-relay traffic we're using. Resolves
      bug 516.
    - As a client, do not believe any server that tells us that an
      address maps to an internal address space.
    - Warn about unsafe ControlPort configurations.
    - Directory authorities now call routers Fast if their bandwidth is
      at least 100KB/s, and consider their bandwidth adequate to be a
      Guard if it is at least 250KB/s, no matter the medians. This fix
      complements proposal 107.
    - Directory authorities now never mark more than 2 servers per IP as
      Valid and Running (or 5 on addresses shared by authorities).
      Implements proposal 109, by Kevin Bauer and Damon McCoy.
    - If we're a relay, avoid picking ourselves as an introduction point,
      a rendezvous point, or as the final hop for internal circuits. Bug
      reported by taranis and lodger.
    - Exit relays that are used as a client can now reach themselves
      using the .exit notation, rather than just launching an infinite
      pile of circuits. Fixes bug 641. Reported by Sebastian Hahn.
    - Fix a bug where, when we were choosing the 'end stream reason' to
      put in our relay end cell that we send to the exit relay, Tor
      clients on Windows were sometimes sending the wrong 'reason'. The
      anonymity problem is that exit relays may be able to guess whether
      the client is running Windows, thus helping partition the anonymity
      set. Down the road we should stop sending reasons to exit relays,
      or otherwise prevent future versions of this bug.
    - Only update guard status (usable / not usable) once we have
      enough directory information. This was causing us to discard all our
      guards on startup if we hadn't been running for a few weeks. Fixes
      bug 448.
    - When our directory information has been expired for a while, stop
      being willing to build circuits using it. Fixes bug 401.

  o Major bugfixes (peace of mind for relay operators)
    - Non-exit relays no longer answer "resolve" relay cells, so they
      can't be induced to do arbitrary DNS requests. (Tor clients already
      avoid using non-exit relays for resolve cells, but now servers
      enforce this too.) Fixes bug 619. Patch from lodger.
    - When we setconf ClientOnly to 1, close any current OR and Dir
      listeners. Reported by mwenge.

  o Major bugfixes (other):
    - If we only ever used Tor for hidden service lookups or posts, we
      would stop building circuits and start refusing connections after
      24 hours, since we falsely believed that Tor was dormant. Reported
      by nwf.
    - Add a new __HashedControlSessionPassword option for controllers
      to use for one-off session password hashes that shouldn't get
      saved to disk by SAVECONF --- Vidalia users were accumulating a
      pile of HashedControlPassword lines in their torrc files, one for
      each time they had restarted Tor and then clicked Save. Make Tor
      automatically convert "HashedControlPassword" to this new option but
      only when it's given on the command line. Partial fix for bug 586.
    - Patch from "Andrew S. Lists" to catch when we contact a directory
      mirror at IP address X and he says we look like we're coming from
      IP address X. Otherwise this would screw up our address detection.
    - Reject uploaded descriptors and extrainfo documents if they're
      huge. Otherwise we'll cache them all over the network and it'll
      clog everything up. Suggested by Aljosha Judmayer.
    - When a hidden service was trying to establish an introduction point,
      and Tor *did* manage to reuse one of the preemptively built
      circuits, it didn't correctly remember which one it used,
      so it asked for another one soon after, until there were no
      more preemptive circuits, at which point it launched one from
      scratch. Bugfix on 0.0.9.x.

  o Rate limiting and load balancing improvements:
    - When we add data to a write buffer in response to the data on that
      write buffer getting low because of a flush, do not consider the
      newly added data as a candidate for immediate flushing, but rather
      make it wait until the next round of writing. Otherwise, we flush
      and refill recursively, and a single greedy TLS connection can
      eat all of our bandwidth.
    - When counting the number of bytes written on a TLS connection,
      look at the BIO actually used for writing to the network, not
      at the BIO used (sometimes) to buffer data for the network.
      Looking at different BIOs could result in write counts on the
      order of ULONG_MAX. Fixes bug 614.
    - If we change our MaxAdvertisedBandwidth and then reload torrc,
      Tor won't realize it should publish a new relay descriptor. Fixes
      bug 688, reported by mfr.
    - Avoid using too little bandwidth when our clock skips a few seconds.
    - Choose which bridge to use proportional to its advertised bandwidth,
      rather than uniformly at random. This should speed up Tor for
      bridge users. Also do this for people who set StrictEntryNodes.

  o Bootstrapping faster and building circuits more intelligently:
    - Fix bug 660 that was preventing us from knowing that we should
      preemptively build circuits to handle expected directory requests.
    - When we're checking if we have enough dir info for each relay
      to begin establishing circuits, make sure that we actually have
      the descriptor listed in the consensus, not just any descriptor.
    - Correctly notify one-hop connections when a circuit build has
      failed. Possible fix for bug 669. Found by lodger.
    - Clients now hold circuitless TLS connections open for 1.5 times
      MaxCircuitDirtiness (15 minutes), since it is likely that they'll
      rebuild a new circuit over them within that timeframe. Previously,
      they held them open only for KeepalivePeriod (5 minutes).

  o Performance improvements (memory):
    - Add OpenBSD malloc code from "phk" as an optional malloc
      replacement on Linux: some glibc libraries do very poorly with
      Tor's memory allocation patterns. Pass --enable-openbsd-malloc to
      ./configure to get the replacement malloc code.
    - Switch our old ring buffer implementation for one more like that
      used by free Unix kernels. The wasted space in a buffer with 1mb
      of data will now be more like 8k than 1mb. The new implementation
      also avoids realloc();realloc(); patterns that can contribute to
      memory fragmentation.
    - Change the way that Tor buffers data that it is waiting to write.
      Instead of queueing data cells in an enormous ring buffer for each
      client->OR or OR->OR connection, we now queue cells on a separate
      queue for each circuit. This lets us use less slack memory, and
      will eventually let us be smarter about prioritizing different kinds
      of traffic.
    - Reference-count and share copies of address policy entries; only 5%
      of them were actually distinct.
    - Tune parameters for cell pool allocation to minimize amount of
      RAM overhead used.
    - Keep unused 4k and 16k buffers on free lists, rather than wasting 8k
      for every single inactive connection_t. Free items from the
      4k/16k-buffer free lists when they haven't been used for a while.
    - Make memory debugging information describe more about history
      of cell allocation, so we can help reduce our memory use.
    - Be even more aggressive about releasing RAM from small
      empty buffers. Thanks to our free-list code, this shouldn't be too
      performance-intensive.
    - Log malloc statistics from mallinfo() on platforms where it exists.
    - Use memory pools to allocate cells with better speed and memory
      efficiency, especially on platforms where malloc() is inefficient.
    - Add a --with-tcmalloc option to the configure script to link
      against tcmalloc (if present). Does not yet search for non-system
      include paths.

  o Performance improvements (socket management):
    - Count the number of open sockets separately from the number of
      active connection_t objects. This will let us avoid underusing
      our allocated connection limit.
    - We no longer use socket pairs to link an edge connection to an
      anonymous directory connection or a DirPort test connection.
      Instead, we track the link internally and transfer the data
      in-process. This saves two sockets per "linked" connection (at the
      client and at the server), and avoids the nasty Windows socketpair()
      workaround.
    - We were leaking a file descriptor if Tor started with a zero-length
      cached-descriptors file. Patch by "freddy77".

  o Performance improvements (CPU use):
    - Never walk through the list of logs if we know that no log target
      is interested in a given message.
    - Call routerlist_remove_old_routers() much less often. This should
      speed startup, especially on directory caches.
    - Base64 decoding was actually showing up on our profile when parsing
      the initial descriptor file; switch to an in-process all-at-once
      implementation that's about 3.5x times faster than calling out to
      OpenSSL.
    - Use a slightly simpler string hashing algorithm (copying Python's
      instead of Java's) and optimize our digest hashing algorithm to take
      advantage of 64-bit platforms and to remove some possibly-costly
      voodoo.
    - When implementing AES counter mode, update only the portions of the
      counter buffer that need to change, and don't keep separate
      network-order and host-order counters on big-endian hosts (where
      they are the same).
    - Add an in-place version of aes_crypt() so that we can avoid doing a
      needless memcpy() call on each cell payload.
    - Use Critical Sections rather than Mutexes for synchronizing threads
      on win32; Mutexes are heavier-weight, and designed for synchronizing
      between processes.

  o Performance improvements (bandwidth use):
    - Don't try to launch new descriptor downloads quite so often when we
      already have enough directory information to build circuits.
    - Version 1 directories are no longer generated in full. Instead,
      authorities generate and serve "stub" v1 directories that list
      no servers. This will stop Tor versions 0.1.0.x and earlier from
      working, but (for security reasons) nobody should be running those
      versions anyway.
    - Avoid going directly to the directory authorities even if you're a
      relay, if you haven't found yourself reachable yet or if you've
      decided not to advertise your dirport yet. Addresses bug 556.
    - If we've gone 12 hours since our last bandwidth check, and we
      estimate we have less than 50KB bandwidth capacity but we could
      handle more, do another bandwidth test.
    - Support "If-Modified-Since" when answering HTTP requests for
      directories, running-routers documents, and v2 and v3 networkstatus
      documents. (There's no need to support it for router descriptors,
      since those are downloaded by descriptor digest.)
    - Stop fetching directory info so aggressively if your DirPort is
      on but your ORPort is off; stop fetching v2 dir info entirely.
      You can override these choices with the new FetchDirInfoEarly
      config option.

  o Changed config option behavior (features):
    - Configuration files now accept C-style strings as values. This
      helps encode characters not allowed in the current configuration
      file format, such as newline or #. Addresses bug 557.
    - Add hidden services and DNSPorts to the list of things that make
      Tor accept that it has running ports. Change starting Tor with no
      ports from a fatal error to a warning; we might change it back if
      this turns out to confuse anybody. Fixes bug 579.
    - Make PublishServerDescriptor default to 1, so the default doesn't
      have to change as we invent new directory protocol versions.
    - Allow people to say PreferTunnelledDirConns rather than
      PreferTunneledDirConns, for those alternate-spellers out there.
    - Raise the default BandwidthRate/BandwidthBurst to 5MB/10MB, to
      accommodate the growing number of servers that use the default
      and are reaching it.
    - Make it possible to enable HashedControlPassword and
      CookieAuthentication at the same time.
    - When a TrackHostExits-chosen exit fails too many times in a row,
      stop using it. Fixes bug 437.

  o Changed config option behavior (bugfixes):
    - Do not read the configuration file when we've only been told to
      generate a password hash. Fixes bug 643. Bugfix on 0.0.9pre5. Fix
      based on patch from Sebastian Hahn.
    - Actually validate the options passed to AuthDirReject,
      AuthDirInvalid, AuthDirBadDir, and AuthDirBadExit.
    - Make "ClientOnly 1" config option disable directory ports too.
    - Don't stop fetching descriptors when FetchUselessDescriptors is
      set, even if we stop asking for circuits. Bug reported by tup
      and ioerror.
    - Servers used to decline to publish their DirPort if their
      BandwidthRate or MaxAdvertisedBandwidth were below a threshold. Now
      they look only at BandwidthRate and RelayBandwidthRate.
    - Treat "2gb" when given in torrc for a bandwidth as meaning 2gb,
      minus 1 byte: the actual maximum declared bandwidth.
    - Make "TrackHostExits ." actually work. Bugfix on 0.1.0.x.
    - Make the NodeFamilies config option work. (Reported by
      lodger -- it has never actually worked, even though we added it
      in Oct 2004.)
    - If Tor is invoked from something that isn't a shell (e.g. Vidalia),
      now we expand "-f ~/.tor/torrc" correctly. Suggested by Matt Edman.

  o New config options:
    - New configuration options AuthDirMaxServersPerAddr and
      AuthDirMaxServersperAuthAddr to override default maximum number
      of servers allowed on a single IP address. This is important for
      running a test network on a single host.
    - Three new config options (AlternateDirAuthority,
      AlternateBridgeAuthority, and AlternateHSAuthority) that let the
      user selectively replace the default directory authorities by type,
      rather than the all-or-nothing replacement that DirServer offers.
    - New config options AuthDirBadDir and AuthDirListBadDirs for
      authorities to mark certain relays as "bad directories" in the
      networkstatus documents. Also supports the "!baddir" directive in
      the approved-routers file.
    - New config option V2AuthoritativeDirectory that all v2 directory
      authorities must set. This lets v3 authorities choose not to serve
      v2 directory information.

  o Minor features (other):
    - When we're not serving v2 directory information, there is no reason
      to actually keep any around. Remove the obsolete files and directory
      on startup if they are very old and we aren't going to serve them.
    - When we negotiate a v2 link-layer connection (not yet implemented),
      accept RELAY_EARLY cells and turn them into RELAY cells if we've
      negotiated a v1 connection for their next step. Initial steps for
      proposal 110.
    - When we have no consensus, check FallbackNetworkstatusFile (defaults
      to $PREFIX/share/tor/fallback-consensus) for a consensus. This way
      we can start out knowing some directory caches. We don't ship with
      a fallback consensus by default though, because it was making
      bootstrapping take too long while we tried many down relays.
    - Authorities send back an X-Descriptor-Not-New header in response to
      an accepted-but-discarded descriptor upload. Partially implements
      fix for bug 535.
    - If we find a cached-routers file that's been sitting around for more
      than 28 days unmodified, then most likely it's a leftover from
      when we upgraded to 0.2.0.8-alpha. Remove it. It has no good
      routers anyway.
    - When we (as a cache) download a descriptor because it was listed
      in a consensus, remember when the consensus was supposed to expire,
      and don't expire the descriptor until then.
    - Optionally (if built with -DEXPORTMALLINFO) export the output
      of mallinfo via http, as tor/mallinfo.txt. Only accessible
      from localhost.
    - Tag every guard node in our state file with the version that
      we believe added it, or with our own version if we add it. This way,
      if a user temporarily runs an old version of Tor and then switches
      back to a new one, she doesn't automatically lose her guards.
    - When somebody requests a list of statuses or servers, and we have
      none of those, return a 404 rather than an empty 200.
    - Merge in some (as-yet-unused) IPv6 address manipulation code. (Patch
      from croup.)
    - Add an HSAuthorityRecordStats option that hidden service authorities
      can use to track statistics of overall hidden service usage without
      logging information that would be as useful to an attacker.
    - Allow multiple HiddenServicePort directives with the same virtual
      port; when they occur, the user is sent round-robin to one
      of the target ports chosen at random.  Partially fixes bug 393 by
      adding limited ad-hoc round-robining.
    - Revamp file-writing logic so we don't need to have the entire
      contents of a file in memory at once before we write to disk. Tor,
      meet stdio.

  o Minor bugfixes (other):
    - Alter the code that tries to recover from unhandled write
      errors, to not try to flush onto a socket that's given us
      unhandled errors.
    - Directory mirrors no longer include a guess at the client's IP
      address if the connection appears to be coming from the same /24
      network; it was producing too many wrong guesses.
    - If we're trying to flush the last bytes on a connection (for
      example, when answering a directory request), reset the
      time-to-give-up timeout every time we manage to write something
      on the socket.
    - Reject router descriptors with out-of-range bandwidthcapacity or
      bandwidthburst values.
    - If we can't expand our list of entry guards (e.g. because we're
      using bridges or we have StrictEntryNodes set), don't mark relays
      down when they fail a directory request. Otherwise we're too quick
      to mark all our entry points down.
    - Authorities no longer send back "400 you're unreachable please fix
      it" errors to Tor servers that aren't online all the time. We're
      supposed to tolerate these servers now.
    - Let directory authorities startup even when they can't generate
      a descriptor immediately, e.g. because they don't know their
      address.
    - Correctly enforce that elements of directory objects do not appear
      more often than they are allowed to appear.
    - Stop allowing hibernating servers to be "stable" or "fast".
    - On Windows, we were preventing other processes from reading
      cached-routers while Tor was running. (Reported by janbar)
    - Check return values from pthread_mutex functions.
    - When opening /dev/null in finish_daemonize(), do not pass the
      O_CREAT flag. Fortify was complaining, and correctly so. Fixes
      bug 742; fix from Michael Scherer. Bugfix on 0.0.2pre19.

  o Controller features:
    - The GETCONF command now escapes and quotes configuration values
      that don't otherwise fit into the torrc file.
    - The SETCONF command now handles quoted values correctly.
    - Add "GETINFO/desc-annotations/id/<OR digest>" so controllers can
      ask about source, timestamp of arrival, purpose, etc. We need
      something like this to help Vidalia not do GeoIP lookups on bridge
      addresses.
    - Allow multiple HashedControlPassword config lines, to support
      multiple controller passwords.
    - Accept LF instead of CRLF on controller, since some software has a
      hard time generating real Internet newlines.
    - Add GETINFO values for the server status events
      "REACHABILITY_SUCCEEDED" and "GOOD_SERVER_DESCRIPTOR". Patch from
      Robert Hogan.
    - There is now an ugly, temporary "desc/all-recent-extrainfo-hack"
      GETINFO for Torstat to use until it can switch to using extrainfos.
    - New config option CookieAuthFile to choose a new location for the
      cookie authentication file, and config option
      CookieAuthFileGroupReadable to make it group-readable.
    - Add a SOURCE_ADDR field to STREAM NEW events so that controllers can
      match requests to applications. Patch from Robert Hogan.
    - Add a RESOLVE command to launch hostname lookups. Original patch
      from Robert Hogan.
    - Add GETINFO status/enough-dir-info to let controllers tell whether
      Tor has downloaded sufficient directory information. Patch from Tup.
    - You can now use the ControlSocket option to tell Tor to listen for
      controller connections on Unix domain sockets on systems that
      support them. Patch from Peter Palfrader.
    - New "GETINFO address-mappings/*" command to get address mappings
      with expiry information. "addr-mappings/*" is now deprecated.
      Patch from Tup.
    - Add a new config option __DisablePredictedCircuits designed for
      use by the controller, when we don't want Tor to build any circuits
      preemptively.
    - Let the controller specify HOP=%d as an argument to ATTACHSTREAM,
      so we can exit from the middle of the circuit.
    - Implement "getinfo status/circuit-established".
    - Implement "getinfo status/version/..." so a controller can tell
      whether the current version is recommended, and whether any versions
      are good, and how many authorities agree. Patch from "shibz".
    - Controllers should now specify cache=no or cache=yes when using
      the +POSTDESCRIPTOR command.
    - Add a "PURPOSE=" argument to "STREAM NEW" events, as suggested by
      Robert Hogan. Fixes the first part of bug 681.
    - When reporting clock skew, and we know that the clock is _at least
      as skewed_ as some value, but we don't know the actual value,
      report the value as a "minimum skew."

  o Controller bugfixes:
    - Generate "STATUS_SERVER" events rather than misspelled
      "STATUS_SEVER" events. Caught by mwenge.
    - Reject controller commands over 1MB in length, so rogue
      processes can't run us out of memory.
    - Change the behavior of "getinfo status/good-server-descriptor"
      so it doesn't return failure when any authority disappears.
    - Send NAMESERVER_STATUS messages for a single failed nameserver
      correctly.
    - When the DANGEROUS_VERSION controller status event told us we're
      running an obsolete version, it used the string "OLD" to describe
      it. Yet the "getinfo" interface used the string "OBSOLETE". Now use
      "OBSOLETE" in both cases.
    - Respond to INT and TERM SIGNAL commands before we execute the
      signal, in case the signal shuts us down. We had a patch in
      0.1.2.1-alpha that tried to do this by queueing the response on
      the connection's buffer before shutting down, but that really
      isn't the same thing at all. Bug located by Matt Edman.
    - Provide DNS expiry times in GMT, not in local time. For backward
      compatibility, ADDRMAP events only provide GMT expiry in an extended
      field. "GETINFO address-mappings" always does the right thing.
    - Use CRLF line endings properly in NS events.
    - Make 'getinfo fingerprint' return a 551 error if we're not a
      server, so we match what the control spec claims we do. Reported
      by daejees.
    - Fix a typo in an error message when extendcircuit fails that
      caused us to not follow the \r\n-based delimiter protocol. Reported
      by daejees.
    - When tunneling an encrypted directory connection, and its first
      circuit fails, do not leave it unattached and ask the controller
      to deal. Fixes the second part of bug 681.
    - Treat some 403 responses from directory servers as INFO rather than
      WARN-severity events.

  o Portability / building / compiling:
    - When building with --enable-gcc-warnings, check for whether Apple's
      warning "-Wshorten-64-to-32" is available.
    - Support compilation to target iPhone; patch from cjacker huang.
      To build for iPhone, pass the --enable-iphone option to configure.
    - Detect non-ASCII platforms (if any still exist) and refuse to
      build there: some of our code assumes that 'A' is 65 and so on.
    - Clear up some MIPSPro compiler warnings.
    - Make autoconf search for libevent, openssl, and zlib consistently.
    - Update deprecated macros in configure.in.
    - When warning about missing headers, tell the user to let us
      know if the compile succeeds anyway, so we can downgrade the
      warning.
    - Include the current subversion revision as part of the version
      string: either fetch it directly if we're in an SVN checkout, do
      some magic to guess it if we're in an SVK checkout, or use
      the last-detected version if we're building from a .tar.gz.
      Use this version consistently in log messages.
    - Correctly report platform name on Windows 95 OSR2 and Windows 98 SE.
    - Read resolv.conf files correctly on platforms where read() returns
      partial results on small file reads.
    - Build without verbose warnings even on gcc 4.2 and 4.3.
    - On Windows, correctly detect errors when listing the contents of
      a directory. Fix from lodger.
    - Run 'make test' as part of 'make dist', so we stop releasing so
      many development snapshots that fail their unit tests.
    - Add support to detect Libevent versions in the 1.4.x series
      on mingw.
    - Add command-line arguments to unit-test executable so that we can
      invoke any chosen test from the command line rather than having
      to run the whole test suite at once; and so that we can turn on
      logging for the unit tests.
    - Do not automatically run configure from autogen.sh. This
      non-standard behavior tended to annoy people who have built other
      programs.
    - Fix a macro/CPP interaction that was confusing some compilers:
      some GCCs don't like #if/#endif pairs inside macro arguments.
      Fixes bug 707.
    - Fix macro collision between OpenSSL 0.9.8h and Windows headers.
      Fixes bug 704; fix from Steven Murdoch.
    - Correctly detect transparent proxy support on Linux hosts that
      require in.h to be included before netfilter_ipv4.h.  Patch
      from coderman.

  o Logging improvements:
    - When we haven't had any application requests lately, don't bother
      logging that we have expired a bunch of descriptors.
    - When attempting to open a logfile fails, tell us why.
    - Only log guard node status when guard node status has changed.
    - Downgrade the 3 most common "INFO" messages to "DEBUG". This will
      make "INFO" 75% less verbose.
    - When SafeLogging is disabled, log addresses along with all TLS
      errors.
    - Report TLS "zero return" case as a "clean close" and "IO error"
      as a "close". Stop calling closes "unexpected closes": existing
      Tors don't use SSL_close(), so having a connection close without
      the TLS shutdown handshake is hardly unexpected.
    - When we receive a consensus from the future, warn about skew.
    - Make "not enough dir info yet" warnings describe *why* Tor feels
      it doesn't have enough directory info yet.
    - On the USR1 signal, when dmalloc is in use, log the top 10 memory
      consumers. (We already do this on HUP.)
    - Give more descriptive well-formedness errors for out-of-range
      hidden service descriptor/protocol versions.
    - Stop recommending that every server operator send mail to tor-ops.
      Resolves bug 597. Bugfix on 0.1.2.x.
    - Improve skew reporting: try to give the user a better log message
      about how skewed they are, and how much this matters.
    - New --quiet command-line option to suppress the default console log.
      Good in combination with --hash-password.
    - Don't complain that "your server has not managed to confirm that its
      ports are reachable" if we haven't been able to build any circuits
      yet.
    - Detect the reason for failing to mmap a descriptor file we just
      wrote, and give a more useful log message.  Fixes bug 533.
    - Always prepend "Bug: " to any log message about a bug.
    - When dumping memory usage, list bytes used in buffer memory
      free-lists.
    - When running with dmalloc, dump more stats on hup and on exit.
    - Put a platform string (e.g. "Linux i686") in the startup log
      message, so when people paste just their logs, we know if it's
      OpenBSD or Windows or what.
    - When logging memory usage, break down memory used in buffers by
      buffer type.
    - When we are reporting the DirServer line we just parsed, we were
      logging the second stanza of the key fingerprint, not the first.
    - Even though Windows is equally happy with / and \ as path separators,
      try to use \ consistently on Windows and / consistently on Unix: it
      makes the log messages nicer.
     - On OSX, stop warning the user that kqueue support in libevent is
      "experimental", since it seems to have worked fine for ages.

  o Contributed scripts and tools:
    - Update linux-tor-prio.sh script to allow QoS based on the uid of
      the Tor process. Patch from Marco Bonetti with tweaks from Mike
      Perry.
    - Include the "tor-ctrl.sh" bash script by Stefan Behte to provide
      Unix users an easy way to script their Tor process (e.g. by
      adjusting bandwidth based on the time of the day).
    - In the exitlist script, only consider the most recently published
      server descriptor for each server. Also, when the user requests
      a list of servers that _reject_ connections to a given address,
      explicitly exclude the IPs that also have servers that accept
      connections to that address. Resolves bug 405.
    - Include a new contrib/tor-exit-notice.html file that exit relay
      operators can put on their website to help reduce abuse queries.

  o Newly deprecated features:
    - The status/version/num-versioning and status/version/num-concurring
      GETINFO controller options are no longer useful in the v3 directory
      protocol: treat them as deprecated, and warn when they're used.
    - The RedirectExits config option is now deprecated.

  o Removed features:
    - Drop the old code to choke directory connections when the
      corresponding OR connections got full: thanks to the cell queue
      feature, OR conns don't get full any more.
    - Remove the old "dns worker" server DNS code: it hasn't been default
      since 0.1.2.2-alpha, and all the servers are using the new
      eventdns code.
    - Remove the code to generate the oldest (v1) directory format.
    - Remove support for the old bw_accounting file: we've been storing
      bandwidth accounting information in the state file since
      0.1.2.5-alpha. This may result in bandwidth accounting errors
      if you try to upgrade from 0.1.1.x or earlier, or if you try to
      downgrade to 0.1.1.x or earlier.
    - Drop support for OpenSSL version 0.9.6. Just about nobody was using
      it, it had no AES, and it hasn't seen any security patches since
      2004.
    - Stop overloading the circuit_t.onionskin field for both "onionskin
      from a CREATE cell that we are waiting for a cpuworker to be
      assigned" and "onionskin from an EXTEND cell that we are going to
      send to an OR as soon as we are connected". Might help with bug 600.
    - Remove the tor_strpartition() function: its logic was confused,
      and it was only used for one thing that could be implemented far
      more easily.
    - Remove the contrib scripts ExerciseServer.py, PathDemo.py,
      and TorControl.py, as they use the old v0 controller protocol,
      and are obsoleted by TorFlow anyway.
    - Drop support for v1 rendezvous descriptors, since we never used
      them anyway, and the code has probably rotted by now. Based on
      patch from Karsten Loesing.
    - Stop allowing address masks that do not correspond to bit prefixes.
      We have warned about these for a really long time; now it's time
      to reject them. (Patch from croup.)
    - Remove an optimization in the AES counter-mode code that assumed
      that the counter never exceeded 2^68. When the counter can be set
      arbitrarily as an IV (as it is by Karsten's new hidden services
      code), this assumption no longer holds.
    - Disable the SETROUTERPURPOSE controller command: it is now
      obsolete.


Changes in version 0.1.2.19 - 2008-01-17
  Tor 0.1.2.19 fixes a huge memory leak on exit relays, makes the default
  exit policy a little bit more conservative so it's safer to run an
  exit relay on a home system, and fixes a variety of smaller issues.

  o Security fixes:
    - Exit policies now reject connections that are addressed to a
      relay's public (external) IP address too, unless
      ExitPolicyRejectPrivate is turned off. We do this because too
      many relays are running nearby to services that trust them based
      on network address.

  o Major bugfixes:
    - When the clock jumps forward a lot, do not allow the bandwidth
      buckets to become negative. Fixes bug 544.
    - Fix a memory leak on exit relays; we were leaking a cached_resolve_t
      on every successful resolve. Reported by Mike Perry.
    - Purge old entries from the "rephist" database and the hidden
      service descriptor database even when DirPort is zero.
    - Stop thinking that 0.1.2.x directory servers can handle "begin_dir"
      requests. Should ease bugs 406 and 419 where 0.1.2.x relays are
      crashing or mis-answering these requests.
    - When we decide to send a 503 response to a request for servers, do
      not then also send the server descriptors: this defeats the whole
      purpose. Fixes bug 539.

  o Minor bugfixes:
    - Changing the ExitPolicyRejectPrivate setting should cause us to
      rebuild our server descriptor.
    - Fix handling of hex nicknames when answering controller requests for
      networkstatus by name, or when deciding whether to warn about
      unknown routers in a config option. (Patch from mwenge.)
    - Fix a couple of hard-to-trigger autoconf problems that could result
      in really weird results on platforms whose sys/types.h files define
      nonstandard integer types.
    - Don't try to create the datadir when running --verify-config or
      --hash-password. Resolves bug 540.
    - If we were having problems getting a particular descriptor from the
      directory caches, and then we learned about a new descriptor for
      that router, we weren't resetting our failure count. Reported
      by lodger.
    - Although we fixed bug 539 (where servers would send HTTP status 503
      responses _and_ send a body too), there are still servers out there
      that haven't upgraded. Therefore, make clients parse such bodies
      when they receive them.
    - Run correctly on systems where rlim_t is larger than unsigned long.
      This includes some 64-bit systems.
    - Run correctly on platforms (like some versions of OS X 10.5) where
      the real limit for number of open files is OPEN_FILES, not rlim_max
      from getrlimit(RLIMIT_NOFILES).
    - Avoid a spurious free on base64 failure.
    - Avoid segfaults on certain complex invocations of
      router_get_by_hexdigest().
    - Fix rare bug on REDIRECTSTREAM control command when called with no
      port set: it could erroneously report an error when none had
      happened.


Changes in version 0.1.2.18 - 2007-10-28
  Tor 0.1.2.18 fixes many problems including crash bugs, problems with
  hidden service introduction that were causing huge delays, and a big
  bug that was causing some servers to disappear from the network status
  lists for a few hours each day.

  o Major bugfixes (crashes):
    - If a connection is shut down abruptly because of something that
      happened inside connection_flushed_some(), do not call
      connection_finished_flushing(). Should fix bug 451:
      "connection_stop_writing: Assertion conn->write_event failed"
      Bugfix on 0.1.2.7-alpha.
    - Fix possible segfaults in functions called from
      rend_process_relay_cell().

  o Major bugfixes (hidden services):
    - Hidden services were choosing introduction points uniquely by
      hexdigest, but when constructing the hidden service descriptor
      they merely wrote the (potentially ambiguous) nickname.
    - Clients now use the v2 intro format for hidden service
      connections: they specify their chosen rendezvous point by identity
      digest rather than by (potentially ambiguous) nickname. These
      changes could speed up hidden service connections dramatically.

  o Major bugfixes (other):
    - Stop publishing a new server descriptor just because we get a
      HUP signal. This led (in a roundabout way) to some servers getting
      dropped from the networkstatus lists for a few hours each day.
    - When looking for a circuit to cannibalize, consider family as well
      as identity. Fixes bug 438. Bugfix on 0.1.0.x (which introduced
      circuit cannibalization).
    - When a router wasn't listed in a new networkstatus, we were leaving
      the flags for that router alone -- meaning it remained Named,
      Running, etc -- even though absence from the networkstatus means
      that it shouldn't be considered to exist at all anymore. Now we
      clear all the flags for routers that fall out of the networkstatus
      consensus. Fixes bug 529.

  o Minor bugfixes:
    - Don't try to access (or alter) the state file when running
      --list-fingerprint or --verify-config or --hash-password. Resolves
      bug 499.
    - When generating information telling us how to extend to a given
      router, do not try to include the nickname if it is
      absent. Resolves bug 467.
    - Fix a user-triggerable segfault in expand_filename(). (There isn't
      a way to trigger this remotely.)
    - When sending a status event to the controller telling it that an
      OR address is reachable, set the port correctly. (Previously we
      were reporting the dir port.)
    - Fix a minor memory leak whenever a controller sends the PROTOCOLINFO
      command. Bugfix on 0.1.2.17.
    - When loading bandwidth history, do not believe any information in
      the future. Fixes bug 434.
    - When loading entry guard information, do not believe any information
      in the future.
    - When we have our clock set far in the future and generate an
      onion key, then re-set our clock to be correct, we should not stop
      the onion key from getting rotated.
    - On some platforms, accept() can return a broken address. Detect
      this more quietly, and deal accordingly. Fixes bug 483.
    - It's not actually an error to find a non-pending entry in the DNS
      cache when canceling a pending resolve. Don't log unless stuff
      is fishy. Resolves bug 463.
    - Don't reset trusted dir server list when we set a configuration
      option. Patch from Robert Hogan.


Changes in version 0.1.2.17 - 2007-08-30
  Tor 0.1.2.17 features a new Vidalia version in the Windows and OS
  X bundles. Vidalia 0.0.14 makes authentication required for the
  ControlPort in the default configuration, which addresses important
  security risks. Everybody who uses Vidalia (or another controller)
  should upgrade.

  In addition, this Tor update fixes major load balancing problems with
  path selection, which should speed things up a lot once many people
  have upgraded.

  o Major bugfixes (security):
    - We removed support for the old (v0) control protocol. It has been
      deprecated since Tor 0.1.1.1-alpha, and keeping it secure has
      become more of a headache than it's worth.

  o Major bugfixes (load balancing):
    - When choosing nodes for non-guard positions, weight guards
      proportionally less, since they already have enough load. Patch
      from Mike Perry.
    - Raise the "max believable bandwidth" from 1.5MB/s to 10MB/s. This
      will allow fast Tor servers to get more attention.
    - When we're upgrading from an old Tor version, forget our current
      guards and pick new ones according to the new weightings. These
      three load balancing patches could raise effective network capacity
      by a factor of four. Thanks to Mike Perry for measurements.

  o Major bugfixes (stream expiration):
    - Expire not-yet-successful application streams in all cases if
      they've been around longer than SocksTimeout. Right now there are
      some cases where the stream will live forever, demanding a new
      circuit every 15 seconds. Fixes bug 454; reported by lodger.

  o Minor features (controller):
    - Add a PROTOCOLINFO controller command. Like AUTHENTICATE, it
      is valid before any authentication has been received. It tells
      a controller what kind of authentication is expected, and what
      protocol is spoken. Implements proposal 119.

  o Minor bugfixes (performance):
    - Save on most routerlist_assert_ok() calls in routerlist.c, thus
      greatly speeding up loading cached-routers from disk on startup.
    - Disable sentinel-based debugging for buffer code: we squashed all
      the bugs that this was supposed to detect a long time ago, and now
      its only effect is to change our buffer sizes from nice powers of
      two (which platform mallocs tend to like) to values slightly over
      powers of two (which make some platform mallocs sad).

  o Minor bugfixes (misc):
    - If exit bandwidth ever exceeds one third of total bandwidth, then
      use the correct formula to weight exit nodes when choosing paths.
      Based on patch from Mike Perry.
    - Choose perfectly fairly among routers when choosing by bandwidth and
      weighting by fraction of bandwidth provided by exits. Previously, we
      would choose with only approximate fairness, and correct ourselves
      if we ran off the end of the list.
    - If we require CookieAuthentication but we fail to write the
      cookie file, we would warn but not exit, and end up in a state
      where no controller could authenticate. Now we exit.
    - If we require CookieAuthentication, stop generating a new cookie
      every time we change any piece of our config.
    - Refuse to start with certain directory authority keys, and
      encourage people using them to stop.
    - Terminate multi-line control events properly. Original patch
      from tup.
    - Fix a minor memory leak when we fail to find enough suitable
      servers to choose a circuit.
    - Stop leaking part of the descriptor when we run into a particularly
      unparseable piece of it.


Changes in version 0.1.2.16 - 2007-08-01
  Tor 0.1.2.16 fixes a critical security vulnerability that allows a
  remote attacker in certain situations to rewrite the user's torrc
  configuration file. This can completely compromise anonymity of users
  in most configurations, including those running the Vidalia bundles,
  TorK, etc. Or worse.

  o Major security fixes:
    - Close immediately after missing authentication on control port;
      do not allow multiple authentication attempts.


Changes in version 0.1.2.15 - 2007-07-17
  Tor 0.1.2.15 fixes several crash bugs, fixes some anonymity-related
  problems, fixes compilation on BSD, and fixes a variety of other
  bugs. Everybody should upgrade.

  o Major bugfixes (compilation):
    - Fix compile on FreeBSD/NetBSD/OpenBSD. Oops.

  o Major bugfixes (crashes):
    - Try even harder not to dereference the first character after
      an mmap(). Reported by lodger.
    - Fix a crash bug in directory authorities when we re-number the
      routerlist while inserting a new router.
    - When the cached-routers file is an even multiple of the page size,
      don't run off the end and crash. (Fixes bug 455; based on idea
      from croup.)
    - Fix eventdns.c behavior on Solaris: It is critical to include
      orconfig.h _before_ sys/types.h, so that we can get the expected
      definition of _FILE_OFFSET_BITS.

  o Major bugfixes (security):
    - Fix a possible buffer overrun when using BSD natd support. Bug
      found by croup.
    - When sending destroy cells from a circuit's origin, don't include
      the reason for tearing down the circuit. The spec says we didn't,
      and now we actually don't. Reported by lodger.
    - Keep streamids from different exits on a circuit separate. This
      bug may have allowed other routers on a given circuit to inject
      cells into streams. Reported by lodger; fixes bug 446.
    - If there's a never-before-connected-to guard node in our list,
      never choose any guards past it. This way we don't expand our
      guard list unless we need to.

  o Minor bugfixes (guard nodes):
    - Weight guard selection by bandwidth, so that low-bandwidth nodes
      don't get overused as guards.

  o Minor bugfixes (directory):
    - Correctly count the number of authorities that recommend each
      version. Previously, we were under-counting by 1.
    - Fix a potential crash bug when we load many server descriptors at
      once and some of them make others of them obsolete. Fixes bug 458.

  o Minor bugfixes (hidden services):
    - Stop tearing down the whole circuit when the user asks for a
      connection to a port that the hidden service didn't configure.
      Resolves bug 444.

  o Minor bugfixes (misc):
    - On Windows, we were preventing other processes from reading
      cached-routers while Tor was running. Reported by janbar.
    - Fix a possible (but very unlikely) bug in picking routers by
      bandwidth. Add a log message to confirm that it is in fact
      unlikely. Patch from lodger.
    - Backport a couple of memory leak fixes.
    - Backport miscellaneous cosmetic bugfixes.


Changes in version 0.1.2.14 - 2007-05-25
  Tor 0.1.2.14 changes the addresses of two directory authorities (this
  change especially affects those who serve or use hidden services),
  and fixes several other crash- and security-related bugs.

  o Directory authority changes:
    - Two directory authorities (moria1 and moria2) just moved to new
      IP addresses. This change will particularly affect those who serve
      or use hidden services.

  o Major bugfixes (crashes):
    - If a directory server runs out of space in the connection table
      as it's processing a begin_dir request, it will free the exit stream
      but leave it attached to the circuit, leading to unpredictable
      behavior. (Reported by seeess, fixes bug 425.)
    - Fix a bug in dirserv_remove_invalid() that would cause authorities
      to corrupt memory under some really unlikely scenarios.
    - Tighten router parsing rules. (Bugs reported by Benedikt Boss.)
    - Avoid segfaults when reading from mmaped descriptor file. (Reported
      by lodger.)

  o Major bugfixes (security):
    - When choosing an entry guard for a circuit, avoid using guards
      that are in the same family as the chosen exit -- not just guards
      that are exactly the chosen exit. (Reported by lodger.)

  o Major bugfixes (resource management):
    - If a directory authority is down, skip it when deciding where to get
      networkstatus objects or descriptors. Otherwise we keep asking
      every 10 seconds forever. Fixes bug 384.
    - Count it as a failure if we fetch a valid network-status but we
      don't want to keep it. Otherwise we'll keep fetching it and keep
      not wanting to keep it. Fixes part of bug 422.
    - If all of our dirservers have given us bad or no networkstatuses
      lately, then stop hammering them once per minute even when we
      think they're failed. Fixes another part of bug 422.

  o Minor bugfixes:
    - Actually set the purpose correctly for descriptors inserted with
      purpose=controller.
    - When we have k non-v2 authorities in our DirServer config,
      we ignored the last k authorities in the list when updating our
      network-statuses.
    - Correctly back-off from requesting router descriptors that we are
      having a hard time downloading.
    - Read resolv.conf files correctly on platforms where read() returns
      partial results on small file reads.
    - Don't rebuild the entire router store every time we get 32K of
      routers: rebuild it when the journal gets very large, or when
      the gaps in the store get very large.

  o Minor features:
    - When routers publish SVN revisions in their router descriptors,
      authorities now include those versions correctly in networkstatus
      documents.
    - Warn when using a version of libevent before 1.3b to run a server on
      OSX or BSD: these versions interact badly with userspace threads.


Changes in version 0.1.2.13 - 2007-04-24
  This release features some major anonymity fixes, such as safer path
  selection; better client performance; faster bootstrapping, better
  address detection, and better DNS support for servers; write limiting as
  well as read limiting to make servers easier to run; and a huge pile of
  other features and bug fixes. The bundles also ship with Vidalia 0.0.11.

  Tor 0.1.2.13 is released in memory of Rob Levin (1955-2006), aka lilo
  of the Freenode IRC network, remembering his patience and vision for
  free speech on the Internet.

  o Major features, client performance:
    - Weight directory requests by advertised bandwidth. Now we can
      let servers enable write limiting but still allow most clients to
      succeed at their directory requests. (We still ignore weights when
      choosing a directory authority; I hope this is a feature.)
    - Stop overloading exit nodes -- avoid choosing them for entry or
      middle hops when the total bandwidth available from non-exit nodes
      is much higher than the total bandwidth available from exit nodes.
    - Rather than waiting a fixed amount of time between retrying
      application connections, we wait only 10 seconds for the first,
      10 seconds for the second, and 15 seconds for each retry after
      that. Hopefully this will improve the expected user experience.
    - Sometimes we didn't bother sending a RELAY_END cell when an attempt
      to open a stream fails; now we do in more cases. This should
      make clients able to find a good exit faster in some cases, since
      unhandleable requests will now get an error rather than timing out.

  o Major features, client functionality:
    - Implement BEGIN_DIR cells, so we can connect to a directory
      server via TLS to do encrypted directory requests rather than
      plaintext. Enable via the TunnelDirConns and PreferTunneledDirConns
      config options if you like. For now, this feature only works if
      you already have a descriptor for the destination dirserver.
    - Add support for transparent application connections: this basically
      bundles the functionality of trans-proxy-tor into the Tor
      mainline. Now hosts with compliant pf/netfilter implementations
      can redirect TCP connections straight to Tor without diverting
      through SOCKS. (Based on patch from tup.)
    - Add support for using natd; this allows FreeBSDs earlier than
      5.1.2 to have ipfw send connections through Tor without using
      SOCKS. (Patch from Zajcev Evgeny with tweaks from tup.)

  o Major features, servers:
    - Setting up a dyndns name for your server is now optional: servers
      with no hostname or IP address will learn their IP address by
      asking the directory authorities. This code only kicks in when you
      would normally have exited with a "no address" error. Nothing's
      authenticated, so use with care.
    - Directory servers now spool server descriptors, v1 directories,
      and v2 networkstatus objects to buffers as needed rather than en
      masse. They also mmap the cached-routers files. These steps save
      lots of memory.
    - Stop requiring clients to have well-formed certificates, and stop
      checking nicknames in certificates. (Clients have certificates so
      that they can look like Tor servers, but in the future we might want
      to allow them to look like regular TLS clients instead. Nicknames
      in certificates serve no purpose other than making our protocol
      easier to recognize on the wire.) Implements proposal 106.

  o Improvements on DNS support:
    - Add "eventdns" asynchronous dns library originally based on code
      from Adam Langley. Now we can discard the old rickety dnsworker
      concept, and support a wider variety of DNS functions. Allows
      multithreaded builds on NetBSD and OpenBSD again.
    - Add server-side support for "reverse" DNS lookups (using PTR
      records so clients can determine the canonical hostname for a given
      IPv4 address). Only supported by servers using eventdns; servers
      now announce in their descriptors if they don't support eventdns.
    - Workaround for name servers (like Earthlink's) that hijack failing
      DNS requests and replace the no-such-server answer with a "helpful"
      redirect to an advertising-driven search portal. Also work around
      DNS hijackers who "helpfully" decline to hijack known-invalid
      RFC2606 addresses. Config option "ServerDNSDetectHijacking 0"
      lets you turn it off.
    - Servers now check for the case when common DNS requests are going to
      wildcarded addresses (i.e. all getting the same answer), and change
      their exit policy to reject *:* if it's happening.
    - When asked to resolve a hostname, don't use non-exit servers unless
      requested to do so. This allows servers with broken DNS to be
      useful to the network.
    - Start passing "ipv4" hints to getaddrinfo(), so servers don't do
      useless IPv6 DNS resolves.
    - Specify and implement client-side SOCKS5 interface for reverse DNS
      lookups (see doc/socks-extensions.txt). Also cache them.
    - When we change nameservers or IP addresses, reset and re-launch
      our tests for DNS hijacking.

  o Improvements on reachability testing:
    - Servers send out a burst of long-range padding cells once they've
      established that they're reachable. Spread them over 4 circuits,
      so hopefully a few will be fast. This exercises bandwidth and
      bootstraps them into the directory more quickly.
    - When we find our DirPort to be reachable, publish a new descriptor
      so we'll tell the world (reported by pnx).
    - Directory authorities now only decide that routers are reachable
      if their identity keys are as expected.
    - Do DirPort reachability tests less often, since a single test
      chews through many circuits before giving up.
    - Avoid some false positives during reachability testing: don't try
      to test via a server that's on the same /24 network as us.
    - Start publishing one minute or so after we find our ORPort
      to be reachable. This will help reduce the number of descriptors
      we have for ourselves floating around, since it's quite likely
      other things (e.g. DirPort) will change during that minute too.
    - Routers no longer try to rebuild long-term connections to directory
      authorities, and directory authorities no longer try to rebuild
      long-term connections to all servers. We still don't hang up
      connections in these two cases though -- we need to look at it
      more carefully to avoid flapping, and we likely need to wait til
      0.1.1.x is obsolete.

  o Improvements on rate limiting:
    - Enable write limiting as well as read limiting. Now we sacrifice
      capacity if we're pushing out lots of directory traffic, rather
      than overrunning the user's intended bandwidth limits.
    - Include TLS overhead when counting bandwidth usage; previously, we
      would count only the bytes sent over TLS, but not the bytes used
      to send them.
    - Servers decline directory requests much more aggressively when
      they're low on bandwidth. Otherwise they end up queueing more and
      more directory responses, which can't be good for latency.
    - But never refuse directory requests from local addresses.
    - Be willing to read or write on local connections (e.g. controller
      connections) even when the global rate limiting buckets are empty.
    - Flush local controller connection buffers periodically as we're
      writing to them, so we avoid queueing 4+ megabytes of data before
      trying to flush.
    - Revise and clean up the torrc.sample that we ship with; add
      a section for BandwidthRate and BandwidthBurst.

  o Major features, NT services:
    - Install as NT_AUTHORITY\LocalService rather than as SYSTEM; add a
      command-line flag so that admins can override the default by saying
      "tor --service install --user "SomeUser"". This will not affect
      existing installed services. Also, warn the user that the service
      will look for its configuration file in the service user's
      %appdata% directory. (We can't do the "hardwire the user's appdata
      directory" trick any more, since we may not have read access to that
      directory.)
    - Support running the Tor service with a torrc not in the same
      directory as tor.exe and default to using the torrc located in
      the %appdata%\Tor\ of the user who installed the service. Patch
      from Matt Edman.
    - Add an --ignore-missing-torrc command-line option so that we can
      get the "use sensible defaults if the configuration file doesn't
      exist" behavior even when specifying a torrc location on the
      command line.
    - When stopping an NT service, wait up to 10 sec for it to actually
      stop. (Patch from Matt Edman; resolves bug 295.)

  o Directory authority improvements:
    - Stop letting hibernating or obsolete servers affect uptime and
      bandwidth cutoffs.
    - Stop listing hibernating servers in the v1 directory.
    - Authorities no longer recommend exits as guards if this would shift
      too much load to the exit nodes.
    - Authorities now specify server versions in networkstatus. This adds
      about 2% to the size of compressed networkstatus docs, and allows
      clients to tell which servers support BEGIN_DIR and which don't.
      The implementation is forward-compatible with a proposed future
      protocol version scheme not tied to Tor versions.
    - DirServer configuration lines now have an orport= option so
      clients can open encrypted tunnels to the authorities without
      having downloaded their descriptors yet. Enabled for moria1,
      moria2, tor26, and lefkada now in the default configuration.
    - Add a BadDirectory flag to network status docs so that authorities
      can (eventually) tell clients about caches they believe to be
      broken. Not used yet.
    - Allow authorities to list nodes as bad exits in their
      approved-routers file by fingerprint or by address. If most
      authorities set a BadExit flag for a server, clients don't think
      of it as a general-purpose exit. Clients only consider authorities
      that advertise themselves as listing bad exits.
    - Patch from Steve Hildrey: Generate network status correctly on
      non-versioning dirservers.
    - Have directory authorities allow larger amounts of drift in uptime
      without replacing the server descriptor: previously, a server that
      restarted every 30 minutes could have 48 "interesting" descriptors
      per day.
    - Reserve the nickname "Unnamed" for routers that can't pick
      a hostname: any router can call itself Unnamed; directory
      authorities will never allocate Unnamed to any particular router;
      clients won't believe that any router is the canonical Unnamed.

  o Directory mirrors and clients:
    - Discard any v1 directory info that's over 1 month old (for
      directories) or over 1 week old (for running-routers lists).
    - Clients track responses with status 503 from dirservers. After a
      dirserver has given us a 503, we try not to use it until an hour has
      gone by, or until we have no dirservers that haven't given us a 503.
    - When we get a 503 from a directory, and we're not a server, we no
      longer count the failure against the total number of failures
      allowed for the object we're trying to download.
    - Prepare for servers to publish descriptors less often: never
      discard a descriptor simply for being too old until either it is
      recommended by no authorities, or until we get a better one for
      the same router. Make caches consider retaining old recommended
      routers for even longer.
    - Directory servers now provide 'Pragma: no-cache' and 'Expires'
      headers for content, so that we can work better in the presence of
      caching HTTP proxies.
    - Stop fetching descriptors if you're not a dir mirror and you
      haven't tried to establish any circuits lately. (This currently
      causes some dangerous behavior, because when you start up again
      you'll use your ancient server descriptors.)

  o Major fixes, crashes:
    - Stop crashing when the controller asks us to resetconf more than
      one config option at once. (Vidalia 0.0.11 does this.)
    - Fix a longstanding obscure crash bug that could occur when we run
      out of DNS worker processes, if we're not using eventdns. (Resolves
      bug 390.)
    - Fix an assert that could trigger if a controller quickly set then
      cleared EntryNodes. (Bug found by Udo van den Heuvel.)
    - Avoid crash when telling controller about stream-status and a
      stream is detached.
    - Avoid sending junk to controllers or segfaulting when a controller
      uses EVENT_NEW_DESC with verbose nicknames.
    - Stop triggering asserts if the controller tries to extend hidden
      service circuits (reported by mwenge).
    - If we start a server with ClientOnly 1, then set ClientOnly to 0
      and hup, stop triggering an assert based on an empty onion_key.
    - Mask out all signals in sub-threads; only the libevent signal
      handler should be processing them. This should prevent some crashes
      on some machines using pthreads. (Patch from coderman.)
    - Disable kqueue on OS X 10.3 and earlier, to fix bug 371.

  o Major fixes, anonymity/security:
    - Automatically avoid picking more than one node from the same
      /16 network when constructing a circuit. Add an
      "EnforceDistinctSubnets" option to let people disable it if they
      want to operate private test networks on a single subnet.
    - When generating bandwidth history, round down to the nearest
      1k. When storing accounting data, round up to the nearest 1k.
    - When we're running as a server, remember when we last rotated onion
      keys, so that we will rotate keys once they're a week old even if
      we never stay up for a week ourselves.
    - If a client asked for a server by name, and there's a named server
      in our network-status but we don't have its descriptor yet, we
      could return an unnamed server instead.
    - Reject (most) attempts to use Tor circuits with length one. (If
      many people start using Tor as a one-hop proxy, exit nodes become
      a more attractive target for compromise.)
    - Just because your DirPort is open doesn't mean people should be
      able to remotely teach you about hidden service descriptors. Now
      only accept rendezvous posts if you've got HSAuthoritativeDir set.
    - Fix a potential race condition in the rpm installer. Found by
      Stefan Nordhausen.
    - Do not log IPs with TLS failures for incoming TLS
      connections. (Fixes bug 382.)

  o Major fixes, other:
    - If our system clock jumps back in time, don't publish a negative
      uptime in the descriptor.
    - When we start during an accounting interval before it's time to wake
      up, remember to wake up at the correct time. (May fix bug 342.)
    - Previously, we would cache up to 16 old networkstatus documents
      indefinitely, if they came from nontrusted authorities. Now we
      discard them if they are more than 10 days old.
    - When we have a state file we cannot parse, tell the user and
      move it aside. Now we avoid situations where the user starts
      Tor in 1904, Tor writes a state file with that timestamp in it,
      the user fixes her clock, and Tor refuses to start.
    - Publish a new descriptor after we hup/reload. This is important
      if our config has changed such that we'll want to start advertising
      our DirPort now, etc.
    - If we are using an exit enclave and we can't connect, e.g. because
      its webserver is misconfigured to not listen on localhost, then
      back off and try connecting from somewhere else before we fail.

  o New config options or behaviors:
    - When EntryNodes are configured, rebuild the guard list to contain,
      in order: the EntryNodes that were guards before; the rest of the
      EntryNodes; the nodes that were guards before.
    - Do not warn when individual nodes in the configuration's EntryNodes,
      ExitNodes, etc are down: warn only when all possible nodes
      are down. (Fixes bug 348.)
    - Put a lower-bound on MaxAdvertisedBandwidth.
    - Start using the state file to store bandwidth accounting data:
      the bw_accounting file is now obsolete. We'll keep generating it
      for a while for people who are still using 0.1.2.4-alpha.
    - Try to batch changes to the state file so that we do as few
      disk writes as possible while still storing important things in
      a timely fashion.
    - The state file and the bw_accounting file get saved less often when
      the AvoidDiskWrites config option is set.
    - Make PIDFile work on Windows.
    - Add internal descriptions for a bunch of configuration options:
      accessible via controller interface and in comments in saved
      options files.
    - Reject *:563 (NNTPS) in the default exit policy. We already reject
      NNTP by default, so this seems like a sensible addition.
    - Clients now reject hostnames with invalid characters. This should
      avoid some inadvertent info leaks. Add an option
      AllowNonRFC953Hostnames to disable this behavior, in case somebody
      is running a private network with hosts called @, !, and #.
    - Check for addresses with invalid characters at the exit as well,
      and warn less verbosely when they fail. You can override this by
      setting ServerDNSAllowNonRFC953Addresses to 1.
    - Remove some options that have been deprecated since at least
      0.1.0.x: AccountingMaxKB, LogFile, DebugLogFile, LogLevel, and
      SysLog. Use AccountingMax instead of AccountingMaxKB, and use Log
      to set log options. Mark PathlenCoinWeight as obsolete.
    - Stop accepting certain malformed ports in configured exit policies.
    - When the user uses bad syntax in the Log config line, stop
      suggesting other bad syntax as a replacement.
    - Add new config option "ResolvConf" to let the server operator
      choose an alternate resolve.conf file when using eventdns.
    - If one of our entry guards is on the ExcludeNodes list, or the
      directory authorities don't think it's a good guard, treat it as
      if it were unlisted: stop using it as a guard, and throw it off
      the guards list if it stays that way for a long time.
    - Allow directory authorities to be marked separately as authorities
      for the v1 directory protocol, the v2 directory protocol, and
      as hidden service directories, to make it easier to retire old
      authorities. V1 authorities should set "HSAuthoritativeDir 1"
      to continue being hidden service authorities too.
    - Remove 8888 as a LongLivedPort, and add 6697 (IRCS).
    - Make TrackExitHosts case-insensitive, and fix the behavior of
      ".suffix" TrackExitHosts items to avoid matching in the middle of
      an address.
    - New DirPort behavior: if you have your dirport set, you download
      descriptors aggressively like a directory mirror, whether or not
      your ORPort is set.

  o Docs:
    - Create a new file ReleaseNotes which was the old ChangeLog. The
      new ChangeLog file now includes the notes for all development
      versions too.
    - Add a new address-spec.txt document to describe our special-case
      addresses: .exit, .onion, and .noconnnect.
    - Fork the v1 directory protocol into its own spec document,
      and mark dir-spec.txt as the currently correct (v2) spec.

  o Packaging, porting, and contrib
    - "tor --verify-config" now exits with -1(255) or 0 depending on
      whether the config options are bad or good.
    - The Debian package now uses --verify-config when (re)starting,
      to distinguish configuration errors from other errors.
    - Adapt a patch from goodell to let the contrib/exitlist script
      take arguments rather than require direct editing.
    - Prevent the contrib/exitlist script from printing the same
      result more than once.
    - Add support to tor-resolve tool for reverse lookups and SOCKS5.
    - In the hidden service example in torrc.sample, stop recommending
      esoteric and discouraged hidden service options.
    - Patch from Michael Mohr to contrib/cross.sh, so it checks more
      values before failing, and always enables eventdns.
    - Try to detect Windows correctly when cross-compiling.
    - Libevent-1.2 exports, but does not define in its headers, strlcpy.
      Try to fix this in configure.in by checking for most functions
      before we check for libevent.
    - Update RPMs to require libevent 1.2.
    - Experimentally re-enable kqueue on OSX when using libevent 1.1b
      or later. Log when we are doing this, so we can diagnose it when
      it fails. (Also, recommend libevent 1.1b for kqueue and
      win32 methods; deprecate libevent 1.0b harder; make libevent
      recommendation system saner.)
    - Build with recent (1.3+) libevents on platforms that do not
      define the nonstandard types "u_int8_t" and friends.
    - Remove architecture from OS X builds. The official builds are
      now universal binaries.
    - Run correctly on OS X platforms with case-sensitive filesystems.
    - Correctly set maximum connection limit on Cygwin. (This time
      for sure!)
    - Start compiling on MinGW on Windows (patches from Mike Chiussi
      and many others).
    - Start compiling on MSVC6 on Windows (patches from Frediano Ziglio).
    - Finally fix the openssl warnings from newer gccs that believe that
      ignoring a return value is okay, but casting a return value and
      then ignoring it is a sign of madness.
    - On architectures where sizeof(int)>4, still clamp declarable
      bandwidth to INT32_MAX.

  o Minor features, controller:
    - Warn the user when an application uses the obsolete binary v0
      control protocol. We're planning to remove support for it during
      the next development series, so it's good to give people some
      advance warning.
    - Add STREAM_BW events to report per-entry-stream bandwidth
      use. (Patch from Robert Hogan.)
    - Rate-limit SIGNEWNYM signals in response to controllers that
      impolitely generate them for every single stream. (Patch from
      mwenge; closes bug 394.)
    - Add a REMAP status to stream events to note that a stream's
      address has changed because of a cached address or a MapAddress
      directive.
    - Make REMAP stream events have a SOURCE (cache or exit), and
      make them generated in every case where we get a successful
      connected or resolved cell.
    - Track reasons for OR connection failure; make these reasons
      available via the controller interface. (Patch from Mike Perry.)
    - Add a SOCKS_BAD_HOSTNAME client status event so controllers
      can learn when clients are sending malformed hostnames to Tor.
    - Specify and implement some of the controller status events.
    - Have GETINFO dir/status/* work on hosts with DirPort disabled.
    - Reimplement GETINFO so that info/names stays in sync with the
      actual keys.
    - Implement "GETINFO fingerprint".
    - Implement "SETEVENTS GUARD" so controllers can get updates on
      entry guard status as it changes.
    - Make all connections to addresses of the form ".noconnect"
      immediately get closed. This lets application/controller combos
      successfully test whether they're talking to the same Tor by
      watching for STREAM events.
    - Add a REASON field to CIRC events; for backward compatibility, this
      field is sent only to controllers that have enabled the extended
      event format. Also, add additional reason codes to explain why
      a given circuit has been destroyed or truncated. (Patches from
      Mike Perry)
    - Add a REMOTE_REASON field to extended CIRC events to tell the
      controller why a remote OR told us to close a circuit.
    - Stream events also now have REASON and REMOTE_REASON fields,
      working much like those for circuit events.
    - There's now a GETINFO ns/... field so that controllers can ask Tor
      about the current status of a router.
    - A new event type "NS" to inform a controller when our opinion of
      a router's status has changed.
    - Add a GETINFO events/names and GETINFO features/names so controllers
      can tell which events and features are supported.
    - A new CLEARDNSCACHE signal to allow controllers to clear the
      client-side DNS cache without expiring circuits.
    - Fix CIRC controller events so that controllers can learn the
      identity digests of non-Named servers used in circuit paths.
    - Let controllers ask for more useful identifiers for servers. Instead
      of learning identity digests for un-Named servers and nicknames
      for Named servers, the new identifiers include digest, nickname,
      and indication of Named status. Off by default; see control-spec.txt
      for more information.
    - Add a "getinfo address" controller command so it can display Tor's
      best guess to the user.
    - New controller event to alert the controller when our server
      descriptor has changed.
    - Give more meaningful errors on controller authentication failure.
    - Export the default exit policy via the control port, so controllers
      don't need to guess what it is / will be later.

  o Minor bugfixes, controller:
    - When creating a circuit via the controller, send a 'launched'
      event when we're done, so we follow the spec better.
    - Correct the control spec to match how the code actually responds
      to 'getinfo addr-mappings/*'. Reported by daejees.
    - The control spec described a GUARDS event, but the code
      implemented a GUARD event. Standardize on GUARD, but let people
      ask for GUARDS too. Reported by daejees.
    - Give the controller END_STREAM_REASON_DESTROY events _before_ we
      clear the corresponding on_circuit variable, and remember later
      that we don't need to send a redundant CLOSED event. (Resolves part
      3 of bug 367.)
    - Report events where a resolve succeeded or where we got a socks
      protocol error correctly, rather than calling both of them
      "INTERNAL".
    - Change reported stream target addresses to IP consistently when
      we finally get the IP from an exit node.
    - Send log messages to the controller even if they happen to be very
      long.
    - Flush ERR-level controller status events just like we currently
      flush ERR-level log events, so that a Tor shutdown doesn't prevent
      the controller from learning about current events.
    - Report the circuit number correctly in STREAM CLOSED events. Bug
      reported by Mike Perry.
    - Do not report bizarre values for results of accounting GETINFOs
      when the last second's write or read exceeds the allotted bandwidth.
    - Report "unrecognized key" rather than an empty string when the
      controller tries to fetch a networkstatus that doesn't exist.
    - When the controller does a "GETINFO network-status", tell it
      about even those routers whose descriptors are very old, and use
      long nicknames where appropriate.
    - Fix handling of verbose nicknames with ORCONN controller events:
      make them show up exactly when requested, rather than exactly when
      not requested.
    - Controller signals now work on non-Unix platforms that don't define
      SIGUSR1 and SIGUSR2 the way we expect.
    - Respond to SIGNAL command before we execute the signal, in case
      the signal shuts us down. Suggested by Karsten Loesing.
    - Handle reporting OR_CONN_EVENT_NEW events to the controller.

  o Minor features, code performance:
    - Major performance improvement on inserting descriptors: change
      algorithm from O(n^2) to O(n).
    - Do not rotate onion key immediately after setting it for the first
      time.
    - Call router_have_min_dir_info half as often. (This is showing up in
      some profiles, but not others.)
    - When using GCC, make log_debug never get called at all, and its
      arguments never get evaluated, when no debug logs are configured.
      (This is showing up in some profiles, but not others.)
    - Statistics dumped by -USR2 now include a breakdown of public key
      operations, for profiling.
    - Make the common memory allocation path faster on machines where
      malloc(0) returns a pointer.
    - Split circuit_t into origin_circuit_t and or_circuit_t, and
      split connection_t into edge, or, dir, control, and base structs.
      These will save quite a bit of memory on busy servers, and they'll
      also help us track down bugs in the code and bugs in the spec.
    - Use OpenSSL's AES implementation on platforms where it's faster.
      This could save us as much as 10% CPU usage.

  o Minor features, descriptors and descriptor handling:
    - Avoid duplicate entries on MyFamily line in server descriptor.
    - When Tor receives a router descriptor that it asked for, but
      no longer wants (because it has received fresh networkstatuses
      in the meantime), do not warn the user. Cache the descriptor if
      we're a cache; drop it if we aren't.
    - Servers no longer ever list themselves in their "family" line,
      even if configured to do so. This makes it easier to configure
      family lists conveniently.

  o Minor fixes, confusing/misleading log messages:
    - Display correct results when reporting which versions are
      recommended, and how recommended they are. (Resolves bug 383.)
    - Inform the server operator when we decide not to advertise a
      DirPort due to AccountingMax enabled or a low BandwidthRate.
    - Only include function names in log messages for info/debug messages.
      For notice/warn/err, the content of the message should be clear on
      its own, and printing the function name only confuses users.
    - Remove even more protocol-related warnings from Tor server logs,
      such as bad TLS handshakes and malformed begin cells.
    - Fix bug 314: Tor clients issued "unsafe socks" warnings even
      when the IP address is mapped through MapAddress to a hostname.
    - Fix misleading log messages: an entry guard that is "unlisted",
      as well as not known to be "down" (because we've never heard
      of it), is not therefore "up".

  o Minor fixes, old/obsolete behavior:
    - Start assuming we can use a create_fast cell if we don't know
      what version a router is running.
    - We no longer look for identity and onion keys in "identity.key" and
      "onion.key" -- these were replaced by secret_id_key and
      secret_onion_key in 0.0.8pre1.
    - We no longer require unrecognized directory entries to be
      preceded by "opt".
    - Drop compatibility with obsolete Tors that permit create cells
      to have the wrong circ_id_type.
    - Remove code to special-case "-cvs" ending, since it has not
      actually mattered since 0.0.9.
    - Don't re-write the fingerprint file every restart, unless it has
      changed.

  o Minor fixes, misc client-side behavior:
    - Always remove expired routers and networkstatus docs before checking
      whether we have enough information to build circuits. (Fixes
      bug 373.)
    - When computing clock skew from directory HTTP headers, consider what
      time it was when we finished asking for the directory, not what
      time it is now.
    - Make our socks5 handling more robust to broken socks clients:
      throw out everything waiting on the buffer in between socks
      handshake phases, since they can't possibly (so the theory
      goes) have predicted what we plan to respond to them.
    - Expire socks connections if they spend too long waiting for the
      handshake to finish. Previously we would let them sit around for
      days, if the connecting application didn't close them either.
    - And if the socks handshake hasn't started, don't send a
      "DNS resolve socks failed" handshake reply; just close it.
    - If the user asks to use invalid exit nodes, be willing to use
      unstable ones.
    - Track unreachable entry guards correctly: don't conflate
      'unreachable by us right now' with 'listed as down by the directory
      authorities'. With the old code, if a guard was unreachable by us
      but listed as running, it would clog our guard list forever.
    - Behave correctly in case we ever have a network with more than
      2GB/s total advertised capacity.
    - Claim a commonname of Tor, rather than TOR, in TLS handshakes.
    - Fix a memory leak when we ask for "all" networkstatuses and we
      get one we don't recognize.


Changes in version 0.1.1.26 - 2006-12-14
  o Security bugfixes:
    - Stop sending the HttpProxyAuthenticator string to directory
      servers when directory connections are tunnelled through Tor.
    - Clients no longer store bandwidth history in the state file.
    - Do not log introduction points for hidden services if SafeLogging
      is set.

  o Minor bugfixes:
    - Fix an assert failure when a directory authority sets
      AuthDirRejectUnlisted and then receives a descriptor from an
      unlisted router (reported by seeess).


Changes in version 0.1.1.25 - 2006-11-04
  o Major bugfixes:
    - When a client asks us to resolve (rather than connect to)
      an address, and we have a cached answer, give them the cached
      answer. Previously, we would give them no answer at all.
    - We were building exactly the wrong circuits when we predict
      hidden service requirements, meaning Tor would have to build all
      its circuits on demand.
    - If none of our live entry guards have a high uptime, but we
      require a guard with a high uptime, try adding a new guard before
      we give up on the requirement. This patch should make long-lived
      connections more stable on average.
    - When testing reachability of our DirPort, don't launch new
      tests when there's already one in progress -- unreachable
      servers were stacking up dozens of testing streams.

  o Security bugfixes:
    - When the user sends a NEWNYM signal, clear the client-side DNS
      cache too. Otherwise we continue to act on previous information.

  o Minor bugfixes:
    - Avoid a memory corruption bug when creating a hash table for
      the first time.
    - Avoid possibility of controller-triggered crash when misusing
      certain commands from a v0 controller on platforms that do not
      handle printf("%s",NULL) gracefully.
    - Avoid infinite loop on unexpected controller input.
    - Don't log spurious warnings when we see a circuit close reason we
      don't recognize; it's probably just from a newer version of Tor.
    - Add Vidalia to the OS X uninstaller script, so when we uninstall
      Tor/Privoxy we also uninstall Vidalia.


Changes in version 0.1.1.24 - 2006-09-29
  o Major bugfixes:
    - Allow really slow clients to not hang up five minutes into their
      directory downloads (suggested by Adam J. Richter).
    - Fix major performance regression from 0.1.0.x: instead of checking
      whether we have enough directory information every time we want to
      do something, only check when the directory information has changed.
      This should improve client CPU usage by 25-50%.
    - Don't crash if, after a server has been running for a while,
      it can't resolve its hostname.
    - When a client asks us to resolve (not connect to) an address,
      and we have a cached answer, give them the cached answer.
      Previously, we would give them no answer at all.

  o Minor bugfixes:
    - Allow Tor to start when RunAsDaemon is set but no logs are set.
    - Don't crash when the controller receives a third argument to an
      "extendcircuit" request.
    - Controller protocol fixes: fix encoding in "getinfo addr-mappings"
      response; fix error code when "getinfo dir/status/" fails.
    - Fix configure.in to not produce broken configure files with
      more recent versions of autoconf. Thanks to Clint for his auto*
      voodoo.
    - Fix security bug on NetBSD that could allow someone to force
      uninitialized RAM to be sent to a server's DNS resolver. This
      only affects NetBSD and other platforms that do not bounds-check
      tolower().
    - Warn user when using libevent 1.1a or earlier with win32 or kqueue
      methods: these are known to be buggy.
    - If we're a directory mirror and we ask for "all" network status
      documents, we would discard status documents from authorities
      we don't recognize.


Changes in version 0.1.1.23 - 2006-07-30
  o Major bugfixes:
    - Fast Tor servers, especially exit nodes, were triggering asserts
      due to a bug in handling the list of pending DNS resolves. Some
      bugs still remain here; we're hunting them.
    - Entry guards could crash clients by sending unexpected input.
    - More fixes on reachability testing: if you find yourself reachable,
      then don't ever make any client requests (so you stop predicting
      circuits), then hup or have your clock jump, then later your IP
      changes, you won't think circuits are working, so you won't try to
      test reachability, so you won't publish.

  o Minor bugfixes:
    - Avoid a crash if the controller does a resetconf firewallports
      and then a setconf fascistfirewall=1.
    - Avoid an integer underflow when the dir authority decides whether
      a router is stable: we might wrongly label it stable, and compute
      a slightly wrong median stability, when a descriptor is published
      later than now.
    - Fix a place where we might trigger an assert if we can't build our
      own server descriptor yet.


Changes in version 0.1.1.22 - 2006-07-05
  o Major bugfixes:
    - Fix a big bug that was causing servers to not find themselves
      reachable if they changed IP addresses. Since only 0.1.1.22+
      servers can do reachability testing correctly, now we automatically
      make sure to test via one of these.
    - Fix to allow clients and mirrors to learn directory info from
      descriptor downloads that get cut off partway through.
    - Directory authorities had a bug in deciding if a newly published
      descriptor was novel enough to make everybody want a copy -- a few
      servers seem to be publishing new descriptors many times a minute.
  o Minor bugfixes:
    - Fix a rare bug that was causing some servers to complain about
      "closing wedged cpuworkers" and skip some circuit create requests.
    - Make the Exit flag in directory status documents actually work.


Changes in version 0.1.1.21 - 2006-06-10
  o Crash and assert fixes from 0.1.1.20:
    - Fix a rare crash on Tor servers that have enabled hibernation.
    - Fix a seg fault on startup for Tor networks that use only one
      directory authority.
    - Fix an assert from a race condition that occurs on Tor servers
      while exiting, where various threads are trying to log that they're
      exiting, and delete the logs, at the same time.
    - Make our unit tests pass again on certain obscure platforms.

  o Other fixes:
    - Add support for building SUSE RPM packages.
    - Speed up initial bootstrapping for clients: if we are making our
      first ever connection to any entry guard, then don't mark it down
      right after that.
    - When only one Tor server in the network is labelled as a guard,
      and we've already picked him, we would cycle endlessly picking him
      again, being unhappy about it, etc. Now we specifically exclude
      current guards when picking a new guard.
    - Servers send create cells more reliably after the TLS connection
      is established: we were sometimes forgetting to send half of them
      when we had more than one pending.
    - If we get a create cell that asks us to extend somewhere, but the
      Tor server there doesn't match the expected digest, we now send
      a destroy cell back, rather than silently doing nothing.
    - Make options->RedirectExit work again.
    - Make cookie authentication for the controller work again.
    - Stop being picky about unusual characters in the arguments to
      mapaddress. It's none of our business.
    - Add a new config option "TestVia" that lets you specify preferred
      middle hops to use for test circuits. Perhaps this will let me
      debug the reachability problems better.

  o Log / documentation fixes:
    - If we're a server and some peer has a broken TLS certificate, don't
      log about it unless ProtocolWarnings is set, i.e., we want to hear
      about protocol violations by others.
    - Fix spelling of VirtualAddrNetwork in man page.
    - Add a better explanation at the top of the autogenerated torrc file
      about what happened to our old torrc.


Changes in version 0.1.1.20 - 2006-05-23
  o Crash and assert fixes from 0.1.0.17:
    - Fix assert bug in close_logs() on exit: when we close and delete
      logs, remove them all from the global "logfiles" list.
    - Fix an assert error when we're out of space in the connection_list
      and we try to post a hidden service descriptor (reported by Peter
      Palfrader).
    - Fix a rare assert error when we've tried all intro points for
      a hidden service and we try fetching the service descriptor again:
      "Assertion conn->state != AP_CONN_STATE_RENDDESC_WAIT failed".
    - Setconf SocksListenAddress kills Tor if it fails to bind. Now back
      out and refuse the setconf if it would fail.
    - If you specify a relative torrc path and you set RunAsDaemon in
      your torrc, then it chdir()'s to the new directory. If you then
      HUP, it tries to load the new torrc location, fails, and exits.
      The fix: no longer allow a relative path to torrc when using -f.
    - Check for integer overflows in more places, when adding elements
      to smartlists. This could possibly prevent a buffer overflow
      on malicious huge inputs.

  o Security fixes, major:
    - When we're printing strings from the network, don't try to print
      non-printable characters. Now we're safer against shell escape
      sequence exploits, and also against attacks to fool users into
      misreading their logs.
    - Implement entry guards: automatically choose a handful of entry
      nodes and stick with them for all circuits. Only pick new guards
      when the ones you have are unsuitable, and if the old guards
      become suitable again, switch back. This will increase security
      dramatically against certain end-point attacks. The EntryNodes
      config option now provides some hints about which entry guards you
      want to use most; and StrictEntryNodes means to only use those.
      Fixes CVE-2006-0414.
    - Implement exit enclaves: if we know an IP address for the
      destination, and there's a running Tor server at that address
      which allows exit to the destination, then extend the circuit to
      that exit first. This provides end-to-end encryption and end-to-end
      authentication. Also, if the user wants a .exit address or enclave,
      use 4 hops rather than 3, and cannibalize a general circ for it
      if you can.
    - Obey our firewall options more faithfully:
      . If we can't get to a dirserver directly, try going via Tor.
      . Don't ever try to connect (as a client) to a place our
        firewall options forbid.
      . If we specify a proxy and also firewall options, obey the
        firewall options even when we're using the proxy: some proxies
        can only proxy to certain destinations.
    - Make clients regenerate their keys when their IP address changes.
    - For the OS X package's modified privoxy config file, comment
      out the "logfile" line so we don't log everything passed
      through privoxy.
    - Our TLS handshakes were generating a single public/private
      keypair for the TLS context, rather than making a new one for
      each new connection. Oops. (But we were still rotating them
      periodically, so it's not so bad.)
    - When we were cannibalizing a circuit with a particular exit
      node in mind, we weren't checking to see if that exit node was
      already present earlier in the circuit. Now we are.
    - Require server descriptors to list IPv4 addresses -- hostnames
      are no longer allowed. This also fixes potential vulnerabilities
      to servers providing hostnames as their address and then
      preferentially resolving them so they can partition users.
    - Our logic to decide if the OR we connected to was the right guy
      was brittle and maybe open to a mitm for invalid routers.

  o Security fixes, minor:
    - Adjust tor-spec.txt to parameterize cell and key lengths. Now
      Ian Goldberg can prove things about our handshake protocol more
      easily.
    - Make directory authorities generate a separate "guard" flag to
      mean "would make a good entry guard". Clients now honor the
      is_guard flag rather than looking at is_fast or is_stable.
    - Try to list MyFamily elements by key, not by nickname, and warn
      if we've not heard of a server.
    - Start using RAND_bytes rather than RAND_pseudo_bytes from
      OpenSSL. Also, reseed our entropy every hour, not just at
      startup. And add entropy in 512-bit chunks, not 160-bit chunks.
    - Refuse server descriptors where the fingerprint line doesn't match
      the included identity key. Tor doesn't care, but other apps (and
      humans) might actually be trusting the fingerprint line.
    - We used to kill the circuit when we receive a relay command we
      don't recognize. Now we just drop that cell.
    - Fix a bug found by Lasse Overlier: when we were making internal
      circuits (intended to be cannibalized later for rendezvous and
      introduction circuits), we were picking them so that they had
      useful exit nodes. There was no need for this, and it actually
      aids some statistical attacks.
    - Start treating internal circuits and exit circuits separately.
      It's important to keep them separate because internal circuits
      have their last hops picked like middle hops, rather than like
      exit hops. So exiting on them will break the user's expectations.
    - Fix a possible way to DoS dirservers.
    - When the client asked for a rendezvous port that the hidden
      service didn't want to provide, we were sending an IP address
      back along with the end cell. Fortunately, it was zero. But stop
      that anyway.

  o Packaging improvements:
    - Implement --with-libevent-dir option to ./configure. Improve
      search techniques to find libevent, and use those for openssl too.
    - Fix a couple of bugs in OpenSSL detection. Deal better when
      there are multiple SSLs installed with different versions.
    - Avoid warnings about machine/limits.h on Debian GNU/kFreeBSD.
    - On non-gcc compilers (e.g. Solaris's cc), use "-g -O" instead of
      "-Wall -g -O2".
    - Make unit tests (and other invocations that aren't the real Tor)
      run without launching listeners, creating subdirectories, and so on.
    - The OS X installer was adding a symlink for tor_resolve but
      the binary was called tor-resolve (reported by Thomas Hardly).
    - Now we can target arch and OS in rpm builds (contributed by
      Phobos). Also make the resulting dist-rpm filename match the
      target arch.
    - Apply Matt Ghali's --with-syslog-facility patch to ./configure
      if you log to syslog and want something other than LOG_DAEMON.
    - Fix the torify (tsocks) config file to not use Tor for localhost
      connections.
    - Start shipping socks-extensions.txt, tor-doc-unix.html,
      tor-doc-server.html, and stylesheet.css in the tarball.
    - Stop shipping tor-doc.html, INSTALL, and README in the tarball.
      They are useless now.
    - Add Peter Palfrader's contributed check-tor script. It lets you
      easily check whether a given server (referenced by nickname)
      is reachable by you.
    - Add BSD-style contributed startup script "rc.subr" from Peter
      Thoenen.

  o Directory improvements -- new directory protocol:
    - See tor/doc/dir-spec.txt for all the juicy details. Key points:
    - Authorities and caches publish individual descriptors (by
      digest, by fingerprint, by "all", and by "tell me yours").
    - Clients don't download or use the old directory anymore. Now they
      download network-statuses from the directory authorities, and
      fetch individual server descriptors as needed from mirrors.
    - Clients don't download descriptors of non-running servers.
    - Download descriptors by digest, not by fingerprint. Caches try to
      download all listed digests from authorities; clients try to
      download "best" digests from caches. This avoids partitioning
      and isolating attacks better.
    - Only upload a new server descriptor when options change, 18
      hours have passed, uptime is reset, or bandwidth changes a lot.
    - Directory authorities silently throw away new descriptors that
      haven't changed much if the timestamps are similar. We do this to
      tolerate older Tor servers that upload a new descriptor every 15
      minutes. (It seemed like a good idea at the time.)
    - Clients choose directory servers from the network status lists,
      not from their internal list of router descriptors. Now they can
      go to caches directly rather than needing to go to authorities
      to bootstrap the first set of descriptors.
    - When picking a random directory, prefer non-authorities if any
      are known.
    - Add a new flag to network-status indicating whether the server
      can answer v2 directory requests too.
    - Directory mirrors now cache up to 16 unrecognized network-status
      docs, so new directory authorities will be cached too.
    - Stop parsing, storing, or using running-routers output (but
      mirrors still cache and serve it).
    - Clients consider a threshold of "versioning" directory authorities
      before deciding whether to warn the user that he's obsolete.
    - Authorities publish separate sorted lists of recommended versions
      for clients and for servers.
    - Change DirServers config line to note which dirs are v1 authorities.
    - Put nicknames on the DirServer line, so we can refer to them
      without requiring all our users to memorize their IP addresses.
    - Remove option when getting directory cache to see whether they
      support running-routers; they all do now. Replace it with one
      to see whether caches support v2 stuff.
    - Stop listing down or invalid nodes in the v1 directory. This
      reduces its bulk by about 1/3, and reduces load on mirrors.
    - Mirrors no longer cache the v1 directory as often.
    - If we as a directory mirror don't know of any v1 directory
      authorities, then don't try to cache any v1 directories.

  o Other directory improvements:
    - Add lefkada.eecs.harvard.edu and tor.dizum.com as fourth and
      fifth authoritative directory servers.
    - Directory authorities no longer require an open connection from
      a server to consider him "reachable". We need this change because
      when we add new directory authorities, old servers won't know not
      to hang up on them.
    - Dir authorities now do their own external reachability testing
      of each server, and only list as running the ones they found to
      be reachable. We also send back warnings to the server's logs if
      it uploads a descriptor that we already believe is unreachable.
    - Spread the directory authorities' reachability testing over the
      entire testing interval, so we don't try to do 500 TLS's at once
      every 20 minutes.
    - Make the "stable" router flag in network-status be the median of
      the uptimes of running valid servers, and make clients pay
      attention to the network-status flags. Thus the cutoff adapts
      to the stability of the network as a whole, making IRC, IM, etc
      connections more reliable.
    - Make the v2 dir's "Fast" flag based on relative capacity, just
      like "Stable" is based on median uptime. Name everything in the
      top 7/8 Fast, and only the top 1/2 gets to be a Guard.
    - Retry directory requests if we fail to get an answer we like
      from a given dirserver (we were retrying before, but only if
      we fail to connect).
    - Return a robots.txt on our dirport to discourage google indexing.

  o Controller protocol improvements:
    - Revised controller protocol (version 1) that uses ascii rather
      than binary: tor/doc/control-spec.txt. Add supporting libraries
      in python and java and c# so you can use the controller from your
      applications without caring how our protocol works.
    - Allow the DEBUG controller event to work again. Mark certain log
      entries as "don't tell this to controllers", so we avoid cycles.
    - New controller function "getinfo accounting", to ask how
      many bytes we've used in this time period.
    - Add a "resetconf" command so you can set config options like
      AllowUnverifiedNodes and LongLivedPorts to "". Also, if you give
      a config option in the torrc with no value, then it clears it
      entirely (rather than setting it to its default).
    - Add a "getinfo config-file" to tell us where torrc is. Also
      expose guard nodes, config options/names.
    - Add a "quit" command (when when using the controller manually).
    - Add a new signal "newnym" to "change pseudonyms" -- that is, to
      stop using any currently-dirty circuits for new streams, so we
      don't link new actions to old actions. This also occurs on HUP
      or "signal reload".
    - If we would close a stream early (e.g. it asks for a .exit that
      we know would refuse it) but the LeaveStreamsUnattached config
      option is set by the controller, then don't close it.
    - Add a new controller event type "authdir_newdescs" that allows
      controllers to get all server descriptors that were uploaded to
      a router in its role as directory authority.
    - New controller option "getinfo desc/all-recent" to fetch the
      latest server descriptor for every router that Tor knows about.
    - Fix the controller's "attachstream 0" command to treat conn like
      it just connected, doing address remapping, handling .exit and
      .onion idioms, and so on. Now we're more uniform in making sure
      that the controller hears about new and closing connections.
    - Permit transitioning from ORPort==0 to ORPort!=0, and back, from
      the controller. Also, rotate dns and cpu workers if the controller
      changes options that will affect them; and initialize the dns
      worker cache tree whether or not we start out as a server.
    - Add a new circuit purpose 'controller' to let the controller ask
      for a circuit that Tor won't try to use. Extend the "extendcircuit"
      controller command to let you specify the purpose if you're starting
      a new circuit.  Add a new "setcircuitpurpose" controller command to
      let you change a circuit's purpose after it's been created.
    - Let the controller ask for "getinfo dir/server/foo" so it can ask
      directly rather than connecting to the dir port. "getinfo
      dir/status/foo" also works, but currently only if your DirPort
      is enabled.
    - Let the controller tell us about certain router descriptors
      that it doesn't want Tor to use in circuits. Implement
      "setrouterpurpose" and modify "+postdescriptor" to do this.
    - If the controller's *setconf commands fail, collect an error
      message in a string and hand it back to the controller -- don't
      just tell them to go read their logs.

  o Scalability, resource management, and performance:
    - Fix a major load balance bug: we were round-robin reading in 16 KB
      chunks, and servers with bandwidthrate of 20 KB, while downloading
      a 600 KB directory, would starve their other connections. Now we
      try to be a bit more fair.
    - Be more conservative about whether to advertise our DirPort.
      The main change is to not advertise if we're running at capacity
      and either a) we could hibernate ever or b) our capacity is low
      and we're using a default DirPort.
    - We weren't cannibalizing circuits correctly for
      CIRCUIT_PURPOSE_C_ESTABLISH_REND and
      CIRCUIT_PURPOSE_S_ESTABLISH_INTRO, so we were being forced to
      build those from scratch. This should make hidden services faster.
    - Predict required circuits better, with an eye toward making hidden
      services faster on the service end.
    - Compress exit policies even more: look for duplicate lines and
      remove them.
    - Generate 18.0.0.0/8 address policy format in descs when we can;
      warn when the mask is not reducible to a bit-prefix.
    - There used to be two ways to specify your listening ports in a
      server descriptor: on the "router" line and with a separate "ports"
      line. Remove support for the "ports" line.
    - Reduce memory requirements in our structs by changing the order
      of fields. Replace balanced trees with hash tables. Inline
      bottleneck smartlist functions. Add a "Map from digest to void*"
      abstraction so we can do less hex encoding/decoding, and use it
      in router_get_by_digest(). Many other CPU and memory improvements.
    - Allow tor_gzip_uncompress to extract as much as possible from
      truncated compressed data. Try to extract as many
      descriptors as possible from truncated http responses (when
      purpose is DIR_PURPOSE_FETCH_ROUTERDESC).
    - Make circ->onionskin a pointer, not a static array. moria2 was using
      125000 circuit_t's after it had been up for a few weeks, which
      translates to 20+ megs of wasted space.
    - The private half of our EDH handshake keys are now chosen out
      of 320 bits, not 1024 bits. (Suggested by Ian Goldberg.)
    - Stop doing the complex voodoo overkill checking for insecure
      Diffie-Hellman keys. Just check if it's in [2,p-2] and be happy.
    - Do round-robin writes for TLS of at most 16 kB per write. This
      might be more fair on loaded Tor servers.
    - Do not use unaligned memory access on alpha, mips, or mipsel.
      It *works*, but is very slow, so we treat them as if it doesn't.

  o Other bugfixes and improvements:
    - Start storing useful information to $DATADIR/state, so we can
      remember things across invocations of Tor. Retain unrecognized
      lines so we can be forward-compatible, and write a TorVersion line
      so we can be backward-compatible.
    - If ORPort is set, Address is not explicitly set, and our hostname
      resolves to a private IP address, try to use an interface address
      if it has a public address. Now Windows machines that think of
      themselves as localhost can guess their address.
    - Regenerate our local descriptor if it's dirty and we try to use
      it locally (e.g. if it changes during reachability detection).
      This was causing some Tor servers to keep publishing the same
      initial descriptor forever.
    - Tor servers with dynamic IP addresses were needing to wait 18
      hours before they could start doing reachability testing using
      the new IP address and ports. This is because they were using
      the internal descriptor to learn what to test, yet they were only
      rebuilding the descriptor once they decided they were reachable.
    - It turns out we couldn't bootstrap a network since we added
      reachability detection in 0.1.0.1-rc. Good thing the Tor network
      has never gone down. Add an AssumeReachable config option to let
      servers and authorities bootstrap. When we're trying to build a
      high-uptime or high-bandwidth circuit but there aren't enough
      suitable servers, try being less picky rather than simply failing.
    - Newly bootstrapped Tor networks couldn't establish hidden service
      circuits until they had nodes with high uptime. Be more tolerant.
    - Really busy servers were keeping enough circuits open on stable
      connections that they were wrapping around the circuit_id
      space. (It's only two bytes.) This exposed a bug where we would
      feel free to reuse a circuit_id even if it still exists but has
      been marked for close. Try to fix this bug. Some bug remains.
    - When we fail to bind or listen on an incoming or outgoing
      socket, we now close it before refusing, rather than just
      leaking it. (Thanks to Peter Palfrader for finding.)
    - Fix a file descriptor leak in start_daemon().
    - On Windows, you can't always reopen a port right after you've
      closed it. So change retry_listeners() to only close and re-open
      ports that have changed.
    - Workaround a problem with some http proxies that refuse GET
      requests that specify "Content-Length: 0". Reported by Adrian.
    - Recover better from TCP connections to Tor servers that are
      broken but don't tell you (it happens!); and rotate TLS
      connections once a week.
    - Fix a scary-looking but apparently harmless bug where circuits
      would sometimes start out in state CIRCUIT_STATE_OR_WAIT at
      servers, and never switch to state CIRCUIT_STATE_OPEN.
    - Check for even more Windows version flags when writing the platform
      string in server descriptors, and note any we don't recognize.
    - Add reasons to DESTROY and RELAY_TRUNCATED cells, so clients can
      get a better idea of why their circuits failed. Not used yet.
    - Add TTLs to RESOLVED, CONNECTED, and END_REASON_EXITPOLICY cells.
      We don't use them yet, but maybe one day our DNS resolver will be
      able to discover them.
    - Let people type "tor --install" as well as "tor -install" when they
      want to make it an NT service.
    - Looks like we were never delivering deflated (i.e. compressed)
      running-routers lists, even when asked. Oops.
    - We were leaking some memory every time the client changed IPs.
    - Clean up more of the OpenSSL memory when exiting, so we can detect
      memory leaks better.
    - Never call free() on tor_malloc()d memory. This will help us
      use dmalloc to detect memory leaks.
    - Some Tor servers process billions of cells per day. These
      statistics are now uint64_t's.
    - Check [X-]Forwarded-For headers in HTTP requests when generating
      log messages. This lets people run dirservers (and caches) behind
      Apache but still know which IP addresses are causing warnings.
    - Fix minor integer overflow in calculating when we expect to use up
      our bandwidth allocation before hibernating.
    - Lower the minimum required number of file descriptors to 1000,
      so we can have some overhead for Valgrind on Linux, where the
      default ulimit -n is 1024.
    - Stop writing the "router.desc" file, ever. Nothing uses it anymore,
      and its existence is confusing some users.

  o Config option fixes:
    - Add a new config option ExitPolicyRejectPrivate which defaults
      to on. Now all exit policies will begin with rejecting private
      addresses, unless the server operator explicitly turns it off.
    - Bump the default bandwidthrate to 3 MB, and burst to 6 MB.
    - Add new ReachableORAddresses and ReachableDirAddresses options
      that understand address policies. FascistFirewall is now a synonym
      for "ReachableORAddresses *:443", "ReachableDirAddresses *:80".
    - Start calling it FooListenAddress rather than FooBindAddress,
      since few of our users know what it means to bind an address
      or port.
    - If the user gave Tor an odd number of command-line arguments,
      we were silently ignoring the last one. Now we complain and fail.
      This wins the oldest-bug prize -- this bug has been present since
      November 2002, as released in Tor 0.0.0.
    - If you write "HiddenServicePort 6667 127.0.0.1 6668" in your
      torrc rather than "HiddenServicePort 6667 127.0.0.1:6668",
      it would silently ignore the 6668.
    - If we get a linelist or linelist_s config option from the torrc,
      e.g. ExitPolicy, and it has no value, warn and skip rather than
      silently resetting it to its default.
    - Setconf was appending items to linelists, not clearing them.
    - Add MyFamily to torrc.sample in the server section, so operators
      will be more likely to learn that it exists.
    - Make ContactInfo mandatory for authoritative directory servers.
    - MaxConn has been obsolete for a while now. Document the ConnLimit
      config option, which is a *minimum* number of file descriptors
      that must be available else Tor refuses to start.
    - Get rid of IgnoreVersion undocumented config option, and make us
      only warn, never exit, when we're running an obsolete version.
    - Make MonthlyAccountingStart config option truly obsolete now.
    - Correct the man page entry on TrackHostExitsExpire.
    - Let directory authorities start even if they don't specify an
      Address config option.
    - Change "AllowUnverifiedNodes" to "AllowInvalidNodes", to
      reflect the updated flags in our v2 dir protocol.

  o Config option features:
    - Add a new config option FastFirstHopPK (on by default) so clients
      do a trivial crypto handshake for their first hop, since TLS has
      already taken care of confidentiality and authentication.
    - Let the user set ControlListenAddress in the torrc. This can be
      dangerous, but there are some cases (like a secured LAN) where it
      makes sense.
    - New config options to help controllers: FetchServerDescriptors
      and FetchHidServDescriptors for whether to fetch server
      info and hidserv info or let the controller do it, and
      PublishServerDescriptor and PublishHidServDescriptors.
    - Also let the controller set the __AllDirActionsPrivate config
      option if you want all directory fetches/publishes to happen via
      Tor (it assumes your controller bootstraps your circuits).
    - Add "HardwareAccel" config option: support for crypto hardware
      accelerators via OpenSSL. Off by default, until we find somebody
      smart who can test it for us. (It appears to produce seg faults
      in at least some cases.)
    - New config option "AuthDirRejectUnlisted" for directory authorities
      as a panic button: if we get flooded with unusable servers we can
      revert to only listing servers in the approved-routers file.
    - Directory authorities can now reject/invalidate by key and IP,
      with the config options "AuthDirInvalid" and "AuthDirReject", or
      by marking a fingerprint as "!reject" or "!invalid" (as its
      nickname) in the approved-routers file. This is useful since
      currently we automatically list servers as running and usable
      even if we know they're jerks.
    - Add a new config option TestSocks so people can see whether their
      applications are using socks4, socks4a, socks5-with-ip, or
      socks5-with-fqdn. This way they don't have to keep mucking
      with tcpdump and wondering if something got cached somewhere.
    - Add "private:*" as an alias in configuration for policies. Now
      you can simplify your exit policy rather than needing to list
      every single internal or nonroutable network space.
    - Accept "private:*" in routerdesc exit policies; not generated yet
      because older Tors do not understand it.
    - Add configuration option "V1AuthoritativeDirectory 1" which
      moria1, moria2, and tor26 have set.
    - Implement an option, VirtualAddrMask, to set which addresses
      get handed out in response to mapaddress requests. This works
      around a bug in tsocks where 127.0.0.0/8 is never socksified.
    - Add a new config option FetchUselessDescriptors, off by default,
      for when you plan to run "exitlist" on your client and you want
      to know about even the non-running descriptors.
    - SocksTimeout: How long do we let a socks connection wait
      unattached before we fail it?
    - CircuitBuildTimeout: Cull non-open circuits that were born
      at least this many seconds ago.
    - CircuitIdleTimeout: Cull open clean circuits that were born
      at least this many seconds ago.
    - New config option SafeSocks to reject all application connections
      using unsafe socks protocols. Defaults to off.

  o Improved and clearer log messages:
    - Reduce clutter in server logs. We're going to try to make
      them actually usable now. New config option ProtocolWarnings that
      lets you hear about how _other Tors_ are breaking the protocol. Off
      by default.
    - Divide log messages into logging domains. Once we put some sort
      of interface on this, it will let people looking at more verbose
      log levels specify the topics they want to hear more about.
    - Log server fingerprint on startup, so new server operators don't
      have to go hunting around their filesystem for it.
    - Provide dire warnings to any users who set DirServer manually;
      move it out of torrc.sample and into torrc.complete.
    - Make the log message less scary when all the dirservers are
      temporarily unreachable.
    - When tor_socketpair() fails in Windows, give a reasonable
      Windows-style errno back.
    - Improve tor_gettimeofday() granularity on windows.
    - We were printing the number of idle dns workers incorrectly when
      culling them.
    - Handle duplicate lines in approved-routers files without warning.
    - We were whining about using socks4 or socks5-with-local-lookup
      even when it's an IP address in the "virtual" range we designed
      exactly for this case.
    - Check for named servers when looking them up by nickname;
      warn when we're calling a non-named server by its nickname;
      don't warn twice about the same name.
    - Downgrade the dirserver log messages when whining about
      unreachability.
    - Correct "your server is reachable" log entries to indicate that
      it was self-testing that told us so.
    - If we're trying to be a Tor server and running Windows 95/98/ME
      as a server, explain that we'll likely crash.
    - Provide a more useful warn message when our onion queue gets full:
      the CPU is too slow or the exit policy is too liberal.
    - Don't warn when we receive a 503 from a dirserver/cache -- this
      will pave the way for them being able to refuse if they're busy.
    - When we fail to bind a listener, try to provide a more useful
      log message: e.g., "Is Tor already running?"
    - Only start testing reachability once we've established a
      circuit. This will make startup on dir authorities less noisy.
    - Don't try to upload hidden service descriptors until we have
      established a circuit.
    - Tor didn't warn when it failed to open a log file.
    - Warn when listening on a public address for socks. We suspect a
      lot of people are setting themselves up as open socks proxies,
      and they have no idea that jerks on the Internet are using them,
      since they simply proxy the traffic into the Tor network.
    - Give a useful message when people run Tor as the wrong user,
      rather than telling them to start chowning random directories.
    - Fix a harmless bug that was causing Tor servers to log
      "Got an end because of misc error, but we're not an AP. Closing."
    - Fix wrong log message when you add a "HiddenServiceNodes" config
      line without any HiddenServiceDir line (reported by Chris Thomas).
    - Directory authorities now stop whining so loudly about bad
      descriptors that they fetch from other dirservers. So when there's
      a log complaint, it's for sure from a freshly uploaded descriptor.
    - When logging via syslog, include the pid whenever we provide
      a log entry. Suggested by Todd Fries.
    - When we're shutting down and we do something like try to post a
      server descriptor or rendezvous descriptor, don't complain that
      we seem to be unreachable. Of course we are, we're shutting down.
    - Change log line for unreachability to explicitly suggest /etc/hosts
      as the culprit. Also make it clearer what IP address and ports we're
      testing for reachability.
    - Put quotes around user-supplied strings when logging so users are
      more likely to realize if they add bad characters (like quotes)
      to the torrc.
    - NT service patch from Matt Edman to improve error messages on Win32.


Changes in version 0.1.0.17 - 2006-02-17
  o Crash bugfixes on 0.1.0.x:
    - When servers with a non-zero DirPort came out of hibernation,
      sometimes they would trigger an assert.

  o Other important bugfixes:
    - On platforms that don't have getrlimit (like Windows), we were
      artificially constraining ourselves to a max of 1024
      connections. Now just assume that we can handle as many as 15000
      connections. Hopefully this won't cause other problems.

  o Backported features:
    - When we're a server, a client asks for an old-style directory,
      and our write bucket is empty, don't give it to him. This way
      small servers can continue to serve the directory *sometimes*,
      without getting overloaded.
    - Whenever you get a 503 in response to a directory fetch, try
      once more. This will become important once servers start sending
      503's whenever they feel busy.
    - Fetch a new directory every 120 minutes, not every 40 minutes.
      Now that we have hundreds of thousands of users running the old
      directory algorithm, it's starting to hurt a lot.
    - Bump up the period for forcing a hidden service descriptor upload
      from 20 minutes to 1 hour.


Changes in version 0.1.0.16 - 2006-01-02
  o Crash bugfixes on 0.1.0.x:
    - On Windows, build with a libevent patch from "I-M Weasel" to avoid
      corrupting the heap, losing FDs, or crashing when we need to resize
      the fd_sets. (This affects the Win32 binaries, not Tor's sources.)
    - It turns out sparc64 platforms crash on unaligned memory access
      too -- so detect and avoid this.
    - Handle truncated compressed data correctly (by detecting it and
      giving an error).
    - Fix possible-but-unlikely free(NULL) in control.c.
    - When we were closing connections, there was a rare case that
      stomped on memory, triggering seg faults and asserts.
    - Avoid potential infinite recursion when building a descriptor. (We
      don't know that it ever happened, but better to fix it anyway.)
    - We were neglecting to unlink marked circuits from soon-to-close OR
      connections, which caused some rare scribbling on freed memory.
    - Fix a memory stomping race bug when closing the joining point of two
      rendezvous circuits.
    - Fix an assert in time parsing found by Steven Murdoch.

  o Other bugfixes on 0.1.0.x:
    - When we're doing reachability testing, provide more useful log
      messages so the operator knows what to expect.
    - Do not check whether DirPort is reachable when we are suppressing
      advertising it because of hibernation.
    - When building with -static or on Solaris, we sometimes needed -ldl.
    - One of the dirservers (tor26) changed its IP address.
    - When we're deciding whether a stream has enough circuits around
      that can handle it, count the freshly dirty ones and not the ones
      that are so dirty they won't be able to handle it.
    - When we're expiring old circuits, we had a logic error that caused
      us to close new rendezvous circuits rather than old ones.
    - Give a more helpful log message when you try to change ORPort via
      the controller: you should upgrade Tor if you want that to work.
    - We were failing to parse Tor versions that start with "Tor ".
    - Tolerate faulty streams better: when a stream fails for reason
      exitpolicy, stop assuming that the router is lying about his exit
      policy. When a stream fails for reason misc, allow it to retry just
      as if it was resolvefailed. When a stream has failed three times,
      reset its failure count so we can try again and get all three tries.


Changes in version 0.1.0.15 - 2005-09-23
  o Bugfixes on 0.1.0.x:
    - Reject ports 465 and 587 (spam targets) in default exit policy.
    - Don't crash when we don't have any spare file descriptors and we
      try to spawn a dns or cpu worker.
    - Get rid of IgnoreVersion undocumented config option, and make us
      only warn, never exit, when we're running an obsolete version.
    - Don't try to print a null string when your server finds itself to
      be unreachable and the Address config option is empty.
    - Make the numbers in read-history and write-history into uint64s,
      so they don't overflow and publish negatives in the descriptor.
    - Fix a minor memory leak in smartlist_string_remove().
    - We were only allowing ourselves to upload a server descriptor at
      most every 20 minutes, even if it changed earlier than that.
    - Clean up log entries that pointed to old URLs.


Changes in version 0.1.0.14 - 2005-08-08
  o Bugfixes on 0.1.0.x:
      - Fix the other half of the bug with crypto handshakes
        (CVE-2005-2643).
      - Fix an assert trigger if you send a 'signal term' via the
        controller when it's listening for 'event info' messages.


Changes in version 0.1.0.13 - 2005-08-04
  o Bugfixes on 0.1.0.x:
    - Fix a critical bug in the security of our crypto handshakes.
    - Fix a size_t underflow in smartlist_join_strings2() that made
      it do bad things when you hand it an empty smartlist.
    - Fix Windows installer to ship Tor license (thanks to Aphex for
      pointing out this oversight) and put a link to the doc directory
      in the start menu.
    - Explicitly set no-unaligned-access for sparc: it turns out the
      new gcc's let you compile broken code, but that doesn't make it
      not-broken.


Changes in version 0.1.0.12 - 2005-07-18
  o New directory servers:
      - tor26 has changed IP address.

  o Bugfixes on 0.1.0.x:
    - Fix a possible double-free in tor_gzip_uncompress().
    - When --disable-threads is set, do not search for or link against
      pthreads libraries.
    - Don't trigger an assert if an authoritative directory server
      claims its dirport is 0.
    - Fix bug with removing Tor as an NT service: some people were
      getting "The service did not return an error." Thanks to Matt
      Edman for the fix.


Changes in version 0.1.0.11 - 2005-06-30
  o Bugfixes on 0.1.0.x:
    - Fix major security bug: servers were disregarding their
      exit policies if clients behaved unexpectedly.
    - Make OS X init script check for missing argument, so we don't
      confuse users who invoke it incorrectly.
    - Fix a seg fault in "tor --hash-password foo".
    - The MAPADDRESS control command was broken.


Changes in version 0.1.0.10 - 2005-06-14
  o Fixes on Win32:
    - Make NT services work and start on startup on Win32 (based on
      patch by Matt Edman). See the FAQ entry for details.
    - Make 'platform' string in descriptor more accurate for Win32
      servers, so it's not just "unknown platform".
    - REUSEADDR on normal platforms means you can rebind to the port
      right after somebody else has let it go. But REUSEADDR on Win32
      means you can bind to the port _even when somebody else already
      has it bound_! So, don't do that on Win32.
    - Clean up the log messages when starting on Win32 with no config
      file.
    - Allow seeding the RNG on Win32 even when you're not running as
      Administrator. If seeding the RNG on Win32 fails, quit.

  o Assert / crash bugs:
    - Refuse relay cells that claim to have a length larger than the
      maximum allowed. This prevents a potential attack that could read
      arbitrary memory (e.g. keys) from an exit server's process
      (CVE-2005-2050).
    - If unofficial Tor clients connect and send weird TLS certs, our
      Tor server triggers an assert. Stop asserting, and start handling
      TLS errors better in other situations too.
    - Fix a race condition that can trigger an assert when we have a
      pending create cell and an OR connection attempt fails.

  o Resource leaks:
    - Use pthreads for worker processes rather than forking. This was
      forced because when we forked, we ended up wasting a lot of
      duplicate ram over time.
      - Also switch to foo_r versions of some library calls to allow
        reentry and threadsafeness.
      - Implement --disable-threads configure option. Disable threads on
        netbsd and openbsd by default, because they have no reentrant
        resolver functions (!), and on solaris since it has other
        threading issues.
    - Fix possible bug on threading platforms (e.g. win32) which was
      leaking a file descriptor whenever a cpuworker or dnsworker died.
    - Fix a minor memory leak when somebody establishes an introduction
      point at your Tor server.
    - Fix possible memory leak in tor_lookup_hostname(). (Thanks to
      Adam Langley.)
    - Add ./configure --with-dmalloc option, to track memory leaks.
    - And try to free all memory on closing, so we can detect what
      we're leaking.

  o Protocol correctness:
    - When we've connected to an OR and handshaked but didn't like
      the result, we were closing the conn without sending destroy
      cells back for pending circuits. Now send those destroys.
    - Start sending 'truncated' cells back rather than destroy cells
      if the circuit closes in front of you. This means we won't have
      to abandon partially built circuits.
    - Handle changed router status correctly when dirserver reloads
      fingerprint file. We used to be dropping all unverified descriptors
      right then. The bug was hidden because we would immediately
      fetch a directory from another dirserver, which would include the
      descriptors we just dropped.
    - Revise tor-spec to add more/better stream end reasons.
    - Revise all calls to connection_edge_end to avoid sending 'misc',
      and to take errno into account where possible.
    - Client now retries when streams end early for 'hibernating' or
      'resource limit' reasons, rather than failing them.
    - Try to be more zealous about calling connection_edge_end when
      things go bad with edge conns in connection.c.

  o Robustness improvements:
    - Better handling for heterogeneous / unreliable nodes:
      - Annotate circuits with whether they aim to contain high uptime
        nodes and/or high capacity nodes. When building circuits, choose
        appropriate nodes.
      - This means that every single node in an intro rend circuit,
        not just the last one, will have a minimum uptime.
      - New config option LongLivedPorts to indicate application streams
        that will want high uptime circuits.
      - Servers reset uptime when a dir fetch entirely fails. This
        hopefully reflects stability of the server's network connectivity.
      - If somebody starts his tor server in Jan 2004 and then fixes his
        clock, don't make his published uptime be a year.
      - Reset published uptime when we wake up from hibernation.
    - Introduce a notion of 'internal' circs, which are chosen without
      regard to the exit policy of the last hop. Intro and rendezvous
      circs must be internal circs, to avoid leaking information. Resolve
      and connect streams can use internal circs if they want.
    - New circuit pooling algorithm: keep track of what destination ports
      we've used recently (start out assuming we'll want to use 80), and
      make sure to have enough circs around to satisfy these ports. Also
      make sure to have 2 internal circs around if we've required internal
      circs lately (and with high uptime if we've seen that lately too).
    - Turn addr_policy_compare from a tristate to a quadstate; this should
      help address our "Ah, you allow 1.2.3.4:80. You are a good choice
      for google.com" problem.
    - When a client asks us for a dir mirror and we don't have one,
      launch an attempt to get a fresh one.
    - First cut at support for "create-fast" cells. Clients can use
      these when extending to their first hop, since the TLS already
      provides forward secrecy and authentication. Not enabled on
      clients yet.

  o Reachability testing.
    - Your Tor server will automatically try to see if its ORPort and
      DirPort are reachable from the outside, and it won't upload its
      descriptor until it decides at least ORPort is reachable (when
      DirPort is not yet found reachable, publish it as zero).
    - When building testing circs for ORPort testing, use only
      high-bandwidth nodes, so fewer circuits fail.
    - Notice when our IP changes, and reset stats/uptime/reachability.
    - Authdirservers don't do ORPort reachability detection, since
      they're in clique mode, so it will be rare to find a server not
      already connected to them.
    - Authdirservers now automatically approve nodes running 0.1.0.2-rc
      or later.

  o Dirserver fixes:
    - Now we allow two unverified servers with the same nickname
      but different keys. But if a nickname is verified, only that
      nickname+key are allowed.
    - If you're an authdirserver connecting to an address:port,
      and it's not the OR you were expecting, forget about that
      descriptor. If he *was* the one you were expecting, then forget
      about all other descriptors for that address:port.
    - Allow servers to publish descriptors from 12 hours in the future.
      Corollary: only whine about clock skew from the dirserver if
      he's a trusted dirserver (since now even verified servers could
      have quite wrong clocks).
    - Require servers that use the default dirservers to have public IP
      addresses. We have too many servers that are configured with private
      IPs and their admins never notice the log entries complaining that
      their descriptors are being rejected.

  o Efficiency improvements:
    - Use libevent. Now we can use faster async cores (like epoll, kpoll,
      and /dev/poll), and hopefully work better on Windows too.
      - Apple's OS X 10.4.0 ships with a broken kqueue API, and using
        kqueue on 10.3.9 causes kernel panics. Don't use kqueue on OS X.
      - Find libevent even if it's hiding in /usr/local/ and your
        CFLAGS and LDFLAGS don't tell you to look there.
      - Be able to link with libevent as a shared library (the default
        after 1.0d), even if it's hiding in /usr/local/lib and even
        if you haven't added /usr/local/lib to your /etc/ld.so.conf,
        assuming you're running gcc. Otherwise fail and give a useful
        error message.
    - Switch to a new buffer management algorithm, which tries to avoid
      reallocing and copying quite as much. In first tests it looks like
      it uses *more* memory on average, but less cpu.
    - Switch our internal buffers implementation to use a ring buffer,
      to hopefully improve performance for fast servers a lot.
    - Reenable the part of the code that tries to flush as soon as an
      OR outbuf has a full TLS record available. Perhaps this will make
      OR outbufs not grow as huge except in rare cases, thus saving lots
      of CPU time plus memory.
    - Improve performance for dirservers: stop re-parsing the whole
      directory every time you regenerate it.
    - Keep a big splay tree of (circid,orconn)->circuit mappings to make
      it much faster to look up a circuit for each relay cell.
    - Remove most calls to assert_all_pending_dns_resolves_ok(),
      since they're eating our cpu on exit nodes.
    - Stop wasting time doing a case insensitive comparison for every
      dns name every time we do any lookup. Canonicalize the names to
      lowercase when you first see them.

  o Hidden services:
    - Handle unavailable hidden services better. Handle slow or busy
      hidden services better.
    - Cannibalize GENERAL circs to be C_REND, C_INTRO, S_INTRO, and S_REND
      circ as necessary, if there are any completed ones lying around
      when we try to launch one.
    - Make hidden services try to establish a rendezvous for 30 seconds
      after fetching the descriptor, rather than for n (where n=3)
      attempts to build a circuit.
    - Adjust maximum skew and age for rendezvous descriptors: let skew
      be 48 hours rather than 90 minutes.
    - Reject malformed .onion addresses rather then passing them on as
      normal web requests.

  o Controller:
    - More Tor controller support. See
      http://tor.eff.org/doc/control-spec.txt for all the new features,
      including signals to emulate unix signals from any platform;
      redirectstream; extendcircuit; mapaddress; getinfo; postdescriptor;
      closestream; closecircuit; etc.
    - Encode hashed controller passwords in hex instead of base64,
      to make it easier to write controllers.
    - Revise control spec and implementation to allow all log messages to
      be sent to controller with their severities intact (suggested by
      Matt Edman). Disable debug-level logs while delivering a debug-level
      log to the controller, to prevent loop. Update TorControl to handle
      new log event types.

  o New config options/defaults:
    - Begin scrubbing sensitive strings from logs by default. Turn off
      the config option SafeLogging if you need to do debugging.
    - New exit policy: accept most low-numbered ports, rather than
      rejecting most low-numbered ports.
    - Put a note in the torrc about abuse potential with the default
      exit policy.
    - Add support for CONNECTing through https proxies, with "HttpsProxy"
      config option.
    - Add HttpProxyAuthenticator and HttpsProxyAuthenticator support
      based on patch from Adam Langley (basic auth only).
    - Bump the default BandwidthRate from 1 MB to 2 MB, to accommodate
      the fast servers that have been joining lately. (Clients are now
      willing to load balance over up to 2 MB of advertised bandwidth
      capacity too.)
    - New config option MaxAdvertisedBandwidth which lets you advertise
      a low bandwidthrate (to not attract as many circuits) while still
      allowing a higher bandwidthrate in reality.
    - Require BandwidthRate to be at least 20kB/s for servers.
    - Add a NoPublish config option, so you can be a server (e.g. for
      testing running Tor servers in other Tor networks) without
      publishing your descriptor to the primary dirservers.
    - Add a new AddressMap config directive to rewrite incoming socks
      addresses. This lets you, for example, declare an implicit
      required exit node for certain sites.
    - Add a new TrackHostExits config directive to trigger addressmaps
      for certain incoming socks addresses -- for sites that break when
      your exit keeps changing (based on patch from Mike Perry).
    - Split NewCircuitPeriod option into NewCircuitPeriod (30 secs),
      which describes how often we retry making new circuits if current
      ones are dirty, and MaxCircuitDirtiness (10 mins), which describes
      how long we're willing to make use of an already-dirty circuit.
    - Change compiled-in SHUTDOWN_WAIT_LENGTH from a fixed 30 secs to
      a config option "ShutdownWaitLength" (when using kill -INT on
      servers).
    - Fix an edge case in parsing config options: if they say "--"
      on the commandline, it's not a config option (thanks weasel).
    - New config option DirAllowPrivateAddresses for authdirservers.
      Now by default they refuse router descriptors that have non-IP or
      private-IP addresses.
    - Change DirFetchPeriod/StatusFetchPeriod to have a special "Be
      smart" default value: low for servers and high for clients.
    - Some people were putting "Address  " in their torrc, and they had
      a buggy resolver that resolved " " to 0.0.0.0. Oops.
    - If DataDir is ~/.tor, and that expands to /.tor, then default to
      LOCALSTATEDIR/tor instead.
    - Implement --verify-config command-line option to check if your torrc
      is valid without actually launching Tor.

  o Logging improvements:
    - When dirservers refuse a server descriptor, we now log its
      contactinfo, platform, and the poster's IP address.
    - Only warn once per nickname from add_nickname_list_to_smartlist()
      per failure, so an entrynode or exitnode choice that's down won't
      yell so much.
    - When we're connecting to an OR and he's got a different nickname/key
      than we were expecting, only complain loudly if we're an OP or a
      dirserver. Complaining loudly to the OR admins just confuses them.
    - Whine at you if you're a server and you don't set your contactinfo.
    - Warn when exit policy implicitly allows local addresses.
    - Give a better warning when some other server advertises an
      ORPort that is actually an apache running ssl.
    - If we get an incredibly skewed timestamp from a dirserver mirror
      that isn't a verified OR, don't warn -- it's probably him that's
      wrong.
    - When a dirserver causes you to give a warn, mention which dirserver
      it was.
    - Initialize libevent later in the startup process, so the logs are
      already established by the time we start logging libevent warns.
    - Use correct errno on win32 if libevent fails.
    - Check and warn about known-bad/slow libevent versions.
    - Stop warning about sigpipes in the logs. We're going to
      pretend that getting these occassionally is normal and fine.

  o New contrib scripts:
    - New experimental script tor/contrib/exitlist: a simple python
      script to parse directories and find Tor nodes that exit to listed
      addresses/ports.
    - New experimental script tor/contrib/ExerciseServer.py (needs more
      work) that uses the controller interface to build circuits and
      fetch pages over them. This will help us bootstrap servers that
      have lots of capacity but haven't noticed it yet.
    - New experimental script tor/contrib/PathDemo.py (needs more work)
      that uses the controller interface to let you choose whole paths
      via addresses like
      "<hostname>.<path,separated by dots>.<length of path>.path"
    - New contributed script "privoxy-tor-toggle" to toggle whether
      Privoxy uses Tor. Seems to be configured for Debian by default.
    - Have torctl.in/tor.sh.in check for location of su binary (needed
      on FreeBSD)

  o Misc bugfixes:
    - chdir() to your datadirectory at the *end* of the daemonize process,
      not the beginning. This was a problem because the first time you
      run tor, if your datadir isn't there, and you have runasdaemon set
      to 1, it will try to chdir to it before it tries to create it. Oops.
    - Fix several double-mark-for-close bugs, e.g. where we were finding
      a conn for a cell even if that conn is already marked for close.
    - Stop most cases of hanging up on a socks connection without sending
      the socks reject.
    - Fix a bug in the RPM package: set home directory for _tor to
      something more reasonable when first installing.
    - Stop putting nodename in the Platform string in server descriptors.
      It doesn't actually help, and it is confusing/upsetting some people.
    - When using preferred entry or exit nodes, ignore whether the
      circuit wants uptime or capacity. They asked for the nodes, they
      get the nodes.
    - Tie MAX_DIR_SIZE to MAX_BUF_SIZE, so now directory sizes won't get
      artificially capped at 500kB.
    - Cache local dns resolves correctly even when they're .exit
      addresses.
    - If we're hibernating and we get a SIGINT, exit immediately.
    - tor-resolve requests were ignoring .exit if there was a working circuit
      they could use instead.
    - Pay more attention to the ClientOnly config option.
    - Resolve OS X installer bugs: stop claiming to be 0.0.9.2 in certain
      installer screens; and don't put stuff into StartupItems unless
      the user asks you to.

  o Misc features:
    - Rewrite address "serifos.exit" to "externalIP.serifos.exit"
      rather than just rejecting it.
    - If our clock jumps forward by 100 seconds or more, assume something
      has gone wrong with our network and abandon all not-yet-used circs.
    - When an application is using socks5, give him the whole variety of
      potential socks5 responses (connect refused, host unreachable, etc),
      rather than just "success" or "failure".
    - A more sane version numbering system. See
      http://tor.eff.org/cvs/tor/doc/version-spec.txt for details.
    - Change version parsing logic: a version is "obsolete" if it is not
      recommended and (1) there is a newer recommended version in the
      same series, or (2) there are no recommended versions in the same
      series, but there are some recommended versions in a newer series.
      A version is "new" if it is newer than any recommended version in
      the same series.
    - Report HTTP reasons to client when getting a response from directory
      servers -- so you can actually know what went wrong.
    - Reject odd-looking addresses at the client (e.g. addresses that
      contain a colon), rather than having the server drop them because
      they're malformed.
    - Stop publishing socksport in the directory, since it's not
      actually meant to be public. For compatibility, publish a 0 there
      for now.
    - Since we ship our own Privoxy on OS X, tweak it so it doesn't write
      cookies to disk and doesn't log each web request to disk. (Thanks
      to Brett Carrington for pointing this out.)
    - Add OSX uninstall instructions. An actual uninstall script will
      come later.
    - Add "opt hibernating 1" to server descriptor to make it clearer
      whether the server is hibernating.


Changes in version 0.0.9.10 - 2005-06-16
  o Bugfixes on 0.0.9.x (backported from 0.1.0.10):
    - Refuse relay cells that claim to have a length larger than the
      maximum allowed. This prevents a potential attack that could read
      arbitrary memory (e.g. keys) from an exit server's process
      (CVE-2005-2050).


Changes in version 0.0.9.9 - 2005-04-23
  o Bugfixes on 0.0.9.x:
    - If unofficial Tor clients connect and send weird TLS certs, our
      Tor server triggers an assert. This release contains a minimal
      backport from the broader fix that we put into 0.1.0.4-rc.


Changes in version 0.0.9.8 - 2005-04-07
  o Bugfixes on 0.0.9.x:
    - We have a bug that I haven't found yet. Sometimes, very rarely,
      cpuworkers get stuck in the 'busy' state, even though the cpuworker
      thinks of itself as idle. This meant that no new circuits ever got
      established. Here's a workaround to kill any cpuworker that's been
      busy for more than 100 seconds.


Changes in version 0.0.9.7 - 2005-04-01
  o Bugfixes on 0.0.9.x:
    - Fix another race crash bug (thanks to Glenn Fink for reporting).
    - Compare identity to identity, not to nickname, when extending to
      a router not already in the directory. This was preventing us from
      extending to unknown routers. Oops.
    - Make sure to create OS X Tor user in <500 range, so we aren't
      creating actual system users.
    - Note where connection-that-hasn't-sent-end was marked, and fix
      a few really loud instances of this harmless bug (it's fixed more
      in 0.1.0.x).


Changes in version 0.0.9.6 - 2005-03-24
  o Bugfixes on 0.0.9.x (crashes and asserts):
    - Add new end stream reasons to maintainance branch. Fix bug where
      reason (8) could trigger an assert.  Prevent bug from recurring.
    - Apparently win32 stat wants paths to not end with a slash.
    - Fix assert triggers in assert_cpath_layer_ok(), where we were
      blowing away the circuit that conn->cpath_layer points to, then
      checking to see if the circ is well-formed. Backport check to make
      sure we dont use the cpath on a closed connection.
    - Prevent circuit_resume_edge_reading_helper() from trying to package
      inbufs for marked-for-close streams.
    - Don't crash on hup if your options->address has become unresolvable.
    - Some systems (like OS X) sometimes accept() a connection and tell
      you the remote host is 0.0.0.0:0. If this happens, due to some
      other mis-features, we get confused; so refuse the conn for now.

  o Bugfixes on 0.0.9.x (other):
    - Fix harmless but scary "Unrecognized content encoding" warn message.
    - Add new stream error reason: TORPROTOCOL reason means "you are not
      speaking a version of Tor I understand; say bye-bye to your stream."
    - Be willing to cache directories from up to ROUTER_MAX_AGE seconds
      into the future, now that we are more tolerant of skew. This
      resolves a bug where a Tor server would refuse to cache a directory
      because all the directories it gets are too far in the future;
      yet the Tor server never logs any complaints about clock skew.
    - Mac packaging magic: make man pages useable, and do not overwrite
      existing torrc files.
    - Make OS X log happily to /var/log/tor/tor.log


Changes in version 0.0.9.5 - 2005-02-22
  o Bugfixes on 0.0.9.x:
    - Fix an assert race at exit nodes when resolve requests fail.
    - Stop picking unverified dir mirrors--it only leads to misery.
    - Patch from Matt Edman to make NT services work better. Service
      support is still not compiled into the executable by default.
    - Patch from Dmitri Bely so the Tor service runs better under
      the win32 SYSTEM account.
    - Make tor-resolve actually work (?) on Win32.
    - Fix a sign bug when getrlimit claims to have 4+ billion
      file descriptors available.
    - Stop refusing to start when bandwidthburst == bandwidthrate.
    - When create cells have been on the onion queue more than five
      seconds, just send back a destroy and take them off the list.


Changes in version 0.0.9.4 - 2005-02-03
  o Bugfixes on 0.0.9:
    - Fix an assert bug that took down most of our servers: when
      a server claims to have 1 GB of bandwidthburst, don't
      freak out.
    - Don't crash as badly if we have spawned the max allowed number
      of dnsworkers, or we're out of file descriptors.
    - Block more file-sharing ports in the default exit policy.
    - MaxConn is now automatically set to the hard limit of max
      file descriptors we're allowed (ulimit -n), minus a few for
      logs, etc.
    - Give a clearer message when servers need to raise their
      ulimit -n when they start running out of file descriptors.
    - SGI Compatibility patches from Jan Schaumann.
    - Tolerate a corrupt cached directory better.
    - When a dirserver hasn't approved your server, list which one.
    - Go into soft hibernation after 95% of the bandwidth is used,
      not 99%. This is especially important for daily hibernators who
      have a small accounting max. Hopefully it will result in fewer
      cut connections when the hard hibernation starts.
    - Load-balance better when using servers that claim more than
      800kB/s of capacity.
    - Make NT services work (experimental, only used if compiled in).


Changes in version 0.0.9.3 - 2005-01-21
  o Bugfixes on 0.0.9:
    - Backport the cpu use fixes from main branch, so busy servers won't
      need as much processor time.
    - Work better when we go offline and then come back, or when we
      run Tor at boot before the network is up. We do this by
      optimistically trying to fetch a new directory whenever an
      application request comes in and we think we're offline -- the
      human is hopefully a good measure of when the network is back.
    - Backport some minimal hidserv bugfixes: keep rend circuits open as
      long as you keep using them; actually publish hidserv descriptors
      shortly after they change, rather than waiting 20-40 minutes.
    - Enable Mac startup script by default.
    - Fix duplicate dns_cancel_pending_resolve reported by Giorgos Pallas.
    - When you update AllowUnverifiedNodes or FirewallPorts via the
      controller's setconf feature, we were always appending, never
      resetting.
    - When you update HiddenServiceDir via setconf, it was screwing up
      the order of reading the lines, making it fail.
    - Do not rewrite a cached directory back to the cache; otherwise we
      will think it is recent and not fetch a newer one on startup.
    - Workaround for webservers that lie about Content-Encoding: Tor
      now tries to autodetect compressed directories and compression
      itself. This lets us Proxypass dir fetches through apache.


Changes in version 0.0.9.2 - 2005-01-04
  o Bugfixes on 0.0.9 (crashes and asserts):
    - Fix an assert on startup when the disk is full and you're logging
      to a file.
    - If you do socks4 with an IP of 0.0.0.x but *don't* provide a socks4a
      style address, then we'd crash.
    - Fix an assert trigger when the running-routers string we get from
      a dirserver is broken.
    - Make worker threads start and run on win32. Now win32 servers
      may work better.
    - Bandaid (not actually fix, but now it doesn't crash) an assert
      where the dns worker dies mysteriously and the main Tor process
      doesn't remember anything about the address it was resolving.

  o Bugfixes on 0.0.9 (Win32):
    - Workaround for brain-damaged __FILE__ handling on MSVC: keep Nick's
      name out of the warning/assert messages.
    - Fix a superficial "unhandled error on read" bug on win32.
    - The win32 installer no longer requires a click-through for our
      license, since our Free Software license grants rights but does not
      take any away.
    - Win32: When connecting to a dirserver fails, try another one
      immediately. (This was already working for non-win32 Tors.)
    - Stop trying to parse $HOME on win32 when hunting for default
      DataDirectory.
    - Make tor-resolve.c work on win32 by calling network_init().

  o Bugfixes on 0.0.9 (other):
    - Make 0.0.9.x build on Solaris again.
    - Due to a fencepost error, we were blowing away the \n when reporting
      confvalue items in the controller. So asking for multiple config
      values at once couldn't work.
    - When listing circuits that are pending on an opening OR connection,
      if we're an OR we were listing circuits that *end* at us as
      being pending on every listener, dns/cpu worker, etc. Stop that.
    - Dirservers were failing to create 'running-routers' or 'directory'
      strings if we had more than some threshold of routers. Fix them so
      they can handle any number of routers.
    - Fix a superficial "Duplicate mark for close" bug.
    - Stop checking for clock skew for OR connections, even for servers.
    - Fix a fencepost error that was chopping off the last letter of any
      nickname that is the maximum allowed nickname length.
    - Update URLs in log messages so they point to the new website.
    - Fix a potential problem in mangling server private keys while
      writing to disk (not triggered yet, as far as we know).
    - Include the licenses for other free software we include in Tor,
      now that we're shipping binary distributions more regularly.


Changes in version 0.0.9.1 - 2004-12-15
  o Bugfixes on 0.0.9:
    - Make hibernation actually work.
    - Make HashedControlPassword config option work.
    - When we're reporting event circuit status to a controller,
      don't use the stream status code.


Changes in version 0.0.9 - 2004-12-12
  o Bugfixes on 0.0.8.1 (Crashes and asserts):
    - Catch and ignore SIGXFSZ signals when log files exceed 2GB; our
      write() call will fail and we handle it there.
    - When we run out of disk space, or other log writing error, don't
      crash. Just stop logging to that log and continue.
    - Fix isspace() and friends so they still make Solaris happy
      but also so they don't trigger asserts on win32.
    - Fix assert failure on malformed socks4a requests.
    - Fix an assert bug where a hidden service provider would fail if
      the first hop of his rendezvous circuit was down.
    - Better handling of size_t vs int, so we're more robust on 64
      bit platforms.

  o Bugfixes on 0.0.8.1 (Win32):
    - Make windows sockets actually non-blocking (oops), and handle
      win32 socket errors better.
    - Fix parse_iso_time on platforms without strptime (eg win32).
    - win32: when being multithreaded, leave parent fdarray open.
    - Better handling of winsock includes on non-MSV win32 compilers.
    - Change our file IO stuff (especially wrt OpenSSL) so win32 is
      happier.
    - Make unit tests work on win32.

  o Bugfixes on 0.0.8.1 (Path selection and streams):
    - Calculate timeout for waiting for a connected cell from the time
      we sent the begin cell, not from the time the stream started. If
      it took a long time to establish the circuit, we would time out
      right after sending the begin cell.
    - Fix router_compare_addr_to_addr_policy: it was not treating a port
      of * as always matching, so we were picking reject *:* nodes as
      exit nodes too. Oops.
    - When read() failed on a stream, we would close it without sending
      back an end. So 'connection refused' would simply be ignored and
      the user would get no response.
    - Stop a sigpipe: when an 'end' cell races with eof from the app,
      we shouldn't hold-open-until-flush if the eof arrived first.
    - Let resolve conns retry/expire also, rather than sticking around
      forever.
    - Fix more dns related bugs: send back resolve_failed and end cells
      more reliably when the resolve fails, rather than closing the
      circuit and then trying to send the cell. Also attach dummy resolve
      connections to a circuit *before* calling dns_resolve(), to fix
      a bug where cached answers would never be sent in RESOLVED cells.

  o Bugfixes on 0.0.8.1 (Circuits):
    - Finally fix a bug that's been plaguing us for a year:
      With high load, circuit package window was reaching 0. Whenever
      we got a circuit-level sendme, we were reading a lot on each
      socket, but only writing out a bit. So we would eventually reach
      eof. This would be noticed and acted on even when there were still
      bytes sitting in the inbuf.
    - Use identity comparison, not nickname comparison, to choose which
      half of circuit-ID-space each side gets to use. This is needed
      because sometimes we think of a router as a nickname, and sometimes
      as a hex ID, and we can't predict what the other side will do.

  o Bugfixes on 0.0.8.1 (Other):
    - Fix a whole slew of memory leaks.
    - Disallow NDEBUG. We don't ever want anybody to turn off debug.
    - If we are using select, make sure we stay within FD_SETSIZE.
    - When poll() is interrupted, we shouldn't believe the revents values.
    - Add a FAST_SMARTLIST define to optionally inline smartlist_get
      and smartlist_len, which are two major profiling offenders.
    - If do_hup fails, actually notice.
    - Flush the log file descriptor after we print "Tor opening log file",
      so we don't see those messages days later.
    - Hidden service operators now correctly handle version 1 style
      INTRODUCE1 cells (nobody generates them still, so not a critical
      bug).
    - Handle more errnos from accept() without closing the listener.
      Some OpenBSD machines were closing their listeners because
      they ran out of file descriptors.
    - Some people had wrapped their tor client/server in a script
      that would restart it whenever it died. This did not play well
      with our "shut down if your version is obsolete" code. Now people
      don't fetch a new directory if their local cached version is
      recent enough.
    - Make our autogen.sh work on ksh as well as bash.
    - Better torrc example lines for dirbindaddress and orbindaddress.
    - Improved bounds checking on parsed ints (e.g. config options and
      the ones we find in directories.)
    - Stop using separate defaults for no-config-file and
      empty-config-file. Now you have to explicitly turn off SocksPort,
      if you don't want it open.
    - We were starting to daemonize before we opened our logs, so if
      there were any problems opening logs, we would complain to stderr,
      which wouldn't work, and then mysteriously exit.
    - If a verified OR connects to us before he's uploaded his descriptor,
      or we verify him and hup but he still has the original TLS
      connection, then conn->nickname is still set like he's unverified.

  o Code security improvements, inspired by Ilja:
    - tor_snprintf wrapper over snprintf with consistent (though not C99)
      overflow behavior.
    - Replace sprintf with tor_snprintf. (I think they were all safe, but
      hey.)
    - Replace strcpy/strncpy with strlcpy in more places.
    - Avoid strcat; use tor_snprintf or strlcat instead.

  o Features (circuits and streams):
    - New circuit building strategy: keep a list of ports that we've
      used in the past 6 hours, and always try to have 2 circuits open
      or on the way that will handle each such port. Seed us with port
      80 so web users won't complain that Tor is "slow to start up".
    - Make kill -USR1 dump more useful stats about circuits.
    - When warning about retrying or giving up, print the address, so
      the user knows which one it's talking about.
    - If you haven't used a clean circuit in an hour, throw it away,
      just to be on the safe side. (This means after 6 hours a totally
      unused Tor client will have no circuits open.)
    - Support "foo.nickname.exit" addresses, to let Alice request the
      address "foo" as viewed by exit node "nickname". Based on a patch
      from Geoff Goodell.
    - If your requested entry or exit node has advertised bandwidth 0,
      pick it anyway.
    - Be more greedy about filling up relay cells -- we try reading again
      once we've processed the stuff we read, in case enough has arrived
      to fill the last cell completely.
    - Refuse application socks connections to port 0.
    - Use only 0.0.9pre1 and later servers for resolve cells.

  o Features (bandwidth):
    - Hibernation: New config option "AccountingMax" lets you
      set how many bytes per month (in each direction) you want to
      allow your server to consume. Rather than spreading those
      bytes out evenly over the month, we instead hibernate for some
      of the month and pop up at a deterministic time, work until
      the bytes are consumed, then hibernate again. Config option
      "MonthlyAccountingStart" lets you specify which day of the month
      your billing cycle starts on.
    - Implement weekly/monthly/daily accounting: now you specify your
      hibernation properties by
      AccountingMax N bytes|KB|MB|GB|TB
      AccountingStart day|week|month [day] HH:MM
        Defaults to "month 1 0:00".
    - Let bandwidth and interval config options be specified as 5 bytes,
      kb, kilobytes, etc; and as seconds, minutes, hours, days, weeks.

  o Features (directories):
    - New "router-status" line in directory, to better bind each verified
      nickname to its identity key.
    - Clients can ask dirservers for /dir.z to get a compressed version
      of the directory. Only works for servers running 0.0.9, of course.
    - Make clients cache directories and use them to seed their router
      lists at startup. This means clients have a datadir again.
    - Respond to content-encoding headers by trying to uncompress as
      appropriate.
    - Clients and servers now fetch running-routers; cache
      running-routers; compress running-routers; serve compressed
      running-routers.z
    - Make moria2 advertise a dirport of 80, so people behind firewalls
      will be able to get a directory.
    - Http proxy support
      - Dirservers translate requests for http://%s:%d/x to /x
      - You can specify "HttpProxy %s[:%d]" and all dir fetches will
        be routed through this host.
      - Clients ask for /tor/x rather than /x for new enough dirservers.
        This way we can one day coexist peacefully with apache.
      - Clients specify a "Host: %s%d" http header, to be compatible
        with more proxies, and so running squid on an exit node can work.
    - Protect dirservers from overzealous descriptor uploading -- wait
      10 seconds after directory gets dirty, before regenerating.

  o Features (packages and install):
    - Add NSI installer contributed by J Doe.
    - Apply NT service patch from Osamu Fujino. Still needs more work.
    - Commit VC6 and VC7 workspace/project files.
    - Commit a tor.spec for making RPM files, with help from jbash.
    - Add contrib/torctl.in contributed by Glenn Fink.
    - Make expand_filename handle ~ and ~username.
    - Use autoconf to enable largefile support where necessary. Use
      ftello where available, since ftell can fail at 2GB.
    - Ship src/win32/ in the tarball, so people can use it to build.
    - Make old win32 fall back to CWD if SHGetSpecialFolderLocation
      is broken.

  o Features (ui controller):
    - Control interface: a separate program can now talk to your
      client/server over a socket, and get/set config options, receive
      notifications of circuits and streams starting/finishing/dying,
      bandwidth used, etc. The next step is to get some GUIs working.
      Let us know if you want to help out. See doc/control-spec.txt .
    - Ship a contrib/tor-control.py as an example script to interact
      with the control port.
    - "tor --hash-password zzyxz" will output a salted password for
      use in authenticating to the control interface.
    - Implement the control-spec's SAVECONF command, to write your
      configuration to torrc.
    - Get cookie authentication for the controller closer to working.
    - When set_conf changes our server descriptor, upload a new copy.
      But don't upload it too often if there are frequent changes.

  o Features (config and command-line):
    - Deprecate unofficial config option abbreviations, and abbreviations
      not on the command line.
    - Configuration infrastructure support for warning on obsolete
      options.
    - Give a slightly more useful output for "tor -h".
    - Break DirFetchPostPeriod into:
      - DirFetchPeriod for fetching full directory,
      - StatusFetchPeriod for fetching running-routers,
      - DirPostPeriod for posting server descriptor,
      - RendPostPeriod for posting hidden service descriptors.
    - New log format in config:
      "Log minsev[-maxsev] stdout|stderr|syslog" or
      "Log minsev[-maxsev] file /var/foo"
    - DirPolicy config option, to let people reject incoming addresses
      from their dirserver.
    - "tor --list-fingerprint" will list your identity key fingerprint
      and then exit.
    - Make tor --version --version dump the cvs Id of every file.
    - New 'MyFamily nick1,...' config option for a server to
      specify other servers that shouldn't be used in the same circuit
      with it. Only believed if nick1 also specifies us.
    - New 'NodeFamily nick1,nick2,...' config option for a client to
      specify nodes that it doesn't want to use in the same circuit.
    - New 'Redirectexit pattern address:port' config option for a
      server to redirect exit connections, e.g. to a local squid.
    - Add "pass" target for RedirectExit, to make it easier to break
      out of a sequence of RedirectExit rules.
    - Make the dirservers file obsolete.
      - Include a dir-signing-key token in directories to tell the
        parsing entity which key is being used to sign.
      - Remove the built-in bulky default dirservers string.
      - New config option "Dirserver %s:%d [fingerprint]", which can be
        repeated as many times as needed. If no dirservers specified,
        default to moria1,moria2,tor26.
      - Make 'Routerfile' config option obsolete.
    - Discourage people from setting their dirfetchpostperiod more often
      than once per minute.

  o Features (other):
    - kill -USR2 now moves all logs to loglevel debug (kill -HUP to
      get back to normal.)
    - Accept *:706 (silc) in default exit policy.
    - Implement new versioning format for post 0.1.
    - Distinguish between TOR_TLS_CLOSE and TOR_TLS_ERROR, so we can
      log more informatively.
    - Check clock skew for verified servers, but allow unverified
      servers and clients to have any clock skew.
    - Make sure the hidden service descriptors are at a random offset
      from each other, to hinder linkability.
    - Clients now generate a TLS cert too, in preparation for having
      them act more like real nodes.
    - Add a pure-C tor-resolve implementation.
    - Use getrlimit and friends to ensure we can reach MaxConn (currently
      1024) file descriptors.
    - Raise the max dns workers from 50 to 100.


Changes in version 0.0.8.1 - 2004-10-13
  o Bugfixes:
    - Fix a seg fault that can be triggered remotely for Tor
      clients/servers with an open dirport.
    - Fix a rare assert trigger, where routerinfos for entries in
      our cpath would expire while we're building the path.
    - Fix a bug in OutboundBindAddress so it (hopefully) works.
    - Fix a rare seg fault for people running hidden services on
      intermittent connections.
    - Fix a bug in parsing opt keywords with objects.
    - Fix a stale pointer assert bug when a stream detaches and
      reattaches.
    - Fix a string format vulnerability (probably not exploitable)
      in reporting stats locally.
    - Fix an assert trigger: sometimes launching circuits can fail
      immediately, e.g. because too many circuits have failed recently.
    - Fix a compile warning on 64 bit platforms.


Changes in version 0.0.8 - 2004-08-25
  o Bugfixes:
    - Made our unit tests compile again on OpenBSD 3.5, and tor
      itself compile again on OpenBSD on a sparc64.
    - We were neglecting milliseconds when logging on win32, so
      everything appeared to happen at the beginning of each second.
    - Check directory signature _before_ you decide whether you're
      you're running an obsolete version and should exit.
    - Check directory signature _before_ you parse the running-routers
      list to decide who's running.
    - Check return value of fclose while writing to disk, so we don't
      end up with broken files when servers run out of disk space.
    - Port it to SunOS 5.9 / Athena
    - Fix two bugs in saving onion keys to disk when rotating, so
      hopefully we'll get fewer people using old onion keys.
    - Remove our mostly unused -- and broken -- hex_encode()
      function. Use base16_encode() instead. (Thanks to Timo Lindfors
      for pointing out this bug.)
    - Only pick and establish intro points after we've gotten a
      directory.
    - Fix assert triggers: if the other side returns an address 0.0.0.0,
      don't put it into the client dns cache.
    - If a begin failed due to exit policy, but we believe the IP
      address should have been allowed, switch that router to exitpolicy
      reject *:* until we get our next directory.

  o Protocol changes:
    - 'Extend' relay cell payloads now include the digest of the
      intended next hop's identity key. Now we can verify that we're
      extending to the right router, and also extend to routers we
      hadn't heard of before.

  o Features:
    - Tor nodes can now act as relays (with an advertised ORPort)
      without being manually verified by the dirserver operators.
      - Uploaded descriptors of unverified routers are now accepted
        by the dirservers, and included in the directory.
      - Verified routers are listed by nickname in the running-routers
        list; unverified routers are listed as "$<fingerprint>".
      - We now use hash-of-identity-key in most places rather than
        nickname or addr:port, for improved security/flexibility.
      - AllowUnverifiedNodes config option to let circuits choose no-name
        routers in entry,middle,exit,introduction,rendezvous positions.
        Allow middle and rendezvous positions by default.
      - When picking unverified routers, skip those with low uptime and/or
        low bandwidth, depending on what properties you care about.
      - ClientOnly option for nodes that never want to become servers.
    - Directory caching.
      - "AuthoritativeDir 1" option for the official dirservers.
      - Now other nodes (clients and servers) will cache the latest
        directory they've pulled down.
      - They can enable their DirPort to serve it to others.
      - Clients will pull down a directory from any node with an open
        DirPort, and check the signature/timestamp correctly.
      - Authoritative dirservers now fetch directories from other
        authdirservers, to stay better synced.
      - Running-routers list tells who's down also, along with noting
        if they're verified (listed by nickname) or unverified (listed
        by hash-of-key).
      - Allow dirservers to serve running-router list separately.
        This isn't used yet.
      - You can now fetch $DIRURL/running-routers to get just the
        running-routers line, not the whole descriptor list. (But
        clients don't use this yet.)
    - Clients choose nodes proportional to advertised bandwidth.
    - Clients avoid using nodes with low uptime as introduction points.
    - Handle servers with dynamic IP addresses: don't just replace
      options->Address with the resolved one at startup, and
      detect our address right before we make a routerinfo each time.
    - 'FascistFirewall' option to pick dirservers and ORs on specific
      ports; plus 'FirewallPorts' config option to tell FascistFirewall
      which ports are open. (Defaults to 80,443)
    - Try other dirservers immediately if the one you try is down. This
      should tolerate down dirservers better now.
    - ORs connect-on-demand to other ORs
      - If you get an extend cell to an OR you're not connected to,
        connect, handshake, and forward the create cell.
      - The authoritative dirservers stay connected to everybody,
        and everybody stays connected to 0.0.7 servers, but otherwise
        clients/servers expire unused connections after 5 minutes.
    - When servers get a sigint, they delay 30 seconds (refusing new
      connections) then exit. A second sigint causes immediate exit.
    - File and name management:
      - Look for .torrc if no CONFDIR "torrc" is found.
      - If no datadir is defined, then choose, make, and secure ~/.tor
        as datadir.
      - If torrc not found, exitpolicy reject *:*.
      - Expands ~/ in filenames to $HOME/ (but doesn't yet expand ~arma).
      - If no nickname is defined, derive default from hostname.
      - Rename secret key files, e.g. identity.key -> secret_id_key,
        to discourage people from mailing their identity key to tor-ops.
    - Refuse to build a circuit before the directory has arrived --
      it won't work anyway, since you won't know the right onion keys
      to use.
    - Parse tor version numbers so we can do an is-newer-than check
      rather than an is-in-the-list check.
    - New socks command 'resolve', to let us shim gethostbyname()
      locally.
      - A 'tor_resolve' script to access the socks resolve functionality.
      - A new socks-extensions.txt doc file to describe our
        interpretation and extensions to the socks protocols.
    - Add a ContactInfo option, which gets published in descriptor.
    - Write tor version at the top of each log file
    - New docs in the tarball:
      - tor-doc.html.
      - Document that you should proxy your SSL traffic too.
    - Log a warning if the user uses an unsafe socks variant, so people
      are more likely to learn about privoxy or socat.
    - Log a warning if you're running an unverified server, to let you
      know you might want to get it verified.
    - Change the default exit policy to reject the default edonkey,
      kazaa, gnutella ports.
    - Add replace_file() to util.[ch] to handle win32's rename().
    - Publish OR uptime in descriptor (and thus in directory) too.
    - Remember used bandwidth (both in and out), and publish 15-minute
      snapshots for the past day into our descriptor.
    - Be more aggressive about trying to make circuits when the network
      has changed (e.g. when you unsuspend your laptop).
    - Check for time skew on http headers; report date in response to
      "GET /".
    - If the entrynode config line has only one node, don't pick it as
      an exitnode.
    - Add strict{entry|exit}nodes config options. If set to 1, then
      we refuse to build circuits that don't include the specified entry
      or exit nodes.
    - OutboundBindAddress config option, to bind to a specific
      IP address for outgoing connect()s.
    - End truncated log entries (e.g. directories) with "[truncated]".


Changes in version 0.0.7.3 - 2004-08-12
  o Stop dnsworkers from triggering an assert failure when you
    ask them to resolve the host "".


Changes in version 0.0.7.2 - 2004-07-07
  o A better fix for the 0.0.0.0 problem, that will hopefully
    eliminate the remaining related assertion failures.


Changes in version 0.0.7.1 - 2004-07-04
  o When an address resolves to 0.0.0.0, treat it as a failed resolve,
    since internally we use 0.0.0.0 to signify "not yet resolved".


Changes in version 0.0.7 - 2004-06-07
  o Fixes for crashes and other obnoxious bugs:
    - Fix an epipe bug: sometimes when directory connections failed
      to connect, we would give them a chance to flush before closing
      them.
    - When we detached from a circuit because of resolvefailed, we
      would immediately try the same circuit twice more, and then
      give up on the resolve thinking we'd tried three different
      exit nodes.
    - Limit the number of intro circuits we'll attempt to build for a
      hidden service per 15-minute period.
    - Check recommended-software string *early*, before actually parsing
      the directory. Thus we can detect an obsolete version and exit,
      even if the new directory format doesn't parse.
  o Fixes for security bugs:
    - Remember which nodes are dirservers when you startup, and if a
      random OR enables his dirport, don't automatically assume he's
      a trusted dirserver.
  o Other bugfixes:
    - Directory connections were asking the wrong poll socket to
      start writing, and not asking themselves to start writing.
    - When we detached from a circuit because we sent a begin but
      didn't get a connected, we would use it again the first time;
      but after that we would correctly switch to a different one.
    - Stop warning when the first onion decrypt attempt fails; they
      will sometimes legitimately fail now that we rotate keys.
    - Override unaligned-access-ok check when $host_cpu is ia64 or
      arm. Apparently they allow it but the kernel whines.
    - Dirservers try to reconnect periodically too, in case connections
      have failed.
    - Fix some memory leaks in directory servers.
    - Allow backslash in Win32 filenames.
    - Made Tor build complain-free on FreeBSD, hopefully without
      breaking other BSD builds. We'll see.
    - Check directory signatures based on name of signer, not on whom
      we got the directory from. This will let us cache directories more
      easily.
    - Rotate dnsworkers and cpuworkers on SIGHUP, so they get new config
      settings too.
  o Features:
    - Doxygen markup on all functions and global variables.
    - Make directory functions update routerlist, not replace it. So
      now directory disagreements are not so critical a problem.
    - Remove the upper limit on number of descriptors in a dirserver's
      directory (not that we were anywhere close).
    - Allow multiple logfiles at different severity ranges.
    - Allow *BindAddress to specify ":port" rather than setting *Port
      separately. Allow multiple instances of each BindAddress config
      option, so you can bind to multiple interfaces if you want.
    - Allow multiple exit policy lines, which are processed in order.
      Now we don't need that huge line with all the commas in it.
    - Enable accept/reject policies on SOCKS connections, so you can bind
      to 0.0.0.0 but still control who can use your OP.
    - Updated the man page to reflect these features.


Changes in version 0.0.6.2 - 2004-05-16
  o Our integrity-checking digest was checking only the most recent cell,
    not the previous cells like we'd thought.
    Thanks to Stefan Mark for finding the flaw!


Changes in version 0.0.6.1 - 2004-05-06
  o Fix two bugs in our AES counter-mode implementation (this affected
    onion-level stream encryption, but not TLS-level). It turns
    out we were doing something much more akin to a 16-character
    polyalphabetic cipher. Oops.
    Thanks to Stefan Mark for finding the flaw!
  o Retire moria3 as a directory server, and add tor26 as a directory
    server.


Changes in version 0.0.6 - 2004-05-02
  o Features:
    - Hidden services and rendezvous points are implemented. Go to
      http://6sxoyfb3h2nvok2d.onion/ for an index of currently available
      hidden services. (This only works via a socks4a proxy such as
      Privoxy, and currently it's quite slow.)
    - We now rotate link (tls context) keys and onion keys.
    - CREATE cells now include oaep padding, so you can tell
      if you decrypted them correctly.
    - Retry stream correctly when we fail to connect because of
      exit-policy-reject (should try another) or can't-resolve-address.
    - When we hup a dirserver and we've *removed* a server from the
      approved-routers list, now we remove that server from the
      in-memory directories too.
    - Add bandwidthburst to server descriptor.
    - Directories now say which dirserver signed them.
    - Use a tor_assert macro that logs failed assertions too.
    - Since we don't support truncateds much, don't bother sending them;
      just close the circ.
    - Fetch randomness from /dev/urandom better (not via fopen/fread)
    - Better debugging for tls errors
    - Set Content-Type on the directory and hidserv descriptor.
    - Remove IVs from cipher code, since AES-ctr has none.
  o Bugfixes:
    - Fix an assert trigger for exit nodes that's been plaguing us since
      the days of 0.0.2prexx (thanks weasel!)
    - Fix a bug where we were closing tls connections intermittently.
      It turns out openssl keeps its errors around -- so if an error
      happens, and you don't ask about it, and then another openssl
      operation happens and succeeds, and you ask if there was an error,
      it tells you about the first error.
    - Fix a bug that's been lurking since 27 may 03 (!)
      When passing back a destroy cell, we would use the wrong circ id.
    - Don't crash if a conn that sent a begin has suddenly lost its circuit.
    - Some versions of openssl have an SSL_pending function that erroneously
      returns bytes when there is a non-application record pending.
    - Win32 fixes. Tor now compiles on win32 with no warnings/errors.
      o We were using an array of length zero in a few places.
      o Win32's gethostbyname can't resolve an IP to an IP.
      o Win32's close can't close a socket.
      o Handle windows socket errors correctly.
  o Portability:
    - check for <sys/limits.h> so we build on FreeBSD again, and
      <machine/limits.h> for NetBSD.


Changes in version 0.0.5 - 2004-03-30
  o Install torrc as torrc.sample -- we no longer clobber your
    torrc. (Woo!)
  o Fix mangled-state bug in directory fetching (was causing sigpipes).
  o Only build circuits after we've fetched the directory: clients were
    using only the directory servers before they'd fetched a directory.
    This also means longer startup time; so it goes.
  o Fix an assert trigger where an OP would fail to handshake, and we'd
    expect it to have a nickname.
  o Work around a tsocks bug: do a socks reject when AP connection dies
    early, else tsocks goes into an infinite loop.
  o Hold socks connection open until reply is flushed (if possible)
  o Make exit nodes resolve IPs to IPs immediately, rather than asking
    the dns farm to do it.
  o Fix c99 aliasing warnings in rephist.c
  o Don't include server descriptors that are older than 24 hours in the
    directory.
  o Give socks 'reject' replies their whole 15s to attempt to flush,
    rather than seeing the 60s timeout and assuming the flush had failed.
  o Clean automake droppings from the cvs repository
  o Add in a 'notice' log level for things the operator should hear
    but that aren't warnings


Changes in version 0.0.4 - 2004-03-26
  o When connecting to a dirserver or OR and the network is down,
    we would crash.


Changes in version 0.0.3 - 2004-03-26
  o Warn and fail if server chose a nickname with illegal characters
  o Port to Solaris and Sparc:
    - include missing header fcntl.h
    - have autoconf find -lsocket -lnsl automatically
    - deal with hardware word alignment
    - make uname() work (solaris has a different return convention)
    - switch from using signal() to sigaction()
  o Preliminary work on reputation system:
    - Keep statistics on success/fail of connect attempts; they're published
      by kill -USR1 currently.
    - Add a RunTesting option to try to learn link state by creating test
      circuits, even when SocksPort is off.
    - Remove unused open circuits when there are too many.


Changes in version 0.0.2 - 2004-03-19
    - Include strlcpy and strlcat for safer string ops
    - define INADDR_NONE so we compile (but still not run) on solaris


Changes in version 0.0.2pre27 - 2004-03-14
  o Bugfixes:
    - Allow internal tor networks (we were rejecting internal IPs,
      now we allow them if they're set explicitly).
    - And fix a few endian issues.


Changes in version 0.0.2pre26 - 2004-03-14
  o New features:
    - If a stream times out after 15s without a connected cell, don't
      try that circuit again: try a new one.
    - Retry streams at most 4 times. Then give up.
    - When a dirserver gets a descriptor from an unknown router, it
      logs its fingerprint (so the dirserver operator can choose to
      accept it even without mail from the server operator).
    - Inform unapproved servers when we reject their descriptors.
    - Make tor build on Windows again. It works as a client, who knows
      about as a server.
    - Clearer instructions in the torrc for how to set up a server.
    - Be more efficient about reading fd's when our global token bucket
      (used for rate limiting) becomes empty.
  o Bugfixes:
    - Stop asserting that computers always go forward in time. It's
      simply not true.
    - When we sent a cell (e.g. destroy) and then marked an OR connection
      expired, we might close it before finishing a flush if the other
      side isn't reading right then.
    - Don't allow dirservers to start if they haven't defined
      RecommendedVersions
    - We were caching transient dns failures. Oops.
    - Prevent servers from publishing an internal IP as their address.
    - Address a strcat vulnerability in circuit.c


Changes in version 0.0.2pre25 - 2004-03-04
  o New features:
    - Put the OR's IP in its router descriptor, not its fqdn. That way
      we'll stop being stalled by gethostbyname for nodes with flaky dns,
      e.g. poblano.
  o Bugfixes:
    - If the user typed in an address that didn't resolve, the server
      crashed.


Changes in version 0.0.2pre24 - 2004-03-03
  o Bugfixes:
    - Fix an assertion failure in dns.c, where we were trying to dequeue
      a pending dns resolve even if it wasn't pending
    - Fix a spurious socks5 warning about still trying to write after the
      connection is finished.
    - Hold certain marked_for_close connections open until they're finished
      flushing, rather than losing bytes by closing them too early.
    - Correctly report the reason for ending a stream
    - Remove some duplicate calls to connection_mark_for_close
    - Put switch_id and start_daemon earlier in the boot sequence, so it
      will actually try to chdir() to options.DataDirectory
    - Make 'make test' exit(1) if a test fails; fix some unit tests
    - Make tor fail when you use a config option it doesn't know about,
      rather than warn and continue.
    - Make --version work
    - Bugfixes on the rpm spec file and tor.sh, so it's more up to date


Changes in version 0.0.2pre23 - 2004-02-29
  o New features:
    - Print a statement when the first circ is finished, so the user
      knows it's working.
    - If a relay cell is unrecognized at the end of the circuit,
      send back a destroy. (So attacks to mutate cells are more
      clearly thwarted.)
    - New config option 'excludenodes' to avoid certain nodes for circuits.
    - When it daemonizes, it chdir's to the DataDirectory rather than "/",
      so you can collect coredumps there.
 o Bugfixes:
    - Fix a bug in tls flushing where sometimes data got wedged and
      didn't flush until more data got sent. Hopefully this bug was
      a big factor in the random delays we were seeing.
    - Make 'connected' cells include the resolved IP, so the client
      dns cache actually gets populated.
    - Disallow changing from ORPort=0 to ORPort>0 on hup.
    - When we time-out on a stream and detach from the circuit, send an
      end cell down it first.
    - Only warn about an unknown router (in exitnodes, entrynodes,
      excludenodes) after we've fetched a directory.


Changes in version 0.0.2pre22 - 2004-02-26
  o New features:
    - Servers publish less revealing uname information in descriptors.
    - More memory tracking and assertions, to crash more usefully when
      errors happen.
    - If the default torrc isn't there, just use some default defaults.
      Plus provide an internal dirservers file if they don't have one.
    - When the user tries to use Tor as an http proxy, give them an http
      501 failure explaining that we're a socks proxy.
    - Dump a new router.desc on hup, to help confused people who change
      their exit policies and then wonder why router.desc doesn't reflect
      it.
    - Clean up the generic tor.sh init script that we ship with.
  o Bugfixes:
    - If the exit stream is pending on the resolve, and a destroy arrives,
      then the stream wasn't getting removed from the pending list. I
      think this was the one causing recent server crashes.
    - Use a more robust poll on OSX 10.3, since their poll is flaky.
    - When it couldn't resolve any dirservers, it was useless from then on.
      Now it reloads the RouterFile (or default dirservers) if it has no
      dirservers.
    - Move the 'tor' binary back to /usr/local/bin/ -- it turns out
      many users don't even *have* a /usr/local/sbin/.


Changes in version 0.0.2pre21 - 2004-02-18
  o New features:
    - There's a ChangeLog file that actually reflects the changelog.
    - There's a 'torify' wrapper script, with an accompanying
      tor-tsocks.conf, that simplifies the process of using tsocks for
      tor. It even has a man page.
    - The tor binary gets installed to sbin rather than bin now.
    - Retry streams where the connected cell hasn't arrived in 15 seconds
    - Clean up exit policy handling -- get the default out of the torrc,
      so we can update it without forcing each server operator to fix
      his/her torrc.
    - Allow imaps and pop3s in default exit policy
  o Bugfixes:
    - Prevent picking middleman nodes as the last node in the circuit


Changes in version 0.0.2pre20 - 2004-01-30
  o New features:
    - We now have a deb package, and it's in debian unstable. Go to
      it, apt-getters. :)
    - I've split the TotalBandwidth option into BandwidthRate (how many
      bytes per second you want to allow, long-term) and
      BandwidthBurst (how many bytes you will allow at once before the cap
      kicks in).  This better token bucket approach lets you, say, set
      BandwidthRate to 10KB/s and BandwidthBurst to 10MB, allowing good
      performance while not exceeding your monthly bandwidth quota.
    - Push out a tls record's worth of data once you've got it, rather
      than waiting until you've read everything waiting to be read. This
      may improve performance by pipelining better. We'll see.
    - Add an AP_CONN_STATE_CONNECTING state, to allow streams to detach
      from failed circuits (if they haven't been connected yet) and attach
      to new ones.
    - Expire old streams that haven't managed to connect. Some day we'll
      have them reattach to new circuits instead.

  o Bugfixes:
    - Fix several memory leaks that were causing servers to become bloated
      after a while.
    - Fix a few very rare assert triggers. A few more remain.
    - Setuid to User _before_ complaining about running as root.


Changes in version 0.0.2pre19 - 2004-01-07
  o Bugfixes:
    - Fix deadlock condition in dns farm. We were telling a child to die by
      closing the parent's file descriptor to him. But newer children were
      inheriting the open file descriptor from the parent, and since they
      weren't closing it, the socket never closed, so the child never read
      eof, so he never knew to exit. Similarly, dns workers were holding
      open other sockets, leading to all sorts of chaos.
    - New cleaner daemon() code for forking and backgrounding.
    - If you log to a file, it now prints an entry at the top of the
      logfile so you know it's working.
    - The onionskin challenge length was 30 bytes longer than necessary.
    - Started to patch up the spec so it's not quite so out of date.


Changes in version 0.0.2pre18 - 2004-01-02
  o Bugfixes:
    - Fix endian issues with the 'integrity' field in the relay header.
    - Fix a potential bug where connections in state
      AP_CONN_STATE_CIRCUIT_WAIT might unexpectedly ask to write.


Changes in version 0.0.2pre17 - 2003-12-30
  o Bugfixes:
    - Made --debuglogfile (or any second log file, actually) work.
    - Resolved an edge case in get_unique_circ_id_by_conn where a smart
      adversary could force us into an infinite loop.

  o Features:
    - Each onionskin handshake now includes a hash of the computed key,
      to prove the server's identity and help perfect forward secrecy.
    - Changed cell size from 256 to 512 bytes (working toward compatibility
      with MorphMix).
    - Changed cell length to 2 bytes, and moved it to the relay header.
    - Implemented end-to-end integrity checking for the payloads of
      relay cells.
    - Separated streamid from 'recognized' (otherwise circuits will get
      messed up when we try to have streams exit from the middle). We
      use the integrity-checking to confirm that a cell is addressed to
      this hop.
    - Randomize the initial circid and streamid values, so an adversary who
      breaks into a node can't learn how many circuits or streams have
      been made so far.


Changes in version 0.0.2pre16 - 2003-12-14
  o Bugfixes:
    - Fixed a bug that made HUP trigger an assert
    - Fixed a bug where a circuit that immediately failed wasn't being
      counted as a failed circuit in counting retries.

  o Features:
    - Now we close the circuit when we get a truncated cell: otherwise we're
      open to an anonymity attack where a bad node in the path truncates
      the circuit and then we open streams at him.
    - Add port ranges to exit policies
    - Add a conservative default exit policy
    - Warn if you're running tor as root
    - on HUP, retry OR connections and close/rebind listeners
    - options.EntryNodes: try these nodes first when picking the first node
    - options.ExitNodes: if your best choices happen to include any of
      your preferred exit nodes, you choose among just those preferred
      exit nodes.
    - options.ExcludedNodes: nodes that are never picked in path building


Changes in version 0.0.2pre15 - 2003-12-03
  o Robustness and bugfixes:
    - Sometimes clients would cache incorrect DNS resolves, which would
      really screw things up.
    - An OP that goes offline would slowly leak all its sockets and stop
      working.
    - A wide variety of bugfixes in exit node selection, exit policy
      handling, and processing pending streams when a new circuit is
      established.
    - Pick nodes for a path only from those the directory says are up
    - Choose randomly from all running dirservers, not always the first one
    - Increase allowed http header size for directory fetch.
    - Stop writing to stderr (if we're daemonized it will be closed).
    - Enable -g always, so cores will be more useful to me.
    - Switch "-lcrypto -lssl" to "-lssl -lcrypto" for broken distributions.

  o Documentation:
    - Wrote a man page. It lists commonly used options.

  o Configuration:
    - Change default loglevel to warn.
    - Make PidFile default to null rather than littering in your CWD.
    - OnionRouter config option is now obsolete. Instead it just checks
      ORPort>0.
    - Moved to a single unified torrc file for both clients and servers.


Changes in version 0.0.2pre14 - 2003-11-29
  o Robustness and bugfixes:
    - Force the admin to make the DataDirectory himself
      - to get ownership/permissions right
      - so clients no longer make a DataDirectory and then never use it
    - fix bug where a client who was offline for 45 minutes would never
      pull down a directory again
    - fix (or at least hide really well) the dns assert bug that was
      causing server crashes
    - warnings and improved robustness wrt clockskew for certs
    - use the native daemon(3) to daemonize, when available
    - exit if bind() fails
    - exit if neither socksport nor orport is defined
    - include our own tor_timegm (Win32 doesn't have its own)
    - bugfix for win32 with lots of connections
    - fix minor bias in PRNG
    - make dirserver more robust to corrupt cached directory

  o Documentation:
    - Wrote the design document (woo)

  o Circuit building and exit policies:
    - Circuits no longer try to use nodes that the directory has told them
      are down.
    - Exit policies now support bitmasks (18.0.0.0/255.0.0.0) and
      bitcounts (18.0.0.0/8).
    - Make AP connections standby for a circuit if no suitable circuit
      exists, rather than failing
    - Circuits choose exit node based on addr/port, exit policies, and
      which AP connections are standing by
    - Bump min pathlen from 2 to 3
    - Relay end cells have a payload to describe why the stream ended.
    - If the stream failed because of exit policy, try again with a new
      circuit.
    - Clients have a dns cache to remember resolved addresses.
    - Notice more quickly when we have no working circuits

  o Configuration:
    - APPort is now called SocksPort
    - SocksBindAddress, ORBindAddress, DirBindAddress let you configure
      where to bind
    - RecommendedVersions is now a config variable rather than
      hardcoded (for dirservers)
    - Reloads config on HUP
    - Usage info on -h or --help
    - If you set User and Group config vars, it'll setu/gid to them.

Changes in version 0.0.2pre13 - 2003-10-19
  o General stability:
    - SSL_write no longer fails when it returns WANTWRITE and the number
      of bytes in the buf has changed by the next SSL_write call.
    - Fix segfault fetching directory when network is down
    - Fix a variety of minor memory leaks
    - Dirservers reload the fingerprints file on HUP, so I don't have
      to take down the network when I approve a new router
    - Default server config file has explicit Address line to specify fqdn

  o Buffers:
    - Buffers grow and shrink as needed (Cut process size from 20M to 2M)
    - Make listener connections not ever alloc bufs

  o Autoconf improvements:
    - don't clobber an external CFLAGS in ./configure
    - Make install now works
    - create var/lib/tor on make install
    - autocreate a tor.sh initscript to help distribs
    - autocreate the torrc and sample-server-torrc with correct paths

  o Log files and Daemonizing now work:
    - If --DebugLogFile is specified, log to it at -l debug
    - If --LogFile is specified, use it instead of commandline
    - If --RunAsDaemon is set, tor forks and backgrounds on startup