When we first implemented TLS, we assumed in conneciton_handle_write
that a TOR_TLS_WANT_WRITE from flush_buf_tls meant that nothing had
been written. But when we moved our buffers to a ring buffer
implementation back in 0.1.0.5-rc (!), we broke that invariant: it's
possible that some bytes have been written but nothing.
That's bad. It means that if we do a sequence of TLS writes that ends
with a WANTWRITE, we don't notice that we flushed any bytes, and we
don't (I think) decrement buckets.
Fixes bug 7708; bugfix on 0.1.0.5-rc
Also, deprecate the torrc options for the scaling values. It's unlikely anyone
but developers will ever tweak them, even if we provided a single ratio value.
This is meant to avoid conflict with the built-in log() function in
math.h. It resolves ticket 7599. First reported by dhill.
This was generated with the following perl script:
#!/usr/bin/perl -w -i -p
s/\blog\(LOG_(ERR|WARN|NOTICE|INFO|DEBUG)\s*,\s*/log_\L$1\(/g;
s/\blog\(/tor_log\(/g;
Instead of hardcoding the minimum fraction of possible paths to 0.6, we
take it from the user, and failing that from the consensus, and
failing that we fall back to 0.6.
Previously we did this based on the fraction of descriptors we
had. But really, we should be going based on what fraction of paths
we're able to build based on weighted bandwidth, since otherwise a
directory guard or two could make us behave quite oddly.
Implementation for feature 5956
If any circuits were opened during a scaling event, we were scaling attempts
and successes by different amounts. This leads to rounding error.
The fix is to record how many circuits are in a state that hasn't been fully
counted yet, and subtract that before scaling, and add it back afterwords.
Since they use RELAY_EARLY (which can be seen by all hops on the path),
it's not safe to say they actually count as a successful use.
There are also problems with trying to allow them to finish extending due to
the circuit purpose state machine logic. It is way less complicated (and
possibly more semantically coherent) to simply wait until we actually try to
do something with them before claiming we 'used' them.
Also, we shouldn't call timed out circuits 'used' either, for semantic
consistency.
An adversary could let the first stream request succeed (ie the resolve), but
then tag and timeout the remainder (via cell dropping), forcing them on new
circuits.
Rolling back the state will cause us to probe such circuits, which should lead
to probe failures in the event of such tagging due to either unrecognized
cells coming in while we wait for the probe, or the cipher state getting out
of sync in the case of dropped cells.
Path use bias measures how often we can actually succeed using the circuits we
actually try to use. It is a subset of path bias accounting, but it is
computed as a separate statistic because the rate of client circuit use may
vary depending on use case.
This is a minimal refactoring to expose the weighted bandwidth
calculations for each node so I can use them to see what fraction of
nodes, weighted by bandwidth, we have descriptors for.
This is ticket 7706, reported by "bugcatcher." The rationale here
is that if somebody says 'ExcludeNodes {tv}', then they probably
don't just want to block definitely Tuvaluan nodes: they also want
to block nodes that have unknown country, since for all they know
such nodes are also in Tuvalu.
This behavior is controlled by a new GeoIPExcludeUnknown autobool
option. With the default (auto) setting, we exclude ?? and A1 if
any country is excluded. If the option is 1, we add ?? and A1
unconditionally; if the option is 0, we never add them.
(Right now our geoip file doesn't actually seem to include A1: I'm
including it here in case it comes back.)
This feature only takes effect if you have a GeoIP file. Otherwise
you'd be excluding every node.
The implementation is pretty straightforward: parse_extended_hostname() is
modified to drop any leading components from an address like
'foo.aaaaaaaaaaaaaaaa.onion'.
This is an automatically generated commit, from the following perl script,
run with the options "-w -i -p".
s/smartlist_string_num_isin/smartlist_contains_int_as_string/g;
s/smartlist_string_isin((?:_case)?)/smartlist_contains_string$1/g;
s/smartlist_digest_isin/smartlist_contains_digest/g;
s/smartlist_isin/smartlist_contains/g;
s/digestset_isin/digestset_contains/g;
In 6fbdf635 we added a couple of statements like:
if (test) {
...
};
The extraneous semicolons there get flagged as worrisome empty
statements by the cparser library, so let's fix them.
Patch by Christian Grothoff; fixes bug 7115.
Otherwise, it's possible to create streams or circuits with these
bogus IDs, leading to orphaned circuits or streams, or to ones that
can cause bandwidth DOS problems.
Fixes bug 7889; bugfix on all released Tors.
In general, if we tried to use a circ for a stream, but then decided to place
that stream on a different circuit, we need to probe the original circuit
before deciding it was a "success".
We also need to do the same for cannibalized circuits that go unused.
This makes removing items from the middle of the queue into an O(1)
operation, which could prove important as we let onionqueues grow
longer.
Doing this actually makes the code slightly smaller, too.
The right way to set "MaxOnionsPending" was to adjust it until the
processing delay was appropriate. So instead, let's measure how long
it takes to process onionskins (sampling them once we have a big
number), and then limit the queue based on its expected time to
finish.
This change is extra-necessary for ntor, since there is no longer a
reasonable way to set MaxOnionsPending without knowing what mix of
onionskins you'll get.
This patch also reserves 1/3 of the onionskin spots for ntor
handshakes, on the theory that TAP handshakes shouldn't be allowed to
starve their speedier cousins. We can change this later if need be.
Resolves 7291.
The unit of work sent to a cpuworker is now a create_cell_t; its
response is now a created_cell_t. Several of the things that call or
get called by this chain of logic now take create_cell_t or
created_cell_t too.
Since all cpuworkers are forked or spawned by Tor, they don't need a
stable wire protocol, so we can just send structs. This saves us some
insanity, and helps p
As elsewhere, it makes sense when adding or extending a cell type to
actually make the code to parse it into a separate tested function.
This commit doesn't actually make anything use these new functions;
that's for a later commit.
The handshake_digest field was never meaningfully a digest *of* the
handshake, but rather is a digest *from* the handshake that we exapted
to prevent replays of ESTABLISH_INTRO cells. The ntor handshake will
generate it as more key material rather than taking it from any part
of the circuit handshake reply..
I'm going to want a generic "onionskin" type and set of wrappers, and
for that, it will be helpful to isolate the different circuit creation
handshakes. Now the original handshake is in onion_tap.[ch], the
CREATE_FAST handshake is in onion_fast.[ch], and onion.[ch] now
handles the onion queue.
This commit does nothing but move code and adjust header files.
Here we try to handle curve25519 onion keys from generating them,
loading and storing them, publishing them in our descriptors, putting
them in microdescriptors, and so on.
This commit is untested and probably buggy like whoa
This patch moves curve25519_keypair_t from src/or/onion_ntor.h to
src/common/crypto_curve25519.h, and adds new functions to generate,
load, and store keypairs.
The ntor handshake--described in proposal 216 and in a paper by
Goldberg, Stebila, and Ustaoglu--gets us much better performance than
our current approach.
Our old warn_nonlocal_client_ports() would give a bogus warning for
every nonlocal port every time it parsed any ports at all. So if it
parsed a nonlocal socksport, it would complain that it had a nonlocal
socksport...and then turn around and complain about the nonlocal
socksport again, calling it a nonlocal transport or nonlocal dnsport,
if it had any of those.
Fixes bug 7836; bugfix on 0.2.3.3-alpha.
mr-4 reports on #7799 that he was seeing it several times per second,
which suggests that things had gone very wrong.
This isn't a real fix, but it should make Tor usable till we can
figure out the real issue.
This implements the server-side of proposal 198 by detecting when
clients lack the magic list of ciphersuites that indicates that
they're lying faking some ciphers they don't really have. When
clients lack this list, we can choose any cipher that we'd actually
like. The newly allowed ciphersuites are, currently, "All ECDHE-RSA
ciphers that openssl supports, except for ECDHE-RSA-RC4".
The code to detect the cipher list relies on on (ab)use of
SSL_set_session_secret_cb.
It's important not to call choose_array_element_by_weight and then
pass its return value unchecked to smartlist_get : it is allowed to
return -1.
Fixes bug 7756; bugfix on 4e3d07a6 (not in any released Tor)
This is good enough to give P_success >= 999,999,999/1,000,000,000 so
long as the address space is less than 97.95 full. It'd be ridiculous
for that to happen for IPv6, and usome reasonable assumptions, it
would also be pretty silly for IPv4.
With an IPv6 virtual address map, we can basically hand out a new
IPv6 address for _every_ address we connect to. That'll be cool, and
will let us maybe get around prop205 issues.
This uses some fancy logic to try to make the code paths in the ipv4
and the ipv6 case as close as possible, and moves to randomly
generated addresses so we don't need to maintain those stupid counters
that will collide if Tor restarts but apps don't.
Also has some XXXX items to fix to make this useful. More design
needed.
This function gives us a single place to set reasonable default flags
for port_cfg_t entries, to avoid bugs like the one where we weren't
setting ipv4_traffic_ok to 1 on SocksPorts initialized in an older
way.
(This is part 2 of making DNS cache use enabled/disabled on a
per-client port basis. This implements the CacheIPv[46]DNS options,
but not the UseCachedIPv[46] ones.)
(This is part 1 of making DNS cache use enabled/disabled on a
per-client port basis. These options are shuffled around correctly,
but don't do anything yet.)
We want to be saying fast_mem{cmp,eq,neq} when we're doing a
comparison that's allowed to exit early, or tor_mem{cmp,eq,neq} when
we need a data-invariant timing. Direct use of memcmp tends to imply
that we haven't thought about the issue.
This has several advantages, including more resilience to ambient failure.
I still need to rename all the first_hop vars tho.. Saving that for a separate
commit.
Turns out there's more than one way to block a tagged circuit.
This seems to successfully handle all of the normal exit circuits. Hidden
services need additional tweaks, still.
This replaces the old FallbackConsensus notion, and should provide a
way -- assuming we pick reasonable nodes! -- to give clients
suggestions of placs to go to get their first consensus.
Now creating a dir_server_t and adding it are separate functions, and
there are frontend functions for adding a trusted dirserver and a
fallback dirserver.
We use trusted_dir_server_t for two pieces of functionality: a list of
all directory authorities, and a list of initial places to look for
a directory. With this patch we start to separate those two roles.
There is as of now no actual way to be a fallback directory without being
an authority.
This is a customizable extract-and-expand HMAC-KDF for deriving keys.
It derives from RFC5869, which derives its rationale from Krawczyk,
H., "Cryptographic Extraction and Key Derivation: The HKDF Scheme",
Proceedings of CRYPTO 2010, 2010, <http://eprint.iacr.org/2010/264>.
I'm also renaming the existing KDF, now that Tor has two of them.
This is the key derivation scheme specified in ntor.
There are also unit tests.
This one is necessary for sending BEGIN cells with sane flags when
self-testing a directory port. All real entry connections were
getting their ipv{4,6}_traffic_ok flags set from their listeners, and
for begindir entry connections we didn't care, but for directory
self-testing, we had a problem.
Fixes at least one more case of 7493; if there are more lingering
cases of 7493, this might fix them too.
Bug not in any released version of Tor.
Looks like when i was writing the code to set the ipv4_traffic flag on
port_cfg_t, I missed some cases, such as the one where the port was
set from its default value.
Fix for 7493. Bug not in any released Tor.
Previously, I was freaking out about passing an unspec address to
dns_found_answer() on an error, since I was using the address type to
determine whether the error was an error on an ipv4 address lookup or
on an ipv6 address lookup. But now dns_found_answer() has a separate
orig_query_type argument to tell what kind of query it is, so there's
no need to freak out.
* If there's an IPv4 and an IPv6 address, return both in the resolved
cell.
* Treat all resolve requests as permitting IPv6, since by the spec they're
allowed to, and by the code that won't break anything.
IPv4-only exits have an implicit "reject [::]/0", which was making
policy_is_reject_star() return 1 for them, making us refuse to do
hostname lookups.
This fix chanes policy_is_reject_star() to ask about which family we meant.
The code previously detected wildcarding and replaced wildcarded
answers with DNS_STATUS_FAILED_PERMANENT. But that status variable
was no longer used! Remove the status variable, and instead change
the value of 'result' in evdns_callback.
Thank goodness for compiler warnings. In this case,
unused-but-set-variable.
Thanks to Linus for finding this one.
Now, every cached_resolve_t can remember an IPv4 result *and* an IPv6
result. As a light protection against timing-based distinguishers for
IPv6 users (and against complexity!), every forward request generates
an IPv4 *and* an IPv6 request, assuming that we're an IPv6 exit. Once
we have answers or errors for both, we act accordingly.
This patch additionally makes some useful refactorings in the dns.c
code, though there is quite a bit more of useful refactoring that could
be done.
Additionally, have a new interface for the argument passed to the
evdns_callback function. Previously, it was just the original address
we were resolving. But it turns out that, on error, evdns doesn't
tell you the type of the query, so on a failure we didn't know whether
IPv4 or IPv6 queries were failing.
The new convention is to have the first byte of that argument include
the query type. I've refactored the code a bit to make that simpler.
This makes it so we can handle getting an IPv6 in the 3 different
formats we specified it for in RESOLVED cells,
END_STREAM_REASON_EXITPOLICY cells, and CONNECTED cells.
We don't cache IPv6 addresses yet, since proposal 205 isn't
implemented.
There's a refactored function for parsing connected cells; it has unit
tests.
These options are for telling the SOCKSPort that it should allow or
not allow connections to IPv4/IPv6 addresses.
These aren't implemented yet; this is just the code to read the
options and get them into the entrey_connection_t.
Also, count ipv6 timeouts vs others. If we have too many ipv6
requests time out, then we could be degrading performance because of a
broken DNS server that ignores AAAA requests. Other cases in which
we never learn an AAAA address aren't so bad, since they don't slow
A (ipv4) answers down very much.
This is a relatively simple set of changes: we mostly need to
remove a few "but not for IPv6" changes. We also needed to tweak
the handling of DNS code to generate RESOLVED cells that could get
an IPv6 answer in return.
Now, "accept *:80" means "accept all addresses on port 80", and not
just IPv4. For just v4, say "accept *4:80"; for just v6 say "accept
*6:80".
We can parse these policies from torrc just fine, and we should be
successfully keeping them out of descriptors for now.
We also now include appropriate IPv6 addresses in "reject private:*"
By default, "*" means "All IPv4 addresses" with
tor_addr_parse_mask_ports, so I won't break anything. But if the new
EXTENDED_STAR flag is provided, then * means "any address", *4 means
"any IPv4 address" (that is, 0.0.0.0/0), and "*6" means "any IPv6
address" (that is, [::]/0).
This is going to let us have a syntax for specifying exit policies in
torrc that won't drive people mad.
Also, add a bunch of unit tests for tor_addr_parse_mask_ports to test
these new features, and to increase coverage.
We'd like these functions to be circuit-relative so that we can
implement a per-circuit DNS cache and per-circuit DNS cache rules for
proposal 205 or its successors. I'm doing this now, as a part of the
IPv6 exits code, since there are about to be a few more instances
of code using this.
This is the simplest possible workaround: make it safe to call
circuit_cell_queue_clear() on a non-attached circuit, and make it
safe-but-a-LD_BUG-warning to call update_circuit_on_cmux() on a
non-attached circuit.
LocalWords: unstage src Untracked
Apparently some compilers like to eliminate memset() operations on
data that's about to go out-of-scope. I've gone with the safest
possible replacement, which might be a bit slow. I don't think this
is critical path in any way that will affect performance, but if it
is, we can work on that in 0.2.4.
Fixes bug 7352.
There is probably no code that can write the 2 bytes at the end of the
packed_cell_t when the cell is only a 512-byte cell, but let's not get
overconfident there.
Instead of warning about low ports that are advertised, we should have
been warning about low ports that we're listening on. Bug 7285, fix
on 0.2.3.9-alpha.
That's not where I'd want to put a $, but apparently the other
foo/id/<identity> things allow it, as does an arguably valid
interpretation of control-spec.txt. So let's be consistent.
Fix for a piece of bug 7059.
We still want to build on compilers w/o c99 support, such as
(notoriously, shamefully) MSVC.
So I'm commenting out the designated initializers in
circuitmux_ewma.c. The alternative would have been to use some kind
of macros to use designated initializers only when they're
supported, but that's error-prone, and can lead to code having
different meanings under different compilers.
Bug 7286; fix on 0.2.4.4-alpha; spotted by Gisle Vanem.
If we completed the handshake for the v2 link protocol but wound up
negotiating the wong protocol version, we'd become so confused about
what part of the handshake we were in that we'd promptly die with an
assertion.
This is a fix for CVE-2012-2250; it's a bugfix on 0.2.3.6-alpha.
All servers running that version or later should really upgrade.
Bug and fix from "some guy from France." I tweaked his code slightly
to make it log the IP of the offending node, and to forward-port it to
0.2.4.
If we completed the handshake for the v2 link protocol but wound up
negotiating the wong protocol version, we'd become so confused about
what part of the handshake we were in that we'd promptly die with an
assertion.
This is a fix for CVE-2012-2250; it's a bugfix on 0.2.3.6-alpha.
All servers running that version or later should really upgrade.
Bug and fix from "some guy from France." I tweaked his code slightly
to make it log the IP of the offending node.
Clients now consider the ClientRejectInternalAddresses config option
when using a microdescriptor consensus stanza to decide whether
an exit relay would allow exiting to an internal address. Fixes
bug 7190; bugfix on 0.2.3.1-alpha.
Our implementation of parse_short_policy was screwed up: it would
ignore the last character of every short policy. Obviously, that's
broken.
This patch fixes the busted behavior, and adds a bunch of unit tests
to make sure the rest of that function is okay.
Fixes bug 7192; fix on 0.2.3.1-alpha.
Conflicts:
src/or/circuitbuild.c
There was a huge-looking conflict in circuitbuild.c, but the only
change that had been made to circuitbuild.c since I forked off the
split_circuitbuild branch was 17442560c4. So I took the
split_circuitbuild version of the conflicting part, and manually
re-applied the change from 17442560c44e8093f9a..
Failure to do so left us open to a remotely triggerable assertion
failure. Fixes CVE-2012-2249; bugfix on 0.2.3.6-alpha. Reported by
"some guy from France".
This patch is a forward-port to 0.2.4, to work with the new channel
logic.
Failure to do so left us open to a remotely triggerable assertion
failure. Fixes CVE-2012-2249; bugfix on 0.2.3.6-alpha. Reported by
"some guy from France".
We were calling channel_get_actual_remote_descr() before we used the
output of a previous channel_get_canonical_remote_descr(), thus
invalidating its output.
When we merged the channel code, we made the 'address' field of linked
directory connections created with begindir (and their associated edge
connections) contain an address:port string, when they should only
have contained the address part.
This patch also tweaks the interface to the get_descr method of
channels so that it takes a set of flags rather than a single flag.
In 4768c0efe3 (not in any released
version of Tor), we removed a little block of code that set the addr
field of an exit connection used in making a tunneled directory
request. Turns out that wasn't right.
