Previously, when we ran low on memory, we'd close whichever circuits
had the most queued cells. Now, we close those that have the
*oldest* queued cells, on the theory that those are most responsible
for us running low on memory, and that those are the least likely to
actually drain on their own if we wait a little longer.
Based on analysis from a forthcoming paper by Jansen, Tschorsch,
Johnson, and Scheuermann. Fixes bug 9093.
As a bridge authority, before we create our networkstatus document, we
should compute the thresholds needed for the various status flags
assigned to each bridge based on the status of all other bridges. We
then add these thresholds to the networkstatus document for easy access.
Fixes for #1117 and #9859.
Also fix a bug where if the guard we choose first doesn't answer, we
would try the second guard, but once we connected to the second guard
we would abandon it and retry the first one, slowing down bootstrapping.
The fix in both cases is to treat all our initially chosen guards as
acceptable to use.
Fixes bug 9946.
We had updated our "do we have enough microdescs to begin building
circuits?" logic most recently in 0.2.4.10-alpha (see bug 5956), but we
left the bootstrap status event logic at "how far through getting 1/4
of them are we?"
Fixes bug 9958; bugfix on 0.2.2.36, which is where they diverged (see
bug 5343).
According to the manpage, bridges use P256 for conformity and relays
use P224 for speed. But skruffy points out that we've gotten it
backwards in the code.
In this patch, we make the default P256 for everybody.
Fixes bug 9780; bugfix on 0.2.4.8-alpha.
The old code had logic to use a shorter path length if we didn't
have enough nodes. But we don't support 2-node networks anwyay.
Fix for #9926. I'm not calling this a bugfix on any particular
version, since a 2-node network would fail to work for you for a lot
of other reasons too, and it's not clear to me when that began, or if
2-node networks would ever have worked.
This is probably not an exploitable bug, since you would need to have
errno be a large negative value in the unix pluggable-transport launcher
case. Still, best avoided.
Fixes bug 9928; bugfix on 0.2.3.18-rc.
By calling circuit_n_chan_done() unconditionally on close, we were
closing pending connections that might not have been pending quite for
the connection we were closing. Fix for bug 9880.
Thanks to skruffy for finding this and explaining it patiently until
we understood.
To fix#6033, we disabled TLS 1.1 and 1.2. Eventually, OpenSSL fixed
the bug behind #6033.
I've considered alternate implementations that do more testing to see
if there's secretly an OpenSSL 1.0.1c or something that secretly has a
backport of the OpenSSL 1.0.1e fix, and decided against it on the
grounds of complexity.
This isn't actually much of an issue, since only relays send
AUTHENTICATE cells, but while we're removing timestamps, we might as
well do this too.
Part of proposal 222. I didn't take the approach in the proposal of
using a time-based HMAC, since that was a bad-prng-mitigation hack
from SSL3, and in real life, if you don't have a good RNG, you're
hopeless as a Tor server.
For now, round down to the nearest 10 minutes. Later, eliminate entirely by
setting a consensus parameter.
(This rounding is safe because, in 0.2.2, where the timestamp mattered,
REND_REPLAY_TIME_INTERVAL was a nice generous 60 minutes.)
We were freeing these on exit, but when we added the dl_status_map
field to them in fddb814f, we forgot to arrange for it to be freed.
I've moved the cert_list_free() code into its own function, and added
an appropriate dsmap_free() call.
Fixes bug 9644; bugfix on 0.2.4.13-alpha.
The problem was that the server_identity_key_is_set() function could
return true under conditions where we don't really have an identity
key -- specifically, where we used to have one, but we stopped being a
server.
This is a fix for 6979; bugfix on 0.2.2.18-alpha where we added that
assertion to get_server_identity_key().
Fall back to SOMAXCONN if INT_MAX doesn't work.
We'd like to do this because the actual maximum is overrideable by the
kernel, and the value in the header file might not be right at all.
All implementations I can find out about claim that this is supported.
Fix for 9716; bugfix on every Tor.
This would make us do testing circuits "even when cbt is disabled by
consensus, or when we're a directory authority, or when we've failed
to write cbt history to our state file lately." (Roger's words.)
This is a fix for 9671 and an improvement in our fix for 5049.
The original misbehavior was in 0.2.2.14-alpha; the incomplete
fix was in 0.2.3.17-beta.
The spec requires them to do so, and not doing so creates a situation
where they can't send-test because relays won't extend to them because
of the other part of bug 9546.
Fixes bug 9546; bugfix on 0.2.3.6-alpha.
The spec requires them to do so, and not doing so creates a situation
where they can't send-test because relays won't extend to them because
of the other part of bug 9546.
Fixes bug 9546; bugfix on 0.2.3.6-alpha.
Relays previously, when initiating a connection, would only send a
NETINFO after sending an AUTHENTICATE. But bridges, when receiving a
connection, would never send AUTH_CHALLENGE. So relays wouldn't
AUTHENTICATE, and wouldn't NETINFO, and then bridges would be
surprised to be receiving CREATE cells on a non-open circuit.
Fixes bug 9546.
Fortunately, later checks mean that uninitialized data can't get sent
to the network by this bug. Unfortunately, reading uninitialized heap
*can* (in some cases, with some allocators) cause a crash if you get
unlucky and go off the end of a page.
Found by asn. Bugfix on 0.2.4.1-alpha.
When we moved channel_matches_target_addr_for_extend() into a separate
function, its sense was inverted from what one might expect, and we
didn't have a ! in one place where we should have.
Found by skruffy.
When we moved channel_matches_target_addr_for_extend() into a separate
function, its sense was inverted from what one might expect, and we
didn't have a ! in one place where we should have.
Found by skruffy.
This is not the most beautiful possible implementation (it requires
decorating mockable functions with ugly macros), but it actually
works, and is portable across multiple compilers and architectures.
If you pass the --enable-coverage flag on the command line, we build
our testing binaries with appropriate options eo enable coverage
testing. We also build a "tor-cov" binary that has coverage enabled,
for integration tests.
On recent OSX versions, test coverage only works with clang, not gcc.
So we warn about that.
Also add a contrib/coverage script to actually run gcov with the
appropriate options to generate useful .gcov files. (Thanks to
automake, the .o files will not have the names that gcov expects to
find.)
Also, remove generated gcda and gcno files on clean.
We previously used FILENAME_PRIVATE identifiers mostly for
identifiers exposed only to the unit tests... but also for
identifiers exposed to the benchmarker, and sometimes for
identifiers exposed to a similar module, and occasionally for no
really good reason at all.
Now, we use FILENAME_PRIVATE identifiers for identifiers shared by
Tor and the unit tests. They should be defined static when we
aren't building the unit test, and globally visible otherwise. (The
STATIC macro will keep us honest here.)
For identifiers used only by the unit tests and never by Tor at all,
on the other hand, we wrap them in #ifdef TOR_UNIT_TESTS.
This is not the motivating use case for the split test/non-test
build system; it's just a test example to see how it works, and to
take a chance to clean up the code a little.
This is mainly a matter of automake trickery: we build each static
library in two versions now: one with the TOR_UNIT_TESTS macro
defined, and one without. When TOR_UNIT_TESTS is defined, we can
enable mocking and expose more functions. When it's not defined, we
can lock the binary down more.
The alternatives would be to have alternate build modes: a "testing
configuration" for building the libraries with test support, and a
"production configuration" for building them without. I don't favor
that approach, since I think it would mean more people runnning
binaries build for testing, or more people not running unit tests.
Fix a bug in the voting algorithm that could yield incorrect results
when a non-naming authority declared too many flags. Fixes bug 9200;
bugfix on 0.2.0.3-alpha.
Found by coverity scan.
This implements "algorithm 1" from my discussion of bug #9072: on OOM,
find the circuits with the longest queues, and kill them. It's also a
fix for #9063 -- without the side-effects of bug #9072.
The memory bounds aren't perfect here, and you need to be sure to
allow some slack for the rest of Tor's usage.
This isn't a perfect fix; the rest of the solutions I describe on
codeable.
doc/TODO and doc/spec/README were placeholders to tell people where to
look for the real TODO and README stuff -- we replaced them years ago,
though.
authority-policy, v3-authority-howto, and torel-design.txt belong in
torspec. I'm putting them in attic there since I think they may be in
large part obsolete, but someone can rescue them if they're not.
translations.txt is outdated, and refers to lots of programs other
than Tor. We have much better translation resources on the website
now.
tor-win32-mingw-creation.txt is pending review of a revised version
for 0.2.5 (see ticket #4520), but there's no reason to ship this one
while we're waiting for an accurate version.
the tor-rpm-creation.txt isn't obsolete AFAIK, but it belongs in
doc/contrib if anywhere.
Resolves bug #8965.
This is a reprise of the fix in bdff7e3299d78; 6905c1f6 reintroduced
that bug. Briefly: windows doesn't seem to like deleting a mapped
file. I tried adding the PROT_SHARED_DELETE flag to the createfile
all, but that didn't actually fix this issue. Fortunately, the unit
test I added in 4f4fc63fea should
prevent us from making this particular screw-up again.
This patch also tries to limit the crash potential of a failure to
write by a little bit, although it could do a better job of retaining
microdescriptor bodies.
Fix for bug 8822, bugfix on 0.2.4.12-alpha.
This reverts commit 884a0e269c.
I'm reverting this because it doesn't actually make the problem go
away. It appears that instead we need to do unmap-then-replace.
A comment by rransom on #8795 taken together with a comment by doorss
recorded on #2077 suggest that *every* attempt to replace the md cache
will fail on Vista/Win7 if we don't have the FILE_SHARE_DELETE flag
passed to CreateFile, and if we try to replace the file ourselves
before unmapping it. I'm adding the FILE_SHARE_DELETE, since that's
this simplest fix. Broken indexers (the favored #2077 hypothesis)
could still cause trouble here, but at least this patch should make us
stop stepping on our own feet.
Likely fix for #2077 and its numerous duplicates. Bugfix on
0.2.2.6-alpha, which first had a microdescriptor cache that would get
replaced before remapping it.
There's an assertion failure that can occur if a connection has
optimistic data waiting, and then the connect() call returns 0 on the
first attempt (rather than -1 and EINPROGRESS). That latter behavior
from connect() appears to be an (Open?)BSDism when dealing with remote
addresses in some cases. (At least, I've only seen it reported with
the BSDs under libevent, even when the address was 127.0.0.1. And
we've only seen this problem in Tor with OpenBSD.)
Fixes bug 9017; bugfix on 0.2.3.1-alpha, which first introduced
optimistic data. (Although you could also argue that the commented-out
connection_start_writing in 155c9b80 back in 2002 is the real source
of the issue.)
A new option TestingV3AuthVotingStartOffset is added which offsets the
starting time of the voting interval. This is possible only when
TestingTorNetwork is set.
This patch makes run_scheduled_events() check for new consensus
downloads every second when TestingTorNetwork, instead of every
minute. This should be fine, see #8532 for reasoning.
This patch also brings MIN_VOTE_SECONDS and MIN_DIST_SECONDS down from
20 to 2 seconds, unconditionally. This makes sanity checking of
misconfiguration slightly less sane.
Addresses #8532.
This is a fix for bug 8844, where eugenis correctly notes that there's
a sentinel value at the end of the list-of-freelists that's never
actually checked. It's a bug since the first version of the chunked
buffer code back in 0.2.0.16-alpha.
This would probably be a crash bug if it ever happens, but nobody's
ever reported something like this, so I'm unsure whether it can occur.
It would require write_to_buf, write_to_buf_zlib, read_to_buf, or
read_to_buf_tls to get an input size of more than 32K. Still, it's a
good idea to fix this kind of thing!
It appears that moria1 crashed because of one instance of this (the
one in router_counts_toward_thresholds). The other instance I fixed
won't actually have broken anything, but I think it's more clear this
way.
Fixes bug 8833; bugfix on 0.2.4.12-alpha.
I believe this was introduced in 6bc071f765, which makes
this a fix on 0.2.0.10-alpha. But my code archeology has not extended
to actually testing that theory.
It seems that some versions of clang that would prefer the
-Wswitch-enum compiler flag to warn about switch statements with
missing enum values, even if those switch statements have a
default.
Fixes bug 8598; bugfix on 0.2.4.10-alpha.
Found while investigating 8093, but probably not the cause of it,
since this bug would result in us sending too few SENDMEs, not in us
receiving SENDMEs unexpectedly.
Bugfix on the fix for 7889, which has appeared in 0.2.4.10-alpha, but
not yet in any released 0.2.3.x version.
It was previously --Test in the help output and --test-commandline in
the getopt call. The man page already had --test.
(Originally by David, who resolved the tie in favor of "--test"; I
chose --test-commandline" instead so that nothing that depended
on it could break. -Nick)
If we get a write error on a SOCKS connection, we can't send a
SOCKS reply, now can we?
This bug has been here since 36baf7219, where we added the "hey, I'm
closing an AP connection but I haven't finished the socks
handshake!" message. It's bug 8427.
Also, don't call the exit node 'reject *' unless our decision to pick
that node was based on a non-summarized version of that node's exit
policy.
rransom and arma came up with the ideas for this fix.
Fix for 7582; the summary-related part is a bugfix on 0.2.3.2-alpha.
When we're hibernating, the main reqason we can't bootstrap will
always be that we're hibernating: reporting anything else at severity
WARN is pointless.
Fixes part of 7302.
This bug affects hosts where time_t is unsigned, which AFAICT does
not include anything we currently support. (It _does_ include
OpenVMS, about a month of BSD4.2's history[1], and a lot of the 1970s.)
There are probably more bugs when time_t is unsigned. This one was
[1] http://mail-index.netbsd.org/tech-userlevel/1998/06/04/0000.html
Inspired by #8042.
As far as I know, OpenVMS is the only place you're likely to hit an
unsigned time_t these days, and Tor's VMS support
is... lacking. Still worth letting people know about it, though.
This should have been 2 bytes all along, since version numbers can
be 16 bits long. This isn't a live bug, since the call to
is_or_protocol_version_known in channel_tls_process_versions_cell
will reject any version number not in the range 1..4. Still, let's
fix this before we accidentally start supporting version 256.
Reported pseudonymously. Fixes bug 8062; bugfix on 0.2.0.10-alpha --
specifically, on commit 6fcda529, where during development I
increased the width of a version to 16 bits without changing the
type of link_proto.
Our ++ should have been += 2. This means that we'd accept version
numbers even when they started at an odd position.
This bug should be harmless in practice for so long as every version
number we allow begins with a 0 byte, but if we ever have a version
number starting with 1, 2, 3, or 4, there will be trouble here.
Fix for bug 8059, reported pseudonymously. Bugfix on 0.2.0.10-alpha
-- specifically, commit 6fcda529, where during development I
increased the width of a version to 16 bits without changing the
loop step.
I have no idea whether b0rken clients will DoS the network if the v2
authorities all turn this on or not. It's experimental. See #6783 for
a description of how to test it more or less safely, and please be
careful!
Now the manpages no longer refer to tsocks or tsocks.conf, and we no
longer have or ship a tor-tsocks.conf. The only remaining instances
of "tsocks" in our repository are old ChangeLog and ReleaseNotes
entries, and the torify script saying that it doesn't support tsocks.
Fixes bug 8290.
In a number of places, we decrement timestamp_dirty by
MaxCircuitDirtiness in order to mark a stream as "unusable for any
new connections.
This pattern sucks for a few reasons:
* It is nonobvious.
* It is error-prone: decrementing 0 can be a bad choice indeed.
* It really wants to have a function.
It can also introduce bugs if the system time jumps backwards, or if
MaxCircuitDirtiness is increased.
So in this patch, I add an unusable_for_new_conns flag to
origin_circuit_t, make it get checked everywhere it should (I looked
for things that tested timestamp_dirty), and add a new function to
frob it.
For now, the new function does still frob timestamp_dirty (after
checking for underflow and whatnot), in case I missed any cases that
should be checking unusable_for_new_conns.
Fixes bug 6174. We first used this pattern in 516ef41ac1,
which I think was in 0.0.2pre26 (but it could have been 0.0.2pre27).
Without this patch, there's no way to know what went wrong when we
fail to parse a torrc line entirely (that is, we can't turn it into
a K,V pair.) This patch introduces a new function that yields an
error message on failure, so we can at least tell the user what to
look for in their nonfunctional torrc.
(Actually, it's the same function as before with a new name:
parse_config_line_from_str is now a wrapper macro that the unit
tests use.)
Fixes bug 7950; fix on 0.2.0.16-alpha (58de695f90) which first
introduced the possibility of a torrc value not parsing correctly.
Instead of capping whenever a router has fewer than 3 measurements,
we cap whenever a router has fewer than 3 measurements *AND* there
are at least 3 authorities publishing measured bandwidths.
We also generate bandwidth lines with a new "Unmeasured=1" flag,
meaning that we didn't have enough observations for a node to use
measured bandwidth values in the authority's input, whether we capped
it or not.
There are two ways to use sysconf to ask about the number of
CPUs. When we're on a VM, we would sometimes get it wrong by asking
for the number of total CPUs (say, 64) when we should have been asking
for the number of CPUs online (say, 1 or 2).
Fix for bug 8002.
Stop marking every relay as having been down for one hour every
time we restart a directory authority. These artificial downtimes
were messing with our Stable and Guard flag calculations.
Fixes bug 8218 (introduced by the fix for 1035). Bugfix on 0.2.2.23-alpha.
Relays used to check every 10 to 60 seconds, as an accidental side effect
of calling directory_fetches_from_authorities() when considering doing
a directory fetch. The fix for bug 1992 removes that side effect. At the
same time, bridge relays never had the side effect, leading to confused
bridge operators who tried crazy tricks to get their bridges to notice
IP address changes (see ticket 1913).
The new behavior is to reinstate an every-60-seconds check for both
public relays and bridge relays, now that the side effect is gone.
For example, we were doing a resolve every time we think about doing a
directory fetch. Now we reuse the cached answer in some cases.
Fixes bugs 1992 (bugfix on 0.2.0.20-rc) and 2410 (bugfix on
0.1.2.2-alpha).
When we compute the estimated microseconds we need to handle our
pending onionskins, we could (in principle) overflow a uint32_t if
we ever had 4 million pending onionskins before we had any data
about how onionskins take. Nevertheless, let's compute it properly.
Fixes bug 8210; bugfix on 0.2.4.10. Found by coverity; this is CID
980651.
The refactoring in commit 471ab34032 wasn't complete enough: we
were checking the auth_len variable, but never actually setting it,
so it would never seem that authentication had been provided.
This commit also removes a bunch of unused variables from
rend_service_introduce, whose unusedness we hadn't noticed because
we were wiping them at the end of the function.
Fix for bug 8207; bugfix on 0.2.4.1-alpha.
