Since svn r1475/git 5b6099e8 in tor-0.0.6, we have responded to an
exhaustion of all 65535 stream IDs on a circuit by marking that
circuit for close. That's not the right response. Instead, we
should mark the circuit as "too dirty for new circuits".
Of course in reality this isn't really right either. If somebody
has managed to cram 65535 streams onto a circuit, the circuit is
probably not going to work well for any of those streams, so maybe
we should be limiting the number of streams on an origin circuit
concurrently.
Also, closing the stream in this case is probably the wrong thing to
do as well, but fixing that can also wait.
We fixed bug 539 (where directories would say "503" but send data
anyway) back in 0.2.0.16-alpha/0.1.2.19. Because most directory
versions were affected, we added workaround to make sure that we
examined the contents of 503-replies to make sure there wasn't any
data for them to find. But now that such routers are nonexistent,
we can remove this code. (Even if somebody fired up an 0.1.2.19
directory cache today, it would still be fine to ignore data in its
erroneous 503 replies.)
The first was genuinely impossible, I think: it could only happen
when the amount we read differed from the amount we wanted to read
by more than INT_MAX.
The second is just very unlikely: it would give incorrect results to
the controller if you somehow wrote or read more than 4GB on one
edge conn in one second. That one is a bugfix on 0.1.2.8-beta.
In afe414 (tor-0.1.0.1-rc~173), when we moved to
connection_edge_end_errno(), we used it in handling errors from
connection_connect(). That's not so good, since by the time
connection_connect() returns, the socket is no longer set, and we're
supposed to be looking at the socket_errno return value from
connection_connect() instead. So do what we should've done, and
look at the socket_errno value that we get from connection_connect().
Right now, we only consider sending stream-level SENDME cells when we
have completely flushed a connection_edge's outbuf, or when it sends
us a DATA cell. Neither of these is ideal for throughput.
This patch changes the behavior so we now call
connection_edge_consider_sending_sendme when we flush _some_ data from
an edge outbuf.
Fix for bug 2756; bugfix on svn r152.
Name the magic value "10" rather than re-deriving it.
Comment more.
Use the pattern that works for periodic timers, not the pattern that
doesn't work. ;)
It is important to verify the uptime claim of a relay instead of just
trusting it, otherwise it becomes too easy to blackhole a specific
hidden service. rephist already has data available that we can use here.
Bugfix on 0.2.0.10-alpha.
Partial backport of daa0326aaa .
Resolves bug 2402. Bugfix on 0.2.1.15 (for the part where we switched to
git) and on 0.2.1.30 (for the part where we dumped micro-revisions.)
The calculation of when to send the logmessage was correct, but we
didn't give the correct number of relays required: We want more than
half of all authorities we know about. Fixes bug 2663.
This fixes a remotely triggerable assert on directory authorities, who
don't handle descriptors with ipv6 contents well yet. We will want to
revert this once we're ready to handle ipv6.
Issue raised by lorth on #tor, who wasn't able to use Tor anymore.
Analyzed with help from Christian Fromme. Fix suggested by arma. Bugfix
on 0.2.1.3-alpha.
We want to use the discard port correctly, so a htons() was missing.
Also we need to set it correctly depending on address family.
Review provided by danieldg
This should fix a bug that special ran into, where if your state file
didn't record period maxima, it would never decide that it had
successfully parsed itself unless you got lucky with your
uninitialized-variable values.
This patch also tries to improve error messags in the case where a
maximum value legitimately doesn't parse.
In private networks, the defaults for some options are changed. This
means that in options_validate(), where we're testing that the defaults
are what we think they are, we fail. Use a workaround by setting a
hidden configuration option _UsingTestingTorNetwork when we have altered
the configuration this way, so that options_validate() can do the right
thing.
Fixes bug 2250, bugfix on 0.2.1.2-alpha (the version introducing private
network options).
Our regular DH parameters that we use for circuit and rendezvous
crypto are unchanged. This is yet another small step on the path of
protocol fingerprinting resistance.
(Backport from 0.2.2's 5ed73e3807)
Patch our implementation of tor_lockfile_lock() to handle this case
correctly. Also add a note that blocking behaviour differs from windows
to *nix. Fixes bug 2504, issue pointed out by mobmix.
rransom noticed that a change of ORPort is just as bad as a change of IP
address from a client's perspective, because both mean that the relay is
not available to them while the new information hasn't propagated.
Change the bug1035 fix accordingly.
Also make sure we don't log a bridge's IP address (which might happen
when we are the bridge authority).
When calling circuit_build_times_shuffle_and_store_array, we were
passing a uint32_t as an int. arma is pretty sure that this can't
actually cause a bug, because of checks elsewhere in the code, but
it's best not to pass a uint32_t as an int anyway.
Found by doorss; fix on 0.2.2.4-alpha.
We detect and reject said attempts if there is no chosen exit node or
circuit: connecting to a private addr via a randomly chosen exit node
will usually fail (if all exits reject private addresses), is always
ill-defined (you're not asking for any particular host or service),
and usually an error (you've configured all requests to go over Tor
when you really wanted to configure all _remote_ requests to go over
Tor).
This can also help detect forwarding loop requests.
Found as part of bug2279.
Left circuit_build_times_get_bw_scale() uncommented because it is in the wrong
place due to an improper bug2317 fix. It needs to be moved and renamed, as it
is not a cbt parameter.
To quote arma: "So instead of stopping your CBT from screaming, you're just
going to throw it in the closet and hope you can't hear it?"
Yep. The log message can happen because at 95% point on the curve, we can be
way beyond the max timeout we've seen, if the curve has few points and is
shallow.
Also applied Nick's rule of thumb for rewriting some other notice log messages
to read like how you would explain them to a raving lunatic on #tor who was
shouting at you demanding what they meant. Hopefully the changes live up to
that standard.
If we got a signed digest that was shorter than the required digest
length, but longer than 20 bytes, we would accept it as long
enough.... and then immediately fail when we want to check it.
Fixes bug 2409; bug in 0.2.2.20-alpha; found by piebeer.
Previously if you wanted to say "All messages except network
messages", you needed to say "[*,~net]" and if you said "[~net]" by
mistake, you would get no messages at all. Now, if you say "[~net]",
you get everything except networking messages.
When we added support for separate client tls certs on bridges in
a2bb0bfdd5 we forgot to correctly initialize this when changing
from relay to bridge or vice versa while Tor is running. Fix that
by always initializing keys when the state changes.
Fixes bug 2433.
When we stopped using svn, 0.2.1.x lost the ability to notice its svn
revision and report it in the version number. However, it kept
looking at the micro-revision.i file... so if you switched to master,
built tor, then switched to 0.2.1.x, you'd get a micro-revision.i file
from master reported as an SVN tag. This patch takes out the "include
the svn tag" logic entirely.
Bugfix on 0.2.1.15-rc; fixes bug 2402.
Our regular DH parameters that we use for circuit and rendezvous
crypto are unchanged. This is yet another small step on the path of
protocol fingerprinting resistance.
We need to make sure that the worst thing that a weird consensus param
can do to us is to break our Tor (and only if the other Tors are
reliably broken in the same way) so that the majority of directory
authorities can't pull any attacks that are worse than the DoS that
they can trigger by simply shutting down.
One of these worse things was the cbtnummodes parameter, which could
lead to heap corruption on some systems if the value was sufficiently
large.
This commit fixes this particular issue and also introduces sanity
checking for all consensus parameters.
Our public key functions assumed that they were always writing into a
large enough buffer. In one case, they weren't.
(Incorporates fixes from sebastian)
In dnsserv_resolved(), we carefully made a nul-terminated copy of the
answer in a PTR RESOLVED cell... then never used that nul-terminated
copy. Ouch.
Surprisingly this one isn't as huge a security problem as it could be.
The only place where the input to dnsserv_resolved wasn't necessarily
nul-terminated was when it was called indirectly from relay.c with the
contents of a relay cell's payload. If the end of the payload was
filled with junk, eventdns.c would take the strdup() of the name [This
part is bad; we might crash there if the cell is in a bad part of the
stack or the heap] and get a name of at least length
495[*]. eventdns.c then rejects any name of length over 255, so the
bogus data would be neither transmitted nor altered.
[*] If the name was less than 495 bytes long, the client wouldn't
actually be reading off the end of the cell.
Nonetheless this is a reasonably annoying bug. Better fix it.
Found while looking at bug 2332, reported by doorss. Bugfix on
0.2.0.1-alpha.
The C standard says that INT32_MAX is supposed to be a signed
integer. On platforms that have it, we get the correct
platform-defined value. Our own replacement, however, was
unsigned. That's going to cause a bug somewhere eventually.
Previously, we only looked at up to 128 bytes. This is a bad idea
since socks messages can be at least 256+x bytes long. Now we look at
up to 512 bytes; this should be enough for 0.2.2.x to handle all valid
SOCKS messages. For 0.2.3.x, we can think about handling trickier
cases.
Fixes 2330. Bugfix on 0.2.0.16-alpha.
Right now, Tor routers don't save the maxima values from the
bw_history_t between sessions. That's no good, since we use those
values to determine bandwidth. This code adds a new BWHist.*Maximum
set of values to the state file. If they're not present, we estimate
them by taking the observed total bandwidth and dividing it by the
period length, which provides a lower bound.
This should fix bug 1863. I'm calling it a feature.
Previously, our state parsing code would fail to parse a bwhist
correctly if the Interval was anything other than the default
hardcoded 15 minutes. This change makes the parsing less incorrect,
though the resulting history array might get strange values in it if
the intervals don't match the one we're using. (That is, if stuff was
generated in 15 minute intervals, and we read it into an array that
expects 30 minute intervals, we're fine, since values can be combined
pairwise. But if we generate data at 30 minute intervals and read it
into 15 minute intervals, alternating buckets will be empty.)
Bugfix on 0.1.1.11-alpha.
The trick of looping from i=0..4 , switching on i to set up some
variables, then running some common code is much better expressed by
just calling a function 4 times with 4 sets of arguments. This should
make the code a little easier to follow and maintain here.
An object, you'll recall, is something between -----BEGIN----- and
-----END----- tags in a directory document. Some of our code, as
doorss has noted in bug 2352, could assert if one of these ever
overflowed SIZE_T_CEILING but not INT_MAX. As a solution, I'm setting
a maximum size on a single object such that neither of these limits
will ever be hit. I'm also fixing the INT_MAX checks, just to be sure.
When using libevent 2, we use evdns_base_resolve_*(). When not, we
fake evdns_base_resolve_*() using evdns_resolve_*().
Our old check was looking for negative values (like libevent 2
returns), but our eventdns.c code returns 1. This code makes the
check just test for nonzero.
Note that this broken check was not for _resolve_ failures or even for
failures to _launch_ a resolve: it was for failures to _create_ or
_encode_ a resolve request.
Bug introduced in 81eee0ecfff3dac1e9438719d2f7dc0ba7e84a71; found by
lodger; uploaded to trac by rransom. Bug 2363. Fix on 0.2.2.6-alpha.
We were not decrementing "available" every time we did
++next_virtual_addr in addressmap_get_virtual_address: we left out the
--available when we skipped .00 and .255 addresses.
This didn't actually cause a bug in most cases, since the failure mode
was to keep looping around the virtual addresses until we found one,
or until available hit zero. It could have given you an infinite loop
rather than a useful message, however, if you said "VirtualAddrNetwork
127.0.0.255/32" or something broken like that.
Spotted by cypherpunks
We were decrementing "available" twice for each in-use address we ran
across. This would make us declare that we ran out of virtual
addresses when the address space was only half full.
On Windows, we never use pthreads, since it doesn't usually exist,
and when it does it tends to be a little weirdly-behaved. But some
mingw installations have a pthreads installed, so autoconf detects
pthread.h and tells us about it. This would make us include
pthread.h, which could make for trouble when the iffy pthread.h
tried to include config.h.
This patch changes compat.h so that we never include pthread.h on
Windows. Fixes bug 2313; bugfix on 0.1.0.1-rc.
It's all too easy in C to convert an unsigned value to a signed one,
which will (on all modern computers) give you a huge signed value. If
you have a size_t value of size greater than SSIZE_T_MAX, that is way
likelier to be an underflow than it is to be an actual request for
more than 2gb of memory in one go. (There's nothing in Tor that
should be trying to allocate >2gb chunks.)
If you had TIME_MAX > INT_MAX, and your "time_to_exhaust_bw =
accountingmax/expected_bandwidth_usage * 60" calculation managed to
overflow INT_MAX, then your time_to_consider value could underflow and
wind up being rediculously low or high. "Low" was no problem;
negative values got caught by the (time_to_consider <= 0) check.
"High", however, would get you a wakeup time somewhere in the distant
future.
The fix is to check for time_to_exhaust_bw overflowing INT_MAX, not
TIME_MAX: We don't allow any accounting interval longer than a month,
so if time_to_exhaust_bw is significantly larger than 31*24*60*60, we
can just clip it.
This is a bugfix on 0.0.9pre6, when accounting was first introduced.
It fixes bug 2146, unless there are other causes there too. The fix
is from boboper. (I tweaked it slightly by removing an assignment
that boboper marked as dead, and lowering a variable that no longer
needed to be function-scoped.)
The old logic would have us fetch from authorities if we were refusing
unknown exits and our exit policy was reject*. Instead, we want to
fetch from authorities if we're refusing unknown exits and our exit
policy is _NOT_ reject*.
Fixed by boboper. Fixes more of 2097. Bugfix on 0.2.2.16-alpha.
We use a hash of the identity key to seed a prng to tell when an
accounting period should end. But thanks to the bug998 changes,
clients no longer have server-identity keys to use as a long-term seed
in accounting calculations. In any case, their identity keys (as used
in TLS) were never never fixed. So we can just set the wakeup time
from a random seed instead there. Still open is whether everybody
should be random.
This patch fixes bug 2235, which was introduced in 0.2.2.18-alpha.
Diagnosed with help from boboper on irc.
The reason the "streams problem" occurs is due to the complicated
interaction between Tor's congestion control and libevent. At some point
during the experiment, the circuit window is exhausted, which blocks all
edge streams. When a circuit level sendme is received at Exit, it
resumes edge reading by looping over linked list of edge streams, and
calling connection_start_reading() to inform libevent to resume reading.
When the streams are activated again, Tor gets the chance to service the
first three streams activated before the circuit window is exhausted
again, which causes all streams to be blocked again. As an experiment,
we reversed the order in which the streams are activated, and indeed the
first three streams, rather than the last three, got service, while the
others starved.
Our solution is to change the order in which streams are activated. We
choose a random edge connection from the linked list, and then we
activate streams starting from that chosen stream. When we reach the end
of the list, then we continue from the head of the list until our chosen
stream (treating the linked list as a circular linked list). It would
probably be better to actually remember which streams have received
service recently, but this way is simple and effective.
Doing so could make Libevent call Libevent from inside a Libevent
logging call, which is a recipe for reentrant confusion and
hard-to-debug crashes. This would especially hurt if Libevent
debug-level logging is enabled AND the user has a controller
watching for low-severity log messages.
Fix bug 2190; fix on 0.1.0.2-rc.
Doing so could make Libevent call Libevent from inside a Libevent
logging call, which is a recipe for reentrant confusion and
hard-to-debug crashes. This would especially hurt if Libevent
debug-level logging is enabled AND the user has a controller
watching for low-severity log messages.
Fix bug 2190; fix on 0.1.0.2-rc.
