Comments below focus on changes, see diff for added code.
New type tor_addr_port_t holding an IP address and a TCP/UDP port.
New flag in routerinfo_t, ipv6_preferred. This should go in the
node_t instead but not now.
Replace node_get_addr() with
- node_get_prim_addr() for primary address, i.e. IPv4 for now
- node_get_pref_addr() for preferred address, IPv4 or IPv6.
Rename node_get_addr_ipv4h() node_get_prim_addr_ipv4h() for
consistency. The primary address will not allways be an IPv4 address.
Same for node_get_orport() -> node_get_prim_orport().
Rewrite node_is_a_configured_bridge() to take all OR ports into account.
Extend argument list to extend_info_from_node and
extend_info_from_router with a flag indicating if we want to use the
routers primary address or the preferred address. Use the preferred
address in as few situtations as possible for allowing clients to
connect to bridges over IPv6.
Instead of only writing the dynamic DH prime modulus to a file, write
the whole DH parameters set for forward compatibility. At the moment
we only accept '2' as the group generator.
The DH parameters gets stored in base64-ed DER format to the
'dynamic_dh_params' file.
Let's *not* expose more cross-platform-compatibility structures, or
expect code to use them right.
Also, don't fclose() stdout_handle and stdin_handle until we do
tor_process_handle_destroy, or we risk a double-fclose.
This shaves about 7% off our per-cell AES crypto time for me; the
effect for accelerated AES crypto should be even more, since the AES
calculation itself will make an even smaller portion of the
counter-mode performance.
(We don't want to do this for pre-1.0.0 OpenSSL, since our AES_CTR
implementation was actually faster than OpenSSL's there, by about
10%.)
Fixes issue #4526.
This is a fancier bug4457 workaround for 0.2.3. In 0.2.2, we could
just tell Libevent "Don't enable locking!" so it wouldn't try to make
the event_base notifiable. But for IOCP, we need a notifiable base.
(Eventually, we'll want a notifiable base for other stuff, like
multithreaded crypto.) So the solution is to try a full-featured
initialization, and then retry with all the options turned off if that
fails.
Conflicts:
src/common/compat_libevent.c
Resolving conflict by not taking 7363eae13c ("Use the
EVENT_BASE_FLAG_NOLOCK flag to prevent socketpair() invocation"): in
Tor 0.2.3.x, we _do_ sometimes use notifiable event bases.
In Tor 0.2.2, we never need the event base to be notifiable, since we
don't call it from other threads. This is a workaround for bug 4457,
which is not actually a Tor bug IMO.
This thing was pretty pointless on versions of OpenSSL 0.9.8 and later,
and almost totally pointless on OpenSSL 1.0.0.
Also, favor EVP by default, since it lets us get hardware acceleration
where present. (See issue 4442)
The old behavior was susceptible to the compiler optimizing out our
assertion check, *and* could still overflow size_t on 32-bit systems
even when it did work.
- Rename tor_tls_got_server_hello() to tor_tls_got_client_hello().
- Replaced some aggressive asserts with LD_BUG logging.
They were the innocent "I believe I understand how these callbacks
work, and this assert proves it" type of callbacks, and not the "If
this statement is not true, computer is exploding." type of
callbacks.
- Added a changes file.
In a2bb0bf we started using a separate client identity key. When we are
in "public server mode" (that means not a bridge) we will use the same
key. Reusing the key without doing the proper refcounting leads to a
segfault on cleanup during shutdown. Fix that.
Also introduce an assert that triggers if our refcount falls below 0.
That should never happen.
* Make tor_tls_context_new internal to tortls.c, and return the new
tor_tls_context_t from it.
* Add a public tor_tls_context_init wrapper function to replace it.
Conflicts:
src/or/main.c
src/or/router.c
SSL_read(), SSL_write() and SSL_do_handshake() can always progress the
SSL protocol instead of their normal operation, this means that we
must be checking for needless renegotiations after they return.
Introduce tor_tls_got_excess_renegotiations() which makes the
tls->server_handshake_count > 2
check for us, and use it in tor_tls_read() and tor_tls_write().
Cases that should not be handled:
* SSL_do_handshake() is only called by tor_tls_renegotiate() which is a
client-only function.
* The SSL_read() in tor_tls_shutdown() does not need to be handled,
since SSL_shutdown() will be called if SSL_read() returns an error.
From the code:
zlib 1.2.4 and 1.2.5 do some "clever" things with macros. Instead of
saying "(defined(FOO) ? FOO : 0)" they like to say "FOO-0", on the theory
that nobody will care if the compile outputs a no-such-identifier warning.
Sorry, but we like -Werror over here, so I guess we need to define these.
I hope that zlib 1.2.6 doesn't break these too.
Possible fix for bug 1526.
Since we check for naughty renegotiations using
tor_tls_t.server_handshake_count we don't need that semi-broken
function (at least till there is a way to disable rfc5746
renegotiations too).
Switch 'server_handshake_count' from a uint8_t to 2 unsigned int bits.
Since we won't ever be doing more than 3 handshakes, we don't need the
extra space.
Toggle tor_tls_t.got_renegotiate based on the server_handshake_count.
Also assert that when we've done two handshakes as a server (the initial
SSL handshake, and the renegotiation handshake) we've just
renegotiated.
Finally, in tor_tls_read() return an error if we see more than 2
handshakes.
The renegotiation callback was called only when the first Application
Data arrived, instead of when the renegotiation took place.
This happened because SSL_read() returns -1 and sets the error to
SSL_ERROR_WANT_READ when a renegotiation happens instead of reading
data [0].
I also added a commented out aggressive assert that I won't enable yet
because I don't feel I understand SSL_ERROR_WANT_READ enough.
[0]: Look at documentation of SSL_read(), SSL_get_error() and
SSL_CTX_set_mode() (SSL_MODE_AUTO_RETRY section).
Introduce tor_tls_state_changed_callback(), which handles every SSL
state change.
The new function tor_tls_got_server_hello() is called every time we
send a ServerHello during a v2 handshake, and plays the role of the
previous tor_tls_server_info_callback() function.
- Add a tor_process_get_pid() function that returns the PID of a
process_handle_t.
- Conform to make check-spaces.
- Add some more documentation.
- Improve some log messages.
It's too risky to have a function where if you leave one parameter
NULL, it splits up address:port strings, but if you set it, it does
hostname resolution.
Under the new convention, having a tor_addr.*lookup function that
doesn't do hostname resolution is too close for comfort.
I used this script here, and have made no other changes.
s/tor_addr_parse_reverse_lookup_name/tor_addr_parse_PTR_name/g;
s/tor_addr_to_reverse_lookup_name/tor_addr_to_PTR_name/g;
Now let's have "lookup" indicate that there can be a hostname
resolution, and "parse" indicate that there wasn't. Previously, we
had one "lookup" function that did resolution; four "parse" functions,
half of which did resolution; and a "from_str()" function that didn't
do resolution. That's confusing and error-prone!
The code changes in this commit are exactly the result of this perl
script, run under "perl -p -i.bak" :
s/tor_addr_port_parse/tor_addr_port_lookup/g;
s/parse_addr_port(?=[^_])/addr_port_lookup/g;
s/tor_addr_from_str/tor_addr_parse/g;
This patch leaves aton and pton alone: their naming convention and
behavior is is determined by the sockets API.
More renaming may be needed.
Right now we can take the digests only of an RSA key, and only expect to
take the digests of an RSA key. The old tor_cert_get_id_digests() would
return a good set of digests for an RSA key, and an all-zero one for a
non-RSA key. This behavior is too error-prone: it carries the risk that
we will someday check two non-RSA keys for equality and conclude that
they must be equal because they both have the same (zero) "digest".
Instead, let's have tor_cert_get_id_digests() return NULL for keys we
can't handle, and make its callers explicitly test for NULL.
Our keys and x.509 certs are proliferating here. Previously we had:
An ID cert (using the main ID key), self-signed
A link cert (using a shorter-term link key), signed by the ID key
Once proposal 176 and 179 are done, we will also have:
Optionally, a presentation cert (using the link key),
signed by whomever.
An authentication cert (using a shorter-term ID key), signed by
the ID key.
These new keys are managed as part of the tls context infrastructure,
since you want to rotate them under exactly the same circumstances,
and since they need X509 certificates.
GCC 4.2 and maybe other compilers optimize away unsigned integer
overflow checks of the form (foo + bar < foo), for all bar.
Fix one such check in `src/common/OpenBSD_malloc_Linux.c'.
After a stream reached eof, we fclose it, but then
test_util_spawn_background_partial_read() reads from it again, which causes
an error and thus another fclose(). Some platforms are fine with this, others
(e.g. debian-sid-i386) trigger a double-free() error. The actual code used by
Tor (log_from_pipe() and tor_check_port_forwarding()) handle this case
correctly.
Mainly used for testing reading from subprocesses. To be more generic
we now pass in a pointer to a process_handle_t rather than a Windows-
specific HANDLE.
Conventionally in Tor, structs are returned as pointers, so change
tor_spawn_background() to return the process handle in a pointer rather
than as return value.
When we're doing filtering ssl bufferevents, we want the rate-limits
to apply to the lowest level of the bufferevent stack, so that we're
actually limiting bytes sent on the network. Otherwise, we'll read
from the network aggressively, and only limit stuff as we process it.
* Use strcmpstart() instead of strcmp(x,y,strlen(y)).
* Warn the user if the managed proxy failed to launch.
* Improve function documentation.
* Use smartlist_len() instead of n_unconfigured_proxies.
* Split managed_proxy_destroy() to managed_proxy_destroy()
and managed_proxy_destroy_with_transports().
* Constification.
Also remove a few other related warnings that could occur during the ssl
handshake. We do this because the relay operator can't do anything about
them, and they aren't their fault.
We'll still need to tweak it so that it looks for includes and
libraries somewhere more sensible than "where we happened to find
them on Erinn's system"; so that tests and tools get built too;
so that it's a bit documented; and so that we actually try running
the output.
Work done with Erinn Clark.
- pid, stdout/stderr_pipe now encapsulated in process_handle
- read_all replaced by tor_read_all_from_process_stdin/stderr
- waitpid replaced by tor_get_exit_code
Untested on *nix
Previously, if tor_addr_to_str() returned NULL, we would reuse the
last value returned by fmt_addr(). (This could happen if we were
erroneously asked to format an AF_UNSPEC address.) Now instead we
return "???".
This is the meat of proposal 171: we change circuit_is_acceptable()
to require that the connection is compatible with every connection
that has been linked to the circuit; we update circuit_is_better to
prefer attaching streams to circuits in the way that decreases the
circuits' usefulness the least; and we update link_apconn_to_circ()
to do the appropriate bookkeeping.
* Create a function that will get input from a stream, so that we can
communicate with the managed proxy.
* Hackish change to tor_spawn_background() so that we can specify an
environ for our spawn.
- We were reporting the _bottom_ N failing states, not the top N.
- With bufferevents enabled, we logged all TLS states as being "in
bufferevent", which isn't actually informative.
- When we had nothing to report, we reported nothing too loudly.
- Also, we needed documentation.
This code lets us record the state of any outgoing OR connection
that fails before it becomes open, so we can notice if they're all
dying in the same SSL state or the same OR handshake state.
More work is still needed:
- We need documentation
- We need to actually call the code that reports the failure when
we realize that we're having a hard time connecting out or
making circuits.
- We need to periodically clear out all this data -- perhaps,
whenever we build a circuit successfully?
- We'll eventually want to expose it to controllers, perhaps.
Partial implementation of feature 3116.
I'm not one to insist on C's miserly stack limits, but allocating a
256K array on the stack is too much even for me.
Bugfix on 0.2.1.7-alpha. Found by coverity. Fixes CID # 450.
For some inexplicable reason, Coverity departs from its usual
standards of avoiding false positives here, and warns about all
sscanf usage, even when the formatting strings are totally safe.
Addresses CID # 447, 446.
Original message from bug3393:
check_private_dir() to ensure that ControlSocketsGroupWritable is
safe to use. Unfortunately, check_private_dir() only checks against
the currently running user… which can be root until privileges are
dropped to the user and group configured by the User config option.
The attached patch fixes the issue by adding a new effective_user
argument to check_private_dir() and updating the callers. It might
not be the best way to fix the issue, but it did in my tests.
(Code by lunar; changelog by nickm)
George Kadianakis notes that if you give crypto_rand_int() a value
above INT_MAX, it can return a negative number, which is not what
the documentation would imply.
The simple solution is to assert that the input is in [1,INT_MAX+1].
If in the future we need a random-value function that can return
values up to UINT_MAX, we can add one.
Fixes bug 3306; bugfix on 0.2.2pre14.
When we added the check for key size, we required that the keys be
128 bytes. But RSA_size (which defers to BN_num_bytes) will return
128 for keys of length 1017..1024. This patch adds a new
crypto_pk_num_bits() that returns the actual number of significant
bits in the modulus, and uses that to enforce key sizes.
Also, credit the original bug3318 in the changes file.
An elusive compile-error (MingW-gcc v4.50 on Win_XP); a missing
comma (!) and a typo ('err_msg' at line 277 changed to 'errmsg').
Aso changed the format for 'err_code' at line 293 into a "%ld" to suppress
a warning. How did this go unnoticed for ~1 month? Btw. This is my 1st ever
'git commit', so it better work.
The conflicts were mainly caused by the routerinfo->node transition.
Conflicts:
src/or/circuitbuild.c
src/or/command.c
src/or/connection_edge.c
src/or/directory.c
src/or/dirserv.c
src/or/relay.c
src/or/rendservice.c
src/or/routerlist.c
On win64, sockets are of type UINT_PTR; on win32 they're u_int;
elsewhere they're int. The correct windows way to check a socket for
being set is to compare it with INVALID_SOCKET; elsewhere you see if
it is negative.
On Libevent 2, all callbacks take sockets as evutil_socket_t; we've
been passing them int.
This patch should fix compilation and correctness when built for
64-bit windows. Fixes bug 3270.
Rename crypto_pk_check_key_public_exponent to crypto_pk_public_exponent_ok:
it's nice to name predicates s.t. you can tell how to interpret true
and false.
This was harmless, since we only used this for checking for lists of
port values, but it's the principle of the thing.
Fixes 3175; bugfix on 0.1.0.1-rc
We'll need this for checking permissions on the directories that hold
control sockets: if somebody says "ControlSocket ~/foo", it would be
pretty rude to do a chmod 700 on their homedir.
Conflicts in various places, mainly node-related. Resolved them in
favor of HEAD, with copying of tor_mem* operations from bug3122_memcmp_022.
src/common/Makefile.am
src/or/circuitlist.c
src/or/connection_edge.c
src/or/directory.c
src/or/microdesc.c
src/or/networkstatus.c
src/or/router.c
src/or/routerlist.c
src/test/test_util.c
Conflicts throughout. All resolved in favor of taking HEAD and
adding tor_mem* or fast_mem* ops as appropriate.
src/common/Makefile.am
src/or/circuitbuild.c
src/or/directory.c
src/or/dirserv.c
src/or/dirvote.c
src/or/networkstatus.c
src/or/rendclient.c
src/or/rendservice.c
src/or/router.c
src/or/routerlist.c
src/or/routerparse.c
src/or/test.c
Here I looked at the results of the automated conversion and cleaned
them up as follows:
If there was a tor_memcmp or tor_memeq that was in fact "safe"[*] I
changed it to a fast_memcmp or fast_memeq.
Otherwise if there was a tor_memcmp that could turn into a
tor_memneq or tor_memeq, I converted it.
This wants close attention.
[*] I'm erring on the side of caution here, and leaving some things
as tor_memcmp that could in my opinion use the data-dependent
fast_memcmp variant.
These functions can return NULL for otherwise-valid values of
time_t. Notably, the glibc gmtime manpage says it can return NULL
if the year if greater than INT_MAX, and the windows MSDN gmtime
page says it can return NULL for negative time_t values.
Also, our formatting code is not guaranteed to correctly handle
years after 9999 CE.
This patch tries to correct this by detecting NULL values from
gmtime/localtime_r, and trying to clip them to a reasonable end of
the scale. If they are in the middle of the scale, we call it a
downright error.
Arguably, it's a bug to get out-of-bounds dates like this to begin
with. But we've had bugs of this kind in the past, and warning when
we see a bug is much kinder than doing a NULL-pointer dereference.
Boboper found this one too.
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
SSL_*_app_data uses ex_data index 0, which will be the first one allocated
by SSL_get_ex_new_index. Thus, if we ever started using the ex_data feature
for some other purpose, or a library linked to Tor ever started using
OpenSSL's ex_data feature, Tor would break in spectacular and mysterious
ways. Using the SSL_*_ex_data functions directly now may save us from
that particular form of breakage in the future.
But I would not be surprised if using OpenSSL's ex_data functions at all
(directly or not) comes back to bite us on our backends quite hard. The
specified behaviour of dup_func in the man page is stupid, and
crypto/ex_data.c is a horrific mess.
Sets:
* Documentation
* Logging domain
* Configuration option
* Scheduled event
* Makefile
It also creates status.c and the log_heartbeat() function.
All code was written by Sebastian Hahn. Commit message was
written by me (George Kadianakis).
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.
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.
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.
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)
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.
C99 allows a syntax for structures whose last element is of
unspecified length:
struct s {
int elt1;
...
char last_element[];
};
Recent (last-5-years) autoconf versions provide an
AC_C_FLEXIBLE_ARRAY_MEMBER test that defines FLEXIBLE_ARRAY_MEMBER
to either no tokens (if you have c99 flexible array support) or to 1
(if you don't). At that point you just use offsetof
[STRUCT_OFFSET() for us] to see where last_element begins, and
allocate your structures like:
struct s {
int elt1;
...
char last_element[FLEXIBLE_ARRAY_MEMBER];
};
tor_malloc(STRUCT_OFFSET(struct s, last_element) +
n_elements*sizeof(char));
The advantages are:
1) It's easier to see which structures and elements are of
unspecified length.
2) The compiler and related checking tools can also see which
structures and elements are of unspecified length, in case they
wants to try weird bounds-checking tricks or something.
3) The compiler can warn us if we do something dumb, like try
to stack-allocate a flexible-length structure.
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.
Libevent 2.0 has a "changelist" feature that avoids making redundant
syscalls if we wind up doing a lot of event_add/event_del operations
on the same fd in a row. Unfortunately, due to a weird design
choice in Linux, it doesn't work right with epoll when multiple fds
refer to the same socket (e.g., one is a dup() of the other). We
don't dup() anything we give to Libevent, though, so it is safe for
us to explicitly turn this feature on.
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.)
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.
