Basic hacks to get TLS handshakes working: remove dead code; fix post-handshake logic; keep servers from writing while the client is supposed to be renegotiating. This may work. Needs testing.
svn:r13122
Here, have some terribly clever new buffer code. It uses a mbuf-like strategy rather than a ring buffer strategy, so it should require far far less extra memory to hold any given amount of data. Also, it avoids access patterns like x=malloc(1024);x=realloc(x,1048576);x=realloc(x,1024);append_to_freelist(x) that might have been contributing to memory fragmentation. I've tested it out a little on peacetime, and it seems to work so far. If you want to benchmark it for speed, make sure to remove the #define PARANOIA; #define NOINLINE macros at the head of the module.
svn:r12983
Add support to get a callback invoked when the client renegotiate a connection. Also, make clients renegotiate. (not enabled yet, until they detect that the server acted like a v2 server)
svn:r12623
Keep track, for each OR connection, of the last time we added a non-padding cell to its outbuf. Use this timestamp, not "lastwritten" to tell if it is time to close a circuitless connection. (We can'tuse lastwritten, since lastwritten is updated when ever the connection flushes anything, and by that point we can no longer tell what is a padding cell and what is not.)
svn:r12437
Fix bug 451. This was a nasty bug, so let's fix it twice: first, by banning recursive calls to connection_handle_write from connection_flushed_some; and second, by not calling connection_finished_flushing() on a closed connection. Backport candidate.
svn:r11882
Add debugging warning to not abort in the case of bug 483. This is probably not an actual error case, so we should figure out what is really causing it and do something more sensible.
svn:r11215
Patch from Robert Hogan: set conn->dns_server_port correctly so that we can close dns server ports when they change, thus avoiding crashes and dangling references and other sources of unhappiness.
svn:r10933
Tweaks on constrained socket buffers patch from coderman: Add a changelog; rename some variables; fix some long lines and whitespace; make ConstrainedSockSize a memunit; pass setsockopt a void.
svn:r10843
Well, that was easier than I thought it would be. Tor is now a DNS proxy as well as a socks proxy. Probably some bugs remain, but since it A) has managed to resolve one address for me successfully, and B) will not affect anybody who leaves DNSPort unset, it feel like a good time to commit.
svn:r10317
Initial version of code to stop using socket pairs for linked connections. Superficially, it seems to work, but it probably needs a lot more testing and attention.
svn:r9995
Track the number of connection_t separately from the number of open sockets. It is already possible to have connections that do not count: resolving conns, for one. Once we move from socketpairs to linked conns, and once we do dns proxying, there will be lots of such connections.
svn:r9994
A surprisingly simple patch to stop reading on edge connections when their circuits get too full, and start again when they empty out. This lets us remove the logic to block begin_dir conns when the corresponding or conns get full: it was already broken by cell queues anyway.
svn:r9905
Remove support for v0 control protocol from 0.2.0.x trunk; send back error when we receive a v0 control message. (Leave "if(v1){...}"blocks indented for now so this patch is easier to read.) ((Finally, the linecount goes _down_ a little.))
svn:r9735
Make all LD_BUG log messsages get prefixed with "Bug: ". Remove manually-generated "Bug: "s from log-messages. (Apparently, we remembered to add them about 40% of the time.)
svn:r9733
Fix connection_get_by_type_state_rendquery(): This has been bogus for most of 0.1.2.x. Thanks to Karsten Loesing for finding the bug; fixes bug 399.
svn:r9651
Resolve some XXXX012 items:
- Remove PathlenCoinWeight: if we want it again, we can add it
back in.
- Ditto with RelayBandwidth*.
- Decide to leave in the "hey, you didn't set end_reason!" BUG log message,
but stop telling people to bug me personally.
- Postpone strengthening assert_connection_ok(): it's important, but
it's also a good way to introduce weird bugs.
- Move some expensive consistency checking from dns_free_all() into
assert_cache_ok().
svn:r9533
This one is a little tricky. Our BEGIN_DIR implementation has a
problem: the dirserv conns will decide they can flush all their data
immediately, since the edge_conns will read greedily.
For our 0.1.2 workaround, we track which or_conn a bridged dirserv
conn is attached to, and stop writing when its outbuf is too full, and
start writing again when the or_conn's outbuf empties out a little.
This requires a bit of pointer management. Let's hope it works.
svn:r9432
Remove redundant check for whether _connection_write_to_buf_impl is called with a zlib and a non-directory connection: TO_DIR_CONN will already assert if it gets a non-dir connection.
svn:r9390
aggressively. my vidalia bandwidth graph, when rate limiting
to 32kB/s, has the "write" line constantly at 32kB. I can't
imagine what's going on with the relay latency but it can't
be good.
svn:r9366
Tidy up ORCONN reason patch from Mike Perry. Changes: make some of the handling of TLS error codes less error prone. Enforce house style wrt spaces. Make it compile with --enable-gcc-warnings. Only set or_conn->tls_error in the case of an actual error. Add a changelog entry.
svn:r9355
Count TLS bytes accurately: previously, we counted only the number of bytes read or transmitted via tls, not the number of extra bytes used to do so. This has been a lonstanding wart. The fix "Works for me".
svn:r9207
Add support for (Free?)BSD's natd, which was an old way to let you
have your firewall automatically redirect traffic. (Original patch
from Zajcev Evgeny, updated for 0.1.2.x by tup.)
svn:r8946
Have connection_about_to_close use an end_reason field in edge_connection_t to tell what reason to tell the controller for closing the stream. Set end_reason in connection_edge_end, connection_mark_unattached_ap, and everwhere we set edge_has_sent_end. Add a changelog entry.
svn:r8779
Start implementing reason extension for stream events to match the one one used by circuit events. (Not a complete implementation yet; actual reasons are not passed to control.c)
svn:r8777
Remove/clarify some XXXs for no longer being accurate; for begin things we do not indend to fix; for already being parts of big todo issues (like "/* XXX ipv6 */"); etc. Also fix some spaces.
svn:r8580
Move is_local_IP to config.c; have it check for same-/24; make it used only for reachability (not for banwidth, because that is probably not what we want). Fixes an XXX.
svn:r8578
Refactor entry guard status logic a lot; allow more factors [like not
having a Guard flag or being listed in ExcludeNodes] to render a guard
"unlisted" (now called "unusable"); track guard down status (now
called "unreachable") separately from is_running.
svn:r8519
Patch from Tup to add support for transparent AP 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.
svn:r7007
Refactor connection_t into edge, or, dir, control, and base subtypes. This might save some RAM on busy exit servers, but really matters most in terms of correctness.
svn:r6906
connecting and for resolving). Now we tolerate applications
that don't follow the RFCs. But continue to block malformed
names at the socks side.
svn:r6193
when they created a network status. so if nobody asked for a
network status, they would never discover that any servers are
is_running, so they could never build a circuit.
svn:r6183
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.
svn:r5906
we screwed up the formatting in wild and unpredictable ways.
fix it before it becomes convention to format logs in wild and
unpredictable ways.
still need to do src/common/ someday.
svn:r5551
Weasel says circuit_get_by_conn is his main timesink. Most of its
users were just checking whether OR conns had circuits, so add a
circuit count to OR conns, and check that. One was
circuit_about_to_close_conn, which was doing an O(n^2) series of calls
to get all circs on an OR conn, so make an O(n) function for that.
Finally, circuit_get_by_edge_conn was using it as a sanity test that
has been around for a while but never found any actualy insanity, so
kill that.
circuit_get_by_conn is finally dead, which is good, since it was never
sane to begin with.
svn:r5460
don't tell you (it happens!); and rotate TLS connections once a week.
1) If an OR conn becomes more than a week old, make it obsolete.
2) If it's obsolete and empty, kill it.
3) When an OR makes a second connection to you, allow it.
4) If we want to send a new create cell, but the best conn we've
got is obsolete, and the router is 0.1.1.9-alpha-cvs or later, ask
for a new conn instead.
5) When we time out on circuit building on the first hop, make that
connection obsolete.
svn:r5429
connection.c:
- Add some more connection accessor functions to make directory
download redundancy checking work.
directory.c, or.h, router.c, routerlist.c:
- Start on logic to note when networkstatus downloads fail.
dirserv.c, routerlist.c, routerparse.c:
- Start maintaining an is_named field in routerstatus_t. Don't
actually look at it yet.
dirserv.c, routerlist.c:
- Remove expired networkstatus objects.
or.h:
- Make some booleans into bitfields
- Add prototypes
routerlist.c:
- Sort networkstatus list by publication time
- Function to remove old (older than 10 days) networkstatus objects.
- Function to set a list of routerinfo_ts' status info from the
current set of networkstatus objects.
- Function to tell which routerinfos we need to download based no the
current set of networkstatus objects.
- Do not launch a networkstatus download if a redundant one is in progress.
routerparse.c:
- Keep router entries in networkstatus sorted by digest.
svn:r5012
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.
svn:r4343
sin_addr.s_addr == 0 || sin->sin_port == 0.
This just happened on moria2, so I guess it happens rarely
on Linux as well as OS X.
We can't afford to accept OR conns from 0.0.0.0:0, since we
send created cells back to the first addr:port that matches,
and we'd better not send them to the wrong place.
So, let's drop them all for now, and see if we can find a pattern
later.
svn:r4028
right after somebody else has let it go. But REUSEADDR on win32
means to let you bind to the port _even when somebody else
already has it bound_. So, don't do that on Win32.
svn:r3867
connection_unregistered it. there's still more work to be done here,
since we need to make sure to send back the socks-reply as soon as we
know what it will be -- we can't just wait until connection-about-to-close
to deal with it.
svn:r3847
until none are left, then we try to refetch the descriptor. If it's
the same one we had before, then close streams right then. Whenever
a new stream arrives, even if it's right after, optimistically try
refetching the descriptor, just in case.
svn:r3379
poll-but-sometimes-select mess. This will let us use faster async cores
(like epoll, kpoll, and /dev/poll), and hopefully work better on Windows
too.
There are some fairly nasty changes to main.c here; this will almost
certainly break something. But hey, that's what alphas are for.
svn:r3341
make it clearer which warns are bugs,
make the control log event match its specification,
point out a bug in how we deal with failure when renewing the tls context.
svn:r3138
close, if we're planning to wait to flush it.
This is important because we were sending a socks reject back if we're
closing and hadn't already sent one, but it wasn't actually getting
written since the conn was already marked-for-close.
svn:r3074
decide what exit node to use; based on a patch by geoff goodell.
needs more work: e.g. it goes bananas building new circuits when the
chosen exit node's exit policy rejects the connection.
svn:r3015
Now we can try setting an option but back out if it fails to parse, or
if it's disallowed (e.g. changing RunAsDaemon from 1 to 0).
Use parse_line_from_str rather than parse_line_from_file.
svn:r2692
- make clients cache directories and use them to seed their router lists
at startup. This means clients have a datadir again.
- Introduce a global_write_bucket. We need to respond better to exhausting
it.
- Remove the last vestiges of LinkPadding and TrafficShaping.
- Configuration infrastructure support for warning on obsolete options.
- Refactor directory header parsing to use smartlist_split_string.
- Respond to content-encoding headers by trying to uncompress as appropriate.
- Reply with a deflated directory when a client asks for "dir.z".
(We could use allow-encodings instead, but allow-encodings isn't
specified in HTTP 1.0.)
svn:r2335
options->Address with the resolved one at startup.
o detect our address right before we make a routerinfo each time.
o external IP vs bind-IP. Already done, just use options->Address.
o OutboundBindAddress config option, to bind to a specific
IP address for outgoing connect()s.
svn:r2241
introduce an authdir_mode() macro to match the others.
don't initialize uptime to the number of seconds since 1970.
non-authoritative dirservers don't cache their directory on disk.
make only authdirservers use clique_mode.
only read approved-routers file if you're an authdirserver.
even authdirservers fetch a new directory in do_hup.
retry_all_connections() is now called retry_all_listeners().
router_parse_list_from_string() correctly reports the router number
it's working on.
only call dirserv_add_own_fingerprint() and
dirserv_add_descriptor() on startup if we're an authdirserver.
if AuthDir and !ORPort then fail.
if AuthDir and ClientOnly then fail.
svn:r2061
now we do.
but i'm not sure it matters, since we also poll for reads, and if
there's an error with the connecting socket, poll is supposed to
return readable, so we should notice it then.
who knows.
svn:r2057
bugfix: actually complain if we duplicate mark-for-close a circuit
add more logging for relay ends that claim dns resolve failed, so we can
find out why they're not being retried.
svn:r1798
choose an intro point, connect to it,
choose a rend point, connect to it and establish a cookie,
get an ack from the rendezvous point,
and know when both circs are ready for her.
APConns don't use conn->purpose anymore
don't initiate a renddesc lookup if one is already in progress
also fix a buffer overflow in nickname parsing (only exploitable
by the operator though)
svn:r1471
* read all the time (before we would ignore eof sometimes, oops)
* we can handle different urls now
* send back 404 for an un-handled url
* commands initiated by the client can handle payloads now
* introduce conn->purpose to avoid exponential state-space explosion
svn:r1400
successful/failed connections, successful/failed extends, and
connection uptimes.
It's still not done: more tests are needed, and not everything calls
connection/circuit_mark_for_close properly. This skews the results.
Also, there needs to be a 'testing' mode for non-OP ORs, where they
periodically build circuits just to test whether extends work.
svn:r1313
we were never writing anything when hold_open_until_flushed was set,
since conn_write returns early if marked_for_conn is set.
seems a bit better now.
svn:r1214
Apparently, when a DNS failure was already cached, then when we tried
to mark the exit connection as closed, we'd try to remove it from the
pending queue anyway, and hit an assert. Now, we put failed-resolve
connections in a separate state so that mark_for_close does the right
thing.
svn:r1196
who wants to shut down a connection calls connection_mark_for_close instead
of setting marked_for_close to 1. This automatically removes the connection
from the DNS cache if needed, sends a RELAY END cell if appropriate, and can
be changed to do whatever else is needed.
Still to do:
- The same for circuits, maybe.
- Add some kind of hold_connection_open_until_flushed flag, maybe.
- Change stuff that closes connections with return -1 to use mark_for_close,
maybe.
svn:r1145
add some more data to be flushed but never turn POLLOUT on. not sure
how commonly this bug was hit, but it would be a doozy.
Also add some asserts to see if it happens elsewhere.
svn:r1142
arrives, then the stream wasn't getting removed from the pending list.
this may have been the lucky-bug.
this commit may also not actually fix the bug. it's darn hard to
reproduce.
svn:r1122
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 them, the socket never closed,
so the child never read eof, so he never knew to exit.
As a side effect to this bug, we were probably failing to properly close
connections to remote hosts, ORs, and OPs, after a dns child was born.
I'm surprised Tor worked at all.
svn:r974
split 7-byte stream_id string into 2-byte recognized and 2-byte stream_id
fix two seg faults in fetch_from_buf_http
fix several lurking seg faults in handling unexpected relay cells
still need to
* clean up relay_crypt
* use relay dummies if there's going to be a conflict with rh.recognized
* check for a conflict when generating stream_ids
svn:r953
never work.
fix vicious bug in choose_good_exit_server that caused it to *skip over*
pending circuits, and look only at *non-pending circuits*, when choosing
a good exit node for the new circuit.
bugfix: remove incorrect asserts in circuit_get_newest()
svn:r876
Also:
- Refactor socks request into a separate struct
- Add a separate 'waiting for circuit' state to AP connections
between 'waiting for socks' and 'open'.
Arma: can you check out the XXX's I've added to connection_edge? I may
be mishandling some async and close logic.
svn:r783
not when we're closing the stream.
this lets us put a payload in the end cell if we want to,
to describe why we're closing the stream.
there are still some places where we don't send the end cell
immediately. i need to track them down. but it's a low priority,
since i've made it send the end cell when we close the stream if
we haven't already sent it.
svn:r640
If DebugLogFile is specified, log to it at -l debug
If LogFile is specified, log to it at the -l from the commandline
(default info)
If no LogFile *and* not a Daemon, then log to stdout.
Make conn->s = -1 by default (this might break things)
When kill -USR1, prefer to log at INFO, but make sure they always see it.
svn:r596
our log() conflicts with log(3)
distribute only the correct files from doc/ and src/config/
sometimes laptops go back in time. i guess that's ok for now.
and bump the version number because we're live.
svn:r544
ERR is if something fatal just happened
WARNING is something bad happened, but we're still running. The bad thing
is either a bug in the code, an attack or buggy protocol/implementation
of the remote peer, etc. The operator should examine the bad thing and
try to correct it.
(No error or warning messages should be expected. I expect most people
to run on -l warning eventually.)
NOTICE is never ever used.
INFO means something happened (maybe bad, maybe ok), but there's nothing
you need to (or can) do about it.
DEBUG is for everything louder than INFO.
svn:r486
redo all the config files for the new format (we'll redo them again soon)
fix (another! yuck) segfault in log_fn when input is too large
tor_tls_context_new() returns -1 for error, not NULL
fix segfault in check_conn_marked() on conn's that die during tls handshake
make ORs also initialize conn from router when we're the receiving node
make non-dirserver ORs upload descriptor to every dirserver on startup
add our local address to the descriptor
add Content-Length field to POST command
revert the Content-Length search in fetch_from_buf_http() to previous code
fix segfault in memmove in fetch_from_buf_http()
raise maximum allowed headers/body size in directory.c
svn:r484
'buf_t' is now an opaque type defined in buffers.c .
Router descriptors now include all keys; routers generate keys as
needed on startup (in a newly defined "data directory"), and generate
their own descriptors. Descriptors are now self-signed.
Implementation is not complete: descriptors are never published; and
upon receiving a descriptor, the directory doesn't do anything with
it.
At least "routers.or" and orkeygen are now obsolete, BTW.
svn:r483
Fixed up the assert_*_ok funcs some (more work remains)
Changed config so it reads either /etc/torrc or the -f arg, never both
Finally tracked down a nasty bug with our use of tls:
It turns out that if you ask SSL_read() for no more than n bytes, it
will read the entire record from the network (and maybe part of the next
record, I'm not sure), give you n bytes of it, and keep the remaining
bytes internally. This is fine, except our poll-for-read looks at the
network, and there are no bytes pending on the network, so we never know
to ask SSL_read() for more bytes. Currently I've hacked it so if we ask
for n bytes and it returns n bytes, then it reads again right then. This
will interact poorly with our rate limiting; we need a cleaner solution.
svn:r481
deal with content-length headers better when reading http
don't assume struct socks4_info is a packed struct
fail the socks handshake if destip is zero
flesh out conn_state_to_string() for dir conn
fix typo (bug) in connection_handle_read()
directory get is now called fetch, post is now upload
reopen logs on sighup
svn:r475
this paves the way for supporting socks5 and other handshakes
it also removes those pesky AP-only variables from connection_t
also hacked a fix for a bug where some streams weren't ending properly --
maybe because marked connections weren't flushing properly?
svn:r472
- signal support
- forking for DNS farm
- changes for async IO
- daemonizing
In other words, some files still don't build, and the ones that do build,
do nonblocking IO incorrectly.
I'm also not checking in the project files till I have a good place
for them.
svn:r380
i've eliminated the master dns process, so now the workers just
act like regular connections and are handled by the normal pollarray.
everything seems to still work. ;)
svn:r327
on startup, it forks off a master dns handler, which forks off dns
slaves (like the apache model). slaves as spawned as load increases,
and then reused. excess slaves are not ever killed, currently.
implemented topics. each topic has a receive window in each direction
at each edge of the circuit, and sends sendme's at the data level, as
per before. each circuit also has receive windows in each direction at
each hop; an edge sends a circuit-level sendme as soon as enough data
cells have arrived (regardless of whether the data cells were flushed
to the exit conns). removed the 'connected' cell type, since it's now
a topic command within data cells.
at the edge of the circuit, there can be multiple connections associated
with a single circuit. you find them via the linked list conn->next_topic.
currently each new ap connection starts its own circuit, so we ought
to see comparable performance to what we had before. but that's only
because i haven't written the code to reattach to old circuits. please
try to break it as-is, and then i'll make it reuse the same circuit and
we'll try to break that.
svn:r152
now tor can be run safely inside nat'ed areas that kill idle
connections; and the proxy can handle when you suspend your laptop
and then emerge hours later from a new domain.
svn:r125
prkey is only fetched when it's needed
tor nodes who aren't dirservers now fetch directories and autoconnect
to new nodes listed in the directory
default role is a non-dirserver node
svn:r120
revamped the router reading section
reference counting for crypto pk env's (so we can dup them)
we now read and write pem pk keys from string rather than from FILE*,
in anticipation of fetching directories over a socket
(so now on startup we slurp in the whole file, then parse it as a string)
fixed a bug in the proxy side, where you could get some circuits
wedged if they showed up while the connection was being made
svn:r110
httpap is obsolete; we support privoxy directly now!
smtpap is obsolete; need to find a good socks4a-enabled smtp proxy/client
I dub thee 0.0.1.
svn:r107
Servers are allowed to send 100 cells initially, and can't send more until
they receive a 'sendme' cell from that direction, indicating that they
can send 10 more cells. As it currently stands, the exit node quickly
runs out of window, and sends bursts of 10 whenever a sendme cell gets
to him. This is much much much faster (and more flexible) than the old
"give each circuit 1 kB/s and hope nothing overflows" approach.
Also divided out the connection_watch_events into stop_reading,
start_writing, etc. That way we can control them separately.
svn:r54
we're now much more robust when bandwidth varies: instead of forcing a
fixed bandwidth on the link, we instead use what the link will give us,
up to our bandwidth.
svn:r53
Each socket reads at most 'bandwidth' bytes per second sustained, but
can handle bursts of up to 10*bandwidth bytes.
Cells are now sent out at evenly-spaced intervals, with padding sent
out otherwise. Set Linkpadding=0 in the rc file to send cells as soon
as they're available (and to never send padding cells).
Added license/copyrights statements at the top of most files.
router->min and router->max have been merged into a single 'bandwidth'
value. We should make the routerinfo_t reflect this (want to do that,
Mat?)
As the bandwidth increases, and we want to stop sleeping more and more
frequently to send a single cell, cpu usage goes up. At 128kB/s we're
pretty much calling poll with a timeout of 1ms or even 0ms. The current
code takes a timeout of 0-9ms and makes it 10ms. prepare_for_poll()
handles everything that should have happened in the past, so as long as
our buffers don't get too full in that 10ms, we're ok.
Speaking of too full, if you run three servers at 100kB/s with -l debug,
it spends too much time printing debugging messages to be able to keep
up with the cells. The outbuf ultimately fills up and it kills that
connection. If you run with -l err, it works fine up through 500kB/s and
probably beyond. Down the road we'll want to teach it to recognize when
an outbuf is getting full, and back off.
svn:r50
basically, a twin is a router which is different except it shares
the same keypair. so in cases where we want to find a "next router"
and all we really care is that it can decrypt the next onion layer,
then a twin is just as good.
we still need to decide how to mark twins in the routerinfo_t and in
the routers config file.
svn:r30
The 'or' process can now be told (by the global_role variable) what
roles this server should play -- connect to all ORs, listen for ORs,
listen for OPs, listen for APs, or any combination.
* everything in /src/op/ is now obsolete.
* connection_ap.c now handles all interactions with application proxies
* "port" is now or_port, op_port, ap_port. But routers are still always
referenced (say, in conn_get_by_addr_port()) by addr / or_port. We
should make routers.c actually read these new ports (currently I've
kludged it so op_port = or_port+10, ap_port=or_port+20)
* circuits currently know if they're at the beginning of the path because
circ->cpath is set. They use this instead for crypts (both ways),
if it's set.
* I still obey the "send a 0 back to the AP when you're ready" protocol,
but I think we should phase it out. I can simply not read from the AP
socket until I'm ready.
I need to do a lot of cleanup work here, but the code appears to work, so
now's a good time for a checkin.
svn:r22
