There is a bug causing busy loops in Libevent and infinite loops in
the Shadow simulator. A connection that is marked for close, wants
to flush, is held open to flush, but is rate limited (the token
bucket is empty) triggers the bug.
This commit fixes the bug. Details are below.
This currently happens on read and write callbacks when the active
socket is marked for close. In this case, Tor doesn't actually try
to complete the read or write (it returns from those methods when
marked), but instead tries to clear the connection with
conn_close_if_marked(). Tor will not close a marked connection that
contains data: it must be flushed first. The bug occurs when this
flush operation on the marked connection can not occur because the
connection is rate-limited (its write token bucket is empty).
The fix is to detect when rate limiting is preventing a marked
connection from properly flushing. In this case, it should be
flagged as read/write_blocked_on_bandwidth and the read/write events
de-registered from Libevent. When the token bucket gets refilled, it
will check the associated read/write_blocked_on_bandwidth flag, and
add the read/write event back to Libevent, which will cause it to
fire. This time, it will be properly flushed and closed.
The reason that both read and write events are both de-registered
when the marked connection can not flush is because both result in
the same behavior. Both read/write events on marked connections will
never again do any actual reads/writes, and are only useful to
trigger the flush and close the connection. By setting the
associated read/write_blocked_on_bandwidth flag, we ensure that the
event will get added back to Libevent, properly flushed, and closed.
Why is this important? Every Shadow event occurs at a discrete time
instant. If Tor does not properly deregister Libevent events that
fire but result in Tor essentially doing nothing, Libevent will
repeatedly fire the event. In Shadow this means infinite loop,
outside of Shadow this means wasted CPU cycles.
Also, try to resolve some doxygen issues. First, define a magic
"This is doxygen!" macro so that we take the correct branch in
various #if/#else/#endifs in order to get the right documentation.
Second, add in a few grouping @{ and @} entries in order to get some
variables and fields to get grouped together.
Previously, we skipped everything that got invoked from
options_init_from_torrc. But some of the stuff in
options_act_reversible and options_act is actually important, like
reopening the logs.
Now, a SIGHUP always makes the effects of an options_set() happen,
even though the options haven't changed.
Fix for bug 5095; bugfix on 0.2.1.9-alpha, which introduced
__ReloadTorrcOnSIGHUP.
This commit is completely mechanical; I used this perl script to make it:
#!/usr/bin/perl -w -i.bak -p
if (/^\s*\#/) {
s/MS_WINDOWS/_WIN32/g;
s/\bWIN32\b/_WIN32/g;
}
Preprocessor directives should not be put inside the arguments
of a macro. This is not supported on older GCC releases (< 3.3)
thus broke compilation on Haiku (running gcc2).
This code handles the new ORPort options, and incidentally makes all
remaining port types use the new port configuration systems.
There are some rough edges! It doesn't do well in the case where your
Address says one thing but you say to Advertise another ORPort. It
doesn't handle AllAddrs. It doesn't actually advertise anything besides
the first listed advertised IPv4 ORPort and DirPort. It doesn't do
port forwarding to them either.
It's not tested either, it needs more documentation, and it probably
forgets to put the milk back in the refrigerator.
Some controllers want this so they can mess with Tor's configuration
for a while via the control port before actually letting Tor out of
the house.
We do this with a new DisableNetwork option, that prevents Tor from
making any outbound connections or binding any non-control
listeners. Additionally, it shuts down the same functionality as
shuts down when we are hibernating, plus the code that launches
directory downloads.
To make sure I didn't miss anything, I added a clause straight to
connection_connect, so that we won't even try to open an outbound
socket when the network is disabled. In my testing, I made this an
assert, but since I probably missed something, I've turned it into a
BUG warning for testing.
We used to do init_keys() if DynamicDHGroups changed after a HUP, so
that the dynamic DH modulus was stored on the disk. Since we are now
doing dynamic DH modulus storing in crypto.c, we can simply initialize
the TLS context and be good with it.
Introduce a new function router_initialize_tls_context() which
initializes the TLS context and use it appropriately.
When running with IOCP, we are in theory able to use userspace-
allocated buffers to avoid filling up the stingy amount of kernel
space allocated for sockets buffers.
The bufferevent_async implementation in Libevent provides this
ability, in theory. (There are likely to be remaining bugs). This
patch adds a new option that, when using IOCP bufferevents, sets
each socket's send and receive buffers to 0, so that we should use
this ability.
When all the bugs are worked out here, if we are right about bug 98,
this might solve or mitigate bug 98.
This option is experimental and will likely require lots of testing
and debugging.
This is used for the bridge authority currently, to get a better
intuition on how many descriptors are actually fetched from it and how
many fetches happen in total.
Implements ticket 4200.
With managed proxies you would always get the error message:
"You have a Bridge line using the X pluggable transport, but there
doesn't seem to be a corresponding ClientTransportPlugin line."
because the check happened directly after parse_client_transport_line()
when managed proxies were not fully configured and their transports
were not registered.
The fix is to move the validation to run_scheduled_events() and make
sure that all managed proxies are configured first.
Right now we only force a new descriptor upload every 18 hours.
This can make servers become unlisted if they upload a descriptor at
time T which the authorities reject as being "too similar" to one
they uploaded before. Nothing will actually make the server upload a
new descriptor later on, until another 18 hours have passed.
This patch changes the upload behavior so that the 18 hour interval
applies only when we're listed in a live consensus with a descriptor
published within the last 18 hours. Otherwise--if we're not listed
in the live consensus, or if we're listed with a publication time
over 18 hours in the past--we upload a new descriptor every 90
minutes.
This is an attempted bugfix for #3327. If we merge it, it should
obsolete #535.
Conflicts:
src/or/connection.c
src/or/connection_edge.c
src/or/connection_edge.h
src/or/dnsserv.c
Some of these were a little tricky, since they touched code that
changed because of the prop171 fixes.
The "nym epoch" of a stream is defined as the number of times that
NEWNYM had been called before the stream was opened. All streams
are isolated by nym epoch.
This feature should be redundant with existing signewnym stuff, but
it provides a good belt-and-suspenders way for us to avoid ever
letting any circuit type bypass signewnym.
* Add some utility transport functions in circuitbuild.[ch] so that we
can use them from pt.c.
* Make the accounting system consider traffic coming from proxies.
* Make sure that we only fetch bridge descriptors when all the
transports are configured.
This lets us make a lot of other stuff const, allows the compiler to
generate (slightly) better code, and will make me get slightly fewer
patches from folks who stick mutable stuff into or_options_t.
const: because not every input is an output!
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.
When we configure a new bridge via the controller, don't wait up to ten
seconds before trying to fetch its descriptor. This wasn't so bad when
you listed your bridges in torrc, but it's dreadful if you configure
your bridges via vidalia.
If you really want to purge the client DNS cache, the TrackHostExits
mappings, and the virtual address mappings, you should be using NEWNYM
instead.
Fixes bug 1345; bugfix on Tor 0.1.0.1-rc.
Note that this needs more work: now that we aren't nuking the
transient addressmap entries on HUP, we need to make sure that
configuration changes to VirtualAddressMap and TrackHostExits actually
have a reasonable effect.
To make sure that a server learns if its IP has changed, the server
sometimes launches authority.z descriptor fetches from
update_router_descriptor_downloads. That's nice, but we're moving
towards a situation where update_router_descriptor_downloads doesn't
always get called. So this patch breaks the authority.z
check-and-fetch into a new function.
This function also renames last_routerdesc_download to a more
appropriate last_descriptor_download, and adds a new
update_all_descriptor_downloads() function.
(For now, this is unnecessary, since servers don't actually use
microdescriptors. But that could change, or bridges could start
using microdescriptors, and then we'll be glad this is refactored
nicely.)
Previously we ensured that it would get called periodically by doing
it from inside the code that added microdescriptors. That won't work
though: it would interfere with our code that tried to read microdescs
from disk initially. Instead, we should consider rebuilding the cache
periodically, and on startup.
Instead of just saying "boogity boogity!" let's actually warn people
that they need to configure stuff right to be safe, and point them
at instructions for how to do that.
Resolves bug 2474.
This is a tweak to the bug2917 fix. Basically, if we want to simulate
a signal arriving in the controller, we shouldn't have to pretend that
we're Libevent, or depend on how Tor sets up its Libevent callbacks.
While doing so, get rid of the now unnecessary function
control_signal_act().
Fixes bug 2917, reported by Robert Ransom. Bugfix on commit
9b4aa8d2ab. This patch is loosely based on
a patch by Robert (Changelog entry).
We've got millisecond timers now, we might as well use them.
This change won't actually make circuits get expiered with microsecond
precision, since we only call the expiry functions once per second.
Still, it should avoid the situation where we have a circuit get
expired too early because of rounding.
A couple of the expiry functions now call tor_gettimeofday: this
should be cheap since we're only doing it once per second. If it gets
to be called more often, though, we should onsider having the current
time be an argument again.
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. ;)
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).
This was originally a patch provided by pipe
(http://www.mail-archive.com/or-talk@freehaven.net/msg13085.html) to
provide a method for controllers to query the total amount of traffic
tor has handled (this is a frequently requested piece of information
by relay operators).
Our old code correctly called bufferevent_flush() on linked
connections to make sure that the other side got an EOF event... but
it didn't call bufferevent_flush() when the connection wasn't
hold_open_until_flushed. Directory connections don't use
hold_open_until_flushed, so the linked exit connection never got an
EOF, so they never sent a RELAY_END cell to the client, and the
client never concluded that data had arrived.
The solution is to make the bufferevent_flush() code apply to _all_
closing linked conns whose partner is not already marked for close.
* 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.
The node_t type is meant to serve two key functions:
1) Abstracting difference between routerinfo_t and microdesc_t
so that clients can use microdesc_t instead of routerinfo_t.
2) Being a central place to hold mutable state about nodes
formerly held in routerstatus_t and routerinfo_t.
This patch implements a nodelist type that holds a node for every
router that we would consider using.
Specifically, a circ attempt that we'd launched while the network was
down could timeout after we've marked our entrynodes up, marking them
back down again. The fix is to annotate as bad the OR conns that were
around before we did the retry, so if a circuit that's attached to them
times out we don't do anything about it.
This requires the latest Git version of Libevent as of 24 March 2010.
In the future, we'll just say it requires Libevent 2.0.5-alpha or
later.
Since Libevent doesn't yet support hierarchical rate limit groups,
there isn't yet support for tracking relayed-bytes separately when
using the bufferevent system. If a future version does add support
for hierarchical buckets, we can add that back in.
When we introduced the code to close non-open OR connections after
KeepalivePeriod had passed, we replaced some code that said
if (!connection_is_open(conn)) {
/* let it keep handshaking forever */
} else if (do other tests here) {
...
with new code that said
if (!connection_is_open(conn) && past_keepalive) {
/* let it keep handshaking forever */
} else if (do other tests here) {
...
This was a mistake, since it made all the other tests start applying
to non-open connections, thus causing bug 1840, where non-open
connections get closed way early.
Fixes bug 1840. Bugfix on 0.2.1.26 (commit 67b38d50).
With this patch we stop scheduling when we should write statistics using a
single timestamp in run_scheduled_events(). Instead, we remember when a
statistics interval starts separately for each statistic type in geoip.c
and rephist.c. Every time run_scheduled_events() tries to write stats to
disk, it learns when it should schedule the next such attempt.
This patch also enables all statistics to be stopped and restarted at a
later time.
This patch comes with a few refactorings, some of which were not easily
doable without the patch.
The next series of commits begins addressing the issue that we're
currently including the complete or.h file in all of our source files.
To change that, we're splitting function definitions into new header
files (one header file per source file).
These timers behave better with non-monotonic clocks than our old
ones, and also try harder to make once-per-second events get called
one second apart, rather than one-plus-epsilon seconds apart.
This fixes bug 943 for everybody using Libevent 2.0 or later.
Most of the changes here are switches to use APIs available on Windows
CE. The most pervasive change is that Windows CE only provides the
wide-character ("FooW") variants of most of the windows function, and
doesn't support the older ASCII verions at all.
This patch will require use of the wcecompat library to get working
versions of the posix-style fd-based file IO functions.
[commit message by nickm]
Fix statistics on client numbers by country as seen by bridges that were
broken in 0.2.2.1-alpha. Also switch to reporting full 24-hour intervals
instead of variable 12-to-48-hour intervals.
The HSAuthorityRecordStats option was used to track statistics of overall
hidden service usage on the version 0 hidden service authorities. With the
version 2 hidden service directories being deployed and version 0
descriptors being phased out, these statistics are not as useful anymore.
Goodbye, you fine piece of software; my first major code contribution to
Tor.
The new rule is: safe_str_X() means "this string is a piece of X
information; make it safe to log." safe_str() on its own means
"this string is a piece of who-knows-what; make it safe to log".
- Avoid memmoving 0 bytes which might lead to compiler warnings.
- Don't require relays to be entry node AND bridge at the same to time to
record clients.
- Fix a memory leak when writing dirreq-stats.
- Don't say in the stats files that measurement intervals are twice as long
as they really are.
- Reduce minimum observation time for requests to 12 hours, or we might
never record usage.
- Clear exit stats correctly after writing them, or we accumulate old stats
over time.
- Reset interval start for buffer stats, too.
We were telling the controller about CHECKING_REACHABILITY and
REACHABILITY_FAILED status events whenever we launch a testing
circuit or notice that one has failed. Instead, only tell the
controller when we want to inform the user of overall success or
overall failure. Bugfix on 0.1.2.6-alpha. Fixes bug 1075. Reported
by SwissTorExit.
when we write out our stability info, detect relays that have slipped
through the cracks. log about them and correct the problem.
if we continue to see a lot of these over time, it means there's another
spot where relays fall out of the routerlist without being marked as
unreachable.
The rest of the code was only including event.h so that it could see
EV_READ and EV_WRITE, which we were using as part of the
connection_watch_events interface for no very good reason.
This patch adds a new compat_libevent.[ch] set of files, and moves our
Libevent compatibility and utilitity functions there. We build them
into a separate .a so that nothing else in src/commmon depends on
Libevent (partially fixing bug 507).
Also, do not use our own built-in evdns copy when we have Libevent
2.0, whose evdns is finally good enough (thus fixing Bug 920).
Apparently all the stuff that does a linear scan over all the DNS
cache entries can get really expensive when your DNS cache is very
large. It's hard to say how much this will help performance, since
gprof doesn't count time spent in OpenSSL or zlib, but I'd guess 10%.
Also, this patch removes calls to assert_connection_ok() from inside
the read and write callbacks, which are similarly unneeded, and a
little costlier than I'm happy with.
This is probably worth backporting to 0.2.0.
- Write geoip stats to disk every 24 hours, not every hour.
- Remove configuration options and define reasonable defaults.
- Clear history of client requests every 24 hours (which wasn't done at
all before).
Specifically if you send SIGUSR1, it will add two lines to the log file:
May 22 07:41:59.576 [notice] Our DNS cache has 3364 entries.
May 22 07:41:59.576 [notice] Our DNS cache size is approximately 1022656
bytes.
[tweaked a bit by nickm]
crypto_global_init gets called. Also have it be crypto_global_init
that calls crypto_seed_rng, so we are not dependent on OpenSSL's
RAND_poll in these fiddly cases.
Should fix bug 907. Bugfix on 0.0.9pre6. Backport candidate.
svn:r18210
The subversion $Id$ fields made every commit force a rebuild of
whatever file got committed. They were not actually useful for
telling the version of Tor files in the wild.
svn:r17867
(The unfixed ones are being downgraded to regular XXXs mainly on the rationale that they don't seem to be exploding Tor, and they were apparently not showstoppers for 0.2.0.x-final.)
svn:r17682
(Many users have no idea what a resolv.conf is, and shouldn't be forced to learn. The old option will keep working for now.)
Also, document it.
svn:r17661
Initial conversion of uint32_t addr to tor_addr_t addr in connection_t and related types. Most of the Tor wire formats using these new types are in, but the code to generate and use it is not. This is a big patch. Let me know what it breaks for you.
svn:r16435
Forward-port: I had apparently broken OSX and Freebsd by not initializing threading before we initialize the logging system. This patch should do so, and fix bug 671.
svn:r14430
Make dumpstats() log the size and fullness of openssl-internal buffers, so I can test my hypothesis that many of them are empty, and my alternative hypothesis that many of them are mostly empty, against the null hypothesis that we really need to be burning 32K per open OR connection on this.
svn:r14350
New --hush command-line option similar to --quiet. While --quiet disables all
logging to the console on startup, --hush limits the output to messages of
warning and error severity.
svn:r14222
Only log a notice that dmalloc has been set up if it fails. Actually, since we have not added a temp log yet, I am not sure this ever does anything.
svn:r14216
example, when answering a directory request), reset the
time-to-give-up timeout every time we manage to write something
on the socket. Bugfix on 0.1.2.x.
svn:r13643
Answer one xxx020 item; move 7 other ones to a new "XXX020rc" category: they should get fixed before we cut a release candidate. arma: please review these to see whether you have fixes/answers for any. Please check out the other 14 XXX020s to see if any look critical for the release candidate.
svn:r13640
As planned, rename networkstatus_vote_t to networkstatus_t, now that v3 networkstatuses are working and standard and v2 networkstatuses are obsolete.
svn:r13383
Periodically check whether we have an expired consensus networkstatus. If we do, and we think we have enough directory info, then call router_dir_info_changed(). Fixes bug 401. This bug was deferred from 0.1.2.x, but fixing it there is nontrivial.
svn:r13342
Fix bug 575: protect the list of logs with a mutex. I couldn't find any appreciable change in logging performance on osx, but ymmv. You can undef USE_LOG_MUTEX to see if stuff gets faster for you.
svn:r13019
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
on but your ORPort is off.
Add a new config option BridgeRelay that specifies you want to
be a bridge relay. Right now the only difference is that it makes
you answer begin_dir requests, and it makes you cache dir info,
even if your DirPort isn't on.
Refactor directory_caches_dir_info() into some more functions.
svn:r12668
