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.
I added the code to pass a destroy cell to a queueing function rather
than writing it immediately, and the code to remember that we
shouldn't reuse the circuit id until the destroy is actually sent, and
the code to release the circuit id once the destroy has been sent...
and then I finished by hooking destroy_cell_queue into the rest of
Tor.
- Move cell_command_to_string from control.c to command.c.
- Use accessor for global_circuitlist instead of extern.
- Add a struct for cell statistics by command instead of six arrays.
- Split up control_event_circuit_cell_stats by using two helper functions.
- Add TestingEnableCellStatsEvent option.
- Prepare functions for testing.
- Rename a few variables and document a few things better.
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.
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.
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).
Coverity is worried that we're checking entry_conn in some cases,
but not in the case where we set entry_conn->pending_optimistic_data.
This commit should calm it down (CID 718623).
This check isn't necessary (see comment on #7801), but it took at
least two smart people a little while to see why it wasn't necessary,
so let's have it in to make the code more readable.
We need a weak RNG in a couple of places where the strong RNG is
both needless and too slow. We had been using the weak RNG from our
platform's libc implementation, but that was problematic (because
many platforms have exceptionally horrible weak RNGs -- like, ones
that only return values between 0 and SHORT_MAX) and because we were
using it in a way that was wrong for LCG-based weak RNGs. (We were
counting on the low bits of the LCG output to be as random as the
high ones, which isn't true.)
This patch adds a separate type for a weak RNG, adds an LCG
implementation for it, and uses that exclusively where we had been
using the platform weak RNG.
This is meant to avoid conflict with the built-in log() function in
math.h. It resolves ticket 7599. First reported by dhill.
This was generated with the following perl script:
#!/usr/bin/perl -w -i -p
s/\blog\(LOG_(ERR|WARN|NOTICE|INFO|DEBUG)\s*,\s*/log_\L$1\(/g;
s/\blog\(/tor_log\(/g;
Path use bias measures how often we can actually succeed using the circuits we
actually try to use. It is a subset of path bias accounting, but it is
computed as a separate statistic because the rate of client circuit use may
vary depending on use case.
Otherwise, it's possible to create streams or circuits with these
bogus IDs, leading to orphaned circuits or streams, or to ones that
can cause bandwidth DOS problems.
Fixes bug 7889; bugfix on all released Tors.
In general, if we tried to use a circ for a stream, but then decided to place
that stream on a different circuit, we need to probe the original circuit
before deciding it was a "success".
We also need to do the same for cannibalized circuits that go unused.
The unit of work sent to a cpuworker is now a create_cell_t; its
response is now a created_cell_t. Several of the things that call or
get called by this chain of logic now take create_cell_t or
created_cell_t too.
Since all cpuworkers are forked or spawned by Tor, they don't need a
stable wire protocol, so we can just send structs. This saves us some
insanity, and helps p
(This is part 2 of making DNS cache use enabled/disabled on a
per-client port basis. This implements the CacheIPv[46]DNS options,
but not the UseCachedIPv[46] ones.)
This makes it so we can handle getting an IPv6 in the 3 different
formats we specified it for in RESOLVED cells,
END_STREAM_REASON_EXITPOLICY cells, and CONNECTED cells.
We don't cache IPv6 addresses yet, since proposal 205 isn't
implemented.
There's a refactored function for parsing connected cells; it has unit
tests.
We'd like these functions to be circuit-relative so that we can
implement a per-circuit DNS cache and per-circuit DNS cache rules for
proposal 205 or its successors. I'm doing this now, as a part of the
IPv6 exits code, since there are about to be a few more instances
of code using this.
This is the simplest possible workaround: make it safe to call
circuit_cell_queue_clear() on a non-attached circuit, and make it
safe-but-a-LD_BUG-warning to call update_circuit_on_cmux() on a
non-attached circuit.
LocalWords: unstage src Untracked
Note: this is a squashed commit; see branch bug6465_rebased_v2 of user/andrea/tor.git for full history of the following 10 commits:
Convert relay.c/relay.h to channel_t
Updating the timestamp if n_flushed > 0 at the end of channel_flush_from_first_active_circuit() was redundant since channel_write_cell() et al. do it themselves.
Get rid of now-unnecessary time parameter in channel_flush_from_first_active_circuit()
Get rid of now-unnecessary time parameter in channel_flush_from_first_active_circuit() in connection_or.c
Add non-inlined external call for channeltls.c to free a packed_cell_t
Appease make check-spaces in relay.c
Replace channel_get_write_queue_len() with sufficient and easier to implement channel_has_queued_writes() in relay.c
Rename channel_touched_by_client() and client_used field for consistency with other timestamps in relay.c
Don't double-free packed cells in relay.c (channel_t Tor now bootstraps and works as a client)
Rearrange channel_t struct to use a union distinguishing listener from cell-bearing channels in relay.c
Extend cells aren't allowed to have a stream_id, but we were only
blocking them when they had a stream_id that corresponded to a
connection. As far as I can tell, this change is harmless: it will
make some kinds of broken clients not work any more, but afaik nobody
actually make a client that was broken in that way.
Found while hunting for other places where we made the same mistake
as in 6271.
Bugfix on d7f50337c1 back from May 2003, which introduced
telescoping circuit construction into 0.0.2pre8.
This avoids a possible crash bug in flush_from_first_active_circuit.
Fixes bug 6341; bugfix on 0.2.2.7-alpha.
Bug reported and fixed by a pseudonymous user on IRC.
I only check on circuits, not streams, since bloating your stream
window past the initial circuit window can't help you much.
Also, I compare to CIRCWINDOW_START_MAX so we don't have surprising
races if we lower CIRCWINDOW_START for an experiment.
This could result in bizarre window values. Report and patch
contributed pseudymously. Fixes part of bug 6271. This bug was
introduced before the first Tor release, in svn commit r152.
(bug 6271, part a.)
This reverts commit c32ec9c425.
It turns out the two sides of the circuit don't actually stay in sync,
so it is perfectly normal for the circuit window on the exit relay to
grow to 2000+. We should fix that bug and then reconsider this patch.
I only check on circuits, not streams, since bloating your stream
window past the initial circuit window can't help you much.
Also, I compare to CIRCWINDOW_START_MAX so we don't have surprising
races if we lower CIRCWINDOW_START for an experiment.
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.
This would happen if the deliver window could become negative
because of an nonexistent connection. (Fortunately, _that_ can't
occur, thanks to circuit_consider_sending_sendme. Still, if we
change our windowing logic at all, we won't want this to become
triggerable.) Fix for bug 5541. Bugfix on 4a66865d, back from
0.0.2pre14. asn found this. Nice catch, asn!
This time, I follow grarpamp's suggestion and move the check for
.exit+AllowDotExit 0 to the top of connection_ap_rewrite_and_attach,
before any rewriting occurs. This way, .exit addresses are
forbidden as they arrive from a socks connection or a DNSPort
request, and not otherwise.
It _is_ a little more complicated than that, though. We need to
treat any .exit addresses whose source is TrackHostExits as meaning
that we can retry without that exit. We also need to treat any
.exit address that comes from an AutomapHostsOnResolve operation as
user-provided (and thus forbidden if AllowDotExits==0), so that
transitioning from AllowDotExits==1 to AllowDotExits==0 will
actually turn off automapped .exit addresses.
We used to do this as a workaround for older Tors, but now it's never
the correct thing to do (especially since anything that didn't
understand RELAY_EARLY is now deprecated hard).
For printf, %f and %lf are synonymous, since floats are promoted to
doubles when passed as varargs. It's only for scanf that we need to
say "%lf" for doubles and "%f" for floats.
Apparenly, some older compilers think it's naughty to say %lf and like
to spew warnings about it.
Found by grarpamp.
This is a little error-prone when the local has a different type
from the parameter, and is very error-prone with both have the same
type. Let's not do this.
Fixes CID #437,438,439,440,441.
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!
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
This patch introduces a few new functions in router.c to produce a
more helpful description of a node than its nickame, and then tweaks
nearly all log messages taking a nickname as an argument to call these
functions instead.
There are a few cases where I left the old log messages alone: in
these cases, the nickname was that of an authority (whose nicknames
are useful and unique), or the message already included an identity
and/or an address. I might have missed a couple more too.
This is a fix for bug 3045.
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.
Ian's original message:
The current code actually correctly handles queued data at the
Exit; if there is queued data in a EXIT_CONN_STATE_CONNECTING
stream, that data will be immediately sent when the connection
succeeds. If the connection fails, the data will be correctly
ignored and freed. The problem with the current server code is
that the server currently drops DATA cells on streams in the
EXIT_CONN_STATE_CONNECTING state. Also, if you try to queue data
in the EXIT_CONN_STATE_RESOLVING state, bad things happen because
streams in that state don't yet have conn->write_event set, and so
some existing sanity checks (any stream with queued data is at
least potentially writable) are no longer sound.
The solution is to simply not drop received DATA cells while in
the EXIT_CONN_STATE_CONNECTING state. Also do not send SENDME
cells in this state, so that the OP cannot send more than one
window's worth of data to be queued at the Exit. Finally, patch
the sanity checks so that streams in the EXIT_CONN_STATE_RESOLVING
state that have buffered data can pass.
[...] Here is a simple patch. It seems to work with both regular
streams and hidden services, but there may be other corner cases
I'm not aware of. (Do streams used for directory fetches, hidden
services, etc. take a different code path?)
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.
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.
