This change makes it so those those APIs will not require prior
inclusion of openssl headers. I've left some APIs alone-- those
will change to be extra-private.
This patch fixes a regression described in bug #21757 that first
appeared after commit 6e78ede73f which was an attempt to fix bug #21654.
When switching from buffered I/O to direct file descriptor I/O our
output strings from get_string_from_pipe() might contain newline
characters (\n). In this patch we modify tor_get_lines_from_handle() to
ensure that the function splits the newly read string at the newline
character and thus might return multiple lines from a single call to
get_string_from_pipe().
Additionally, we add a test case to test_util_string_from_pipe() to
ensure that get_string_from_pipe() correctly returns multiple lines in a
single call.
See: https://bugs.torproject.org/21757
See: https://bugs.torproject.org/21654
We could use one of these for holding "junk" descriptors and
unparseable things -- but we'll _need_ it for having cached
consensuses and diffs between them.
(This commit was extracted by nickm based on the final outcome of
the project, taking only the changes in the files touched by this
commit from the consdiff_rebased branch. The directory-system
changes are going to get worked on separately.)
Windows doesn't let you check the socket error for a socket with
WSAGetLastError() and getsockopt(SO_ERROR). But
getsockopt(SO_ERROR) clears the error on the socket, so you can't
call it more than once per error.
When we introduced recv_ni to help drain alert sockets, back in
0.2.6.3-alpha, we had the failure path for recv_ni call getsockopt()
twice, though: once to check for EINTR and one to check for EAGAIN.
Of course, we never got the eagain, so we treated it as an error,
and warned about: "No error".
The fix here is to have these functions return -errno on failure.
Fixes bug 21540; bugfix on 0.2.6.3-alpha.
So we require that SMARTLIST_FOREACH_END() have the name of the loop
variable in it. But right now the only enforcement for that is to
clear the variable at the end of the loop, which is really not
sufficient: I spent 45 minutes earlier today debugging an issue
where I had said:
SMARTLIST_FOREACH_BEGIN(spool, spooled_resource_t *, spooled) {
...
} SMARTLIST_FOREACH_END(spool);
This patch makes it so that ONLY loop variables can be used, by
referring to the _sl_idx variable.
This patch removes the `tor_fgets()` wrapper around `fgets(3)` since it
is no longer needed. The function was created due to inconsistency
between the returned values of `fgets(3)` on different versions of Unix
when using `fgets(3)` on non-blocking file descriptors, but with the
recent changes in bug #21654 we switch from unbuffered to direct I/O on
non-blocking file descriptors in our utility module.
We continue to use `fgets(3)` directly in the geoip and dirserv module
since this usage is considered safe.
This patch also removes the test-case that was created to detect
differences in the implementation of `fgets(3)` as well as the changes
file since these changes was not included in any releases yet.
See: https://bugs.torproject.org/21654
This patch changes a number of read loops in the util module to use
less-than comparison instead of not-equal-to comparison. We do this in
the case that we have a bug elsewhere that might cause `numread` to
become larger than `count` and thus become an infinite loop.
This patch removes the buffered I/O stream usage in process_handle_t and
its related utility functions. This simplifies the code and avoids racy
code where we used buffered I/O on non-blocking file descriptors.
See: https://bugs.torproject.org/21654
This patch modifies `tor_read_all_handle()` to use read(2) instead of
fgets(3) when reading the stdout from the child process. This should
eliminate the race condition that can be triggered in the 'slow/util/*'
tests on slower machines running OpenBSD, FreeBSD and HardenedBSD.
See: https://bugs.torproject.org/21654
This patch adds the `tor_fgets()` function to our compatibility layer.
`tor_fgets()` adds an additional check for whether the error-bit have
been enabled for the given file stream, if that is the case and `errno`
is set to `EAGAIN` we make sure that we always return NULL.
Unfortunately `fgets(3)` behaves differently on different versions of
the C library.
See: https://bugs.torproject.org/21416
See: https://bugs.torproject.org/20988
This feature makes it possible to turn off memory sentinels (like
those used for safety in buffers.c and memarea.c) when fuzzing, so
that we can catch bugs that they would otherwise prevent.
Since 0.2.4.11-alpha (in 0196647970) we've tried to randomize
the start time to up to some time in the past. But unfortunately we
allowed the start time to be in the future as well, which isn't
really legit.
The new behavior lets the start time be be up to
MAX(cert_lifetime-2days, 0) in the past, but never in the future.
Fixes bug 21420; bugfix on 0.2.4.11-alpha.
Fixes bug 20894; bugfix on 0.2.0.16-alpha.
We already applied a workaround for this as 20834, so no need to
freak out (unless you didn't apply 20384 yet).
maint-0.2.7-redux is an attempt to try to re-create a plausible
maint-0.2.7 branch. I've started from the tor-0.2.7.6, and then I
merged maint-0.2.6 into the branch.
This has produced 2 conflicts: one related to the
rendcommon->rendcache move, and one to the authority refactoring.
Check size argument to memwipe() for underflow.
Closes bug #18089. Reported by "gk", patch by "teor".
Bugfix on 0.2.3.25 and 0.2.4.6-alpha (#7352),
commit 49dd5ef3 on 7 Nov 2012.
If tor_mmap_file is called with a file which is larger than SIZE_MAX,
only a small part of the file will be memory-mapped due to integer
truncation.
This can only realistically happen on 32 bit architectures with large
file support.
Because <unset> makes more sense than AAAAAAAAAAAAAAAAAAA...
(I have indeed verified that ed25519_fmt() is only used for
logging. This patch also clarifies the intention that ed25519_fmt()
is only for logging.
Closes ticket 21037.
This came up on #21035, where somebody tried to build on a linux
system with kernel headers including CLOCK_MONOTONIC_COARSE, then
run on a kernel that didn't support it.
I've adopted a belt-and-suspenders approach here: we detect failures
at initialization time, and we also detect (loudly) failures later on.
Fixes bug 21035; bugfix on 0.2.9.1-alpha when we started using
monotonic time.
I got confused when I saw my Tor saying it was opening a file
that doesn't exist. It turns out it isn't opening it, it's just
calling open() on it and then moving on when it's not there.
This includes:
* making bridge_info_t exposed but opaque
* allowing guards where we don't know an identity
* making it possible to learn the identity of a guard
* creating a guard that lacks a node_t
* remembering a guard's address and port.
* Looking up a guard by address and port.
* Only enforcing the rule that we need a live consensus to update
the "listed" status for guards when we are not using bridges.
They added clock_gettime(), but with tv_nsec as a long, whereas
tv_usec is a __darwin_suseconds_t (a.k.a. 'int'). Now, why would
they do that? Are they preparing for a world where there are more
than 2 billion nanoseconds per second? Are they planning for having
int be less than 32 bits again? Or are they just not paying
attention to the Darwin API?
Also, they forgot to mark clock_gettime() as Sierra-only, so even
if we fixed the issue here, we'd still be stick with portability
breakage like we were for 0.2.9.
So, just disable clock_gettime() on apple.
Tor 0.2.9 has a broader range of fixes and workarounds here, but for
0.2.8, we're just going to maintain the existing behavior.
(The alternative would be to backport both
1eba088054 and
16fcbd21c9 , but the latter is kind of
a subtle kludge in the configure.ac script, and I'm not a fan of
backporting that kind of thing.)
(OpenSSL 1.1 makes EVP_CIPHER_CTX opaque, _and_ adds acceleration
for counter mode on more architectures. So it won't work if we try
the older approach, and it might help if we try the newer one.)
Fixes bug 20588.
It's not okay to use the same varargs list twice, and apparently
some windows build environments produce code here that would leave
tor_asprintf() broken. Fix for bug 20560; bugfix on 0.2.2.11-alpha
when tor_asprintf() was introduced.
(OpenSSL 1.1 makes EVP_CIPHER_CTX opaque, _and_ adds acceleration
for counter mode on more architectures. So it won't work if we try
the older approach, and it might help if we try the newer one.)
Fixes bug 20588.
In our code to write public keys to a string, for some unfathomable
reason since 253f0f160e, we would allocate a memory BIO, then
set the NOCLOSE flag on it, extract its memory buffer, and free it.
Then a little while later we'd free the memory buffer with
BUF_MEM_free().
As of openssl 1.1 this doesn't work any more, since there is now a
BIO_BUF_MEM structure that wraps the BUF_MEM structure. This
BIO_BUF_MEM doesn't get freed in our code.
So, we had a memory leak!
Is this an openssl bug? Maybe. But our code was already pretty
silly. Why mess around with the NOCLOSE flag here when we can just
keep the BIO object around until we don't need the buffer any more?
Fixes bug 20553; bugfix on 0.0.2pre8
This function is allowed to return NULL if the certified key isn't
RSA. But in a couple of places we were treating this as a bug or
internal error, and in one other place we weren't checking for it at
all!
Caught by Isis during code review for #15055. The serious bug was
only on the 15055 branch, thank goodness.
All supported Tors (0.2.4+) require versions of openssl that can
handle this.
Now that our link certificates are RSA2048, this might actually help
vs fingerprinting a little.
See proposal 244. This feature lets us stop looking at the internals
of SSL objects, *and* should let us port better to more SSL libraries,
if they have RFC5705 support.
Preparatory for #19156
Use the following coccinelle script to change uses of
smartlist_add(sl, tor_strdup(str)) to
smartlist_add_strdup(sl, string) (coccinelle script from nickm
via bug 20048):
@@
expression a;
expression b;
@@
- smartlist_add
+ smartlist_add_strdup
(a,
- tor_strdup(
b
- )
)
This commit adds or improves the module-level documenation for:
buffers.c circuitstats.c command.c connection_edge.c control.c
cpuworker.c crypto_curve25519.c crypto_curve25519.h
crypto_ed25519.c crypto_format.c dircollate.c dirserv.c dns.c
dns_structs.h fp_pair.c geoip.c hibernate.c keypin.c ntmain.c
onion.c onion_fast.c onion_ntor.c onion_tap.c periodic.c
protover.c protover.h reasons.c rephist.c replaycache.c
routerlist.c routerparse.c routerset.c statefile.c status.c
tor_main.c workqueue.c
In particular, I've tried to explain (for each documented module)
what each module does, what's in it, what the big idea is, why it
belongs in Tor, and who calls it. In a few cases, I've added TODO
notes about refactoring opportunities.
I've also renamed an argument, and fixed a few DOCDOC comments.
(Specifically, carriage return after a quoted value in a config
line. Fixes bug 19167; bugfix on 0.2.0.16-alpha when we introduced
support for quoted values. Unit tests, changes file, and this
parenthetical by nickm.)
This is a kludge to deal with the fact that `tor_addr_t` doesn't contain
`sun_path`. This currently ONLY happens when circuit isolation is being
checked, for an isolation mode that is force disabled anyway, so the
kludge is "ugly but adequate", but realistically, making `tor_addr_t`
and the AF_UNIX SocksPort code do the right thing is probably the better
option.
Closes ticket 20303.
The LIBRESSL_VERSION_NUMBER check is needed because if our openssl
is really libressl, it will have an openssl version number we can't
really believe.
When deleting unsuitable addresses in get_interface_address6_list(), to
avoid reordering IPv6 interface addresses and keep the order returned by
the OS, use SMARTLIST_DEL_CURRENT_KEEPORDER() instead of
SMARTLIST_DEL_CURRENT().
This issue was reported by René Mayrhofer.
[Closes ticket 20163; changes file written by teor. This paragraph
added by nickm]
Sierra provides clock_gettime(), but not pthread_condattr_setclock.
So we had better lot try to use CLOCK_MONOTONIC as our source for
time when waiting, since we ccan never actually tell the condition
that we mean CLOCK_MONOTONIC.
This isn't a tor bug yet, since we never actually pass a timeout to
tor_cond_wait() outside of the unit tests.
Previously, the IV and key were stored in the structure, even though
they mostly weren't needed. The only purpose they had was to
support a seldom-used API where you could pass NULL when creating
a cipher in order to get a random key/IV, and then pull that key/IV
back out.
This saves 32 bytes per AES instance, and makes it easier to support
different key lengths.
We have a mock for our RSA key generation function, so we now wire
it to pk_generate(). This covers all the cases that were not using
pk_generate() before -- all ~93 of them.
The functions it warns about are:
assert, memcmp, strcat, strcpy, sprintf, malloc, free, realloc,
strdup, strndup, calloc.
Also, fix a few lingering instances of these in the code. Use other
conventions to indicate _intended_ use of assert and
malloc/realloc/etc.
Previously setup_capture_of_logs would prevent log messages from
going to the console entirely. That's a problem, since sometimes
log messages are bugs! Now setup_capture_of_logs() acts sensibly.
If you really do need to keep a message from going to the console
entirely, there is setup_full_capture_of_logs(). But only use that
if you're prepared to make sure that there are no extraneous
messages generated at all.
OpenBSD removes this function, and now that Tor requires Libevent 2,
we should also support the OpenBSD Libevent 2.
Fixes bug 19904; bugfix on 0.2.5.4-alpha.
To maintain precision, to get nanoseconds, we were multiplying our
tick count by a billion, then dividing by ticks-per-second. But
that apparently isn't such a great idea, since ticks-per-second is
sometimes a billion on its own, so our intermediate result was
giving us attoseconds.
When you're counting in attoseconds, you can only fit about 9
seconds into an int64_t, which is not so great for our purposes.
Instead, we now simplify the 1000000000/1000000000 fraction before
we start messing with nanoseconds. This has potential to mess us
up if some future MS version declares that performance counters will
use 1,000,000,007 units per second, but let's burn that bridge when
we come to it.
* Raise limit: 16k isn't all that high.
* Don't log when limit exceded; log later on.
* Say "over" when we log more than we say we log.
* Add target version to changes file
This code uses QueryPerformanceCounter() [**] on Windows,
mach_absolute_time() on OSX, clock_gettime() where available, and
gettimeofday() [*] elsewhere.
Timer types are stored in an opaque OS-specific format; the only
supported operation is to compute the difference between two timers.
[*] As you know, gettimeofday() isn't monotonic, so we include
a simple ratchet function to ensure that it only moves forward.
[**] As you may not know, QueryPerformanceCounter() isn't actually
always as monotonic as you might like it to be, so we ratchet that
one too.
We also include a "coarse monotonic timer" for cases where we don't
actually need high-resolution time. This is GetTickCount{,64}() on
Windows, clock_gettime(CLOCK_MONOTONIC_COARSE) on Linux, and falls
back to regular monotonic time elsewhere.
If we did not find a non-private IPaddress by iterating over interfaces,
we would try to get one via
get_interface_address6_via_udp_socket_hack(). This opens a datagram
socket with IPPROTO_UDP. Previously all our datagram sockets (via
libevent) used IPPROTO_IP, so we did not have that in the sandboxing
whitelist. Add (SOCK_DGRAM, IPPROTO_UDP) sockets to the sandboxing
whitelist. Fixes bug 19660.
We introduded a shadowed variable, thereby causing a log message to
be wrong. Fixes 19578. I believe the bug was introduced by
54d7d31cba in 0.2.2.29-beta.
I grepped and hand-inspected the "it's" instances, to see if any
were supposed to be possessive. While doing that, I found a
"the the", so I grepped to see if there were any more.
Only some very ancient distributions don't ship with Libevent 2 anymore,
even the oldest supported Ubuntu LTS version has it. This allows us to
get rid of a lot of compat code.
Validate that tv_usec inputs to tv_udiff and tv_mdiff are in range.
Do internal calculations in tv_udiff and tv_mdiff in 64-bit,
which makes the function less prone to integer overflow,
particularly on platforms where long and time_t are 32-bit,
but tv_sec is 64-bit, like some BSD configurations.
Check every addition and subtraction that could overflow.
Make sure to memset(0) the destination buffer so we don't leave any
uninitialized data.
Fixes#19462
Signed-off-by: David Goulet <dgoulet@torproject.org>
base16_decodes() now returns the number of decoded bytes. It's interface
changes from returning a "int" to a "ssize_t". Every callsite now checks the
returned value.
Fixes#14013
Signed-off-by: David Goulet <dgoulet@torproject.org>
realloc()ing a thing in order to try to save memory on it just
doesn't make sense with today's allocators. Instead, let's use the
fact that whenever we decompress something, either it isn't too big,
or we chop it up, or we reallocate it.
zlib 1.2 came out in 2003; earlier versions should be dead by now.
Our workaround code was only preventing us from using the gzip
encoding (if we decide to do so), and having some dead code linger
around in torgzip.c
The Autoconf macro AC_USE_SYSTEM_EXTENSIONS defines preprocessor macros
which turn on extensions to C and POSIX. The macro also makes it easier
for developers to use the extensions without needing (or forgetting) to
define them manually.
The macro can be safely used because it was introduced in Autoconf 2.60
and Tor requires Autoconf 2.63 and above.
There's accessors to get at things, but it ends up being rather
cumbersome. The only place where behavior should change is that the
code will fail instead of attempting to generate a new DH key if our
internal sanity check fails.
Like the previous commit, this probably breaks snapshots prior to pre5.
Instead of `ERR_remove_thread_state()` having a modified prototype, it
now has the old prototype and a deprecation annotation. Since it's
pointless to add extra complexity just to remain compatible with an old
OpenSSL development snapshot, update the code to work with 1.1.0pre5
and later.
This is a big-ish patch, but it's very straightforward. Under this
clang warning, we're not actually allowed to have a global variable
without a previous extern declaration for it. The cases where we
violated this rule fall into three roughly equal groups:
* Stuff that should have been static.
* Stuff that was global but where the extern was local to some
other C file.
* Stuff that was only global when built for the unit tests, that
needed a conditional extern in the headers.
The first two were IMO genuine problems; the last is a wart of how
we build tests.
This warning, IIUC, means that the compiler doesn't like it when it
sees a NULL check _after_ we've already dereferenced the
variable. In such cases, it considers itself free to eliminate the
NULL check.
There are a couple of tricky cases:
One was the case related to the fact that tor_addr_to_in6() can
return NULL if it gets a non-AF_INET6 address. The fix was to
create a variant which asserts on the address type, and never
returns NULL.
So, back long ago, XXX012 meant, "before Tor 0.1.2 is released, we
had better revisit this comment and fix it!"
But we have a huge pile of such comments accumulated for a large
number of released versions! Not cool.
So, here's what I tried to do:
* 0.2.9 and 0.2.8 are retained, since those are not yet released.
* XXX+ or XXX++ or XXX++++ or whatever means, "This one looks
quite important!"
* The others, after one-by-one examination, are downgraded to
plain old XXX. Which doesn't mean they aren't a problem -- just
that they cannot possibly be a release-blocking problem.
There are a few places where we want to disable a warning: for
example, when it's impossible to call a legacy API without
triggering it, or when it's impossible to include an external header
without triggering it.
This pile of macros uses GCC's c99 _Pragma support, plus the usual
macro trickery, to enable and disable warnings.
If OpenSSL fails to generate an RSA key, do not retain a dangling
pointer to the previous (uninitialized) key value. The impact here
should be limited to a difficult-to-trigger crash, if OpenSSL is
running an engine that makes key generation failures possible, or if
OpenSSL runs out of memory. Fixes bug 19152; bugfix on
0.2.1.10-alpha. Found by Yuan Jochen Kang, Suman Jana, and Baishakhi
Ray.
This is potentially scary stuff, so let me walk through my analysis.
I think this is a bug, and a backport candidate, but not remotely
triggerable in any useful way.
Observation 1a:
Looking over the OpenSSL code here, the only way we can really fail in
the non-engine case is if malloc() fails. But if malloc() is failing,
then tor_malloc() calls should be tor_asserting -- the only way that an
attacker could do an exploit here would be to figure out some way to
make malloc() fail when openssl does it, but work whenever Tor does it.
(Also ordinary malloc() doesn't fail on platforms like Linux that
overcommit.)
Observation 1b:
Although engines are _allowed_ to fail in extra ways, I can't find much
evidence online that they actually _do_ fail in practice. More evidence
would be nice, though.
Observation 2:
We don't call crypto_pk_generate*() all that often, and we don't do it
in response to external inputs. The only way to get it to happen
remotely would be by causing a hidden service to build new introduction
points.
Observation 3a:
So, let's assume that both of the above observations are wrong, and the
attacker can make us generate a crypto_pk_env_t with a dangling pointer
in its 'key' field, and not immediately crash.
This dangling pointer will point to what used to be an RSA structure,
with the fields all set to NULL. Actually using this RSA structure,
before the memory is reused for anything else, will cause a crash.
In nearly every function where we call crypto_pk_generate*(), we quickly
use the RSA key pointer -- either to sign something, or to encode the
key, or to free the key. The only exception is when we generate an
intro key in rend_consider_services_intro_points(). In that case, we
don't actually use the key until the intro circuit is opened -- at which
point we encode it, and use it to sign an introduction request.
So in order to exploit this bug to do anything besides crash Tor, the
attacker needs to make sure that by the time the introduction circuit
completes, either:
* the e, d, and n BNs look valid, and at least one of the other BNs is
still NULL.
OR
* all 8 of the BNs must look valid.
To look like a valid BN, *they* all need to have their 'top' index plus
their 'd' pointer indicate an addressable region in memory.
So actually getting useful data of of this, rather than a crash, is
going to be pretty damn hard. You'd have to force an introduction point
to be created (or wait for one to be created), and force that particular
crypto_pk_generate*() to fail, and then arrange for the memory that the
RSA points to to in turn point to 3...8 valid BNs, all by the time the
introduction circuit completes.
Naturally, the signature won't check as valid [*], so the intro point
will reject the ESTABLISH_INTRO cell. So you need to _be_ the
introduction point, or you don't actually see this information.
[*] Okay, so if you could somehow make the 'rsa' pointer point to a
different valid RSA key, then you'd get a valid signature of an
ESTABLISH_INTRO cell using a key that was supposed to be used for
something else ... but nothing else looks like that, so you can't use
that signature elsewhere.
Observation 3b:
Your best bet as an attacker would be to make the dangling RSA pointer
actually contain a fake method, with a fake RSA_private_encrypt
function that actually pointed to code you wanted to execute. You'd
still need to transit 3 or 4 pointers deep though in order to make that
work.
Conclusion:
By 1, you probably can't trigger this without Tor crashing from OOM.
By 2, you probably can't trigger this reliably.
By 3, even if I'm wrong about 1 and 2, you have to jump through a pretty
big array of hoops in order to get any kind of data leak or code
execution.
So I'm calling it a bug, but not a security hole. Still worth
patching.
Fortunately, the arithmetic cannot actually overflow, so long as we
*always* check for the size of potentially hostile input before
copying it. I think we do, though. We do check each line against
MAX_LINE_LENGTH, and each object name or object against
MAX_UNPARSED_OBJECT_SIZE, both of which are 128k. So to get this
overflow, we need to have our memarea allocated way way too high up
in RAM, which most allocators won't actually do.
Bugfix on 0.2.1.1-alpha, where memarea was introduced.
Found by Guido Vranken.
Previously, if the header was present, we'd proceed even if the
function wasn't there.
Easy fix for bug 19161. A better fix would involve trying harder to
find libscrypt_scrypt.
We use a pretty specific pair of autoconf tests here to make sure
that we only add this code when:
a) a 64-bit signed multiply fails to link,
AND
b) the same 64-bit signed multiply DOES link correctly when
__mulodi4 is defined.
Closes ticket 19079.
