DIT, for 'Data-Independent Timing', is a bit you can set in the
processor state on sufficiently new Arm CPUs, which promises that a
long list of instructions will deliberately avoid varying their timing
based on the input register values. Just what you want for keeping
your constant-time crypto primitives constant-time.
As far as I'm aware, no CPU has _yet_ implemented any data-dependent
optimisations, so DIT is a safety precaution against them doing so in
future. It would be embarrassing to be caught without it if a future
CPU does do that, so we now turn on DIT in the PuTTY process state.
I've put a call to the new enable_dit() function at the start of every
main() and WinMain() belonging to a program that might do
cryptography (even testcrypt, in case someone uses it for something!),
and in case I missed one there, also added a second call at the first
moment that any cryptography-using part of the code looks as if it
might become active: when an instance of the SSH protocol object is
configured, when the system PRNG is initialised, and when selecting
any cryptographic authentication protocol in an HTTP or SOCKS proxy
connection. With any luck those precautions between them should ensure
it's on whenever we need it.
Arm's own recommendation is that you should carefully choose the
granularity at which you enable and disable DIT: there's a potential
time cost to turning it on and off (I'm not sure what, but plausibly
something of the order of a pipeline flush), so it's a performance hit
to do it _inside_ each individual crypto function, but if CPUs start
supporting significant data-dependent optimisation in future, then it
will also become a noticeable performance hit to just leave it on
across the whole process. So you'd like to do it somewhere in the
middle: for example, you might turn on DIT once around the whole
process of verifying and decrypting an SSH packet, instead of once for
decryption and once for MAC.
With all respect to that recommendation as a strategy for maximum
performance, I'm not following it here. I turn on DIT at the start of
the PuTTY process, and then leave it on. Rationale:
1. PuTTY is not otherwise a performance-critical application: it's
not likely to max out your CPU for any purpose _other_ than
cryptography. The most CPU-intensive non-cryptographic thing I can
imagine a PuTTY process doing is the complicated computation of
font rendering in the terminal, and that will normally be cached
(you don't recompute each glyph from its outline and hints for
every time you display it).
2. I think a bigger risk lies in accidental side channels from having
DIT turned off when it should have been on. I can imagine lots of
causes for that. Missing a crypto operation in some unswept corner
of the code; confusing control flow (like my coroutine macros)
jumping with DIT clear into the middle of a region of code that
expected DIT to have been set at the beginning; having a reference
counter of DIT requests and getting it out of sync.
In a more sophisticated programming language, it might be possible to
avoid the risk in #2 by cleverness with the type system. For example,
in Rust, you could have a zero-sized type that acts as a proof token
for DIT being enabled (it would be constructed by a function that also
sets DIT, have a Drop implementation that clears DIT, and be !Send so
you couldn't use it in a thread other than the one where DIT was set),
and then you could require all the actual crypto functions to take a
DitToken as an extra parameter, at zero runtime cost. Then "oops I
forgot to set DIT around this piece of crypto" would become a compile
error. Even so, you'd have to take some care with coroutine-structured
code (what happens if a Rust async function yields while holding a DIT
token?) and with nesting (if you have two DIT tokens, you don't want
dropping the inner one to clear DIT while the outer one is still there
to wrongly convince callees that it's set). Maybe in Rust you could
get this all to work reliably. But not in C!
DIT is an optional feature of the Arm architecture, so we must first
test to see if it's supported. This is done the same way as we already
do for the various Arm crypto accelerators: on ELF-based systems,
check the appropriate bit in the 'hwcap' words in the ELF aux vector;
on Mac, look for an appropriate sysctl flag.
On Windows I don't know of a way to query the DIT feature, _or_ of a
way to write the necessary enabling instruction in an MSVC-compatible
way. I've _heard_ that it might not be necessary, because Windows
might just turn on DIT unconditionally and leave it on, in an even
more extreme version of my own strategy. I don't have a source for
that - I heard it by word of mouth - but I _hope_ it's true, because
that would suit me very well! Certainly I can't write code to enable
DIT without knowing (a) how to do it, (b) how to know if it's safe.
Nonetheless, I've put the enable_dit() call in all the right places in
the Windows main programs as well as the Unix and cross-platform code,
so that if I later find out that I _can_ put in an explicit enable of
DIT in some way, I'll only have to arrange to set HAVE_ARM_DIT and
compile the enable_dit() function appropriately.
Converting a CmdlineArg straight to a Filename allows us to make the
filename out of the wide-character version of the string on Windows.
So now filenames specified on the command line should generally be
able to handle pathnames containing Unicode characters not in the
system code page.
This change also involves making some char pointers _into_ Filename
structs where they weren't previously: for example, the
'openssh_config_file' variable in Windows Pageant's WinMain().
This begins the process of enabling our Windows applications to handle
Unicode characters on their command lines which don't fit in the
system code page.
Instead of passing plain strings to cmdline_process_param, we now pass
a partially opaque and platform-specific thing called a CmdlineArg.
This has a method that extracts the argument word as a default-encoded
string, and another one that tries to extract it as UTF-8 (though it
may fail if the UTF-8 isn't available).
On Windows, the command line is now constructed by calling
split_into_argv_w on the Unicode command line returned by
GetCommandLineW(), and the UTF-8 method returns text converted
directly from that wide-character form, not going via the system code
page. So it _can_ include UTF-8 characters that wouldn't have
round-tripped via CP_ACP.
This commit introduces the abstraction and switches over the
cross-platform and Windows argv-handling code to use it, with minimal
functional change. Nothing yet tries to call cmdline_arg_get_utf8().
I say 'cross-platform and Windows' because on the Unix side there's
still a lot of use of plain old argv which I haven't converted. That
would be a much larger project, and isn't currently needed: the
_current_ aim of this abstraction is to get the right things to happen
relating to Unicode on Windows, so for code that doesn't run on
Windows anyway, it's not adding value. (Also there's a tension with
GTK, which wants to talk to standard argv and extract arguments _it_
knows about, so at the very least we'd have to let it munge argv
before importing it into this new system.)
Constructing a FontSpec in platform-independent code is awkward,
because you can't call fontspec_new() outside the platform subdirs
(since its prototype varies per platform). But sometimes you just need
_some_ valid FontSpec, e.g. to put in a Conf that will be used in some
place where you don't actually care about font settings, such as a
purely CLI program.
Both Unix and Windows _have_ an idiom for this, but they're different,
because their FontSpec constructors have different prototypes. The
existing CLI tools have always had per-platform main source files, so
they just use the locally appropriate method of constructing a boring
don't-care FontSpec.
But if you want a _platform-independent_ main source file, such as you
might find in a test program, then that's rather awkward. Better to
have a platform-independent API for making a default FontSpec.
I mentioned recently (in commit 9e7d4c53d8) message that I'm no
longer fond of the variable name 'ret', because it's used in two quite
different contexts: it's the return value from a subroutine you just
called (e.g. 'int ret = read(fd, buf, len);' and then check for error
or EOF), or it's the value you're preparing to return from the
_containing_ routine (maybe by assigning it a default value and then
conditionally modifying it, or by starting at NULL and reallocating,
or setting it just before using the 'goto out' cleanup idiom). In the
past I've occasionally made mistakes by forgetting which meaning the
variable had, or accidentally conflating both uses.
If all else fails, I now prefer 'retd' (short for 'returned') in the
former situation, and 'toret' (obviously, the value 'to return') in
the latter case. But even better is to pick a name that actually says
something more specific about what the thing actually is.
One particular bad habit throughout this codebase is to have a set of
functions that deal with some object type (say 'Foo'), all *but one*
of which take a 'Foo *foo' parameter, but the foo_new() function
starts with 'Foo *ret = snew(Foo)'. If all the rest of them think the
canonical name for the ambient Foo is 'foo', so should foo_new()!
So here's a no-brainer start on cutting down on the uses of 'ret': I
looked for all the cases where it was being assigned the result of an
allocation, and renamed the variable to be a description of the thing
being allocated. In the case of a new() function belonging to a
family, I picked the same name as the rest of the functions in its own
family, for consistency. In other cases I picked something sensible.
One case where it _does_ make sense not to use your usual name for the
variable type is when you're cloning an existing object. In that case,
_neither_ of the Foo objects involved should be called 'foo', because
it's ambiguous! They should be named so you can see which is which. In
the two cases I found here, I've called them 'orig' and 'copy'.
As in the previous refactoring, many thanks to clang-rename for the
help.
I only recently found out that OpenSSH defined their own protocol IDs
for AES-GCM, defined to work the same as the standard ones except that
they fixed the semantics for how you select the linked cipher+MAC pair
during key exchange.
(RFC 5647 defines protocol ids for AES-GCM in both the cipher and MAC
namespaces, and requires that you MUST select both or neither - but
this contradicts the selection policy set out in the base SSH RFCs,
and there's no discussion of how you resolve a conflict between them!
OpenSSH's answer is to do it the same way ChaCha20-Poly1305 works,
because that will ensure the two suites don't fight.)
People do occasionally ask us for this linked cipher/MAC pair, and now
I know it's actually feasible, I've implemented it, including a pair
of vector implementations for x86 and Arm using their respective
architecture extensions for multiplying polynomials over GF(2).
Unlike ChaCha20-Poly1305, I've kept the cipher and MAC implementations
in separate objects, with an arm's-length link between them that the
MAC uses when it needs to encrypt single cipher blocks to use as the
inputs to the MAC algorithm. That enables the cipher and the MAC to be
independently selected from their hardware-accelerated versions, just
in case someone runs on a system that has polynomial multiplication
instructions but not AES acceleration, or vice versa.
There's a fourth implementation of the GCM MAC, which is a pure
software implementation of the same algorithm used in the vectorised
versions. It's too slow to use live, but I've kept it in the code for
future testing needs, and because it's a convenient place to dump my
design comments.
The vectorised implementations are fairly crude as far as optimisation
goes. I'm sure serious x86 _or_ Arm optimisation engineers would look
at them and laugh. But GCM is a fast MAC compared to HMAC-SHA-256
(indeed compared to HMAC-anything-at-all), so it should at least be
good enough to use. And we've got a working version with some tests
now, so if someone else wants to improve them, they can.
I booted my M1 Mac into macOS rather than Asahi for the first time in
a while, and discovered that an OS update seems to have added some
sysctl flags indicating the presence of the CPU extensions that I
previously knew of no way to check for! Added them checks to
arm_arch_queries.c, though I've also retained backwards compat with
the previous OS version which didn't have them at all.
If the function name (or expression) in a function call or declaration
is itself so long that even the first argument doesn't fit after it on
the same line, or if that would leave so little space that it would be
silly to try to wrap all the run-on lines into a tall thin column,
then I used to do this
ludicrously_long_function_name
(arg1, arg2, arg3);
and now prefer this
ludicrously_long_function_name(
arg1, arg2, arg3);
I picked up the habit from Python, where the latter idiom is required
by Python's syntactic significance of newlines (you can write the
former if you use a backslash-continuation, but pretty much everyone
seems to agree that that's much uglier). But I've found it works well
in C as well: it makes it more obvious that the previous line is
incomplete, it gives you a tiny bit more space to wrap the following
lines into (the old idiom indents the _third_ line one space beyond
the second), and I generally turn out to agree with the knock-on
indentation decisions made by at least Emacs if you do it in the
middle of a complex expression. Plus, of course, using the _same_
idiom between C and Python means less state-switching.
So, while I'm making annoying indentation changes in general, this
seems like a good time to dig out all the cases of the old idiom in
this code, and switch them over to the new.
Correcting a source file name in the docs just now reminded me that
I've seen a lot of outdated source file names elsewhere in the code,
due to all the reorganisation since we moved to cmake. Here's a giant
pass of trying to make them all accurate again.
All the seat functions that request an interactive prompt of some kind
to the user - both the main seat_get_userpass_input and the various
confirmation dialogs for things like host keys - were using a simple
int return value, with the general semantics of 0 = "fail", 1 =
"proceed" (and in the case of seat_get_userpass_input, answers to the
prompts were provided), and -1 = "request in progress, wait for a
callback".
In this commit I change all those functions' return types to a new
struct called SeatPromptResult, whose primary field is an enum
replacing those simple integer values.
The main purpose is that the enum has not three but _four_ values: the
"fail" result has been split into 'user abort' and 'software abort'.
The distinction is that a user abort occurs as a result of an
interactive UI action, such as the user clicking 'cancel' in a dialog
box or hitting ^D or ^C at a terminal password prompt - and therefore,
there's no need to display an error message telling the user that the
interactive operation has failed, because the user already knows,
because they _did_ it. 'Software abort' is from any other cause, where
PuTTY is the first to know there was a problem, and has to tell the
user.
We already had this 'user abort' vs 'software abort' distinction in
other parts of the code - the SSH backend has separate termination
functions which protocol layers can call. But we assumed that any
failure from an interactive prompt request fell into the 'user abort'
category, which is not true. A couple of examples: if you configure a
host key fingerprint in your saved session via the SSH > Host keys
pane, and the server presents a host key that doesn't match it, then
verify_ssh_host_key would report that the user had aborted the
connection, and feel no need to tell the user what had gone wrong!
Similarly, if a password provided on the command line was not
accepted, then (after I fixed the semantics of that in the previous
commit) the same wrong handling would occur.
So now, those Seat prompt functions too can communicate whether the
user or the software originated a connection abort. And in the latter
case, we also provide an error message to present to the user. Result:
in those two example cases (and others), error messages should no
longer go missing.
Implementation note: to avoid the hassle of having the error message
in a SeatPromptResult being a dynamically allocated string (and hence,
every recipient of one must always check whether it's non-NULL and
free it on every exit path, plus being careful about copying the
struct around), I've instead arranged that the structure contains a
function pointer and a couple of parameters, so that the string form
of the message can be constructed on demand. That way, the only users
who need to free it are the ones who actually _asked_ for it in the
first place, which is a much smaller set.
(This is one of the rare occasions that I regret not having C++'s
extra features available in this code base - a unique_ptr or
shared_ptr to a string would have been just the thing here, and the
compiler would have done all the hard work for me of remembering where
to insert the frees!)
It's another file that should have been subdivided into lots of tiny
separate things in the utils library - especially since for some
reason I made a completely separate 'guimisc' cmake-level library for
it when there was no need.
This applies to all of AES, SHA-1, SHA-256 and SHA-512. All those
source files previously contained multiple implementations of the
algorithm, enabled or disabled by ifdefs detecting whether they would
work on a given compiler. And in order to get advanced machine
instructions like AES-NI or NEON crypto into the output file when the
compile flags hadn't enabled them, we had to do nasty stuff with
compiler-specific pragmas or attributes.
Now we can do the detection at cmake time, and enable advanced
instructions in the more sensible way, by compile-time flags. So I've
broken up each of these modules into lots of sub-pieces: a file called
(e.g.) 'foo-common.c' containing common definitions across all
implementations (such as round constants), one called 'foo-select.c'
containing the top-level vtable(s), and a separate file for each
implementation exporting just the vtable(s) for that implementation.
One advantage of this is that it depends a lot less on compiler-
specific bodgery. My particular least favourite part of the previous
setup was the part where I had to _manually_ define some Arm ACLE
feature macros before including <arm_neon.h>, so that it would define
the intrinsics I wanted. Now I'm enabling interesting architecture
features in the normal way, on the compiler command line, there's no
need for that kind of trick: the right feature macros are already
defined and <arm_neon.h> does the right thing.
Another change in this reorganisation is that I've stopped assuming
there's just one hardware implementation per platform. Previously, the
accelerated vtables were called things like sha256_hw, and varied
between FOO-NI and NEON depending on platform; and the selection code
would simply ask 'is hw available? if so, use hw, else sw'. Now, each
HW acceleration strategy names its vtable its own way, and the
selection vtable has a whole list of possibilities to iterate over
looking for a supported one. So if someone feels like writing a second
accelerated implementation of something for a given platform - for
example, I've heard you can use plain NEON to speed up AES somewhat
even without the crypto extension - then it will now have somewhere to
drop in alongside the existing ones.
Now that the new CMake build system is encouraging us to lay out the
code like a set of libraries, it seems like a good idea to make them
look more _like_ libraries, by putting things into separate modules as
far as possible.
This fixes several previous annoyances in which you had to link
against some object in order to get a function you needed, but that
object also contained other functions you didn't need which included
link-time symbol references you didn't want to have to deal with. The
usual offender was subsidiary supporting programs including misc.c for
some innocuous function and then finding they had to deal with the
requirements of buildinfo().
This big reorganisation introduces three new subdirectories called
'utils', one at the top level and one in each platform subdir. In each
case, the directory contains basically the same files that were
previously placed in the 'utils' build-time library, except that the
ones that were extremely miscellaneous (misc.c, utils.c, uxmisc.c,
winmisc.c, winmiscs.c, winutils.c) have been split up into much
smaller pieces.
