The previous code used comparison operators >= and == that are quite likely to
be compiled to branches by some compilers on some architectures (with some
optimisation levels).
For example, take the following function:
void old_update( size_t data_len, size_t *padlen )
{
*padlen *= ( data_len >= *padlen + 1 );
}
With Clang 3.8, let's compile it for the Arm v6-M architecture:
% clang --target=arm-none-eabi -march=armv6-m -Os foo.c -S -o - |
sed -n '/^old_update:$/,/\.size/p'
old_update:
.fnstart
@ BB#0:
.save {r4, lr}
push {r4, lr}
ldr r2, [r1]
adds r4, r2, #1
movs r3, #0
cmp r4, r0
bls .LBB0_2
@ BB#1:
mov r2, r3
.LBB0_2:
str r2, [r1]
pop {r4, pc}
.Lfunc_end0:
.size old_update, .Lfunc_end0-old_update
We can see an unbalanced secret-dependant branch, resulting in a total
execution time depends on the value of the secret (here padlen) in a
straightforward way.
The new version, based on bit operations, doesn't have this issue:
new_update:
.fnstart
@ BB#0:
ldr r2, [r1]
subs r0, r0, #1
subs r0, r0, r2
asrs r0, r0, #31
bics r2, r0
str r2, [r1]
bx lr
.Lfunc_end1:
.size new_update, .Lfunc_end1-new_update
(As a bonus, it's smaller and uses less stack.)
While there's no formal guarantee that the version based on bit operations in
C won't be translated using branches by the compiler, experiments tend to show
that's the case [1], and it is commonly accepted knowledge in the practical
crypto community that if we want to sick to C, bit operations are the safest
bet [2].
[1] https://github.com/mpg/ct/blob/master/results
[2] https://github.com/veorq/cryptocoding
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
* development:
Update copyright notices to use Linux Foundation guidance
Undef ASSERT before defining it to ensure that no previous definition has sneaked in through included files.
Add ChangeLog entry for X.509 CN-type vulnerability
Improve documentation of cn in x509_crt_verify()
Fix comparison between different name types
Add test: DNS names should not match IP addresses
Remove obsolete buildbot reference in compat.sh
Fix misuse of printf in shell script
Fix added proxy command when IPv6 is used
Simplify test syntax
Fix logic error in setting client port
ssl-opt.sh: include test name in log files
ssl-opt.sh: remove old buildbot-specific condition
ssl-opt.sh: add proxy to all DTLS tests
Signed-off-by: Dan Handley <dan.handley@arm.com>
The tests are supposed to be failing now (in all.sh component
test_memsan_constant_flow), but they don't as apparently MemSan doesn't
complain when the src argument of memcpy() is uninitialized, see
https://github.com/google/sanitizers/issues/1296
The next commit will add an option to test constant flow with valgrind, which
will hopefully correctly flag the current non-constant-flow implementation.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
This paves the way for a constant-flow implementation of HMAC checking, by
making sure that the comparison happens at a constant address. The missing
step is obviously to copy the HMAC from the secret offset to this temporary
buffer with constant flow, which will be done in the next few commits.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
As a result, the copyright of contributors other than Arm is now
acknowledged, and the years of publishing are no longer tracked in the
source files.
Also remove the now-redundant lines declaring that the files are part of
MbedTLS.
This commit was generated using the following script:
# ========================
#!/bin/sh
# Find files
find '(' -path './.git' -o -path './3rdparty' ')' -prune -o -type f -print | xargs sed -bi '
# Replace copyright attribution line
s/Copyright.*Arm.*/Copyright The Mbed TLS Contributors/I
# Remove redundant declaration and the preceding line
$!N
/This file is part of Mbed TLS/Id
P
D
'
# ========================
Signed-off-by: Bence Szépkúti <bence.szepkuti@arm.com>
The test function now depends on MBEDTLS_TEST_HOOKS, which is enabled by
config.py full, and since there are already components in all.sh exercising
the full config, this test function is sill exercised even with this new
dependency.
Since this is the first time a test function depends on MBEDTLS_TEST_HOOKS,
fix a bug in check-names.sh that wasn't apparent so far: headers from
library/*.h were not considered when looking for macro definitions. This
became apparent because MBEDTLS_STATIC_TESTABLE is defined in library/common.h
and started being used in library/ssl_msg.c, so was flagged as a likely typo.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Just move code from ssl_decrypt_buf() to the new cf_hmac() function and then
call cf_hmac() from there.
This makes the new cf_hmac() function used, opening the door for making it
static in the next commit. It also validates that its interface works for
using it in ssl_decrypt_buf().
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
The dummy implementation is not constant-flow at all for now, it's just
here as a starting point and a support for developing the tests and putting
the infrastructure in place.
Depending on the implementation strategy, there might be various corner cases
depending on where the lengths fall relative to block boundaries. So it seems
safer to just test all possible lengths in a given range than to use only a
few randomly-chosen values.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
The condition is a complex and repeated a few times. There were already some
inconsistencies in the repetitions as some of them forgot about DES.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
In library source files, include "common.h", which takes care of
including "mbedtls/config.h" (or the alternative MBEDTLS_CONFIG_FILE)
and other things that are used throughout the library.
FROM=$'#if !defined(MBEDTLS_CONFIG_FILE)\n#include "mbedtls/config.h"\n#else\n#include MBEDTLS_CONFIG_FILE\n#endif' perl -i -0777 -pe 's~\Q$ENV{FROM}~#include "common.h"~' library/*.c 3rdparty/*/library/*.c scripts/data_files/error.fmt scripts/data_files/version_features.fmt
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
The previous version attempted to write the explicit IV from
the destination buffer before it has been written there.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
This is to avoid confusion with the class of macros
MBEDTLS_SSL_PROTO_TLS1_X
which have an underscore between major and minor version number.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
TLS 1.3 record protection allows the addition of an arbitrary amount
of padding.
This commit introduces a configuration option
```
MBEDTLS_SSL_TLS13_PADDING_GRANULARITY
```
The semantics of this option is that padding is chosen in a minimal
way so that the padded plaintext has a length which is a multiple of
MBEDTLS_SSL_TLS13_PADDING_GRANULARITY.
For example, setting MBEDTLS_SSL_TLS13_PADDING_GRANULARITY to 1024
means that padded plaintexts will have length 1024, 2048, ..., while
setting it to 1 means that no padding will be used.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
The structure `mbedtls_ssl_transform` representing record protection
transformations should ideally be used through a function-based
interface only, as this will ease change of implementation as well
as the addition of new record protection routines in the future.
This commit makes a step in that direction by introducing the
helper function `ssl_transform_get_explicit_iv_len()` which
returns the size of the pre-expansion during record encryption
due to the potential addition of an explicit IV.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
This commit simplifies nonce derivation for AEAD based record protection
routines in the following way.
So far, code distinguished between the cases of GCM+CCM and ChachaPoly:
- In the case of GCM+CCM, the AEAD nonce is the concatentation
of a 4-byte Fixed IV and a dynamically chosen 8-byte IV which is prepended
to the record. In Mbed TLS, this is always chosen to be the record sequence
number, but it need not to.
- In the case of ChaChaPoly, the AEAD nonce is derived as
`( 12-byte Fixed IV ) XOR ( 0 || 8-byte dynamic IV == record seq nr )`
and the dynamically chosen IV is no longer prepended to the record.
This commit removes this distinction by always computing the record nonce
via the formula
`IV == ( Fixed IV || 0 ) XOR ( 0 || Dynamic IV )`
The ChaChaPoly case is recovered in case `Len(Fixed IV) == Len(IV)`, and
GCM+CCM is recovered when `Len(IV) == Len(Fixed IV) + Len(Dynamic IV)`.
Moreover, a getter stub `ssl_transform_aead_dynamic_iv_is_explicit()`
is introduced which infers from a transform whether the dynamically
chosen part of the IV is explicit, which in the current implementation
of `mbedtls_ssl_transform` can be derived from the helper field
`mbedtls_ssl_transform::fixed_ivlen`.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
The computation of the per-record nonce for AEAD record protection
varies with the AEAD algorithm and the TLS version in use.
This commit introduces a helper function for the nonce computation
to ease readability of the quite monolithic record encrytion routine.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
The previous record protection code added the explicit part of the
record nonce prior to encrypting the record. This temporarily leaves
the record structure in the undesireable state that the data outsie
of the interval `rec->data_offset, .., rec->data_offset + rec->data_len`
has already been written.
This commit moves the addition of the explicit IV past record encryption.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
The internal functions
`ssl_cid_{build/parse}_inner_plaintext()`
implement the TLSInnerPlaintext mechanism used by DTLS 1.2 + CID
in order to allow for flexible length padding and to protect the
true content type of a record.
This feature is also present in TLS 1.3 support for which is under
development. As a preparatory step towards sharing the code between
the case of DTLS 1.2 + CID and TLS 1.3, this commit renames
`ssl_cid_{build/parse}_inner_plaintext()`
to
`ssl_{build/parse}_inner_plaintext()`.
Signed-off-by: Hanno Becker <hanno.becker@arm.com>
Since the server might want to have a different maximum fragment length
for the outgoing messages than the negotiated one - introduce a new way of
computing it. This commit also adds additional ssl-opt.sh tests ensuring
that the maximum fragment lengths are set as expected.
mbedtls_ssl_get_max_frag_len() is now a deprecated function,
being an alias to mbedtls_ssl_get_output_max_frag_len(). The behaviour
of this function is the same as before.
Signed-off-by: Andrzej Kurek <andrzej.kurek@arm.com>
The current logging was sub-standard, in particular there was no trace
whatsoever of the HelloVerifyRequest being sent. Now it's being logged with
the usual levels: 4 for full content, 2 return of f_send, 1 decision about
sending it (or taking other branches in the same function) because that's the
same level as state changes in the handshake, and also same as the "possible
client reconnect" message" to which it's the logical continuation (what are we
doing about it?).
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
Section 4.2.8 of RFC 6347 describes how to handle the case of a DTLS client
establishing a new connection using the same UDP quartet as an already active
connection, which we implement under the compile option
MBEDTLS_SSL_DLTS_CLIENT_PORT_REUSE. Relevant excerpts:
[the server] MUST NOT destroy the existing
association until the client has demonstrated reachability either by
completing a cookie exchange or by completing a complete handshake
including delivering a verifiable Finished message.
[...]
The reachability requirement prevents
off-path/blind attackers from destroying associations merely by
sending forged ClientHellos.
Our code chooses to use a cookie exchange for establishing reachability, but
unfortunately that check was effectively removed in a recent refactoring,
which changed what value ssl_handle_possible_reconnect() needs to return in
order for ssl_get_next_record() (introduced in that refactoring) to take the
proper action. Unfortunately, in addition to changing the value, the
refactoring also changed a return statement to an assignment to the ret
variable, causing the function to reach the code for a valid cookie, which
immediately destroys the existing association, effectively bypassing the
cookie verification.
This commit fixes that by immediately returning after sending a
HelloVerifyRequest when a ClientHello without a valid cookie is found. It also
updates the description of the function to reflect the new return value
convention (the refactoring updated the code but not the documentation).
The commit that changed the return value convention (and introduced the bug)
is 2fddd3765e, whose commit message explains the
change.
Note: this bug also indirectly caused the ssl-opt.sh test case "DTLS client
reconnect from same port: reconnect" to occasionally fail due to a race
condition between the reception of the ClientHello carrying a valid cookie and
the closure of the connection by the server after noticing the ClientHello
didn't carry a valid cookie after it incorrectly destroyed the previous
connection, that could cause that ClientHello to be invisible to the server
(if that message reaches the server just before it does `net_close()`). A
welcome side effect of this commit is to remove that race condition, as the
new connection will immediately start with a ClientHello carrying a valid
cookie in the SSL input buffer, so the server will not call `net_close()` and
not risk discarding a better ClientHello that arrived in the meantime.
Signed-off-by: Manuel Pégourié-Gonnard <manuel.pegourie-gonnard@arm.com>
This commit is the final step in separating the functionality of
what was originally ssl_tls.c into both ssl_tls.c and ssl_msg.c.
So far, ssl_msg.c has been created as an identical copy of ssl_tls.c.
For each block of code in these files, this commit removes it from
precisely one of the two files, depending on where the respective
functionality belongs.
The splitting separates the following functionalities:
1) An implementation of the TLS and DTLS messaging layer, that is,
the record layer as well as the DTLS retransmission state machine.
This is now contained in ssl_msg.c
2) Handshake parsing and writing functions shared between client and
server (functions specific to either client or server are implemented
in ssl_cli.c and ssl_srv.c, respectively).
This is remains in ssl_tls.c.
This commit adds the newly created copy ssl_msg.c of ssl_tls.c
to the build system but guards its content by an `#if 0 ... #endif`
preprocessor guard in order to avoid compilation failures resulting
from code duplication. This guard will be removed once the contents
of ssl_tls.c and ssl_msg.c have been made disjoint.
This commit is the first in a series of commits aiming to split
the content of ssl_tls.c in two files ssl_tls.c and ssl_msg.c.
As a first step, this commit replaces ssl_tls.c by two identical
copies ssl_tls_old.c and ssl_msg.c. Even though the file
ssl_tls_old.c will subsequently be renamed back into ssl_tls.c,
this approach retains the git history in both files.
