This commit is the first in a series demonstrating how code-size
can be reduced by hardcoding parts of the SSL configuration at
compile-time, focusing on the example of the configuration of
the ExtendedMasterSecret extension.
The flexibility of an SSL configuration defined a runtime vs.
compile-time is necessary for the use of Mbed TLS as a
dynamically linked library, but is undesirable in constrained
environments because it introduces the following overhead:
- Definition of SSL configuration API (code-size overhead)
(and on the application-side: The API needs to be called)
- Additional fields in the SSL configuration (RAM overhead,
and potentially code-size overhead if structures grow
beyond immediate-offset bounds).
- Dereferencing is needed to obtain configuration settings.
- Code contains branches and potentially additional structure
fields to distinguish between different configurations.
Considering the example of the ExtendedMasterSecret extension,
this instantiates as follows:
- mbedtls_ssl_conf_extended_master_secret() and
mbedtls_ssl_conf_extended_master_secret_enforced()
are introduced to configure the ExtendedMasterSecret extension.
- mbedtls_ssl_config contains bitflags `extended_ms` and
`enforce_extended_master_secret` reflecting the runtime
configuration of the ExtendedMasterSecret extension.
- Whenever we need to access these fields, we need a chain
of dereferences `ssl->conf->extended_ms`.
- Determining whether Client/Server should write the
ExtendedMasterSecret extension needs a branch
depending on `extended_ms`, and the state of the
ExtendedMasterSecret negotiation needs to be stored in a new
handshake-local variable mbedtls_ssl_handshake_params::extended_ms.
Finally (that's the point of ExtendedMasterSecret) key derivation
depends on this handshake-local state of ExtendedMasterSecret.
All this is unnecessary if it is known at compile-time that the
ExtendedMasterSecret extension is used and enforced:
- No API calls are necessary because the configuration is fixed
at compile-time.
- No SSL config fields are necessary because there are corresponding
compile-time constants instead.
- Accordingly, no dereferences for field accesses are necessary,
and these accesses can instead be replaced by the corresponding
compile-time constants.
- Branches can be eliminated at compile-time because the compiler
knows the configuration. Also, specifically for the ExtendedMasterSecret
extension, the field `extended_ms` in the handshake structure
is unnecessary, because we can fail immediately during the Hello-
stage of the handshake if the ExtendedMasterSecret extension
is not negotiated; accordingly, the non-ExtendedMS code-path
can be eliminated from the key derivation logic.
A way needs to be found to allow fixing parts of the SSL configuration
at compile-time which removes this overhead in case it is used,
while at the same time maintaining readability and backwards
compatibility.
This commit proposes the following approach:
From the user perspective, for aspect of the SSL configuration
mbedtls_ssl_config that should be configurable at compile-time,
introduce a compile-time option MBEDTLS_SSL_CONF_FIELD_NAME.
If this option is not defined, the field is kept and configurable
at runtime as usual. If the option is defined, the field is logically
forced to the value of the option at compile time.
Internally, read-access to fields in the SSL configuration which are
configurable at compile-time gets replaced by new `static inline` getter
functions which evaluate to the corresponding field access or to the
constant MBEDTLS_SSL_CONF_FIELD_NAME, depending on whether the latter
is defined or not.
Write-access to fields which are configurable at compile-time needs
to be removed: Specifically, the corresponding API itself either
needs to be removed or replaced by a stub function without effect.
This commit takes the latter approach, which has the benefit of
not requiring any change on the example applications, but introducing
the risk of mismatching API calls and compile-time configuration,
in case a user doesn't correctly keep track of which parts of the
configuration have been fixed at compile-time, and which haven't.
Write-access for the purpose of setting defaults is simply omitted.
So far, `ssl_client2` printed the CRT info for the peer's CRT
by requesting the latter through `mbedtls_ssl_get_peer_cert()`
at the end of the handshake, and printing it via
`mbedtls_x509_crt_info()`. When `MBEDTLS_SSL_KEEP_PEER_CERTIFICATE`
is disabled, this does no longer work because the peer's CRT
isn't stored beyond the handshake.
This makes some tests in `ssl-opt.sh` fail which rely on the CRT
info output for the peer certificate.
This commit modifies `ssl_client2` to extract the peer CRT info
from the verification callback, which is always called at a time
when the peer's CRT is available. This way, the peer's CRT info
is still printed if `MBEDTLS_SSL_KEEP_PEER_CERTIFICATE` is disabled.
Introduce MBEDTLS_X509_INFO to indicate the availability of the
mbedtls_x509_*_info() function and closely related APIs. When this is
not defined, also omit name and description from
mbedtls_oid_descriptor_t, and omit OID arrays, macros, and types that
are entirely unused. This saves several KB of code space.
Change-Id: I056312613379890e0d70e1d08c34171287c0aa17
This is enabled by default as we generally enable things by default unless
there's a reason not to (experimental, deprecated, security risk).
We need a compile-time option because, even though the functions themselves
can be easily garbage-collected by the linker, implementing them will require
saving 64 bytes of Client/ServerHello.random values after the handshake, that
would otherwise not be needed, and people who don't need this feature
shouldn't have to pay the price of increased RAM usage.
A positive option looks better, but comes with the following compatibility
issue: people using a custom config.h that is not based on the default
config.h and need TLS support would need to manually change their config in
order to still get TLS.
Work around that by making the public option negative. Internally the positive
option is used, though.
In the future (when preparing the next major version), we might want to switch
back to a positive option as this would be more consistent with other options
we have.
* origin/pr/2481:
Document support for MD2 and MD4 in programs/x509/cert_write
Correct name of X.509 parsing test for well-formed, ill-signed CRT
Add test cases exercising successful verification of MD2/MD4/MD5 CRT
Add test case exercising verification of valid MD2 CRT
Add MD[245] test CRTs to tree
Add instructions for MD[245] test CRTs to tests/data_files/Makefile
Add suppport for MD2 to CSR and CRT writing example programs
Convert further x509parse tests to use lower-case hex data
Correct placement of ChangeLog entry
Adapt ChangeLog
Use SHA-256 instead of MD2 in X.509 CRT parsing tests
Consistently use lower case hex data in X.509 parsing tests
To prevent dropping the same message over and over again, the UDP proxy
test application programs/test/udp_proxy _logically_ maintains a mapping
from records to the number of times the record has already been dropped,
and stops dropping once a configurable threshold (currently 2) is passed.
However, the actual implementation deviates from this logical view
in two crucial respects:
- To keep the implementation simple and independent of
implementations of suitable map interfaces, it only counts how
many times a record of a given _size_ has been dropped, and
stops dropping further records of that size once the configurable
threshold is passed. Of course, this is not fail-proof, but a
good enough approximation for the proxy, and it allows to use
an inefficient but simple array for the required map.
- The implementation mixes datagram lengths and record lengths:
When deciding whether it is allowed to drop a datagram, it
uses the total datagram size as a lookup index into the map
counting the number of times a package has been dropped. However,
when updating this map, the UDP proxy traverses the datagram
record by record, and updates the mapping at the level of record
lengths.
Apart from this inconsistency, the current implementation suffers
from a lack of bounds checking for the parsed length of incoming
DTLS records that can lead to a buffer overflow when facing
malformed records.
This commit removes the inconsistency in datagram vs. record length
and resolves the buffer overflow issue by not attempting any dissection
of datagrams into records, and instead only counting how often _datagrams_
of a particular size have been dropped.
There is only one practical situation where this makes a difference:
If datagram packing is used by default but disabled on retransmission
(which OpenSSL has been seen to do), it can happen that we drop a
datagram in its initial transmission, then also drop some of its records
when they retransmitted one-by-one afterwards, yet still keeping the
drop-counter at 1 instead of 2. However, even in this situation, we'll
correctly count the number of droppings from that point on and eventually
stop dropping, because the peer will not fall back to using packing
and hence use stable record lengths.
For now the option has no effect.
Adapted existing example config files. The fact that I needed to do this
highlights that this is a slightly incompatible change: existing users need to
update their existing custom configs (if standalone as opposed to based on the
default config) in order to still get the same behaviour.
The alternative would be to have a negative config option (eg NO_TLS or
DTLS_ONLY) but this doesn't fit as nicely with the existing options, so
hopefully the minor incompatibility is acceptable.
I don't think it's worth adding a new component to all.sh:
- builds with both DTLS and TLS are done in the default (and full) config
- TLS-only builds are done with eg config-suite-b.h in test-ref-configs
- a DTLS-only build is done with config-thread.h in test-ref-configs
- builds with none of them (and SSL_TLS_C enabled) are forbidden
To prevent dropping the same message over and over again, the UDP proxy
test application programs/test/udp_proxy _logically_ maintains a mapping
from records to the number of times the record has already been dropped,
and stops dropping once a configurable threshold (currently 2) is passed.
However, the actual implementation deviates from this logical view
in two crucial respects:
- To keep the implementation simple and independent of
implementations of suitable map interfaces, it only counts how
many times a record of a given _size_ has been dropped, and
stops dropping further records of that size once the configurable
threshold is passed. Of course, this is not fail-proof, but a
good enough approximation for the proxy, and it allows to use
an inefficient but simple array for the required map.
- The implementation mixes datagram lengths and record lengths:
When deciding whether it is allowed to drop a datagram, it
uses the total datagram size as a lookup index into the map
counting the number of times a package has been dropped. However,
when updating this map, the UDP proxy traverses the datagram
record by record, and updates the mapping at the level of record
lengths.
Apart from this inconsistency, the introduction of the Connection ID
feature leads to yet another problem: The CID length is not part of
the record header but dynamically negotiated during (potentially
encrypted!) handshakes, and it is hence impossible for a passive traffic
analyzer (in this case our UDP proxy) to reliably parse record headers;
especially, it isn't possible to reliably infer the length of a record,
nor to dissect a datagram into records.
The previous implementation of the UDP proxy was not CID-aware and
assumed that the record length would always reside at offsets 11, 12
in the DTLS record header, which would allow it to iterate through
the datagram record by record. As mentioned, this is no longer possible
for CID-based records, and the current implementation can run into
a buffer overflow in this case (because it doesn't validate that
the record length is not larger than what remains in the datagram).
This commit removes the inconsistency in datagram vs. record length
and resolves the buffer overflow issue by not attempting any dissection
of datagrams into records, and instead only counting how often _datagrams_
of a particular size have been dropped.
There is only one practical situation where this makes a difference:
If datagram packing is used by default but disabled on retransmission
(which OpenSSL has been seen to do), it can happen that we drop a
datagram in its initial transmission, then also drop some of its records
when they retransmitted one-by-one afterwards, yet still keeping the
drop-counter at 1 instead of 2. However, even in this situation, we'll
correctly count the number of droppings from that point on and eventually
stop dropping, because the peer will not fall back to using packing
and hence use stable record lengths.
The example programs programs/x509/cert_req and programs/x509/cert_write
(demonstrating the use of X.509 CSR and CRT writing functionality)
previously didn't support MD2 signatures.
For testing purposes, this commit adds support for MD2 to cert_req,
and support for MD2 and MD4 to cert_write.
We have explicit recommendations to use US spelling for technical writing, so
let's apply this to code as well for uniformity. (My fingers tend to prefer UK
spelling, so this needs to be fixed in many places.)
sed -i 's/\([Ss]eriali\)s/\1z/g' **/*.[ch] **/*.function **/*.data ChangeLog
This allows callers to discover what an appropriate size is. Otherwise they'd
have to either try repeatedly, or allocate an overly large buffer (or some
combination of those).
Adapt documentation an example usage in ssl_client2.
Avoid useless copy with mbedtls_ssl_get_session() before serialising.
Used in ssl_client2 for testing and demonstrating usage, but unfortunately
that means mbedtls_ssl_get_session() is no longer tested, which will be fixed
in the next commit.
This provides basic testing for the session (de)serialisation functions, as
well as an example of how to use them.
Tested locally with tests/ssl-opt.sh -f '^Session resume'.
This allows to test PSK-based ciphersuites via ssl_server2 in builds
which have MBEDTLS_X509_CRT_PARSE_C enabled but both MBEDTLS_FS_IO and
MBEDTLS_CERTS_C disabled.
