It happens regularly in test runs that the server example application
shuts down a connection, goes into waiting mode for a new connection,
and then receives the encrypted ClosureAlert from the client. The only
reason why this does currently not trigger the 'record from another epoch'
message is that we handle ClientHello parsing outside of the main record
stack because we want to be able to detect SSLv2 ClientHellos. However,
this is likely to go away, and once it happens, we'll see the log message.
Further, when record checking is used, every record, including the mentioned
closure alert, is passed to the record checking API before being passed to
the rest of the stack, which leads to the log message being printed.
In summary, grepping for 'record from another epoch' is a fragile way
of checking whether a reordered message has arrived. A more reliable
way is to grep for 'Buffer record from epoch' which is printed when
a record from a future epoch is actually buffered, and 'ssl_buffer_message'
which is the function buffering a future handshake message.
This commit implements the record checking API
mbedtls_ssl_check_record()
on top of the restructured incoming record stack.
Specifically, it makes use of the fact that the core processing routines
ssl_parse_record_header()
mbedtls_ssl_decrypt_buf()
now operate on instances of the SSL record structure mbedtls_record
instead of the previous mbedtls_ssl_context::in_xxx fields.
After the rewrite of incoming record processing to use the internal
SSL record structure mbedtls_record (which contains the data_offset
field to indicate where the IV resides), this field is no longer
necessary.
Note: This is an API break.
The function mbedtls_ssl_in_hdr_len() is supposed to return the length
of the record header of the current incoming record. With the advent
of the DTLS Connection ID, this length is only known at runtime and
hence so far needed to be derived from the internal in_iv pointer
pointing to the beginning of the payload of the current incooing
record.
By now, however, those uses of mbedtls_ssl_in_hdr_len() where the
presence of a CID would need to be detected have been removed
(specifically, ssl_parse_record_header() doesn't use it anymore
when checking that the current datagram is large enough to hold
the record header, including the CID), and it's sufficient to
statically return the default record header sizes of 5 / 13 Bytes
for TLS / DTLS.
ssl_get_next_record() updates the legacy in_xxx fields in two places,
once before record decryption and once after. Now that record decryption
doesn't use or affect the in_xxx fields anymore, setting up the these
legacy fields can entirely be moved to the end of ssl_get_next_record(),
which is what this comit does.
This commit solely moves existing code, but doesn't yet simplify the
now partially redundant settings of the in_xxx fields. This will be
done in a separate commit.
Multiple record attributes such as content type and payload length
may change during record decryption, and the legacy in_xxx fields
in the SSL context therefore need to be updated after the record
decryption routine ssl_decrypt_buf() has been called.
After the previous commit has made ssl_prepare_record_content()
independent of the in_xxx fields, setting them can be moved
outside of ssl_prepare_record_content(), which is what this
commit does.
Previously, ssl_update_in_pointers() ensured that the in_xxx pointers
in the SSL context are set to their default state so that the record
header parsing function ssl_parse_record_header() could make use of them.
By now, the latter is independent of these pointers, so they don't need
to be setup before calling ssl_parse_record_header() anymore.
However, other parts of the messaging stack might still depend on it
(to be studied), and hence this commit does not yet reomve
ssl_update_in_pointers() entirely.
The stack maintains pointers mbedtls_ssl_context::in_xxx pointing to
various parts of the [D]TLS record header. Originally, these fields
were determined and set in ssl_parse_record_header(). By now,
ssl_parse_record_header() has been modularized to setup an instance
of the internal SSL record structure mbedtls_record, and to derive
the old in_xxx fields from that.
This commit takes a further step towards removing the in_xxx fields
by deriving them from the established record structure _outside_ of
ssl_parse_record_header() after the latter has succeeded.
One exception is the handling of possible client reconnects,
which happens in the case then ssl_parse_record_header() returns
MBEDTLS_ERR_SSL_UNEXPECTED_RECORD; since ssl_check_client_reconnect()
so far uses the in_xxx fields, they need to be derived from the
record structure beforehand.
This commit makes a first step towards modularizing the incoming record
processing by having it operate on instances of the structure mbedtls_record
representing SSL records.
So far, only record encryption/decryption operate in terms of record
instances, but the rest of the parsing doesn't. In particular,
ssl_parse_record_header() operates directly on the fixed input buffer,
setting the various ssl->in_xxx pointers and fields, and only directly
before/after calling ssl_decrypt_buf() these fields a converted to/from
mbedtls_record instances.
This commit does not yet remove the ssl->in_xxx fields, but makes a step
towards extending the lifetime of mbedtls_record structure representing
incoming records, by modifying ssl_parse_record_header() to setup an
instance of mbedtls_record, and setting the ssl->in_xxx fields from that
instance. The instance so-constructed isn't used further so far, and in
particular it is not yet consolidated with the instance set up for use
in ssl_decrypt_record(). That's for a later commit.
Previously, ssl_parse_record_header() did not check whether the current
datagram is large enough to hold a record of the advertised size. This
could lead to records being silently skipped over or backed up on the
basis of an invalid record length. Concretely, the following would happen:
1) In the case of a record from an old epoch, the record would be
'skipped over' by setting next_record_offset according to the advertised
but non-validated length, and only in the subsequent mbedtls_ssl_fetch_input()
it would be noticed in an assertion failure if the record length is too
large for the current incoming datagram.
While not critical, this is fragile, and also contrary to the intend
that MBEDTLS_ERR_SSL_INTERNAL_ERROR should never be trigger-able by
external input.
2) In the case of a future record being buffered, it might be that we
backup a record before we have validated its length, hence copying
parts of the input buffer that don't belong to the current record.
This is a bug, and it's by luck that it doesn't seem to have critical
consequences.
This commit fixes this by modifying ssl_parse_record_header() to check that
the current incoming datagram is large enough to hold a record of the
advertised length, returning MBEDTLS_ERR_SSL_INVALID_RECORD otherwise.
We don't send alerts on other instances of ill-formed records,
so why should we do it here? If we want to keep it, the alerts
should rather be sent ssl_get_next_record().
As explained in the previous commit, if mbedtls_ssl_fetch_input()
is called multiple times, all but the first call are equivalent to
bounds checks in the incoming datagram.
In DTLS, if mbedtls_ssl_fetch_input() is called multiple times without
resetting the input buffer in between, the non-initial calls are functionally
equivalent to mere bounds checks ensuring that the incoming datagram is
large enough to hold the requested data. In the interest of code-size
and modularity (removing a call to a non-const function which is logically
const in this instance), this commit replaces such a call to
mbedtls_ssl_fetch_input() by an explicit bounds check in
ssl_parse_record_header().
Previously, `ssl_handle_possible_reconnect()` was part of
`ssl_parse_record_header()`, which was required to return a non-zero error
code to indicate a record which should not be further processed because it
was invalid, unexpected, duplicate, .... In this case, some error codes
would lead to some actions to be taken, e.g. `MBEDTLS_ERR_SSL_EARLY_MESSAGE`
to potential buffering of the record, but eventually, the record would be
dropped regardless of the precise value of the error code. The error code
`MBEDTLS_ERR_SSL_HELLO_VERIFY_REQUIRED` returned from
`ssl_handle_possible_reconnect()` did not receive any special treatment and
lead to silent dopping of the record - in particular, it was never returned
to the user.
In the new logic this commit introduces, `ssl_handle_possible_reconnect()` is
part of `ssl_check_client_reconnect()` which is triggered _after_
`ssl_parse_record_header()` found an unexpected record, which is already in
the code-path eventually dropping the record; we want to leave this code-path
only if a valid cookie has been found and we want to reset, but do nothing
otherwise. That's why `ssl_handle_possible_reconnect()` now returns `0` unless
a valid cookie has been found or a fatal error occurred.
Availability of sufficient incoming data should be checked when
it is needed, which is in mbedtls_ssl_fetch_input(), and this
function has the necessary bounds checks in place.
mbedtls_ssl_decrypt_buf() asserts that the passed transform is not NULL,
but the function is only invoked in a single place, and this invocation
is clearly visible to be within a branch ensuring that the incoming
transform isn't NULL. Remove the assertion for the benefit of code-size.
The previous code performed architectural maximum record length checks
both before and after record decryption. Since MBEDTLS_SSL_IN_CONTENT_LEN
bounds the maximum length of the record plaintext, it suffices to check
only once after (potential) decryption.
This must not be confused with the internal check that the record
length is small enough to make the record fit into the internal input
buffer; this is done in mbedtls_ssl_fetch_input().
The check is in terms of the internal input buffer length and is
hence likely to be originally intended to protect against overflow
of the input buffer when fetching data from the underlying
transport in mbedtls_ssl_fetch_input(). For locality of reasoning,
it's better to perform such a check close to where it's needed,
and in fact, mbedtls_ssl_fetch_input() _does_ contain an equivalent
bounds check, too, rendering the bounds check in question redundant.
When looking for a parent, all candidates were considered time-invalid due to
the #ifdef incorrectly including the `parent_valid = 1` line.
When MBEDTLS_HAVE_TIME_DATE is unset the time-validity of certificates is
never checked and always treated as valid. This is usually achieved by proper
usage of mbedtls_x509_time_is_past() and mbedtls_x509_time_is_future() (and
their definition when we don't HAVE_TIME_DATE).
Here the calls to these functions needs to be guarded by
MBEDTLS_X509_CRT_REMOVE_TIME as they access struct members whose presence is
controlled by this option. But the "valid" branch should still always be taken.
(Note: MBEDTLS_X509_CRT_REMOVE_TIME being set forces MBEDTLS_HAVE_TIME_DATE to
be unset, as enforce by check_config.h.)
This bug was found by `all.sh test_baremetal` - no need for a new test.
This was found as a warning when running scripts/baremetal.sh --ram
--build-only manually, but it should have been found in a more automated way.
Adding -Werror so that future such issues will be caught by all.sh
(component_test_baremetal already invokes baremetal.sh --ram --build-only).
Asserting `*p == end` right after setting `end = *p + len` will always fail
unless `len == 0`, which is never the case with properly-formed certificates.
The function x509_skip_dates() is modelled after x509_get_dates() which between
setting `end` and comparing it to `*p` calls mbedtls_x509_get_time() which
advances `*p` to the expected value, which is why this test works in
get_dates().
Since `skip_dates()` has `skip`, not `validate` in its name, and the entire
point of `MBEDTLS_X509_CRT_REMOVE_TIME` is to save code, we don't want to
call the relatively large functions needed to properly parse (and validate)
dates before throwing the parsed dates away, we can just fast-forward to the
end of the sequence.
This makes updating `end` and comparing it to `*p` after the fast-forward
redundant, as the comparison will always be true (unlike the case where we
actually parse the contents of the sequence).
This bug was found by `all.sh test_baremetal` - no need for a new test.
