This uncovered a bug that led to a double-free (in practice, in general could
be free() on any invalid value): initially the session structure is loaded
with `memcpy()` which copies the previous values of pointers peer_cert and
ticket to heap-allocated buffers (or any other value if the input is
attacker-controlled). Now if we exit before we got a chance to replace those
invalid values with valid ones (for example because the input buffer is too
small, or because the second malloc() failed), then the next call to
session_free() is going to call free() on invalid pointers.
This bug is fixed in this commit by always setting the pointers to NULL right
after they've been read from the serialised state, so that the invalid values
can never be used.
(An alternative would be to NULL-ify them when writing, which was rejected
mostly because we need to do it when reading anyway (as the consequences of
free(invalid) are too severe to take any risk), so doing it when writing as
well is redundant and a waste of code size.)
Also, while thinking about what happens in case of errors, it became apparent
to me that it was bad practice to leave the session structure in an
half-initialised state and rely on the caller to call session_free(), so this
commit also ensures we always clear the structure when loading failed.
This test appeared to be passing for the wrong reason, it's actually not
appropriate for the current implementation. The serialised data contains
values of pointers to heap-allocated buffers. There is no reason these should
be identical after a load-save pair. They just happened to be identical when I
first ran the test due to the place of session_free() in the test code and the
fact that the libc's malloc() reused the same buffers. The test no longer
passes if other malloc() implementations are used (for example, when compiling
with asan which avoids re-using the buffer, probably for better error
detection).
So, disable this test for now (we can re-enable it when we changed how
sessions are serialised, which will be done in a future PR, hence the name of
the dummy macro in depends_on). In the next commit we're going to add a test
that save-load is the identity instead - which will be more work in testing as
it will require checking each field manually, but at least is reliable.
This initial test ensures that a load-save function is the identity. It is so
far incomplete in that it only tests sessions without tickets or certificate.
This will be improved in the next commits.
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'.
On client side, this is required for the main use case where of serialising a
session for later resumption, in case tickets are used.
On server side, this doesn't change much as ticket_len will always be 0.
This unblocks testing the functions by using them in ssl_client2, which will
be done in the next commit.
This finishes making these functions public. Next step is to get them tested,
but there's currently a blocker for that, see next commit (and the commit
after it for tests).
The next commit with make the implementation publicly available as well.
For now the API is kept unchanged. The save function API has a serious drawback in that the user
must guess what an appropriate buffer size is.
Internally so far this didn't matter because we were only using that API for
ticket creation, and tickets are written to the SSL output buffer whose size
is fixed anyway, but for external users this might not be suitable. Improving
that is left for later.
Also, so far the functions are defined unconditionally. Whether we want to
re-use existing flags or introduce a new one is left for later.
Finally, currently suggested usage of calling get_session() then
session_save() is memory-inefficient in that get_session() already makes a
copy. I don't want to recommend accessing `ssl->session` directly as we want
to prohibit direct access to struct member in the future. Providing a clean
and efficient way is also left to a later commit.
While 'session hash' is currently unique, so suitable to prove that the
intended code path has been taken, it's a generic enough phrase that in the
future we might add other debug messages containing it in completely unrelated
code paths. In order to future-proof the accuracy of the test, let's use a
more specific string.
The previous comment used "TLS" as a shortcut for "TLS 1.0/1.1" which was
confusing. This partially reflected the names of the calc_verify/finished that
go ssl, tls (for 1.0/1.1) tls_shaxxx (for 1.2), but still it's clearer to be
explicit in the comment - and perhaps in the long term the function names
could be clarified instead.
When using this function to deserialize, it's not a problem to have a session
structure as input as we'll have one around anyway (most probably freshly
deserialised).
However for tests it's convenient to be able to build a transform without
having a session structure around.
Also, removing this structure from parameters makes the function signature
more uniform, the only exception left being the ssl param at the end that's
hard to avoid for now.
Configs with no DEBUG_C are used for example in test-ref-configs.pl, which also
runs parts of compat.sh or ssl-opt.sh on them, so the added 'ssl = NULL'
statements will be exercised in those tests at least.
Make it more explicit what's used. Unfortunately, we still need ssl as a
parameter for debugging, and because calc_verify wants it as a parameter (for
all TLS versions except SSL3 it would actually only need handshake, but SSL3
also accesses session_negotiate).
It's also because of calc_verify that we can't make it const yet, but see next
commit.
* origin/pr/2790: (40 commits)
Fix possibly-lossy conversion warning from MSVC
Reintroduce length 0 check for records
Don't use memcpy() for 2-byte copy operation
Remove integer parsing macro
Fix alignment in record header parsing routine
Don't disallow 'record from another epoch' log msg in proxy ref test
Make sure 'record from another epoch' is displayed for next epoch
Implement record checking API
Mark ssl_parse_record_header() as `const` in SSL context
Make mbedtls_ssl_in_hdr_len() CID-unaware
Remove duplicate setting of ssl->in_msgtype and ssl->in_msglen
Move update of in_xxx fields in ssl_get_next_record()
Move update of in_xxx fields outside of ssl_prepare_record_content()
Reduce dependency of ssl_prepare_record_content() on in_xxx fields
Move ssl_update_in_pointers() to after record hdr parsing
Mark DTLS replay check as `const` on the SSL context
Move updating the internal rec ptrs to outside of rec hdr parsing
Mark ssl_decrypt_buf() as `const in the input SSL context
Adapt ssl_prepare_record_content() to use SSL record structure
Use record length from record structure when fetching content in TLS
...
* origin/pr/2781:
Documentation fixes according to review
Remove unused label in ssl_client2/ssl_server2
Add missing word in documentation of mbedtls_ssl_check_record()
cli/srv ex: Add dbg msg if record checking gives inconsistent result
Fix minor issues in documentation of mbedtls_ssl_check_record()
State that record checking is DTLS only and doesn't check content type
Update version_features.c
Pass dgrams to mbedtls_ssl_check_record in ssl_client2/server2
Add IO wrappers to ssl_server2 as interm's between NET and SSL layer
Add IO wrappers to ssl_client2 as interm's between NET and SSL layer
Introduce configuration option and API for SSL record checking
* origin/pr/2767:
Rename local variables
Update submodule
Update Visual studio project file
Move the examples to PSA 1.0
Use psa_raw_key_agreement
Remove calls to psa_allocate_key
Make variable naming consistent
Update psa_create_key to PSA 1.0
Update psa_import_key to PSA 1.0
Update psa_generator_abort to PSA 1.0
Update psa_generator_read to PSA 1.0
Update psa_crypto_generator_t to PSA 1.0
Update psa_key_agreement to PSA 1.0
Update GENERATOR_INIT macro to PSA 1.0
Update KEYPAIR macros to PSA 1.0
