Conflict resolution:
* ChangeLog: put the new entries in their rightful place.
* library/x509write_crt.c: the change in development was whitespace
only, so use the one from the iotssl-1251 feature branch.
This commit adds some explicit downcasts from `size_t` to `uint8_t` in
the RSASSA signature encoding function `rsa_rsassa_pkcs1_v15_encode`.
The truncation is safe as it has been checked beforehand that the
respective values are in the range of a `uint8_t`.
1) `mbedtls_rsa_import_raw` used an uninitialized return
value when it was called without any input parameters.
While not sensible, this is allowed and should be a
succeeding no-op.
2) The MPI test for prime generation missed a return value
check for a call to `mbedtls_mpi_shift_r`. This is neither
critical nor new but should be fixed.
3) Both the RSA keygeneration example program and the
RSA test suites contained code initializing an RSA context
after a potentially failing call to CTR DRBG initialization,
leaving the corresponding RSA context free call in the
cleanup section of the respective function orphaned.
While this defect existed before, Coverity picked up on
it again because of newly introduced MPI's that were
also wrongly initialized only after the call to CTR DRBG
init. The commit fixes both the old and the new issue
by moving the initializtion of both the RSA context and
all MPI's prior to the first potentially failing call.
The function `mbedtls_rsa_complete` is supposed to guarantee that
RSA operations will complete without failure. In contrast, it does
not ensure consistency of parameters, which is the task of the
checking functions `rsa_check_pubkey` and `rsa_check_privkey`.
Previously, the maximum allowed size of the RSA modulus was checked
in `mbedtls_rsa_check_pubkey`. However, exceeding this size would lead
to failure of some RSA operations, hence this check belongs to
`mbedtls_rsa_complete` rather than `mbedtls_rsa_check_pubkey`.
This commit moves it accordingly.
The function `pk_get_rsapubkey` originally performed some basic
sanity checks (e.g. on the size of public exponent) on the parsed
RSA public key by a call to `mbedtls_rsa_check_pubkey`.
This check was dropped because it is not possible to thoroughly
check full parameter sanity (i.e. that (-)^E is a bijection on Z/NZ).
Still, for the sake of not silently changing existing behavior,
this commit puts back the call to `mbedtls_rsa_check_pubkey`.
- Adapt the change in all.sh to the new keep-going mode
- Restore alphabetical order of configuration flags for
alternative implementations in config.h and rebuild
library/version_features.c
`mbedtls_rsa_deduce_primes` implicitly casts the result of a call to
`mbedtls_mpi_lsb` to a `uint16_t`. This is safe because of the size
of MPI's used in the library, but still may have compilers complain
about it. This commit makes the cast explicit.
Conflict resolution: additions in the same places as
upstream-public/pr/865, both adding into lexicographically sorted
lists, resolved by taking the additions in lexicographic order.
* development:
Timing self test: shorten redundant tests
Timing self test: increased duration
Timing self test: increased tolerance
Timing unit tests: more protection against infinite loops
Unit test for mbedtls_timing_hardclock
New timing unit tests
selftest: allow excluding a subset of the tests
selftest: allow running a subset of the tests
selftest: refactor to separate the list of tests from the logic
Timing self test: print some diagnosis information
mbedtls_timing_get_timer: don't use uninitialized memory
timing interface documentation: minor clarifications
Timing: fix mbedtls_set_alarm(0) on Unix/POSIX
* public/pr/1136:
Timing self test: shorten redundant tests
Timing self test: increased duration
Timing self test: increased tolerance
Timing unit tests: more protection against infinite loops
Unit test for mbedtls_timing_hardclock
New timing unit tests
selftest: allow excluding a subset of the tests
selftest: allow running a subset of the tests
selftest: refactor to separate the list of tests from the logic
Timing self test: print some diagnosis information
mbedtls_timing_get_timer: don't use uninitialized memory
timing interface documentation: minor clarifications
Timing: fix mbedtls_set_alarm(0) on Unix/POSIX
1. Surround the generate keys with
`#if ! defined(MBEDTLS_CMAC_ALT) || defined(MBEDTLS_SELF_TEST)`
to resolve build issue when `MBEDTLS_SELF_TEST` is defined for
alternative CMAC as well
2. Update ChangeLog
Increase the duration of the self test, otherwise it tends to fail on
a busy machine even with the recently upped tolerance. But run the
loop only once, it's enough for a simple smoke test.
mbedtls_timing_self_test fails annoyingly often when running on a busy
machine such as can be expected of a continous integration system.
Increase the tolerances in the delay test, to reduce the chance of
failures that are only due to missing a deadline on a busy machine.
Print some not-very-nice-looking but helpful diagnosis information if
the timing selftest fails. Since the failures tend to be due to heavy
system load that's hard to reproduce, this information is necessary to
understand what's going on.
mbedtls_timing_get_timer with reset=1 is called both to initialize a
timer object and to reset an already-initialized object. In an
initial call, the content of the data structure is indeterminate, so
the code should not read from it. This could crash if signed overflows
trap, for example.
As a consequence, on reset, we can't return the previously elapsed
time as was previously done on Windows. Return 0 as was done on Unix.
The POSIX/Unix implementation of mbedtls_set_alarm did not set the
mbedtls_timing_alarmed flag when called with 0, which was inconsistent
with what the documentation implied and with the Windows behavior.
* restricted/pr/403:
Correct record header size in case of TLS
Don't allocate space for DTLS header if DTLS is disabled
Improve debugging output
Adapt ChangeLog
Add run-time check for handshake message size in ssl_write_record
Add run-time check for record content size in ssl_encrypt_buf
Add compile-time checks for size of record content and payload
* development:
Don't split error code description across multiple lines
Register new error code in error.h
Move deprecation to separate section in ChangeLog
Extend scope of ERR_RSA_UNSUPPORTED_OPERATION error code
Adapt RSA test suite
Adapt ChangeLog
Deprecate usage of RSA primitives with wrong key type
* restricted/pr/397:
Don't split error code description across multiple lines
Register new error code in error.h
Move deprecation to separate section in ChangeLog
Extend scope of ERR_RSA_UNSUPPORTED_OPERATION error code
Adapt RSA test suite
Adapt ChangeLog
Deprecate usage of RSA primitives with wrong key type
In a previous PR (Fix heap corruption in implementation of truncated HMAC
extension #425) the place where MAC is computed was changed from the end of
the SSL I/O buffer to a local buffer (then (part of) the content of the local
buffer is either copied to the output buffer of compare to the input buffer).
Unfortunately, this change was made only for TLS 1.0 and later, leaving SSL
3.0 in an inconsistent state due to ssl_mac() still writing to the old,
hard-coded location, which, for MAC verification, resulted in later comparing
the end of the input buffer (containing the computed MAC) to the local buffer
(uninitialised), most likely resulting in MAC verification failure, hence no
interop (even with ourselves).
This commit completes the move to using a local buffer by using this strategy
for SSL 3.0 too. Fortunately ssl_mac() was static so it's not a problem to
change its signature.
Fix missing definition of mbedtls_zeroize when MBEDTLS_FS_IO is
disabled in the configuration.
Introduced by e7707228b4
Merge remote-tracking branch 'upstream-public/pr/1062' into development
In case truncated HMAC must be used but the Mbed TLS peer hasn't been updated
yet, one can use the compile-time option MBEDTLS_SSL_TRUNCATED_HMAC_COMPAT to
temporarily fall back to the old, non-compliant implementation of the truncated
HMAC extension.
The truncated HMAC extension as described in
https://tools.ietf.org/html/rfc6066.html#section-7 specifies that when truncated
HMAC is used, only the HMAC output should be truncated, while the HMAC key
generation stays unmodified. This commit fixes Mbed TLS's behavior of also
truncating the key, potentially leading to compatibility issues with peers
running other stacks than Mbed TLS.
Details:
The keys for the MAC are pieces of the keyblock that's generated from the
master secret in `mbedtls_ssl_derive_keys` through the PRF, their size being
specified as the size of the digest used for the MAC, regardless of whether
truncated HMAC is enabled or not.
/----- MD size ------\ /------- MD size ----\
Keyblock +----------------------+----------------------+------------------+---
now | MAC enc key | MAC dec key | Enc key | ...
(correct) +----------------------+----------------------+------------------+---
In the previous code, when truncated HMAC was enabled, the HMAC keys
were truncated to 10 bytes:
/-10 bytes-\ /-10 bytes-\
Keyblock +-------------+-------------+------------------+---
previously | MAC enc key | MAC dec key | Enc key | ...
(wrong) +-------------+-------------+------------------+---
The reason for this was that a single variable `transform->maclen` was used for
both the keysize and the size of the final MAC, and its value was reduced from
the MD size to 10 bytes in case truncated HMAC was negotiated.
This commit fixes this by introducing a temporary variable `mac_key_len` which
permanently holds the MD size irrespective of the presence of truncated HMAC,
and using this temporary to obtain the MAC key chunks from the keyblock.
Previously, MAC validation for an incoming record proceeded as follows:
1) Make a copy of the MAC contained in the record;
2) Compute the expected MAC in place, overwriting the presented one;
3) Compare both.
This resulted in a record buffer overflow if truncated MAC was used, as in this
case the record buffer only reserved 10 bytes for the MAC, but the MAC
computation routine in 2) always wrote a full digest.
For specially crafted records, this could be used to perform a controlled write of
up to 6 bytes past the boundary of the heap buffer holding the record, thereby
corrupting the heap structures and potentially leading to a crash or remote code
execution.
This commit fixes this by making the following change:
1) Compute the expected MAC in a temporary buffer that has the size of the
underlying message digest.
2) Compare to this to the MAC contained in the record, potentially
restricting to the first 10 bytes if truncated HMAC is used.
A similar fix is applied to the encryption routine `ssl_encrypt_buf`.
* development: (30 commits)
update README file (#1144)
Fix typo in asn1.h
Improve leap year test names in x509parse.data
Correctly handle leap year in x509_date_is_valid()
Renegotiation: Add tests for SigAlg ext parsing
Parse Signature Algorithm ext when renegotiating
Minor style fix
config.pl get: be better behaved
config.pl get: don't rewrite config.h; detect write errors
Fixed "config.pl get" for options with no value
Fix typo and bracketing in macro args
Ensure failed test_suite output is sent to stdout
Remove use of GNU sed features from ssl-opt.sh
Fix typos in ssl-opt.sh comments
Add ssl-opt.sh test to check gmt_unix_time is good
Extend ssl-opt.h so that run_test takes function
Always print gmt_unix_time in TLS client
Restored note about using minimum functionality in makefiles
Note in README that GNU make is required
Fix changelog for ssl_server2.c usage fix
...
Fix the x509_get_subject_alt_name() function to not accept invalid
tags. The problem was that the ASN.1 class for tags consists of two
bits. Simply doing bit-wise and of the CONTEXT_SPECIFIC macro with the
input tag has the potential of accepting tag values 0x10 (private)
which would indicate that the certificate has an incorrect format.
This commit fixes a comparison of ssl_session->encrypt_then_mac against the
ETM-unrelated constant MBEDTLS_SSL_EXTENDED_MS_DISABLED. Instead,
MBEDTLS_SSL_ETM_DISABLED should be used.
The typo is has no functional effect since both constants have the same value 0.
Remove a check introduced in the previous buffer overflow fix with keys of
size 8N+1 which the subsequent fix for buffer start calculations made
redundant.
Added a changelog entry for the buffer start calculation fix.
For a key of size 8N+1, check that the first byte after applying the
public key operation is 0 (it could have been 1 instead). The code was
incorrectly doing a no-op check instead, which led to invalid
signatures being accepted. Not a security flaw, since you would need the
private key to craft such an invalid signature, but a bug nonetheless.
The check introduced by the previous security fix was off by one. It
fixed the buffer overflow but was not compliant with the definition of
PSS which technically led to accepting some invalid signatures (but
not signatures made without the private key).
I don't think this can cause a crash as the member accessed is in the
beginning of the context, so wouldn't be outside of valid memory if the actual
context was RSA.
Also, the mismatch will be caught later when checking signature, so the cert
chain will be rejected anyway.
Fix buffer overflow in RSA-PSS signature verification when the hash is
too large for the key size. Found by Seth Terashima, Qualcomm.
Added a non-regression test and a positive test with the smallest
permitted key size for a SHA-512 hash.
The function mbedtls_ecp_gen_keypair_base did not wipe the stack buffer used to
hold the private exponent before returning. This commit fixes this by not using
a stack buffer in the first place but instead calling mpi_fill_random directly
to acquire the necessary random MPI.
This commit modifies mpi_read_binary to always allocate the minimum number of
limbs required to hold the entire buffer provided to the function, regardless of
its content. Previously, leading zero bytes in the input data were detected and
used to reduce memory footprint and time, but this non-constant behavior turned
out to be non-tolerable for the cryptographic applications this function is used
for.
Previously, if `MBEDTLS_SSL_RENEGOTIATION` was disabled, incoming handshake
messages in `mbedtls_ssl_read` (expecting application data) lead to the
connection being closed. This commit fixes this, restricting the
`MBEDTLS_SSL_RENEGOTIATION`-guard to the code-paths responsible for accepting
renegotiation requests and aborting renegotiation attempts after too many
unexpected records have been received.
1) use `pk_get_rsapubkey` instead of reimplementing the parsing
2) rename the key files, according to their type and key size
3) comment in the data_files/Makefile hoe the keys were generated
4) Fix issue of failure parsing pkcs#1 DER format parsing, missed in previous commit
Signature algorithm extension was skipped when renegotiation was in
progress, causing the signature algorithm not to be known when
renegotiating, and failing the handshake. Fix removes the renegotiation
step check before parsing the extension.
As the optional RSA parameters DP, DQ and QP are effectively discarded (they are only considered for their length to
ensure that the key fills the entire buffer), it is not necessary to read them into separate MPI's.
The number of loop iterations per candidate in `mbedtls_deduce_primes` was off
by one. This commit corrects this and removes a toy non-example from the RSA
test suite, as it seems difficult to have the function fail on small values of N
even if D,E are corrupted.
Signature algorithm extension was skipped when renegotiation was in
progress, causing the signature algorithm not to be known when
renegotiating, and failing the handshake. Fix removes the renegotiation
step check before parsing the extension.
This commit splits off the RSA helper functions into separate headers and
compilation units to have a clearer separation of the public RSA interface,
intended to be used by end-users, and the helper functions which are publicly
provided only for the benefit of designers of alternative RSA implementations.
1) move the change into Features from Changes, in the changLog
2) Change the feature alternative configuration MBEDTLS_ECDH_ALT
definition to function alternative defintions
MBEDTLS_ECDH_COMPUTE_SHARED_ALT and MBEDTLS_ECDH_GEN_PUBLIC_ALT
1) update ChangLog to have new feature in Features instead of Changes
2) Change MBEDTLS_ECDSA_ALT to function specific alternative definitions:
MBEDTLS_ECDSA_SIGN_ALT, MBEDTLS_ECDSA_VERIFY_ALT and MBEDTLS_ECDSA_GENKEY_ALT
It is not necessary to pass a CSPRNG to `mbedtls_rsa_deduce_moduli`, as there
exist well-working static strategies, and even if a PRNG is preferred, a
non-secure one would be sufficient.
Further, the implementation is changed to use a static strategy for the choice
of candidates which according to some benchmarks even performs better than the
previous one using random candidate choices.
This commit reconciles the code path responsible for resending the
final DTLS handshake flight with the path for handling resending of
the other flights.
This commit restricts WANT_READ to indicate that no data is available on the
underlying transport. To signal the need for further processing - which was
previously also handled through this error code - a new internal error code
MBEDTLS_ERR_SSL_CONTINUE_PROCESSING is introduced.
DTLS records from previous epochs were incorrectly checked against the
current epoch transform's minimal content length, leading to the
rejection of entire datagrams. This commit fixed that and adapts two
test cases accordingly.
Internal reference: IOTSSL-1417
- Enhances the documentation of mbedtls_ssl_get_bytes_avail (return
the number of bytes left in the current application data record, if
there is any).
- Introduces a new public function mbedtls_ssl_check_pending for
checking whether any data in the internal buffers still needs to be
processed. This is necessary for users implementing event-driven IO
to decide when they can safely idle until they receive further
events from the underlying transport.
Give a note on the debugging output on the following occasions:
(1) The timer expires in mbedtls_ssl_fetch_input
(2) There's more than one records within a single datagram
Change ssl_parse_server_hello() so that the parsed first four random
bytes from the ServerHello message are printed by the TLS client as
a Unix timestamp regardless of whether MBEDTLS_DEBUG_C is defined. The
debug message will only be printed if debug_level is 3 or higher.
Unconditionally enabling the debug print enabled testing of this value.
Change ssl_parse_server_hello() so that the parsed first four random
bytes from the ServerHello message are printed by the TLS client as
a Unix timestamp regardless of whether MBEDTLS_DEBUG_C is defined. The
debug message will only be printed if debug_level is 3 or higher.
Unconditionally enabling the debug print enabled testing of this value.
Further, state explicitly that wrong key types need not be supported by alternative RSA implementations, and that those
may instead return the newly introduced error code MBEDTLS_ERR_RSA_UNSUPPORTED_OPERATION.
This commit returns to using constant macros instead of global variables for the DHM group constants. Further, macros
providing the binary encoding of the primes from RFC 3526 and RFC 7919 are added. The hex-string macros are deprecated.
This commit modifies the PKCS1 v1.5 signature verification function `mbedtls_rsa_rsassa_pkcs1_v15_verify` to prepare the
expected PKCS1-v1.5-encoded hash using the function also used by the signing routine `mbedtls_rsa_rsassa_pkcs1_v15_sign`
and comparing it to the provided byte-string afterwards. This comes at the benefits of (1) avoiding any error-prone
parsing, (2) removing the dependency of the RSA module on the ASN.1 parsing module, and (3) reducing code size.
This commit moves the code preparing PKCS1 v1.5 encoded hashes from `mbedtls_rsa_rsassa_pkcs1_v15_sign` to a separate
non-public function `rsa_rsassa_pkcs1_v15_encode`. This code-path will then be re-used by the signature verification function
`mbetls_rsa_rsassa_pkcs1_v15_verify` in a later commit.
Original intention was to be allowed to perform in-place operations like changing the byte-order before importing
parameters into an HSM. Now a copy is needed in this case, but there's no more danger of a user expecting the arguments
to be left untouched.
State explicitly that `pk_parse_pkcs8_undencrypted_der` and `pk_parse_key_pkcs8_encrypted_der` are not responsible for
zeroizing and freeing the provided key buffer.
This commit changes the implementation of `mbedtls_rsa_get_len` to return
`ctx->len` instead of always re-computing the modulus' byte-size via
`mbedtls_mpi_size`.
Although the variable ret was initialised to an error, the
MBEDTLS_MPI_CHK macro was overwriting it. Therefore it ended up being
0 whenewer the bignum computation was successfull and stayed 0
independently of the actual check.
This commit renames the test-only flag MBEDTLS_ENTROPY_HAVE_STRONG to ENTROPY_HAVE_STRONG to make it more transparent
that it's an internal flag, and also to content the testscript tests/scripts/check-names.pl which previously complained
about the macro occurring in a comment in `entropy.c` without being defined in a library file.
This commit removes extension-writing code for X.509 non-v3 certificates from
mbedtls_x509write_crt_der. Previously, even if no extensions were present an
empty sequence would have been added.
Fix compilation error on Mingw32 when `_TRUNCATE` is defined. Use
`_TRUNCATE` only if `__MINGW32__` not defined. Fix suggested by
Thomas Glanzmann and Nick Wilson on issue #355
* mbedtls-2.6: (27 commits)
Update version number to 2.6.0
Fix language in Changelog for clarity
Improve documentation of PKCS1 decryption functions
Fix style and missing item in ChangeLog
Add credit to Changelog to fix for #666
Fix naked call to time() with platform call
Fix ChangeLog for duplication after merge
Rename time and index parameter to avoid name conflict.
Correct comment
Adapt ChangeLog
Reliably zeroize sensitive data in AES sample application
Reliably zeroize sensitive data in Crypt-and-Hash sample application
Fix potential integer overflow parsing DER CRT
Fix potential integer overflow parsing DER CRL
Move the git scripts to correct path
Update after @sbutcher-arm comments
Fix slash direction for linux path
Add note for the git_hoos README file
Pre push hook script
Check return code of mbedtls_mpi_fill_random
...
The stack buffer used to hold the decrypted key in pk_parse_pkcs8_encrypted_der
was statically sized to 2048 bytes, which is not enough for DER encoded 4096bit
RSA keys.
This commit resolves the problem by performing the key-decryption in-place,
circumventing the introduction of another stack or heap copy of the key.
There are two situations where pk_parse_pkcs8_encrypted_der is invoked:
1. When processing a PEM-encoded encrypted key in mbedtls_pk_parse_key.
This does not need adaption since the PEM context used to hold the decoded
key is already constructed and owned by mbedtls_pk_parse_key.
2. When processing a DER-encoded encrypted key in mbedtls_pk_parse_key.
In this case, mbedtls_pk_parse_key calls pk_parse_pkcs8_encrypted_der with
the buffer provided by the user, which is declared const. The commit
therefore adds a small code paths making a copy of the keybuffer before
calling pk_parse_pkcs8_encrypted_der.
If CRT is not used, the helper fields CRT are not assumed to be present in the
RSA context structure, so do the verification directly in this case. If CRT is
used, verification could be done using CRT, but we're sticking to ordinary
verification for uniformity.
This commit adds the function mbedtls_rsa_validate_crt for validating a set of CRT parameters. The function
mbedtls_rsa_check_crt is simplified accordingly.
If rsm != NULL then the curve type has to be Short Weierstrass, as we don't
implement restartable Montgomery now. If and when we do, then it's better to
check for the subcontext only, and not for the curve type.
Exactly one of three ways will be used, so make that clear by using an
if 1 else if 2 else 3 structure.
While at it, don't initialize variables at declaration, just to make extra
sure they're properly initialized afterwards in all code paths.
As done by previous commits for ECC and ECDSA:
- use explicit state assignments rather than increment
- always place the state update right before the operation label
This will make it easier to add restart support for other operations later if
desired.
SSL-specific changes:
- remove useless states: when the last restartable operation on a message is
complete, ssl->state is incremented already, so we don't need any additional
state update: ecrs_state is only meant to complement ssl->state
- rename remaining states consistently as <message>_<operation>
- move some labels closer to the actual operation when possible (no assignment
to variables used after the label between its previous and current position)
Systematically assign state just before the next operation that may return,
rather that just after the previous one. This makes things more local. (For
example, previously precompute_comb() has to handle a state reset for
mul_comb_core(), a kind of coupling that's best avoided.)
Note that this change doesn't move the location of state updates relative
to any potential return point, which is all that matters.
Incrementing the state is error-prone as we can end up doing it too many times
(loops) or not enough (skipped branches), or just make programming mistakes
(eg. the state was incremented twice at the end, so it ended up with a value
not in the enum...)
This is the first step of the rework, the next one will rationalize where the
state assignments are done.
Primality testing is guarded by the configuration flag MBEDTLS_GENPRIME and used in the new RSA helper functions. This
commit adds a corresponding preprocessor directive.
The call would anyway check for pointer equality and return early, but it
doesn't hurt to save a function call, and also this follows more uniformly the
pattern that those two lines go together:
#if defined(MBEDTLS_ECP_RESTARTBLE)
if( rs_ctx != NULL && ...
Alternative RSA implementations can be provided by defining MBEDTLS_RSA_ALT in
config.h, defining an mbedtls_rsa_context struct in a new file rsa_alt.h and
re-implementing the RSA interface specified in rsa.h.
Through the previous reworkings, the adherence to the interface is the only
implementation obligation - in particular, implementors are free to use a
different layout for the RSA context structure.
Child was almost redundant as it's already saved in ver_chain, except it was
multiplexed to also indicate whether an operation is in progress. This commit
removes it and introduces an explicit state variable instead.
This state can be useful later if we start returning IN_PROGRESS at other
points than find_parent() (for example when checking CRL).
Note that the state goes none -> find_parent and stays there until the context
is free(), as it's only on the first call that nothing was in progress.
Some parts were already implicitly using this as the two ifdefs were nested,
and some others didn't, which resulted in compile errors in some configs. This
fixes those errors and saves a bit of code+RAM that was previously wasted when
ECP_RESTARTABLE was defined but ECDSA_C wasn't
Previously we kept the ecdsa context created by the PK layer for ECDSA
operations on ECKEY in the ecdsa_restart_ctx structure, which was wrong, and
caused by the fact that we didn't have a proper handling of restart
sub-contexts in the PK layer.
The fact that you needed to pass a pointer to mbedtls_ecdsa_restart_ctx (or
that you needed to know the key type of the PK context) was a breach of
abstraction.
Change the API (and callers) now, and the implementation will be changed in
the next commit.
Goals include:
- reducing the number of local variables in the main function (so that we
don't have to worry about saving/restoring them)
- reducing the number exit points in the main function, making it easier to
update ssl->state only right before we return
- more consistent naming with ecrs prefix for everything
- always check it enabled before touching the rest
- rm duplicated code in parse_server_hello()
For selection of test cases, see comments added in the commit.
It makes the most sense to test with chains using ECC only, so for the chain
of length 2 we use server10 -> int-ca3 -> int-ca2 and trust int-ca2 directly.
Note: server10.crt was created by copying server10_int3_int-ca2.crt and
manually truncating it to remove the intermediates. That base can now be used
to create derived certs (without or with a chain) in a programmatic way.
This is mainly for the benefit of SSL modules, which only supports restart in
a limited number of cases. In the other cases (ECDHE_PSK) it would currently
return ERR_ECP_IN_PROGRESS and the user would thus call ssl_handshake() again,
but the SSL code wouldn't handle state properly and things would go wrong in
possibly unexpected ways. This is undesirable, so it should be possible for
the SSL module to choose if ECDHE should behave the old or the new way.
Not that it also brings ECDHE more in line with the other modules which
already have that choice available (by passing a NULL or valid restart
context).
For RSA, we could either have the function return an error code like
NOT_IMPLEMENTED or just run while disregarding ecp_max_ops. IMO the second
option makes more sense, as otherwise the caller would need to check whether
the key is EC or RSA before deciding to call either sign() or
sign_restartable(), and having to do this kind of check feels contrary to the
goal of the PK layer.
Two different changes:
- the first one will allow us to store k in the restart context while
restarting the following ecp_mul() operation
- the second one is an simplification, unrelated to restartability, made
possible by the fact that ecp_gen_privkey() is now public
(Unrelated to restartable work, just noticed while staring at the code.)
Checking at the end is inefficient as we might give up when we just generated
a valid signature or key.
Otherwise code that uses these functions in other modules will have to do:
#if defined(MBEDTLS_ECP_RESTARTABLE)
ret = do_stuff( there, may, be, many, args );
#else
ret = do_stuff( their, may, be, namy, args, rs_ctx );
#fi
and there is a risk that the arg list will differ when code is updated, and
this might not be caught immediately by tests because this depends on a
config.h compile-time option which are harder to test.
Always declaring the restartable variants of the API functions avoids this
problem; the cost in ROM size should be negligible.
This will be useful for restartable ECDH and ECDSA. Currently they call
mbedtls_ecp_gen_keypair(); one could make that one restartable, but that means
adding its own sub-context, while ECDH and ECDSA (will) have their own
contexts already, so switching to this saves one extra context.
This should only be done in the top-level function.
Also, we need to know if we indeed are the top-level function or not: for
example, when mbedtls_ecp_muladd() calls mbedtls_ecp_mul(), the later should
not reset ops_done. This is handled by the "depth" parameter in the restart
context.
When a restartable function calls another restartable function, the current
ops_count needs to be shared to avoid either doing too many operations or
returning IN_PROGRESS uselessly. So it needs to be in the top-level context
rather than a specific sub-context.
This was intended to detect aborted operations, but now that case is handled
by the caller freeing the restart context.
Also, as the internal sub-context is managed by the callee, no need for the
caller to free/reset the restart context between successful calls.
Following discussion in the team, it was deemed preferable for the restart
context to be explicitly managed by the caller.
This commits in the first in a series moving in that directly: it starts by
only changing the public API, while still internally using the old design.
Future commits in that series will change to the new design internally.
The test function was simplified as it no longer makes sense to test for some
memory management errors since that responsibility shifted to the caller.
It's going to be convenient for each function that can generate a
MBEDTLS_ERR_ECP_IN_PROGRESS on its own (as opposed to just passing it around)
to have its own restart context that they can allocate and free as needed
independently of the restart context of other functions.
For example ecp_muladd() is going to have its own restart_muladd context that
in can managed, then when it calls ecp_mul() this will manage a restart_mul
context without interfering with the caller's context.
So, things need to be renames to avoid future name clashes.
From a user's perspective, you want a "basic operation" to take approximately
the same amount of time regardless of the curve size, especially since max_ops
is a global setting: otherwise if you pick a limit suitable for P-384 then
when you do an operation on P-256 it will return way more often than needed.
Said otherwise, a user is actually interested in actual running time, and we
do the API in terms of "basic ops" for practical reasons (no timers) but then
we should make sure it's a good proxy for running time.
Ok, so the original plan was to make mpi_inv_mod() the smallest block that
could not be divided. Updated plan is that the smallest block will be either:
- ecp_normalize_jac_many() (one mpi_inv_mod() + a number or mpi_mul_mpi()s)
- or the second loop in ecp_precompute_comb()
With default settings, the minimum non-restartable sequence is:
- for P-256: 222M
- for P-384: 341M
This is within a 2-3x factor of originally planned value of 120M. However,
that value can be approached, at the cost of some performance, by setting
ECP_WINDOW_SIZE (w below) lower than the default of 6. For example:
- w=4 -> 166M for any curve (perf. impact < 10%)
- w=2 -> 130M for any curve (perf. impact ~ 30%)
My opinion is that the current state with w=4 is a good compromise, and the
code complexity need to attain 120M is not warranted by the 1.4 factor between
that and the current minimum with w=4 (which is close to optimal perf).
This is the easy part: with the current steps, all information between steps
is passed via T which is already saved. Next we'll need to split at least the
first loop, and maybe calls to normalize_jac_many() and/or the second loop.
Separating main computation from filling of the auxiliary array makes things
clearer and easier to restart as we don't have to remember the in-progress
auxiliary array.
Previously there were only two states:
- T unallocated
- T allocated and valid
Now there are three:
- T unallocated
- T allocated and in progress
- T allocated and valid
Introduce new bool T_ok to distinguish the last two states.
Free it as soon as it's no longer needed, but as a backup free it in
ecp_group_free(), in case ecp_mul() is not called again after returning
ECP_IN_PROGRESS.
So far we only remember it when it's fully computed, next step is to be able
to compute it in multiple steps.
In case of argument change, freeing everything is not the most efficient
(wastes one free()+calloc()) but makes the code simpler, which is probably
more important here
We'll need to store MPIs and other things that allocate memory in this
context, so we need a place to free it. We can't rely on doing it before
returning from ecp_mul() as we might return MBEDTLS_ERR_ECP_IN_PROGRESS (thus
preserving the context) and never be called again (for example, TLS handshake
aborted for another reason). So, ecp_group_free() looks like a good place to
do this, if the restart context is part of struct ecp_group.
This means it's not possible to use the same ecp_group structure in different
threads concurrently, but:
- that's already the case (and documented) for other reasons
- this feature is precisely intended for environments that lack threading
An alternative option would be for the caller to have to allocate/free the
restart context and pass it explicitly, but this means creating new functions
that take a context argument, and putting a burden on the user.
