Really the only thing we should be doing with these ciphers is hastening
their demise, but it was the weekend and this seemed like fun.
EVP_tls_cbc_copy_mac needs to rotate a buffer by a secret amount. (It
extracts the MAC, but rotated.) We have two codepaths for this. If
CBC_MAC_ROTATE_IN_PLACE is defined (always on), we make some assumptions
abuot cache lines, play games with volatile, and hope that doesn't leak
anything. Otherwise, we do O(N^2) work to constant-time select the
rotation incidences.
But we can do O(N lg N). Rotate by powers of two and constant-time
select by the offset's bit positions. (Handwaivy lower-bound: an array
position has N possible values, so, armed with only a constant-time
select, we need O(lg N) work to resolve it. There's N array positions,
so O(N lg N).)
A microbenchmark of EVP_tls_cbc_copy_mac shows this is 27% faster than
the old one, but still 32% slower than the in-place version.
in-place:
Did 15724000 CopyFromMAC operations in 20000744us (786170.8 ops/sec)
N^2:
Did 8443000 CopyFromMAC operations in 20001582us (422116.6 ops/sec)
N lg N:
Did 10718000 CopyFromMAC operations in 20000763us (535879.6 ops/sec)
This results in the following the CBC ciphers. I measured
AES-128-CBC-SHA1 and AES-256-CBC-SHA384 which are, respectively, the
cipher where the other bits are the fastest and the cipher where N is
largest.
in-place:
Did 2634000 AES-128-CBC-SHA1 (16 bytes) open operations in 10000739us (263380.5 ops/sec): 4.2 MB/s
Did 1424000 AES-128-CBC-SHA1 (1350 bytes) open operations in 10002782us (142360.4 ops/sec): 192.2 MB/s
Did 531000 AES-128-CBC-SHA1 (8192 bytes) open operations in 10002460us (53086.9 ops/sec): 434.9 MB/s
N^2:
Did 2529000 AES-128-CBC-SHA1 (16 bytes) open operations in 10001474us (252862.7 ops/sec): 4.0 MB/s
Did 1392000 AES-128-CBC-SHA1 (1350 bytes) open operations in 10006659us (139107.4 ops/sec): 187.8 MB/s
Did 528000 AES-128-CBC-SHA1 (8192 bytes) open operations in 10001276us (52793.3 ops/sec): 432.5 MB/s
N lg N:
Did 2531000 AES-128-CBC-SHA1 (16 bytes) open operations in 10003057us (253022.7 ops/sec): 4.0 MB/s
Did 1390000 AES-128-CBC-SHA1 (1350 bytes) open operations in 10003287us (138954.3 ops/sec): 187.6 MB/s
Did 531000 AES-128-CBC-SHA1 (8192 bytes) open operations in 10002448us (53087.0 ops/sec): 434.9 MB/s
in-place:
Did 1249000 AES-256-CBC-SHA384 (16 bytes) open operations in 10001767us (124877.9 ops/sec): 2.0 MB/s
Did 879000 AES-256-CBC-SHA384 (1350 bytes) open operations in 10009244us (87818.8 ops/sec): 118.6 MB/s
Did 344000 AES-256-CBC-SHA384 (8192 bytes) open operations in 10025897us (34311.1 ops/sec): 281.1 MB/s
N^2:
Did 1072000 AES-256-CBC-SHA384 (16 bytes) open operations in 10008090us (107113.3 ops/sec): 1.7 MB/s
Did 780000 AES-256-CBC-SHA384 (1350 bytes) open operations in 10007787us (77939.3 ops/sec): 105.2 MB/s
Did 333000 AES-256-CBC-SHA384 (8192 bytes) open operations in 10016332us (33245.7 ops/sec): 272.3 MB/s
N lg N:
Did 1168000 AES-256-CBC-SHA384 (16 bytes) open operations in 10007671us (116710.5 ops/sec): 1.9 MB/s
Did 836000 AES-256-CBC-SHA384 (1350 bytes) open operations in 10001536us (83587.2 ops/sec): 112.8 MB/s
Did 339000 AES-256-CBC-SHA384 (8192 bytes) open operations in 10018522us (33837.3 ops/sec): 277.2 MB/s
TLS CBC performance isn't as important as it was before, and the costs
aren't that high, so avoid making assumptions about cache lines. (If we
care much about CBC open performance, we probably should get the malloc
out of EVP_tls_cbc_digest_record at the end.)
Change-Id: Ib8d8271be4b09e5635062cd3b039e1e96f0d9d3d
Reviewed-on: https://boringssl-review.googlesource.com/11003
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
The old one was rather confusing. Switch to returning 1/0 for whether
the padding is publicly invalid and then add an output argument which
returns a constant_time_eq-style boolean.
Change-Id: Ieba89d352faf80e9bcea993b716f4b2df5439d4b
Reviewed-on: https://boringssl-review.googlesource.com/10222
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
A lot of consumers of obj.h only want the NID values. Others didn't need
it at all. This also removes some OBJ_nid2sn and OBJ_nid2ln calls in EVP
error paths which isn't worth pulling a large table in for.
BUG=chromium:499653
Change-Id: Id6dff578f993012e35b740a13b8e4f9c2edc0744
Reviewed-on: https://boringssl-review.googlesource.com/7563
Reviewed-by: David Benjamin <davidben@google.com>
Partially fixes build with -Wmissing-prototypes -Wmissing-declarations.
Change-Id: I51209c30f532899f57cfdd9a50cff0a8ee3da5b5
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Reviewed-on: https://boringssl-review.googlesource.com/7512
Reviewed-by: David Benjamin <davidben@google.com>
Division isn't constant-time on Intel chips so the code was adding a
large multiple of md_size to try and force the operation to always take
the maximum amount of time.
I'm less convinced, these days, that compilers aren't going to get smart
enough to optimise that away so use Barrett reduction instead.
Change-Id: Ib8c514192682a2fcb4b1fb7e7c6dd1301d9888d0
Reviewed-on: https://boringssl-review.googlesource.com/6906
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <alangley@gmail.com>
The documentation in md32_common.h is now (more) correct with respect
to the most important details of the layout of |HASH_CTX|. The
documentation explaining why sha512.c doesn't use md32_common.h is now
more accurate as well.
Before, the C implementations of HASH_BLOCK_DATA_ORDER took a pointer
to the |HASH_CTX| and the assembly language implementations took a
pointer to the hash state |h| member of |HASH_CTX|. (This worked
because |h| is always the first member of |HASH_CTX|.) Now, the C
implementations take a pointer directly to |h| too.
The definitions of |MD4_CTX|, |MD5_CTX|, and |SHA1_CTX| were changed to
be consistent with |SHA256_CTX| and |SHA512_CTX| in storing the hash
state in an array. This will break source compatibility with any
external code that accesses the hash state directly, but will not
affect binary compatibility.
The second parameter of |HASH_BLOCK_DATA_ORDER| is now of type
|const uint8_t *|; previously it was |void *| and all implementations
had a |uint8_t *data| variable to access it as an array of bytes.
This change paves the way for future refactorings such as automatically
generating the |*_Init| functions and/or sharing one I-U-F
implementation across all digest algorithms.
Change-Id: I6e9dd09ff057c67941021d324a4fa1d39f58b0db
Reviewed-on: https://boringssl-review.googlesource.com/6405
Reviewed-by: Adam Langley <agl@google.com>
The documentation in md32_common.h is now (more) correct with respect
to the most important details of the layout of |HASH_CTX|. The
documentation explaining why sha512.c doesn't use md32_common.h is now
more accurate as well.
Before, the C implementations of HASH_BLOCK_DATA_ORDER took a pointer
to the |HASH_CTX| and the assembly language implementations tool a
pointer to the hash state |h| member of |HASH_CTX|. (This worked
because |h| is always the first member of |HASH_CTX|.) Now, the C
implementations take a pointer directly to |h| too.
The definitions of |MD4_CTX|, |MD5_CTX|, and |SHA1_CTX| were changed to
be consistent with |SHA256_CTX| and |SHA512_CTX| in storing the hash
state in an array. This will break source compatibility with any
external code that accesses the hash state directly, but will not
affect binary compatibility.
The second parameter of |HASH_BLOCK_DATA_ORDER| is now of type
|const uint8_t *|; previously it was |void *| and all implementations
had a |uint8_t *data| variable to access it as an array of bytes.
This change paves the way for future refactorings such as automatically
generating the |*_Init| functions and/or sharing one I-U-F
implementation across all digest algorithms.
Change-Id: I30513bb40b5f1d2c8932551d54073c35484b3f8b
Reviewed-on: https://boringssl-review.googlesource.com/6401
Reviewed-by: Adam Langley <agl@google.com>
Including string.h in base.h causes any file that includes a BoringSSL
header to include string.h. Generally this wouldn't be a problem,
although string.h might slow down the compile if it wasn't otherwise
needed. However, it also causes problems for ipsec-tools in Android
because OpenSSL didn't have this behaviour.
This change removes string.h from base.h and, instead, adds it to each
.c file that requires it.
Change-Id: I5968e50b0e230fd3adf9b72dd2836e6f52d6fb37
Reviewed-on: https://boringssl-review.googlesource.com/3200
Reviewed-by: David Benjamin <davidben@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
CBC modes in SSLv3 are bust already with POODLE and we're moving away from it.
Align all the names from 'ssl3' and 'tls1' to 'tls', to match the names of the
TLS-only AEADs.
Change-Id: If742296a8e2633ef42a484e4d873b4a83558b6aa
Reviewed-on: https://boringssl-review.googlesource.com/2693
Reviewed-by: Adam Langley <agl@google.com>
The EVP_CIPHER codepath should no longer be used with TLS. It still exists for
DTLS and SSLv3. The AEAD construction in TLS does not allow for
variable-overhead AEADs, so stateful AEADs do not include the length in the ad
parameter. Rather the AEADs internally append the unpadded length once it is
known. EVP_aead_rc4_md5_tls is modified to account for this.
Tests are added (and RC4-MD5's regenerated) for each of the new AEADs. The
cipher tests are all moved into crypto/cipher/test because there's now a lot of
them and they clutter the directory listing.
In ssl/, the stateful AEAD logic is also modified to account for stateful AEADs
with a fixed IV component, and for AEADs which use a random nonce (for the
explicit-IV CBC mode ciphers).
The new implementation fixes a bug/quirk in stateless CBC mode ciphers where
the fixed IV portion of the keyblock was generated regardless. This is at the
end, so it's only relevant for EAP-TLS which generates a MSK from the end of
the key block.
Change-Id: I2d8b8aa11deb43bde2fd733f4f90b5d5b8cb1334
Reviewed-on: https://boringssl-review.googlesource.com/2692
Reviewed-by: Adam Langley <agl@google.com>
These helper functions will be used in the implementation of the legacy CBC
mode AEADs. The file is copied as-is and then modified to remove the dependency
on ssl/. Notably explicit IV logic is removed (that's a side effect of how
explicit IVs are currently implemented) and the padding length is returned
directly rather than smuggled into rec->type.
(Diffing tls_cbc.c and s3_cbc.c is probably the easiest for a review.)
The helpers are currently unused.
Change-Id: Ib703f4d3620196c9f2921cb3b8bf985f2d1777db
Reviewed-on: https://boringssl-review.googlesource.com/2691
Reviewed-by: Adam Langley <agl@google.com>
