Commit graph

5544 commits

Author SHA1 Message Date
Manuel Pégourié-Gonnard 085b1dff40 Allow T to be computed in multiple steps
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.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard c9c0aa6306 Remember precomputed table
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.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard c5d844b999 Full restart support in ecp_mul_comb_core()
Still recomputing table every time, though.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 2fad7ae02a Start actually splitting computation
Temporary state is quite inefficient: pre-computed table is recomputed every
single time. This is WIP obviously.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 8962ddbb23 Don't write to destination until we're done 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 78d564a841 Add check for changing arguments
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
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 77af79a324 Add proper allocation of restart context
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.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 62738e9b17 Further restrict variable scope by moving code 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 391f44153d Move more code to separate function
This reduces the scope of some variables (M, k), clarifying where they're
used.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 24be79588d Group related code together 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 4b2336d7f6 Move some more code to new function 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 22be635d13 Re-order some more code 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard ec5606ad0c Extract code to separate function
ecp_mul_comb() is already 110 lines long and we're going to add complexity
with the early-return+restart code, so let's try to make it simpler first.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 7306dff01f Group related code together
This will be split to a new function next.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 510d5caece Add early return test + fake implementation 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 054433c493 Add mbedtls_ecp_set_max_ops()
The plan is to count basic operations as follows:
- call to ecp_add_mixed()   -> 11
- call to ecp_double_jac()  -> 8
- call to mpi_mul_mpi()     -> 1
- call to mpi_inv_mod()     -> 120
- everything else           -> not counted

The counts for ecp_add_mixed() and ecp_double_jac() are based on the actual
number of calls to mpi_mul_mpi() they they make.

The count for mpi_inv_mod() is based on timing measurements on K64F and
LPC1768 boards, and are consistent with the usual very rough estimate of one
inversion = 100 multiplications. It could be useful to repeat that measurement
on a Cortex-M0 board as those have smaller divider and multipliers, so the
result could be a bit different but should be the same order of magnitude.

The documented limitation of 120 basic ops is due to the calls to mpi_inv_mod()
which are currently not interruptible nor planned to be so far.
2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 5e3c62fd1d Add MBEDTLS_ERR_ECP_IN_PROGRESS 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard c3a3bc7636 Add config flag MBEDTLS_ECP_EARLY_RETURN 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard 225b37a543 Fix typos in comment 2017-08-09 11:44:53 +02:00
Manuel Pégourié-Gonnard a54f6cc874 Unify name of default profile in X.509 tests 2017-08-09 10:41:42 +02:00
Manuel Pégourié-Gonnard 24611f9383 Remove redundant variable
path_cnt was always chain_len - 1 in the loop body
2017-08-09 10:28:07 +02:00
Manuel Pégourié-Gonnard 7ff243a87c Add missing dependency in test-certs Makefile 2017-08-08 18:54:13 +02:00
Manuel Pégourié-Gonnard 562df401d3 Improve some comments, fix some typos+whitespace 2017-08-08 18:17:53 +02:00
Manuel Pégourié-Gonnard ea2dc14c0c Fix some whitespace 2017-08-08 11:10:37 +02:00
Manuel Pégourié-Gonnard a7c4c8a46c Make some perl scripts usable with git bisect run
For that they need to return between 0 and 124 on error, while die returns
255, causing bisect-run to abort.
2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard 66a36b03c6 Update comments 2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard 505c3953c7 Make the ver_chain length explicit 2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard a707e1d1ef Extract code to separate function for readablity 2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard ce6e52ff42 Make verify_chain() iterative 2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard f86f491f25 Rm unneeded function arguments & update comments 2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard c547d1ab1f Start using an explicit stack for callback info
This is the first step towards making verify_chain() iterative. While from a
readability point of view the current recursive version is fine, one of the
goals of this refactoring is to prepare for restartable ECC integration, which
will need the explicit stack anyway.
2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard a468eb1764 verify_name(): factor duplicated code to function 2017-08-08 11:06:51 +02:00
Manuel Pégourié-Gonnard 1300e99eb1 Extract name checking to separate function
Just copy-paste and unindent
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 6368612a8f Move code to separate function for readability 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 27e94797aa Simplify handling of locally trusted EE certs
Though this might require one more walk of the list in some cases,
this avoid having a check for that deep inside check_parent().
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard bdc5440232 Update comments 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard cb39610093 Finally merge the remains of top() into child() 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 63642776b1 Let verify_top() handle only the parent
It felt wrong for it to call the vrfy callback on two certs.
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 6e786747fb Move top()'s checks on child to child() 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 784aee3366 Move other special case from top() to child() 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard b9983be73a Move one special case from verify_top() to child() 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 66fac75f8b Merge duplicated checks between child() and top() 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 58dcd2d9b2 Get rid of unused variables/arguments 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 8f8c282de9 Merge near-duplicated (grand)parent finding code
Besides avoiding near-duplication, this avoids having three generations of
certificate (child, parent, grandparent) in one function, with all the
off-by-one opportunities that come with it.

This also allows to simplify the signature of verify_child(), which will be
done in next commit.
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard f82a4d5aba Factor duplicated code into function 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 32fdc60c7b Unnest code in verify_top()
We now know that trust_ca != NULL till the end of the function
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 6038cb6909 Remove duplicate parent-searching in verify_top() 2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 2f09d59456 Add badkey-skipping to find_parent()
This is the last step towards removing the now-duplicated parent-searching
code in verify_top()
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 3e329b8e8d Add badtime-skipping feature to new function
This is from the morally 5th (and soon obsolete) invocation of this function
in verify_top().

Doing this badtime-skipping when we search for a parent in the provided chain
is a change of behaviour, but it's backwards-compatible: it can only cause us
to accept valid chains that we used to reject before. Eg if the peer has a
chain with two version of an intermediate certificate with different validity
periods, the first non valid and the second valid - such cases are probably
rare or users would have complained already, but it doesn't hurt to handle it
properly as it allows for more uniform code.
2017-08-08 11:06:50 +02:00
Manuel Pégourié-Gonnard 9c6118c498 Factor one more occurrence of code into function
This may look like a behaviour change because one check has been added to the
function that was previously done in only one of the 3 call sites. However it
is not, because:
- for the 2 call sites in verify(), the test always succeeds as path_cnt is 0.
- for the call site in verify_child(), the same test was done later anyway in
  verify_top()
2017-08-08 11:06:50 +02:00