IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Autobuild-User(master): Joseph Sutton <jsutton@samba.org>
Autobuild-Date(master): Mon Jan 15 01:56:53 UTC 2024 on atb-devel-224
These and more are also defined in security_token.h, which is later included
from security.h anyway.
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Reviewed-by: Günther Deschner <gd@samba.org>
Reviewed-by: Andreas Schneider <asn@samba.org>
As with the previous commit, though not addressing the particular fuzz
case, zero hex numbers need to be explicitly written as "0x0", or the
round-trip will fail.
Credit to OSS-Fuzz.
REF: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=62929
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
The string "00" will decode into an integer tagged as octal, but
`snprintf("%#oll")` will write the string "0", which would decode as
decimal, so the in the SDDL1->SD1->SDDL2->SD2 round trip, SD1 would
not be the same as SD2.
The effect is really only relevant to SDDL, which wants to remember
what base the numbers were presented in, though the fuzzers and tests
don't directly compare SDDL, which can have extra spaces and so forth.
Credit to OSS-Fuzz.
REF: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=62929
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
The `failed = failed || ok` did the same thing, obscurely.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
They should be tightly packed, allowing conditional ACEs to
round-trip.
Credit to OSS-Fuzz.
REF: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=64197
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
For this purpose, "too many" means we know for sure that it won't fit
in packet format, even if all the ACEs are minimum size. This would
fail anyway.
Credit to OSS-Fuzz, who found that 50 thousand ACEs that took more
than 60 seconds to decode. This will now fail after 4096 ACEs which
should be about 150 times faster than 50k (because the realloc loop in
quadratic), so ~0.5 seconds in the fuzz context with sanitisers
enabled. That is still slowish, but SDDL parsing is not a critical
path and without address sanitisers it will be many times faster.
REF: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=62511
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Our code won’t use this, but NIST’s test vectors are based on handing a
fixed buffer to the key derivation function.
View with ‘git show -b’.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Rename smb2_key_derivation() to samba_gnutls_sp800_108_derive_key() and
move it to GNUTLS_HELPERS.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
The Contains and Any_of operators could use a sorted comparison like
compare_composites_via_sort(), rather than O(n²) nested loops. But
that would involve amount of quite fiddly work that I am not starting
on now.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Autobuild-User(master): Andrew Bartlett <abartlet@samba.org>
Autobuild-Date(master): Mon Nov 27 23:38:13 UTC 2023 on atb-devel-224
We know from the way claims are defined, and from the code that checks
sortedness and sets the flag.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
The ordinary comparison path, using the sorted arrays, already implicitly
checks for comparability. We only need this when we're leaving early.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
If the number of members does not match in certain ways we can
say the sets are not equal without comparing the members.
We first need to check for comparability, though, so that we can return
an error if things aren't comparable.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
We had the comparison method wrong. Composites are compared as sets or
flabby sets, depending on their origin. Until now we compared them as
something a bit like sets, but not quite, in a maximally inefficient way.
Claims are always sets, and the left hand side is always a claim, but
literal composites on the right hand side can be multi-sets
(containing duplicate values). When it comes to comparison, composites
are reduced down to sets. To do the comparison we sort each side and
compare in order.
The fact that either side might ask for case-sensitive comparison (if
it is a claim) is an interesting complication.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
For SDDL Resource ACE conversions we don't want to check too much
claim validity so that a semi-invalid ACE can round-trip through
deserialisation and serialisation. This is because Windows allows it,
but also because if the check puts the values in a sorted order that
makes the round-trip less round (that is, the return string is
semantically the same but possibly different in byte order).
The validity we're talking about is mostly uniqueness. For example
`S:(RA;;;;;WD;("foo",TU,0,7,5,7))` has two 7s, and that would be
invalid as a claim, but this is not checked while in ACE form.
On the other hand `S:(RA;;;;;WD;("foo",TU,0,3,2))` is valid, but the
return string will have 3 and 2 reversed when the check is made. We
prefer the ACE to stay the same while it is just being an ACE.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
If it is a wire claim (which is probably most common), the checking
and sorting has already happened. We don't need to make a copy to
sort and check.
In either case, there is still a copy step to make the conditional ACE
token.
This shuffles around some knownfails because the claim_v1_copy()
function we were using is checking for duplicates, which we don't
always want. That will be fixed soon.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
This function is used in tests and fuzzing.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Because RA ACEs live a double life, sometimes being ACEs and sometimes
being claims, we make a copy of the claim strucutre for sorting and
further use in conditional ACEs.
We don't need to do that for wire claims, because they are not
persistent or forwarded on to somewhere else.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
This roughly returns things to where they were a few commits ago, with
the claims being checked for uniqueness.
The difference is the claims will be sorted afterwards, and the
uniqueness check will be far more efficient on large claims.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
To manage this sort we need a qsort_r-like sort context which holds:
a) the value type,
b) a case sensitive flag for the string compare, and
c) a return flag indicating a failure. Failures are not picked up until
after the sort finishes.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
claim_v1_check_and_sort() is meant to sort the claim values and check
that there are no duplicates, as well as making some value checks.
In order to ease into the idea, we look first at the case where the claim
has Boolean values. There are only two values allowed, which limits the
length of a valid claim set and means we only really need to "sort" in
the {1, 0} case, which we rewrite in place as {0, 1}.
That's what will happen with other types: we'll sort in-place, make
some checks on values, set flags, and return an error if there are
duplicates or value errors.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
This restores the behaviour with regard to duplicate NULL strings that
existed before the last commit. I'm putting it separately, because it
seems so strange, and I not entirely certain the behaviour is
intentional.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
This changes the behaviour when one of the strings is NULL. Previously
a single NULL string would be ignored, and two would cause an error.
That will be restored in the next commit.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
And we allocate all the values together as an array, because
we might as well.
This and the next couple of commits might look like steps backwards,
and they are, but they allow us to get a run-up to leap over a big
fence.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
The interstitial tmp_ctx now does nothing but be interstitial, so
let's get rid of it.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
These values would have leaked in the event of failure (but only onto
the caller mem_ctx, which might be fleeting -- especially as its
security token is now failing).
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
The reason for this, apart from weighing up possible over-allocations
vs realloc costs, is in the first iteration of the loop,
claim_values = talloc_array(claims,
would allocate onto NULL, which leaks.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
We don't change these when writing the SDDL.
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
