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.
The point of this realloc is that we are not using this array, but
keeping it around to remain a node the talloc tree. We'd prefer to
reduce it to nothing.
Coverity rightly spotted that it was reallocing an array of `struct
ldb_val` to an array of `struct ldb_val *`, which has a different size
and all. But it doesn't matter in this case, because we will never use
it.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15590
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Jennifer Sutton <josutton@catalyst.net.nz>
(cherry picked from commit e2a74963fb)
We can't just free it, because has the GUID index list as a child, and
these are shared by the new dn list (from the subtransaction we are
committing). But if the dn list is long and the main transaction is
long-lived, we can save a lot of memory by turning this dn list into
an almost empty node in the talloc tree. This returns us to roughly
the situation we had prior to the last commit.
For example, with the repro.sh script on bug 15590 in indexes mode
with 10000 rules, The last 3 commits use this much memory at the end
of an unusually large transaction:
full talloc report on 'struct ldb_context' (total 4012222 bytes in 90058 blocks)
full talloc report on 'struct ldb_context' (total 2405482219 bytes in 90058 blocks)
full talloc report on 'struct ldb_context' (total 4282195 bytes in 90058 blocks)
That is, the last commit increased usage 500 fold, and this commit
brings it back to normal.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15590
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Stefan Metzmacher <metze@samba.org>
(cherry picked from commit 1bf9ede94f)
We don't want to modify the original list, but we can reuse the keys
if we treat them as immutable and don't free them. That makes it a lot
quicker if there are many keys (i.e. where an index is useful) and may
sub-transactions. In particular, it avoids O(n²) talloc_memdups.
A removed comment that says "We have to free the top level index
memory otherwise we would leak", and this will be addressed in the
next commit.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15590
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Stefan Metzmacher <metze@samba.org>
(cherry picked from commit 5f0198d698)
* BUG 15569: Many qsort() comparison functions are non-transitive, which
can lead to out-of-bounds access in some circumstances.
Signed-off-by: Jule Anger <janger@samba.org>
If two strings are invalid UTF-8, the string is first compared with
memcmp(), which compares as unsigned char.
If the strings are of different lengths and one is a substring of the
other, the memcmp() returns 0 and a second comparison is made which
assumes the next character in the shorter string is '\0' -- but this
comparison was done using SIGNED chars (on most systems). That leads
to non-transitive comparisons.
Consider the strings {"a\xff", "a", "ab\xff"} under that system.
"a\xff" < "a", because (char)0xff == -1.
"ab\xff" > "a", because 'b' == 98.
"ab\xff" < "a\xff", because memcmp("ab\xff", "a\xff", 2) avoiding the
signed char tiebreaker.
(Before c49c48afe0, the final character
might br arbitrarily cast into another character -- in latin-1, for
example, the 0xff here would have been seen as 'ÿ', which would be
uppercased to 'Ÿ', which is U+0178, which would be truncated to
'\x78', a positive char.
On the other hand e.g. 0xfe, 'þ', would have mapped to 0xde, 'Þ',
remaining negative).
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit e2051eebd4)
This also sorts NULLs after invalid DNs, which matches the comment
above.
CID 1596622.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit af7654331f)
This isn't supposed to be used for sorting, but it is hard to say it
won't be, so we might as well make it sort properly.
Following long-standing behaviour, we try to sort "FALSE" > "TRUE", by
length, then switch to using strncasecmp().
strncasecmp would sort the other way, so we swap the operands. This is
to make e.g. "TRUE\0" sort the same as "TRUE".
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit a75c98ad68)
We were returning -1 in all these cases:
ldb_dn_compare(dn, NULL);
ldb_dn_compare(NULL, dn);
ldb_dn_compare(NULL, NULL);
which would give strange results in sort, where this is often used.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit 5fe488d515)
We assume no values is unlikely, since we have been dereferencing
->values[0] forever, with no known reports of trouble.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit d4e69734c6)
There are further changes coming here.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit d785c1991c)
We can still have inconsistent comparisons, because two elements with
the same number of values will always return -1 if they are unequal,
which means they will sort differently depending on the order in which
they are compared.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit 21a071e486)
In other places we tend to include tsort.h, which also has TYPESAFE_QSORT.
ldb.h already has TYPESAFE_QSORT, so it might as well have NUMERIC_CMP.
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit de1b94f79e)
If a compare function is non-transitive (for example, if it evaluates
A > B and B > C, but A < C), this implementation of qsort could access
out-of-bounds memory. This was found in glibc's qsort by Qualys, and
their write-up for OSS-Security explains it very well:
https://www.openwall.com/lists/oss-security/2024/01/30/7
An example of a non-transitive compare is one in which does this
int cmp(const void *_a, const void *_b)
{
int a = *(int *)_a;
int b = *(int *)_b;
return a - b;
}
which does the right thing when the magnitude of the numbers is small,
but which will go wrong if a is INT_MIN and b is INT_MAX. Likewise, if
a and b are e.g. uint32_t, the value can wrap when cast to int.
We have functions that are non-transitive regardless of subtraction.
For example, here (which is not used with ldb_qsort):
int codepoint_cmpi(codepoint_t c1, codepoint_t c2)
if (c1 == c2 ||
toupper_m(c1) == toupper_m(c2)) {
return 0;
}
return c1 - c2;
}
The toupper_m() is only called on equality case. Consider {'a', 'A', 'B'}.
'a' == 'A'
'a' > 'B' (lowercase letters come after upper)
'A' < 'B'
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15569
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625
Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
(cherry picked from commit 73e4f6026a)
- documentation fixes
- build with Python 3.12 (bug #15513)
- a lot of additional error checking in
the python bindings
- minor code fixes
BUG: https://bugzilla.samba.org/show_bug.cgi?id=15513
Signed-off-by: Jule Anger <janger@samba.org>
Signed-off-by: Stefan Metzmacher <metze@samba.org>
When we added a second run of the Python testsuite, the return code from
the first run began to go ignored.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
If we’re going to call PyList_Size() on an object, we should be sure
that it is a list first.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Accessing the wrong member of a union invokes undefined behaviour.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
PyObject_AsMessageElement() has ‘flags’ and ‘attr_name’ parameters to
set properties of the returned MessageElement, but they apply only
*sometimes*.
‘attr_name’ not being set can result in cryptic and misleading error
messages from various ldb operations.
Changing the function’s behaviour to be more consistent could break
existing code, so we work around the issue instead.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
If ‘tmp’ happens to be garbage-collected, ‘name’ will become invalid.
Signed-off-by: Joseph Sutton <josephsutton@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
