The json-c library is used in cryptsetup for LUKS2 header information.
Since cryptsetup can be called very early during boot, the developers
avoid getrandom() calls in their own code base for now. [1]
Introducing a blocking getrandom() call in json-c therefore introduces
this issue for cryptsetup as well. Even though cryptsetup issues do not
have to be json-c issues, here is my proposal:
Let's use a non-blocking call, falling back to other sources if the call
would block. Since getrandom() accesses urandom, it must mean that we
are in an early boot phase -- otherwise the call would not block
according to its manual page.
As stated in manual page of random(4), accessing /dev/urandom won't
block but return weak random numbers, therefore this fallback would work
for json-c.
While at it, fixed the debug message.
[1] https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/47
which references to https://lwn.net/Articles/800509/
Whilst working on the Reproducible Builds effort [0] I noticed that
json-c could not be built reproducibly.
This is because it used the full, absolute path name as an (sanitised)
input to a filename, resulting in some binary package containing, for
example:
/usr/share/doc/libjson-c-dev/html/md__build_1st_json-c-0_815_issues_closed_for_0_813.html
^^^^^^^^^^^^^^^^^^^^^^
or
/usr/share/doc/libjson-c-dev/html/md__build_2_json-c-0_815_2nd_issues_closed_for_0_813.html
^^^^^^^^^^^^^^^^^^^^^^^^
These differing values are based on the path in which json-c is built. This was
originally filed in Debian as #966657 [1].
[0] https://reproducible-builds.org/
[1] https://bugs.debian.org/966657
Lower overhead than opening & reading from /dev/urandom, and works
in chroots and other situtations where /dev/urandom is not available.
Falls back to existing methods when kernel doesn't support the syscall.
With this version script, newly-linked binaries that depend on the
json-c shared library will refer to its symbols in a versioned form,
preventing their references from being resolved to a symbol of the same
name exported by json-glib or libjansson if those libraries appear in
dependency search order before json-c, which will usually result in
a crash. This is necessary because ELF symbol resolution normally uses
a single flat namespace, not a tree like Windows symbol resolution.
At least one symbol (json_object_iter_next()) is exported by all three
JSON libraries.
Linking with -Bsymbolic is not enough to have this effect in all cases,
because -Bsymbolic only affects symbol lookup within a shared object,
for example when json_object_set_serializer() calls
json_object_set_userdata(). It does not affect calls from external
code into json-c, unless json-c was statically linked into the
external caller.
This change will also not prevent code that depends on json-glib or
libjansson from finding json-c's symbols and crashing; to prevent
that, a corresponding change in json-glib or libjansson would be needed.
Adding a symbol-version is a backwards-compatible change, but once
added, removing or changing the symbol-version on a symbol would be an
incompatible change that requires a SONAME bump.
Resolves: https://github.com/json-c/json-c/issues/621
(when combined with an equivalent change to libjansson).
Signed-off-by: Simon McVittie <smcv@collabora.com>
This causes previously failing strings like "123-456" to return a valid json_object with the appropriate value. If you care about the trailing content, call json_tokener_parse_ex() and check the parse end point with json_tokener_get_parse_end().
Add an optional way to use an incremental chunk size ($TEST_PARSE_CHUNKSIZE)
for all of the single_basic_parse tests, in additional to the regular way.
While here, drop the utf8_replacement_char that is unnecesarily added if we run out of input in the middle of a unicode escape. No other functional changes (yet).
