mpi: Explicitly limit the allowed input length for gcry_mpi_scan.

* mpi/mpicoder.c (MAX_EXTERN_SCAN_BYTES): New.
(mpi_fromstr): Check against this limit.
(_gcry_mpi_scan): Ditto.
* tests/mpitests.c (test_maxsize): New.
(main): Cal that test.
--

A too large buffer length may lead to an unsigned integer overflow on
systems where size_t > unsigned int (ie. 64 bit systems).  The
computation of the required number of nlimbs may also be affected by
this.  However this is not a real world case because any processing
which has allocated such a long buffer from an external source would
be prone to other DoS attacks: The required buffer length to exhibit
this overflow is at least 2^32 - 8 bytes.

Signed-off-by: Werner Koch <wk@gnupg.org>

1 files changed, 3 insertions, 1 deletions

diff --git a/doc/gcrypt.texi b/doc/gcrypt.texi
index 437dddb7..c710765b 100644
--- a/doc/gcrypt.texi
+++ b/doc/gcrypt.texi
@@ -4526,7 +4526,8 @@ representation of an MPI and the internal one of Libgcrypt.
 Convert the external representation of an integer stored in @var{buffer}
 with a length of @var{buflen} into a newly created MPI returned which
 will be stored at the address of @var{r_mpi}.  For certain formats the
-length argument is not required and should be passed as @code{0}.  After a
+length argument is not required and should be passed as @code{0}. A
+@var{buflen} larger than 16 MiByte will be rejected.  After a
 successful operation the variable @var{nscanned} receives the number of
 bytes actually scanned unless @var{nscanned} was given as
 @code{NULL}. @var{format} describes the format of the MPI as stored in
@@ -4540,6 +4541,7 @@ bytes actually scanned unless @var{nscanned} was given as
 @item GCRYMPI_FMT_PGP
 As used by OpenPGP (only defined as unsigned). This is basically
 @code{GCRYMPI_FMT_STD} with a 2 byte big endian length header.
+A length header indicating a length of more than 16384 is not allowed.
 
 @item GCRYMPI_FMT_SSH
 As used in the Secure Shell protocol.  This is @code{GCRYMPI_FMT_STD}