[Docs] [txt|pdf] [Tracker] [Email] [Diff1] [Diff2] [Nits]

Versions: 00 01 02

Network Working Group                  Richard Price, Siemens/Roke Manor
INTERNET-DRAFT                       Abigail Surtees, Siemens/Roke Manor
Expires: July 2003
                                                        January 14, 2003


                           SigComp Torture Tests
              <draft-price-rohc-sigcomp-torture-tests-01.txt>


Status of this memo

   This document is an Internet-Draft and is in full conformance with
   all provisions of Section 10 of RFC2026.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-
   Drafts.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or cite them other than as "work in progress".

   The list of current Internet-Drafts can be accessed at
   http://www.ietf.org/ietf/lid-abstracts.txt

   The list of Internet-Draft Shadow Directories can be accessed at
   http://www.ietf.org/shadow.html

   This document is a submission of the IETF ROHC WG.  Comments should
   be directed to its mailing list, rohc@ietf.org.


Abstract

   This document provides a set of "torture tests" for implementers of
   the SigComp protocol.  The torture tests check each of the SigComp
   Universal Decompressor Virtual Machine instructions in turn, focusing
   in particular on the boundary and error cases that are not generally
   encountered when running well-behaved compression algorithms.  Tests
   are also provided for other SigComp entities such as the dispatcher
   and the state handler.










Price et al.                                                    [Page 1]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


Change history

   Changes relative to <draft-price-rohc-sigcomp-torture-tests-00.txt>:

   1. Added tests for the SigComp dispatcher (covering the SigComp
      Useful Values, the SigComp header for message-based transports,
      and the record marking scheme for stream-based transports).

   2. Added tests for the SigComp state handler (covering the SigComp
      feedback mechanism, the state memory management and the
      interaction between multiple compartments).

   3. Updated the cost of the sorting instructions based on the new
      values used in SigComp [RFC-3320].

   4. Updated the stack manipulation test to work correctly when the
      decompression_memory_size is only 2048 bytes.


Table of contents

   1.  Introduction..................................................2
   2.  Torture tests for UDVM........................................3
   3.  Torture tests for dispatcher..................................20
   4.  Torture tests for state handler...............................25
   5.  Security considerations.......................................35
   6.  Authors' addresses............................................35
   7.  References....................................................36
   Appendix A: UDVM bytecode for the torture tests...................37


1.  Introduction

   This document provides a set of torture tests for implementers of the
   SigComp protocol [RFC-3320].  The idea behind SigComp is to
   standardize a Universal Decompressor Virtual Machine (UDVM) that can
   be programmed to understand the output of many well-known compressors
   including DEFLATE and LZW.  The bytecode for the chosen decompressor
   is uploaded to the UDVM as part of the SigComp message flow.

   The SigComp User Guide [USERGUIDE] offers a number of different
   algorithms that can be used by the SigComp protocol.  However, the
   bytecode for the corresponding decompressors is relatively well
   behaved and does not test the boundary and error cases that may
   potentially be exploited by malicious SigComp messages.

   The draft is divided into a number of sections, each containing a
   piece of code designed to test a particular function of one of the
   SigComp entities (UDVM, dispatcher and state handler).  The specific
   boundary and error cases tested by the bytecode are also listed, as
   is the expected output of the code.




Price et al.                                                    [Page 2]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


2.  Torture tests for UDVM

   The following sections each provide code to test one or more UDVM
   instructions.  In the interests of readability the code is given
   using the SigComp assembly language: a description of how to convert
   this assembly code into UDVM bytecode can be found in the SigComp
   User Guide [USERGUIDE].

   The raw UDVM bytecode for each torture test is given in Appendix A.

   Each section also lists the number of UDVM cycles required to execute
   the code.  Note that this figure only takes into account the cost of
   executing each UDVM instruction (in particular it ignores the fact
   that the UDVM can gain extra cycles as a result of inputting more
   data).

2.1.  Bit manipulation

   This section gives assembly code to test the AND, OR, NOT, LSHIFT and
   RSHIFT instructions.  When the instructions have a multitype operand
   the code tests the case where the multitype contains a fixed integer
   value, and the case where it contains a memory address at which the
   2-byte operand value can be found.  In addition the code is designed
   to test that the following boundary cases have been correctly
   implemented:

   1. The instructions overwrite themselves with the result of the bit
   manipulation operation.

   2. The LSHIFT or RSHIFT instructions shift bits beyond the 2-byte
   boundary, in which case the bits must be discarded.

   3. The UDVM registers byte_copy_left and byte_copy_right are used to
   store the results of the bit manipulation operations.  Since no byte
   copying is taking place these registers should behave in exactly the
   same manner as ordinary UDVM memory addresses.

   at (64)

   :a                              pad (2)
   :b                              pad (2)

   at (128)

   JUMP (start)

   at (255)

   :start

   AND ($start, 21845)




Price et al.                                                    [Page 3]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   OR ($a, 42)
   NOT ($b)
   LSHIFT ($a, 3)
   RSHIFT ($b, 65535)

   OUTPUT (64, 4)

   AND ($a, $start)
   OR ($a, $a)
   NOT ($a)
   LSHIFT ($b, $a)
   RSHIFT ($a, $b)

   OUTPUT (64, 4)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   The expected output of the code is 0x0150 0000 febf 0000.  Executing
   the code should cost a total of 22 UDVM cycles.

2.2.  Arithmetic

   This section gives assembly code to test the ADD, SUBTRACT, MULTIPLY,
   DIVIDE and REMAINDER instructions.  The code is designed to test that
   the following boundary cases have been correctly implemented:

   1. The instructions overwrite themselves with the result of the
   arithmetic operation.

   2. The result does not lie between 0 and 2^16 - 1 inclusive, in which
   case it must be taken modulo 2^16.

   3. The divisor in the DIVIDE or REMAINDER instructions is 0 (in which
   case decompression failure should occur).

   at (64)

   :a                              pad (2)
   :b                              pad (2)
   :type                           pad (1)
   :type_lsb                       pad (1)

   at (128)

   INPUT-BYTES (1, type_lsb, !)
   SUBTRACT ($type, 1)
   JUMP (start)

   at (255)

   :start





Price et al.                                                    [Page 4]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   ADD ($start, 63809)
   SUBTRACT ($a, 1)
   MULTIPLY ($a, 1001)
   DIVIDE ($a, 101)
   REMAINDER ($a, 11)

   OUTPUT (64, 4)

   ADD ($b, $start)
   SUBTRACT ($b, $type)
   MULTIPLY ($b, $b)
   DIVIDE ($a, $b)
   REMAINDER ($b, $type)

   OUTPUT (64, 4)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   If the compressed message is 0x00 then the expected output of the
   code is 0x0000 0000 0000 0004 and the execution cost should be 25
   UDVM cycles.  However, if the compressed message is 0x01 or 0x02 then
   decompression failure should occur.

2.3.  Sorting

   This section gives assembly code to test the SORT-ASCENDING and SORT-
   DESCENDING instructions.  The code is designed to test that the
   following boundary cases have been correctly implemented:

   1. The sorting instructions sort integers with the same value, in
   which case the original ordering of the integers must be preserved.

   at (128)

   SORT-DESCENDING (256, 2, 23)
   SORT-ASCENDING (256, 2, 23)

   OUTPUT (302, 45)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   at (256)

   word (10, 10, 17, 7, 22, 3, 3, 3, 19, 1, 16, 14, 8, 2, 13, 20, 18,
   23, 15, 21, 12, 6, 9)

   word (28263, 8297, 30057, 8308, 26996, 11296, 31087, 29991, 8275,
   18031, 28263, 24864, 30066, 29284, 28448, 29807, 28206, 11776, 28773,
   28704, 28276, 29285, 28265)

   The expected output of the code is 0x466f 7264 2c20 796f 7527 7265
   2074 7572 6e69 6e67 2069 6e74 6f20 6120 7065 6e67 7569 6e2e 2053 746f
   7020 6974 2e, and the expected number of cycles required is 371.




Price et al.                                                    [Page 5]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   N.B.  This uses the corrected cost for the sorting instructions,
   which is 1 + k * (ceiling(log2(k)) + n) not 1 + k * ceiling(log2(k)).

2.4.  SHA-1

   This section gives assembly code to test the SHA-1 instruction.  The
   code performs four tests on the SHA-1 algorithm itself, and
   additionally checks the following boundary cases specific to the
   UDVM:

   1. The input string for the SHA-1 hash is obtained by byte copying
   over an area of the UDVM memory.

   2. The SHA-1 hash overwrites its own input string.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :hash_value                     pad (20)

   at (128)

   SHA-1 (test_one, 3, hash_value)
   OUTPUT (hash_value, 20)

   SHA-1 (test_two, 56, hash_value)
   OUTPUT (hash_value, 20)

   LOAD (byte_copy_left, test_three)
   LOAD (byte_copy_right, test_four)

   SHA-1 (test_three, 65535, hash_value)
   OUTPUT (hash_value, 20)

   LOAD (byte_copy_left, test_four)
   LOAD (byte_copy_right, test_end)

   SHA-1 (test_four, 640, test_four)
   OUTPUT (test_four, 20)

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   :test_one

   byte (97, 98, 99)

   :test_two

   byte (97, 98, 99, 100, 98, 99, 100, 101, 99, 100, 101, 102, 100, 101,
   102, 103, 101, 102, 103, 104, 102, 103, 104, 105, 103, 104, 105, 106,




Price et al.                                                    [Page 6]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   104, 105, 106, 107, 105, 106, 107, 108, 106, 107, 108, 109, 107, 108,
   109, 110, 108, 109, 110, 111, 109, 110, 111, 112, 110, 111, 112, 113)

   :test_three

   byte (97)

   :test_four

   byte (48, 49, 50, 51, 52, 53, 54, 55)

   :test_end

   The expected output of the code is as follows:

   0xa999 3e36 4706 816a ba3e 2571 7850 c26c 9cd0 d89d
   0x8498 3e44 1c3b d26e baae 4aa1 f951 29e5 e546 70f1
   0xe1d0 a18d 43d3 a689 af08 8e15 6bd0 434a a0c8 31fc
   0x4f46 0452 ebb5 6393 4f46 0452 ebb5 6393 4f46 0452

   Executing the code is expected to cost a total of 66327 UDVM cycles.

2.5.  LOAD and MULTILOAD

   This section gives assembly code to test the LOAD and MULTILOAD
   instructions.  The code is designed to test the following boundary
   cases:

   1. The MULTILOAD instruction overwrites itself, any of its operands,
   or any memory addresses referenced by its operands (in which case
   decompression failure should occur).

   at (64)

   :start                          pad (1)
   :start_lsb                      pad (1)

   at (128)

   set (location_a, 128)
   set (location_b, 132)

   LOAD (128, 132)
   LOAD (130, $location_a)
   LOAD ($location_a, 134)
   LOAD ($location_b, $location_b)
   OUTPUT (128, 8)

   INPUT-BYTES (1, start_lsb, !)
   MULTIPLY ($start, 2)
   ADD ($start, 60)




Price et al.                                                    [Page 7]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   MULTILOAD ($start, 3, overlap_start, overlap_end, 128)

   :position

   set (overlap_start, (position - 7))

   MULTILOAD ($start, 4, 42, 128, $location_a, $location_b)

   :end

   set (overlap_end, (end - 1))

   OUTPUT (128, 8)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   If the compressed message is 0x00 then the expected output of the
   code is 0x0084 0084 0086 0086 002a 0080 002a 002a, and the expected
   cost of executing the code is 36 UDVM cycles.  However, if the
   compressed message is 0x01 or 0x02 then decompression failure is
   expected to occur while executing the second MULTILOAD instruction.

2.6. COPY

   This section gives assembly code to test the COPY instruction.  The
   code is designed to test that the following boundary cases have been
   correctly implemented:

   1. The COPY instruction copies data from both outside the circular
   buffer and inside the circular buffer within the same operation.

   2. The COPY instruction performs byte-by-byte copying (i.e. some of
   the later bytes to be copied are themselves written into the UDVM
   memory by the COPY instruction currently being executed).

   3. The COPY instruction overwrites itself.

   4. The COPY instruction overwrites the UDVM registers byte_copy_left
   and byte_copy_right.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)

   at (128)

   LOAD (32, 16384)
   LOAD (byte_copy_left, 64)
   LOAD (byte_copy_right, 128)

   COPY (32, 128, 33)




Price et al.                                                    [Page 8]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   LOAD (64, 16640)
   COPY (64, 76, 65)

   OUTPUT (32, 109)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   The expected output of the code is 32 consecutive instances of 0x40
   (the ASCII character "@") followed by 77 consecutive instances of
   0x41 (the ASCII character "A").  Executing the code should cost a
   total of 321 UDVM cycles.

2.7.  COPY-LITERAL and COPY-OFFSET

   This section gives assembly code to test the COPY-LITERAL and COPY-
   OFFSET instructions.  The code is designed to test similar boundary
   cases to the code for the COPY instruction, as well as the following
   condition specific to COPY-LITERAL and COPY-OFFSET:

   1. The COPY-LITERAL or COPY-OFFSET instruction overwrites the value
   of its destination or offset operand.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :destination                    pad (2)
   :offset                         pad (2)

   at (128)

   LOAD (32, 16384)
   LOAD (byte_copy_left, 64)
   LOAD (byte_copy_right, 128)
   LOAD (destination, 33)

   COPY-LITERAL (32, 128, $destination)
   COPY-LITERAL (68, 8, $destination)

   LOAD (byte_copy_left, 66)
   LOAD (byte_copy_right, 74)

   COPY-OFFSET (8, 6, $destination)

   LOAD ($offset, 1)

   COPY-OFFSET ($offset, 5 ,$destination)

   OUTPUT (32, 48)

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)





Price et al.                                                    [Page 9]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   The expected output of the code is 32 instances of 0x40 followed by
   0x0042 004a 0074 4040 4040 004a 0074 4040.  The expected cost of
   executing the code is 208 UDVM cycles.

   N.B.  This uses the corrected cost for COPY-OFFSET, which is 1 +
   length not 1 + length + offset.

2.8.  MEMSET

   This section gives assembly code to test the MEMSET instruction.  The
   code is designed to test that the following boundary cases have been
   correctly implemented:

   1. The MEMSET instruction overwrites the registers byte_copy_left and
   byte_copy_right.

   2. The output values of the MEMSET instruction do not lie between 0
   and 255 inclusive (in which case they must be taken modulo 2^8).

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)

   at (128)

   LOAD (byte_copy_left, 128)
   LOAD (byte_copy_right, 129)
   MEMSET (64, 129, 0, 1)
   MEMSET (129, 15, 64, 15)

   OUTPUT (128, 16)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   The expected output of the code is 0x8040 4f5e 6d7c 8b9a a9b8 c7d6
   e5f4 0312.  Executing the code is expected to cost 166 UDVM cycles.

2.9.  CRC

   This section gives assembly code to test the CRC instruction.  The
   code does not test any specific boundary cases (as there do not
   appear to be any) but focuses instead on verifying the CRC algorithm.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :crc_value                      pad (2)
   :crc_string_a                   pad (24)
   :crc_string_b                   pad (20)





Price et al.                                                   [Page 10]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   at (128)

   MEMSET (crc_string_a, 24, 1, 1)
   MEMSET (crc_string_b, 20, 128, 1)

   INPUT-BYTES (2, crc_value, !)

   CRC ($crc_value, crc_string_a, 44, !)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   If the compressed message is 0x62cb then the code should successfully
   terminate with no output, and with a total execution cost of 95 UDVM
   cycles.  For different 2-byte compressed messages the code should
   terminate with a decompression failure.

2.10.  INPUT-BITS

   This section gives assembly code to test the INPUT-BITS instruction.
   The code is designed to test that the following boundary cases have
   been correctly implemented:

   1. The INPUT-BITS instruction changes between any of the four
   possible bit orderings defined by the input_bit_order register.

   2. The INPUT-BITS instruction inputs 0 bits.

   3. The INPUT-BITS instruction requests data that lies beyond the end
   of the compressed message.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :input_bit_order                pad (2)
   :result                         pad (2)

   at (128)

   :start

   INPUT-BITS ($input_bit_order, result, end_of_message)
   OUTPUT (result, 2)

   ADD ($input_bit_order, 1)
   REMAINDER ($input_bit_order, 7)
   ADD ($input_bit_order, 1)
   JUMP (start)

   :end_of_message

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)




Price et al.                                                   [Page 11]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   An example compressed message is 0x932e ac71, which decompresses to
   give the output 0x0000 0002 0002 0013 0000 0003 001a 0038.  Executing
   the code should cost 66 UDVM cycles.

2.11.  INPUT-HUFFMAN

   This section gives assembly code to test the INPUT-HUFFMAN
   instruction.  The code is designed to test that the following
   boundary cases have been correctly implemented:

   1. The INPUT-HUFFMAN instruction changes between any of the four
   possible bit orderings defined by the input_bit_order register.

   2. The INPUT-HUFFMAN instruction inputs 0 bits.

   3. The INPUT-HUFFMAN instruction requests data that lies beyond the
   end of the compressed message.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :input_bit_order                pad (2)
   :result                         pad (2)

   at (128)

   :start

   INPUT-HUFFMAN (result, end_of_message, 2, $input_bit_order, 0,
   $input_bit_order, $input_bit_order, $input_bit_order, 0, 65535, 0)
   OUTPUT (result, 2)

   ADD ($input_bit_order, 1)
   REMAINDER ($input_bit_order, 7)
   ADD ($input_bit_order, 1)
   JUMP (start)

   :end_of_message

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   An example compressed message is 0x932e ac71 66d8 6f, which
   decompresses to give the output 0x0000 0003 0008 04d7 0002 0003 0399
   30fe.  Executing the code should cost 84 UDVM cycles.

2.12.  INPUT-BYTES

   This section gives assembly code to test the INPUT-BYTES instruction.
   The code is designed to test that the following boundary cases have
   been correctly implemented:




Price et al.                                                   [Page 12]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   1. The INPUT-BYTES instruction inputs 0 bytes.

   2. The INPUT-BYTES instruction requests data that lies beyond the end
   of the compressed message.

   3. The INPUT-BYTES instruction is used after part of a byte has been
   inputted (e.g. by the INPUT-BITS instruction).

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :input_bit_order                pad (2)
   :result                         pad (2)
   :output_start                   pad (4)
   :output_end

   at (128)

   LOAD (byte_copy_left, output_start)
   LOAD (byte_copy_right, output_end)

   :start

   INPUT-BITS ($input_bit_order, result, end_of_message)
   OUTPUT (result, 2)

   ADD ($input_bit_order, 2)
   REMAINDER ($input_bit_order, 7)

   INPUT-BYTES ($input_bit_order, output_start, end_of_message)
   OUTPUT (output_start, $input_bit_order)

   ADD ($input_bit_order, 1)
   JUMP (start)

   :end_of_message

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   An example compressed message is 0x932e ac71 66d8 6fb1 592b dc9a 9734
   d847 a733 874e 1bcb cd51 b5dc 9659 9d6a, which decompresses to give
   the output 0x0000 932e 0001 b166 d86f b100 1a2b 0003 9a97 34d8 0007
   0001 3387 4e00 08dc 9651 b5dc 9600 599d 6a.  Executing the code
   should cost 130 UDVM cycles.

2.13.  Stack manipulation

   This section gives assembly code to test the PUSH, POP, CALL and
   RETURN instructions.  The code is designed to test that the following
   boundary cases have been correctly implemented:




Price et al.                                                   [Page 13]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   1. The stack manipulation instructions overwrite the UDVM register
   stack_location.

   2. The stack manipulation instructions overwrite themselves.

   3. The CALL instruction specifies a reference operand rather than an
   absolute value.

   4. The PUSH instruction pushes the value contained in stack_fill onto
   the stack.

   5. The stack_location register contains an odd integer.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :input_bit_order                pad (2)
   :stack_location                 pad (2)
   :next_address                   pad (2)

   at (128)

   LOAD (stack_location, 64)
   PUSH (2)
   PUSH ($64)
   PUSH (66)

   OUTPUT (64, 8)

   POP (64)
   POP ($stack_location)
   POP (stack_location)

   OUTPUT (64, 8)
   JUMP (address_a)

   at (192)

   :address_a

   LOAD (stack_location, 32)
   LOAD (next_address, address_c)
   SUBTRACT ($next_address, address_b)
   CALL (address_b)

   at (256)

   :address_b

   CALL ($next_address)




Price et al.                                                   [Page 14]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   at (320)

   :address_c

   LOAD (stack_location, 383)
   LOAD (383, 26)
   MULTILOAD (432, 3, 1, 49153, 32768)
   RETURN

   at (448)

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   The expected output of the code is 0x0003 0002 0001 0042 0042 0000
   0001 0001, and a total of 40 UDVM cycles are expected to be used.

2.14.  Program flow

   This section gives assembly code to test the JUMP, COMPARE and SWITCH
   instructions.  The code is designed to test that the following
   boundary cases have been correctly implemented:

   1. The address operands are specified as references to memory
   addresses rather than as absolute values.

   at (64)

   :next_address                   pad (2)
   :counter                        pad (1)
   :counter_lsb                    pad (1)
   :switch_counter                 pad (2)

   at (128)

   LOAD (switch_counter, 4)

   :address_a

   LOAD (next_address, address_c)
   SUBTRACT ($next_address, address_b)
   OUTPUT (counter_lsb, 1)

   :address_b

   JUMP ($next_address)

   :address_c

   ADD ($counter, 1)
   LOAD (next_address, address_a)




Price et al.                                                   [Page 15]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   SUBTRACT ($next_address, address_d)
   OUTPUT (counter_lsb, 1)

   :address_d

   COMPARE ($counter, 6, $next_address, address_c, address_e)

   :address_e

   SUBTRACT ($switch_counter, 1)
   LOAD (next_address, address_a)
   SUBTRACT ($next_address, address_f)
   OUTPUT (counter_lsb, 1)

   :address_f

   SWITCH (4, $switch_counter, address_g, $next_address, address_c,
   address_e)

   :address_g

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   The expected output of the code is 0x0001 0102 0203 0304 0405 0506
   0707 0708 0808 0909, and a total of 131 UDVM cycles are expected to
   be used.

2.15.  State creation

   This section gives assembly code to test the STATE-CREATE and STATE-
   FREE instructions.  The code is designed to test that the following
   boundary cases have been correctly implemented:

   1. An item of state is created that duplicates an existing state
   item.

   2. An item of state is freed when the state has not been created.

   3. An item of state is created and then freed by the same message.

   4. The STATE-FREE instruction frees a state item by sending fewer
   bytes of state_identifier than the minimum_access_length.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :states                         pad (1)
   :states_lsb                     pad (1)

   set (state_length, 10)




Price et al.                                                   [Page 16]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   at (128)

   INPUT-BYTES (1, states_lsb, !)

   :test_one

   LSHIFT ($states, 13)
   COMPARE ($states, 32768, test_two, create_state_a, create_state_a)

   :create_state_a

   STATE-CREATE (state_length, state_address, 0, 20, 0)

   :test_two

   LSHIFT ($states, 1)
   COMPARE ($states, 32768, test_three, free_state, free_state)

   :free_state

   STATE-FREE (state_identifier, 6)

   :test_three

   LSHIFT ($states, 1)
   COMPARE ($states, 32768, end, create_state_b, create_state_b)

   :create_state_b

   END-MESSAGE (0, 0, state_length, state_address, 0, 20, 0)

   :end

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   at (512)

   :state_address

   byte (34, 162, 6, 4, 22, 224, 116, 101, 115, 116)

   :state_identifier

   byte (32, 84, 55, 65, 83, 248, 254, 122, 106, 151, 203, 121, 224, 24,
   194, 221, 214, 143, 254, 155)

   Upon reaching the END-MESSAGE instruction the UDVM does not output
   any decompressed data, but instead may make one or more state
   creation or state free requests to the state handler.  Assuming that
   the application does not veto the state creation request (and that




Price et al.                                                   [Page 17]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   sufficient state memory is available) the code should result in
   either 0 or 1 new state items being created in the chosen
   compartment.

   The following table lists eight different 1-byte compressed messages
   and whether the message should cause a new state item to be created
   in the compartment.  The number of UDVM cycles required to execute
   the code is also given:

    Compressed message:     State item in compartment:     UDVM cycles:

         0x00                          No                       9
         0x01                          Yes                      19
         0x02                          No                       10
         0x03                          Yes                      20
         0x04                          Yes                      20
         0x05                          Yes                      30
         0x06                          No                       21
         0x07                          Yes                      31

2.16.  STATE-ACCESS

   This section gives assembly code to test the STATE-ACCESS
   instruction.  The code is designed to test that the following
   boundary cases have been correctly implemented:

   1. A subset of the bytes contained in a state item is copied to the
   UDVM memory.

   2. Bytes are copied from beyond the end of the state value.

   3. The state_instruction operand is set to 0.

   4. The state cannot be accessed because the partial state identifier
   is too short.

   5. The state identifier is overwritten by the state item being
   accessed.

   The code assumes that the state item created in the previous section
   is available to the state handler.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :type                           pad (1)
   :type_lsb                       pad (1)
   :state_value                    pad (4)

   at (128)




Price et al.                                                   [Page 18]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   INPUT-BYTES (1, type_lsb, !)
   COMPARE ($type, 1, execute_state, extract_state, error_conditions)

   :execute_state

   STATE-ACCESS (state_identifier, 20, 0, 0, 0, 512)

   :extract_state

   STATE-ACCESS (state_identifier, 20, 6, 4, state_value, 0)
   OUTPUT (state_value, 4)
   JUMP (end)

   :error_conditions

   COMPARE ($type, 3, state_not_found, id_too_short, state_too_short)

   :state_not_found

   STATE-ACCESS (128, 20, 0, 0, 0, 0)
   JUMP (end)

   :id_too_short

   STATE-ACCESS (state_identifier, 19, 6, 4, state_value, 0)
   JUMP (end)

   :state_too_short

   STATE-ACCESS (state_identifier, 20, 6, 5, state_value, 0)
   JUMP (end)

   at (484)

   :end

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   at (512)

   :state_identifier

   byte (32, 84, 55, 65, 83, 248, 254, 122, 106, 151, 203, 121, 224, 24,
   194, 221, 214, 143, 254, 155)

   If the compressed message is 0x00 then the expected output of the
   code is 0x7465 7374 and a total of 21 UDVM cycles are expected to be
   used.  If the compressed message is 0x01 then the code should also
   output 0x7465 7374 but in this case using a total of 15 UDVM cycles.





Price et al.                                                   [Page 19]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   If the compressed message is 0x03, 0x04 or 0x05 then decompression
   failure should occur.

3.  Torture tests for dispatcher

   The following sections give code to test the various functions of the
   SigComp dispatcher.

3.1.  Useful Values

   This section gives assembly code to test that the SigComp "Useful
   Values" are correctly initialized in the UDVM memory.  It also tests
   that the UDVM is correctly terminated if the bytecode uses too many
   UDVM cycles or tries to write beyond the end of the available memory.

   The code tests that the following boundary cases have been correctly
   implemented:

   1. The bytecode uses exactly as many UDVM cycles as are available (in
   which case no problems should arise) or one cycle too many (in which
   case decompression failure should occur).

   2. The bytecode writes to the highest memory address available (in
   which case no problems should arise) or to the memory address
   immediately following the highest available address (in which case
   decompression failure should occur).

   :udvm_memory_size               pad (2)
   :cycles_per_bit                 pad (2)
   :sigcomp_version                pad (2)
   :partial_state_id_length        pad (2)
   :state_length                   pad (2)

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :remaining_cycles               pad (2)
   :check_memory                   pad (1)
   :check_memory_lsb               pad (1)
   :check_cycles                   pad (1)
   :check_cycles_lsb               pad (1)

   at (128)

   LOAD (byte_copy_left, 32)
   LOAD (byte_copy_right, 33)

   :test_version

   COMPARE ($sigcomp_version, 1, !, test_state_access, !)




Price et al.                                                   [Page 20]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   :test_state_access

   COMPARE ($partial_state_id_length, 0, !, test_length_equals_zero,
   test_state_length)

   :test_length_equals_zero

   COMPARE ($state_length, 0, !, end, !)

   :test_state_length

   COMPARE ($state_length, 960, !, test_udvm_memory, !)

   :test_udvm_memory

   INPUT-BYTES (1, check_memory_lsb, !)
   ADD ($check_memory, $udvm_memory_size)
   SUBTRACT ($check_memory, 1)
   COPY (32, 1, $check_memory)

   :test_udvm_cycles

   INPUT-BYTES (1, check_cycles_lsb, !)

   ; total_UDVM_cycles = cycles_per_bit * (8 * message_size + 1000)
   ;
   ;       = cycles_per_bit * (8 * (partial_state_id_length + 3) + 1000)

   LOAD (remaining_cycles, $partial_state_id_length)
   ADD ($remaining_cycles, 3)
   MULTIPLY ($remaining_cycles, 8)
   ADD ($remaining_cycles, 1000)
   MULTIPLY ($remaining_cycles, $cycles_per_bit)
   ADD ($remaining_cycles, $check_cycles)

   set (cycles_used_by_bytecode, 982)

   SUBTRACT ($remaining_cycles, cycles_used_by_bytecode)
   COPY (32, $remaining_cycles, 32)

   :end

   END-MESSAGE (0, 0, 960, 64, 128, 6, 0)

   The bytecode must be executed a total of four times in order to fully
   test the SigComp Useful Values.  In the first case the bytecode
   should be uploaded as part of the SigComp message (no compressed data
   is required in this case).  This should cause the UDVM to request
   creation of a new state item, and should use a total of 966 UDVM
   cycles.




Price et al.                                                   [Page 21]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   Subsequent tests should access this state by uploading the state
   identifier as part of the SigComp message.  Note that the SigComp
   message should not contain a returned feedback item (as this would
   cause the bytecode to calculate the total number of available UDVM
   cycles incorrectly).

   A 2-byte compressed message is required for the second and subsequent
   cases: if the message is 0x0000 then the UDVM should successfully
   terminate using exactly the number of available UDVM cycles.
   However, if the message is 0x0001 then the UDVM should use too many
   cycles and hence terminate with decompression failure.  Furthermore
   if the message is 0x0100 then decompression failure should occur
   because the UDVM attempts to write beyond its available memory.

3.2.  Message-based transport

   This section provides a set of messages to test the SigComp header
   over a message-based transport such as UDP.  The messages test that
   the following boundary cases have been correctly implemented:

   1. The UDVM bytecode is copied to different areas of the UDVM memory.

   2. The decompression memory size is set to an incorrect value.

   3. The SigComp message is too short.

   4. The destination address is invalid.

   The basic version of the code used in the test is given below.  Note
   that the code is designed to calculate the decompression memory size
   based on the Useful Values provided to the UDVM:

   :udvm_memory_size               pad (2)
   :cycles_per_bit                 pad (2)
   :sigcomp_version                pad (2)
   :partial_state_id_length        pad (2)
   :state_length                   pad (2)

   at (128)

   :code_start

   ADD ($udvm_memory_size, total_message_size)
   OUTPUT (udvm_memory_size, 2)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 1)

   :code_end

   set (header_size, 3)
   set (code_size, (code_end - code_start))
   set (total_message_size, (header_size + code_size))




Price et al.                                                   [Page 22]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   A number of complete SigComp messages are given below, each
   containing some or all of the above code.  In each case it is
   indicated whether the message should successfully output the
   decompression memory size or whether it should cause a decompression
   failure to occur (together with the reason for the failure):

   SigComp message:                Effect:

   0xf8                            Fails (message too short)

   0xf800                          Fails (message too short)

   0xf800 e106 0011 2200 0223      Outputs the decompression_memory_size
   0x0000 0000 0000 01

   0xf800 f106 0011 2200 0223      Fails (message too short)
   0x0000 0000 0000 01

   0xf800 e006 0011 2200 0223      Fails (invalid destination address)
   0x0000 0000 0000 01

   0xf800 ee06 0011 2200 0223      Outputs the decompression_memory_size
   0x0000 0000 0000 01

   The messages should be decompressed in the order given to check that
   an error in one message does not interfere with the successful
   decompression of subsequent messages.

   The two messages that successfully decompress should each use a total
   of 5 UDVM cycles.

3.3.  Stream-based transport

   This section provides a byte stream to test the SigComp header and
   delimiters over a stream-based transport such as TCP.  The byte
   stream tests all of the boundary cases covered in Section 3.2, as
   well as the following cases specific to stream-based transports:

   1. Quoted bytes are used by the record marking scheme.

   2. Multiple delimiters are used between the same pair of messages.

   3. Unnecessary delimiters are included at the start of the stream.

   The basic version of the code used in the test is given below.  Note
   that the code is designed to calculate the decompression memory size
   based on the Useful Values provided to the UDVM:

   :udvm_memory_size               pad (2)
   :cycles_per_bit                 pad (2)
   :sigcomp_version                pad (2)




Price et al.                                                   [Page 23]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   :partial_state_id_length        pad (2)
   :state_length                   pad (2)

   at (128)

   MULTIPLY ($udvm_memory_size, 2)
   OUTPUT (udvm_memory_size, 2)
   OUTPUT (test_record_marking, 5)
   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   :test_record_marking

   byte (255, 255, 255, 255, 255)

   The above assembly code has been compiled and used to generate the
   following byte stream:

   0xffff f801 7108 0002 2200 0222 a092 0523 0000 0000 0000 00ff 00ff
   0x03ff ffff ffff ffff f801 7e08 0002 2200 0222 a3d2 0523 0000 0000
   0x0000 00ff 04ff ffff ffff ffff ffff ff

   Note that this byte stream can be divided into five distinct portions
   (two SigComp messages and three sets of delimiters) as illustrated
   below:

   Portion of byte stream:                                Meaning:

   0xffff                                                 Delimiter

   0xf801 7108 0002 2200 0222 a092 0523                   First message
   0x0000 0000 0000 00ff 00ff 03ff ffff

   0xffff ffff                                            Delimiter

   0xf801 7e08 0002 2200 0222 a3d2 0523                   Second message
   0x0000 0000 0000 00ff 04ff ffff ff

   0xffff ffff ffff                                       Delimiter

   When the complete byte stream is supplied to the decompressor
   dispatcher, the record marking scheme should use the delimiters to
   partition the stream into two distinct SigComp messages.  Both of
   these messages should successfully output the decompression memory
   size (as a 2-byte value), followed by five consecutive 0xff bytes to
   test that the record marking scheme is working correctly.  A total of
   11 UDVM cycles should be used in each case.

   It must also be checked that the dispatcher can handle the same error
   cases as covered in Section 3.2.  Each of the following byte streams
   should cause a decompression failure to occur for the reason stated:





Price et al.                                                   [Page 24]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   Byte stream:                                      Reason for failure:

   0xf8ff ff                                         Message too short

   0xf800 ffff                                       Message too short

   0xf801 8108 0002 2200 0222 a092 0523 ffff         Message too short
   0x0000 0000 0000 00ff 00ff 03ff ffff

   0xf801 7008 0002 2200 0222 a092 0523 ffff         Invalid destination
   0x0000 0000 0000 00ff 04ff ffff ff

   Note that when a decompression failure occurs it is an implementation
   decision whether to close the entire stream or whether to ignore the
   error and attempt to decompress subsequent messages in the stream.

4.  Torture tests for state handler

   The following sections give code to test the various functions of the
   SigComp state handler.

4.1.  SigComp feedback mechanism

   This section gives assembly code to test the SigComp feedback
   mechanism.  The code is designed to test that the following boundary
   cases have been correctly implemented:

   1. Both the short and the long versions of the SigComp feedback item
   are used.

   2. The chain of returned SigComp parameters is terminated by a non-
   zero value.

   at (64)

   :type                           pad (1)
   :type_lsb                       pad (1)

   :requested_feedback_location    pad (1)
   :requested_feedback_length      pad (1)
   :requested_feedback_bytes       pad (127)

   :returned_parameters_location   pad (2)
   :length_of_partial_state_id_a   pad (1)
   :partial_state_identifier_a     pad (6)
   :length_of_partial_state_id_b   pad (1)
   :partial_state_identifier_b     pad (12)
   :length_of_partial_state_id_c   pad (1)
   :partial_state_identifier_c     pad (20)
   :terminate_returned_parameters  pad (1)





Price et al.                                                   [Page 25]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   align (128)

   set (q_bit, 1)
   set (s_bit, 0)
   set (i_bit, 0)
   set (flags, (((4 * q_bit) + (2 * s_bit)) + i_bit))

   INPUT-BYTES (1, type_lsb, !)
   COMPARE ($type, 1, short_feedback_item, long_feedback_item, !)

   :short_feedback_item

   set (requested_feedback_data, 127)
   set (short_feedback_value, ((flags * 256) + requested_feedback_data))

   LOAD (requested_feedback_location, short_feedback_value)
   JUMP (return_sigcomp_parameters)

   :long_feedback_item

   set (requested_feedback_field, 255)
   set (long_feedback_value, ((flags * 256) + requested_feedback_field))

   LOAD (requested_feedback_location, long_feedback_value)
   MEMSET (requested_feedback_bytes, 127, 1, 1)

   :return_sigcomp_parameters

   set (cpb, 0)
   set (dms, 1)
   set (sms, 0)
   set (sigcomp_version, 1)

   set (parameters_msb, (((64 * cpb) + (8 * dms)) + sms))
   set (sigcomp_parameters, ((256 * parameters_msb) + sigcomp_version))

   LOAD (returned_parameters_location, sigcomp_parameters)

   LOAD (length_of_partial_state_id_a, 1536)
   LOAD (length_of_partial_state_id_b, 3072)
   LOAD (length_of_partial_state_id_c, 5120)
   LOAD (terminate_returned_parameters, 5376)

   MEMSET (partial_state_identifier_a, 6, 0, 1)
   MEMSET (partial_state_identifier_b, 12, 0, 1)
   MEMSET (partial_state_identifier_c, 20, 0, 1)

   END-MESSAGE (requested_feedback_location,
   returned_parameters_location, 0, 0, 0, 0, 0)






Price et al.                                                   [Page 26]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   When the above code is executed it supplies a requested feedback item
   to the state handler.  If the compressed message is 0x00 then the
   short (1-byte) version of the feedback is used.  Assuming that the
   feedback request is successful the feedback item should be returned
   in the first SigComp message to be sent in the reverse direction.
   The SigComp message returning the feedback should begin as follows:

   +---+---+---+---+---+---+---+---+
   | 1   1   1   1   1   1 |   X   |   first header byte
   +---+---+---+---+---+---+---+---+
   | 0 |            127            |   returned feedback field
   +---+---+---+---+---+---+---+---+

   So the first 2 bytes of the returning SigComp message should be
   0xfn7f where n = c, d, e or f (the choice of n is determined by the
   compressor generating the returning SigComp message, which is not
   under the control of the above code).  Executing the bytecode in this
   case should cost a total of 52 UDVM cycles.

   If the compressed message is 0x01 then the long version of the
   feedback item is used.  In this case the SigComp message returning
   the feedback should begin as follows:

   +---+---+---+---+---+---+---+---+
   | 1   1   1   1   1   1 |   X   |   first header byte
   +---+---+---+---+---+---+---+---+
   | 1 |            127            |   returned feedback length
   +---+---+---+---+---+---+---+---+
   |               1               |              ^
   +---+---+---+---+---+---+---+---+              |
   |               2               |              |
   +---+---+---+---+---+---+---+---+
   |               3               |   returned feedback field
   +---+---+---+---+---+---+---+---+
             :           :                        |
   +---+---+---+---+---+---+---+---+              |
   |              127              |              v
   +---+---+---+---+---+---+---+---+

   So the first 129 bytes of the SigComp message should be 0xfnff 0102
   0304 ... 7e7f where n = c, d, e or f.  Executing the bytecode in this
   case should cost a total of 179 UDVM cycles.

   As well as testing the requested and returned feedback items, the
   above code also announces values for each of the SigComp parameters.
   The supplied version of the code announces only the minimum possible
   values for the cycles_per_bit, decompression_memory_size,
   state_memory_size and SigComp_version (although this can easily be
   adjusted to test different values for these parameters).






Price et al.                                                   [Page 27]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   The code should also announce the availability of state items with
   the following partial state identifiers:

   0x0001 0203 0405
   0x0001 0203 0405 0607 0809 0a0b
   0x0001 0203 0405 0607 0809 0a0b 0c0d 0e0f 1011 1213

   Note that different implementations may make use of the announcement
   information in different ways.  It is a valid implementation choice
   to simply ignore all of the announcement data and use only the
   minimum resources that are guaranteed to be available to all
   endpoints.  However the above code is useful for checking that an
   endpoint interprets the announcement data correctly (in particular
   ensuring that it does not mistakenly use resources that have not in
   fact been announced).

4.2.  State memory management

   The following section gives assembly code to test the memory
   management features of the state handler.  The code checks that the
   correct states are retained by the state handler when insufficient
   memory is available to store all of the requested states.

   The code is designed to test that the following boundary cases have
   been correctly implemented:

   1. A state item is created that exceeds the total state_memory_size
   for the compartment.

   2. States are created with a non-zero state_retention_priority.

   3. A new state item is created that has a lower
   state_retention_priority than existing state items in the
   compartment.

   For the duration of this test it is assumed that all states will be
   saved in a single compartment with a state_memory_size of 2048 bytes.

   at (64)

   :byte_copy_left                 pad (2)
   :byte_copy_right                pad (2)
   :order                          pad (2)
   :type                           pad (1)
   :type_lsb                       pad (1)
   :state_length                   pad (2)
   :state_retention_priority       pad (2)

   at (128)

   MULTILOAD (byte_copy_left, 2, state_start, order_data)




Price et al.                                                   [Page 28]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   INPUT-BYTES (1, type_lsb, !)
   COMPARE ($type, 5, general_test, large_state, verify_state)

   :general_test

   COMPARE ($type, 3, start, state_present, state_not_present)

   :start

   MULTIPLY ($type, 6)
   ADD ($type, order_data)
   LOAD (order, $type)
   ADD ($type, 6)

   :loop

   COPY ($order, 2, state_retention_priority)
   COMPARE ($order, $type, continue, end, !)

   :continue

   LOAD (state_length, $state_retention_priority)
   MULTIPLY ($state_length, 256)
   STATE-CREATE ($state_length, state_start, 0, 6,
   $state_retention_priority)

   ADD ($order, 2)
   JUMP (loop)

   :state_present

   STATE-ACCESS (state_identifier_a, 6, 0, 0, 0, 0)
   STATE-ACCESS (state_identifier_b, 6, 0, 0, 0, 0)
   STATE-ACCESS (state_identifier_c, 6, 0, 0, 0, 0)
   STATE-ACCESS (state_identifier_e, 6, 0, 0, 0, 0)
   JUMP (end)

   :state_not_present

   STATE-ACCESS (state_identifier_d, 6, 0, 0, 0, 0)
   JUMP (end)

   :large_state

   STATE-CREATE (2048, state_start, 0, 6, 0)
   JUMP (end)

   :verify_state

   STATE-ACCESS (large_state_identifier, 6, 0, 0, 0, 0)
   JUMP (end)




Price et al.                                                   [Page 29]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   :end

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   at (512)

   :state_start

   byte (116, 101, 115, 116)

   :order_data

   word (0, 1, 2, 3, 4, 3, 2, 1, 0)

   :state_identifier_a

   byte (142, 234, 75, 67, 167, 135)

   :state_identifier_b

   byte (249, 1, 14, 239, 86, 123)

   :state_identifier_c

   byte (35, 154, 52, 107, 21, 166)

   :state_identifier_d

   byte (180, 15, 192, 228, 77, 44)

   :state_identifier_e

   byte (212, 162, 33, 71, 230, 10)

   :large_state_identifier

   byte (239, 242, 188, 15, 182, 175)

   The above code must be executed a total of 7 times in order to
   complete the test.  Each time the code is executed a 1-byte
   compressed message should be provided, taking the values 0x00 to 0x06
   in ascending order (so the compressed message should be 0x00 the
   first time the code is run, 0x01 the second and so on).

   When the compressed message is 0x00, 0x01 or 0x02 the code makes
   three state creation requests per message, establishing a total of
   nine states in the compartment.  Note however that as new states are
   created some of the existing states should be pushed out of the
   compartment due to lack of memory.





Price et al.                                                   [Page 30]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   When the compressed message is 0x03 the code checks that the correct
   state items remain in the compartment.  Decompression should
   successfully terminate in this case.

   When the compressed message is 0x04 the code attempts to access a
   state that has been pushed out of the compartment by states of higher
   priority.  Decompression failure should occur in this case because
   the relevant state is no longer available.

   When the compressed message is 0x05 the code attempts to create a
   state that is larger than the entire compartment.  In this case the
   state handler should save only the first part of the requested state.

   When the compressed message is 0x06 the code verifies that the first
   part of the large state item created by the previous message has been
   successfully saved.

   The cost in UDVM cycles for each compressed message is given below
   (except for message 0x04 where decompression failure should occur):

   Compressed message:   0x00   0x01   0x02   0x03   0x04   0x05   0x06

   Cost in UDVM cycles:   811   2603    811   1805    N/A   2057   1993

4.3.  Multiple compartments

   This section gives assembly code to test the interaction between
   multiple SigComp compartments.  The code is designed to test that the
   following boundary cases have been correctly implemented:

   1. The same state item is saved in more than one compartment.

   2. A state item stored in multiple compartments has the same state
   identifier but a different state_retention_priority in each case.

   3. A state item is deleted from one compartment but still belongs to
   a different compartment.

   4. A state item belonging to multiple compartments is deleted from
   every compartment to which it belongs.

   The test requires a total of three compartments to be available,
   which will be referred to as Compartment 0, Compartment 1 and
   Compartment 2.  Each of the three compartments should have a
   state_memory_size of 2048 bytes.

   The assembly code for the test is given below:

   at (64)

   :byte_copy_left                 pad (2)




Price et al.                                                   [Page 31]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   :byte_copy_right                pad (2)
   :type                           pad (1)
   :type_lsb                       pad (1)

   at (128)

   MULTILOAD (byte_copy_left, 2, state_start, state_end)
   INPUT-BYTES (1, type_lsb, !)
   COMPARE ($type, 3, create_state, overwrite_state, temp)

   :temp

   COMPARE ($type, 5, overwrite_state, access_state, error_conditions)

   :create_state

   ADD ($type, state_start)
   STATE-CREATE (448, $type, 0, 6, 0)

   :duplicate_state

   ADD ($type, 3)
   STATE-CREATE (448, $type, 0, 6, 0)

   SUBTRACT ($type, temp_one)
   REMAINDER ($type, 3)
   ADD ($type, temp_two)
   STATE-CREATE (448, $type, 0, 6, 0)

   :common_state

   STATE-CREATE (448, temp_three, 0, 6, $type)
   JUMP (end)

   :overwrite_state

   STATE-CREATE (1984, 32, 0, 6, 0)
   JUMP (end)

   :access_state

   STATE-ACCESS (state_identifier_c, 6, 0, 0, 0, 0)
   STATE-ACCESS (state_identifier_d, 6, 0, 0, 0, 0)
   STATE-ACCESS (state_identifier_f, 6, 0, 0, 0, 0)
   STATE-ACCESS (state_identifier_g, 6, 0, 0, 0, 0)

   :end

   END-MESSAGE (0, 0, 0, 0, 0, 0, 0)

   :error_conditions




Price et al.                                                   [Page 32]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   COMPARE ($type, 7, access_a, access_b, access_e)

   :access_a

   STATE-ACCESS (state_identifier_a, 6, 0, 0, 0, 0)
   JUMP (end)

   :access_b

   STATE-ACCESS (state_identifier_b, 6, 0, 0, 0, 0)
   JUMP (end)

   :access_e

   STATE-ACCESS (state_identifier_e, 6, 0, 0, 0, 0)
   JUMP (end)

   at (512)

   :state_start

   byte (0, 1, 2, 3, 4, 5, 6)

   :state_end

   set (temp_one, (state_start + 2))
   set (temp_two, (state_start + 3))
   set (temp_three, (state_end - 1))

   :state_identifier_a

   byte (172, 166, 11, 142, 178, 131)

   :state_identifier_b

   byte (157, 191, 175, 198, 61, 210)

   :state_identifier_c

   byte (52, 197, 217, 29, 83, 97)

   :state_identifier_d

   byte (189, 214, 186, 42, 198, 90)

   :state_identifier_e

   byte (71, 194, 24, 20, 238, 7)

   :state_identifier_f




Price et al.                                                   [Page 33]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003



   byte (194, 117, 148, 29, 215, 161)

   :state_identifier_g

   byte (72, 135, 156, 141, 233, 14)

   The above code must be executed a total of 9 times in order to
   complete the test.  Each time the code is executed a 1-byte
   compressed message N should be provided, taking the values 0x00 to
   0x08 in ascending order (so the compressed message should be 0x00 the
   first time the code is run, 0x01 the second and so on).

   If the code makes a state creation request then the state must be
   saved in Compartment (N modulo 3).

   When the compressed message is 0x00, 0x01 or 0x02 the code makes four
   state creation requests in compartments 0, 1 and 2 respectively.
   This creates a total of seven distinct state items referred to as
   State A through to State G.  The states should be distributed amongst
   the three compartments as illustrated in Figure 1 (note that some
   states belong to more than one compartment).

   When the compressed message is 0x03 or 0x04 the code overwrites all
   of the states in compartments 0 and 1 respectively.  This means that
   states A, B and E should be unavailable because they are no longer
   present in any of the three compartments.

   When the compressed message is 0x05 the code checks that the states
   C, D, F and G are still available.  Decompression should successfully
   terminate in this case.

   When the compressed message is 0x06, 0x07 or 0x08 the code attempts
   to access states A, B and E respectively.  Decompression failure
   should occur in this case because the relevant states are no longer
   available.

   The cost in UDVM cycles for each compressed message is given below
   (except for messages 0x06, 0x07 and 0x08 where decompression failure
   is expected to occur):

   Compressed message:  0x00 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08

   Cost in UDVM cycles: 1809 1809 1809 1993 1994 1804  N/A  N/A  N/A











Price et al.                                                   [Page 34]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


                     +-----------------------------+
                     |        Compartment 0        |
                     |                             |
                     |                             |
                     |           State A           |
                     |                             |
                     |         +-------------------+---------+
                     |         |                   |         |
                     |         |                   |         |
                     |         |           State D |         |
                     |         |                   |         |
                     |         |                   |         |
           +---------+---------+---------+         |         |
           |         |         |         |         |         |
           |         |         |         |         |         |
           |         | State E | State G |         | State C |
           |         |         |         |         |         |
           |         |         |         |         |         |
           |         +---------+---------+---------+         |
           |                   |         |                   |
           |                   |         |                   |
           |           State B | State F |                   |
           |                   |         |                   |
           |                   |         |   Compartment 2   |
           |                   +---------+-------------------+
           |                             |
           |                             |
           |                             |
           |                             |
           |        Compartment 1        |
           +-----------------------------+

            Figure 1: States created in the three compartments

5.  Security considerations

   This draft describes implementation options for the SigComp protocol
   [RFC-3320].  Consequently the security considerations for this draft
   match those of SigComp.

6.  Authors' addresses

   Richard Price        Tel: +44 1794 833681
   Email:               richard.price@roke.co.uk

   Abigail Surtees      Tel: +44 1794 833131
   Email:               abigail.surtees@roke.co.uk

   Roke Manor Research Ltd
   Romsey, Hants, SO51 0ZN
   United Kingdom




Price et al.                                                   [Page 35]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


7.  References

   [USERGUIDE] "SigComp User Guide", R. Price et al.,
               <draft-price-rohc-sigcomp-user-guide-01.txt>, October
               2002

   [RFC-2026]  "The Internet Standards Process - Revision 3", Scott
               Bradner, Internet Engineering Task Force, October 1996

   [RFC-2119]  "Key words for use in RFCs to Indicate Requirement
               Levels", Scott Bradner, Internet Engineering Task Force,
               March 1997

   [RFC-3320]  "Signaling Compression (SigComp)", R. Price et al.,
               Internet Engineering Task Force, January 2003








































Price et al.                                                   [Page 36]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


Appendix A: UDVM bytecode for the torture tests

   The following sections list the raw UDVM bytecode generated for each
   test.  The bytecode is presented in the form of a complete SigComp
   message, including the appropriate header and any compressed message
   required by the code.

   In some cases the test is designed to be run several times with
   different compressed messages appended to the code; for each of these
   tests the first compressed message is always supplied.

   Note that the different assemblers can output different bytecode for
   the same piece of assembly code, so a valid assembler can produce
   results different from those presented below.  However, the following
   bytecode should always generate the same results on any UDVM.

A.1.1.  Bit manipulation

   0xf80a 7116 a07f 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x01c0 00ff 8055 5502 202a 0321 0420 0305 21ff 2286 0401 20c0 ff02
   0x2060 0320 0421 6005 2061 2286 0423

A.1.2.  Arithmetic

   0xf80a a11c 01a0 459f 9f07 2201 16a0 7600 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x06c0 00ff 9941 0720 0108 20a3 e909 20a0 650a 200b 2286 0406 21c0
   0xff07 2162 0821 6109 2061 0a21 6222 8604 2300

A.1.3.  Sorting

   0xf80d c10c 8802 170b 8802 1722 a12e 2d23 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0a00 0a00 1100 0700 1600 0300 0300 0300 1300 0100 1000 0e00
   0x0800 0200 0d00 1400 1200 1700 0f00 1500 0c00 0600 096e 6720 6975
   0x6920 7469 742c 2079 6f75 2720 5346 6f6e 6761 2075 7272 646f 2074
   0x6f6e 2e2e 0070 6570 206e 7472 656e 69








Price et al.                                                   [Page 37]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


A.1.4.  SHA-1

   0xf808 710d a0c3 03a0 4422 a044 140d a0c6 38a0 4422 a044 140e 86a0
   0xfe0e a042 a0ff 0da0 feff a044 22a0 4414 0e86 a0ff 0ea0 42a1 070d
   0xa0ff a280 a0ff 22a0 ff14 2300 0000 0000 0000 6162 6361 6263 6462
   0x6364 6563 6465 6664 6566 6765 6667 6866 6768 6967 6869 6a68 696a
   0x6b69 6a6b 6c6a 6b6c 6d6b 6c6d 6e6c 6d6e 6f6d 6e6f 706e 6f70 7161
   0x3031 3233 3435 3637

A.1.5.  LOAD and MULTILOAD

   0xf803 710e 87a0 840e a082 c080 0ec0 80a0 860e c084 c084 2287 081c
   0x01a0 419f 8908 2002 0620 3c0f 6003 a0a3 a0b2 870f 6004 2a87 c080
   0xc084 2287 0823 00

A.1.6.  COPY

   0xf801 e10e 208e 0e86 860e a042 8712 2087 210e 8680 4100 1286 a04c
   0xa041 2220 a06d 23

A.1.7.  COPY-LITERAL and COPY-OFFSET

   0xf802 f10e 208e 0e86 860e a042 870e a044 2113 2087 2213 a044 0822
   0x0e86 a042 0ea0 42a0 4a14 0806 220e 6301 1463 0522 2220 3023

A.1.8.  MEMSET

   0xf801 810e 8687 0ea0 42a0 8115 86a0 8100 0115 a081 0f86 0f22 8710
   0x23

A.1.9.  CRC

   0xf801 a115 a046 1801 0115 a05e 1487 011c 02a0 449f 931b 62a0 462c
   0x9f8d 2362 cb

A.1.10.  INPUT-BITS

   0xf801 511d 62a0 4614 22a0 4602 0622 010a 2207 0622 0116 ee23 932e
   0xac71

A.1.11.  INPUT-HUFFMAN

   0xf801 d11e a046 1c02 6200 6262 6200 ff00 22a0 4602 0622 010a 2207
   0x0622 0116 e623 932e ac71 66d8 6f

A.1.12.  INPUT-BYTES

   0xf802 710e 86a0 480e a042 a04c 1d62 a046 1d22 a046 0206 2202 0a22
   0x071c 62a0 480e 22a0 4862 0622 0116 e523 932e ac71 66d8 6fb1 592b
   0xdc9a 9734 d847 a733 874e 1bcb cd51 b5dc 9659 9d6a





Price et al.                                                   [Page 38]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


A.1.13.  Stack manipulation

   0xf814 110e a046 8610 0210 6010 a042 2286 0811 8611 6311 a046 2286
   0x0816 2800 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 000e a048 a140 0724 8818 3800
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0018 6400 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 000e a046 a17f 0ea1 7f1a 0fa1 b003
   0x0180 c001 8f19 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0023

A.1.14.  Program flow

   0xf803 f10e a044 040e 86a0 9207 20a0 9022 a043 0116 6006 2101 0e86
   0xa084 0720 a0a1 22a0 4301 1761 0660 f106 0722 010e 86a0 8407 20a0
   0xb622 a043 011a 0462 0860 9fdc f123

A.1.15.  State creation

   0xf819 e11c 01a0 459f 9f04 220d 1762 8f0c 0606 200a 8900 1400 0422
   0x0117 628f 0a06 0621 a20a 0604 2201 1762 8f0e 0606 2300 000a 8900
   0x1400 2300 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0022 a206
   0x0416 e074 6573 7420 5437 4153 f8fe 7a6a 97cb 79e0 18c2 ddd6 8ffe
   0x9b00

A.1.16.  STATE-ACCESS

   0xf819 411c 01a0 459f 9f17 6201 060d 1c1f 8914 0000 0089 1f89 1406
   0x04a0 4600 22a0 4604 16a1 4517 6203 0610 1b1f 8714 0000 0000 16a1
   0x351f 8913 0604 a046 0016 a12a 1f89 1406 05a0 4600 16a1 1f00 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000




Price et al.                                                   [Page 39]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0023 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0020 5437
   0x4153 f8fe 7a6a 97cb 79e0 18c2 ddd6 8ffe 9b00

A.2.1.  Useful Values

   0xf805 b10e 8620 0ea0 4221 1742 019f 9808 9f98 1743 009f 9007 0d17
   0x4400 fb3d fb17 44a3 c0fc 07fc 1c01 a047 f506 2340 0723 0112 2001
   0x631c 01a0 49e6 0ea0 4443 0622 0308 2208 0622 a3e8 0822 4106 2264
   0x0722 a3d6 1220 6220 2300 00a3 c086 8706 0000

A.2.2.  Message-based transport

   The bytecode for this test is given in Section 3.2.

A.2.3.  Stream-based transport

   The bytecode for this test is given in Section 3.3.

A.3.1.  SigComp feedback mechanism

   0xf805 031c 01a0 419f 1f17 6001 070e 9f19 0ea0 42a4 7f16 0e0e a042
   0xa4ff 15a0 44a0 7f01 010e a0c3 a801 0ea0 c5a6 000e a0cc ac00 0ea0
   0xd9b4 000e a0ee b500 15a0 c606 0001 15a0 cd0c 0001 15a0 da14 0001
   0x23a0 42a0 c300

A.3.2.  State memory management

   0xf81b a10f 8602 89a2 041c 01a0 479f 9917 6305 08a0 68a0 7017 6303
   0x0734 a056 0823 0606 23a2 040e a044 6306 2306 1262 02a0 4a17 6263
   0x08a0 589f 710e a048 6508 2488 2064 8900 0665 0622 0216 e31f a216
   0x0600 0000 001f a21c 0600 0000 001f a222 0600 0000 001f a22e 0600
   0x0000 0016 1e1f a228 0600 0000 0016 1420 8b89 0006 0016 0c1f a234
   0x0600 0000 0016 0223 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0074 6573
   0x7400 0000 0100 0200 0300 0400 0300 0200 0100 008e ea4b 43a7 87f9
   0x010e ef56 7b23 9a34 6b15 a6b4 0fc0 e44d 2cd4 a221 47e6 0aef f2bc




Price et al.                                                   [Page 40]


INTERNET-DRAFT           SigComp Torture Tests         January 14, 2003


   0x0fb6 af00

A.3.3.  Multiple compartments

   0xf81b 110f 8602 89a2 071c 01a0 459f 9917 6203 0d3d 0617 6205 3786
   0xa068 0622 8920 a1c0 6200 0600 0622 0320 a1c0 6200 0600 0722 a202
   0x0a22 0306 22a2 0320 a1c0 6200 0600 20a1 c0a2 0600 0662 162b 20a7
   0xc020 0006 0016 221f a213 0600 0000 001f a219 0600 0000 001f a225
   0x0600 0000 001f a22b 0600 0000 0023 0000 0000 0000 0017 6207 0610
   0x1a1f a207 0600 0000 0016 ea1f a20d 0600 0000 0016 e01f a21f 0600
   0x0000 0016 9fd6 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
   0x0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0102
   0x0304 0506 aca6 0b8e b283 9dbf afc6 3dd2 34c5 d91d 5361 bdd6 ba2a
   0xc65a 47c2 1814 ee07 c275 941d d7a1 4887 9c8d e90e 00


































Price et al.                                                   [Page 41]


Html markup produced by rfcmarkup 1.129d, available from https://tools.ietf.org/tools/rfcmarkup/