RFC 8251






Internet Engineering Task Force (IETF)                         JM. Valin
Request for Comments: 8251                           Mozilla Corporation
Updates: 6716                                                     K. Vos
Category: Standards Track                                        vocTone
ISSN: 2070-1721                                             October 2017


                    Updates to the Opus Audio Codec

Abstract



   This document addresses minor issues that were found in the
   specification of the Opus audio codec in RFC 6716.  It updates the
   normative decoder implementation included in Appendix A of RFC 6716.
   The changes fix real and potential security-related issues, as well
   as minor quality-related issues.

Status of This Memo



   This is an Internet Standards Track document.

   This document is a product of the Internet Engineering Task Force
   (IETF).  It represents the consensus of the IETF community.  It has
   received public review and has been approved for publication by the
   Internet Engineering Steering Group (IESG).  Further information on
   Internet Standards is available in Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   https://www.rfc-editor.org/info/rfc8251.

Copyright Notice



   Copyright (c) 2017 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (https://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.






Valin & Vos                  Standards Track                    [Page 1]

RFC 8251                       Opus Update                  October 2017


Table of Contents



   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2
   2.  Terminology . . . . . . . . . . . . . . . . . . . . . . . . .   3
   3.  Stereo State Reset in SILK  . . . . . . . . . . . . . . . . .   3
   4.  Parsing of the Opus Packet Padding  . . . . . . . . . . . . .   4
   5.  Resampler Buffer  . . . . . . . . . . . . . . . . . . . . . .   4
   6.  Integer Wrap-Around in Inverse Gain Computation . . . . . . .   6
   7.  Integer Wrap-Around in LSF Decoding . . . . . . . . . . . . .   7
   8.  Cap on Band Energy  . . . . . . . . . . . . . . . . . . . . .   7
   9.  Hybrid Folding  . . . . . . . . . . . . . . . . . . . . . . .   8
   10. Downmix to Mono . . . . . . . . . . . . . . . . . . . . . . .   9
   11. New Test Vectors  . . . . . . . . . . . . . . . . . . . . . .   9
   12. Security Considerations . . . . . . . . . . . . . . . . . . .  11
   13. IANA Considerations . . . . . . . . . . . . . . . . . . . . .  11
   14. Normative References  . . . . . . . . . . . . . . . . . . . .  11
   Acknowledgements  . . . . . . . . . . . . . . . . . . . . . . . .  11
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  12

1.  Introduction



   This document addresses minor issues that were discovered in the
   reference implementation of the Opus codec.  Unlike most IETF
   specifications, RFC 6716 [RFC6716] defines Opus in terms of a
   normative reference decoder implementation rather than from the
   associated text description.  Appendix A of that RFC includes the
   reference decoder implementation, which is why only issues affecting
   the decoder are listed here.  An up-to-date implementation of the
   Opus encoder can be found at <https://opus-codec.org/>.

   Some of the changes in this document update normative behavior in a
   way that requires new test vectors.  Only the C implementation is
   affected, not the English text of the specification.  This
   specification remains fully compatible with RFC 6716 [RFC6716].

   Note: Due to RFC formatting conventions, lines exceeding the column
   width in the patch are split using a backslash character.  The
   backslashes at the end of a line and the white space at the beginning
   of the following line are not part of the patch.  Referenced line
   numbers are approximations.  A properly formatted patch including all
   changes is available at <https://www.ietf.org/proceedings/98/slides/
   materials-98-codec-opus-update-00.patch> and has a SHA-1 hash of
   029e3aa88fc342c91e67a21e7bfbc9458661cd5f.








Valin & Vos                  Standards Track                    [Page 2]

RFC 8251                       Opus Update                  October 2017


2.  Terminology



   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
   "OPTIONAL" in this document are to be interpreted as described in
   BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all
   capitals, as shown here.

3.  Stereo State Reset in SILK



   The reference implementation does not reinitialize the stereo state
   during a mode switch.  The old stereo memory can produce a brief
   impulse (i.e., single sample) in the decoded audio.  This can be
   fixed by changing silk/dec_API.c around line 72:

   <CODE BEGINS>
        for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) {
            ret  = silk_init_decoder( &channel_state[ n ] );
        }
   +    silk_memset(&((silk_decoder *)decState)->sStereo, 0,
   +                sizeof(((silk_decoder *)decState)->sStereo));
   +    /* Not strictly needed, but it's cleaner that way */
   +    ((silk_decoder *)decState)->prev_decode_only_middle = 0;

        return ret;
    }
   <CODE ENDS>

   This change affects the normative output of the decoder, but the
   amount of change is within the tolerance and is too small to make the
   test vector check fail.




















Valin & Vos                  Standards Track                    [Page 3]

RFC 8251                       Opus Update                  October 2017


4.  Parsing of the Opus Packet Padding



   It was discovered that some invalid packets of a very large size
   could trigger an out-of-bounds read in the Opus packet parsing code
   responsible for padding.  This is due to an integer overflow if the
   signaled padding exceeds 2^31-1 bytes (the actual packet may be
   smaller).  The code can be fixed by decrementing the (signed) len
   value, instead of incrementing a separate padding counter.  This is
   done by applying the following changes around line 596 of
   src/opus_decoder.c:

   <CODE BEGINS>
          /* Padding flag is bit 6 */
          if (ch&0x40)
          {
   -         int padding=0;
             int p;
             do {
                if (len<=0)
                   return OPUS_INVALID_PACKET;
                p = *data++;
                len--;
   -            padding += p==255 ? 254: p;
   +            len -= p==255 ? 254: p;
             } while (p==255);
   -         len -= padding;
          }
   <CODE ENDS>

   This packet-parsing issue is limited to reading memory up to about 60
   KB beyond the compressed buffer.  This can only be triggered by a
   compressed packet more than about 16 MB long, so it's not a problem
   for RTP.  In theory, it could crash a file decoder (e.g., Opus in
   Ogg) if the memory just after the incoming packet is out of range,
   but our attempts to trigger such a crash in a production application
   built using an affected version of the Opus decoder failed.

5.  Resampler Buffer



   The SILK resampler had the following issues:

   1.  The calls to memcpy() were using sizeof(opus_int32), but the type
       of the local buffer was opus_int16.








Valin & Vos                  Standards Track                    [Page 4]

RFC 8251                       Opus Update                  October 2017


   2.  Because the size was wrong, this potentially allowed the source
       and destination regions of the memcpy() to overlap on the copy
       from "buf" to "buf".  We believe that nSamplesIn (number of input
       samples) is at least fs_in_khZ (sampling rate in kHz), which is
       at least 8.  Since RESAMPLER_ORDER_FIR_12 is only 8, that should
       not be a problem once the type size is fixed.

   3.  The size of the buffer used RESAMPLER_MAX_BATCH_SIZE_IN, but the
       data stored in it was actually twice the input batch size
       (nSamplesIn<<1).

   The code can be fixed by applying the following changes around line
   78 of silk/resampler_private_IIR_FIR.c:

   <CODE BEGINS>

    )
    {
        silk_resampler_state_struct *S = \
   (silk_resampler_state_struct *)SS;
        opus_int32 nSamplesIn;
        opus_int32 max_index_Q16, index_increment_Q16;
   -    opus_int16 buf[ RESAMPLER_MAX_BATCH_SIZE_IN + \
   RESAMPLER_ORDER_FIR_12 ];
   +    opus_int16 buf[ 2*RESAMPLER_MAX_BATCH_SIZE_IN + \
   RESAMPLER_ORDER_FIR_12 ];

        /* Copy buffered samples to start of buffer */
   -    silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 \
   * sizeof( opus_int32 ) );
   +    silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 \
   * sizeof( opus_int16 ) );

        /* Iterate over blocks of frameSizeIn input samples */
        index_increment_Q16 = S->invRatio_Q16;
        while( 1 ) {
            nSamplesIn = silk_min( inLen, S->batchSize );

            /* Upsample 2x */
            silk_resampler_private_up2_HQ( S->sIIR, &buf[ \
   RESAMPLER_ORDER_FIR_12 ], in, nSamplesIn );

            max_index_Q16 = silk_LSHIFT32( nSamplesIn, 16 + 1 \
   );         /* + 1 because 2x upsampling */
            out = silk_resampler_private_IIR_FIR_INTERPOL( out, \
   buf, max_index_Q16, index_increment_Q16 );
            in += nSamplesIn;
            inLen -= nSamplesIn;



Valin & Vos                  Standards Track                    [Page 5]

RFC 8251                       Opus Update                  October 2017


            if( inLen > 0 ) {
                /* More iterations to do; copy last part of \
   filtered signal to beginning of buffer */
   -            silk_memcpy( buf, &buf[ nSamplesIn << 1 ], \
   RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) );
   +            silk_memmove( buf, &buf[ nSamplesIn << 1 ], \
   RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) );
            } else {
                break;
            }
        }

        /* Copy last part of filtered signal to the state for \
   the next call */
   -    silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], \
   RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) );
   +    silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], \
   RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) );
    }
   <CODE ENDS>

6.  Integer Wrap-Around in Inverse Gain Computation



   It was discovered through decoder fuzzing that some bitstreams could
   produce integer values exceeding 32 bits in
   LPC_inverse_pred_gain_QA(), causing a wrap-around.  The C standard
   considers this behavior as undefined.  The following patch around
   line 87 of silk/LPC_inv_pred_gain.c detects values that do not fit in
   a 32-bit integer and considers the corresponding filters unstable:

  <CODE BEGINS>
           /* Update AR coefficient */
           for( n = 0; n < k; n++ ) {
  -            tmp_QA = Aold_QA[ n ] - MUL32_FRAC_Q( \
  Aold_QA[ k - n - 1 ], rc_Q31, 31 );
  -            Anew_QA[ n ] = MUL32_FRAC_Q( tmp_QA, rc_mult2 , mult2Q );
  +            opus_int64 tmp64;
  +            tmp_QA = silk_SUB_SAT32( Aold_QA[ n ], MUL32_FRAC_Q( \
  Aold_QA[ k - n - 1 ], rc_Q31, 31 ) );
  +            tmp64 = silk_RSHIFT_ROUND64( silk_SMULL( tmp_QA, \
  rc_mult2 ), mult2Q);
  +            if( tmp64 > silk_int32_MAX || tmp64 < silk_int32_MIN ) {
  +               return 0;
  +            }
  +            Anew_QA[ n ] = ( opus_int32 )tmp64;
           }
  <CODE ENDS>




Valin & Vos                  Standards Track                    [Page 6]

RFC 8251                       Opus Update                  October 2017


7.  Integer Wrap-Around in LSF Decoding



   It was discovered -- also from decoder fuzzing -- that an integer
   wrap-around could occur when decoding bitstreams with extremely large
   values for the high Line Spectral Frequency (LSF) parameters.  The
   end result of the wrap-around is an illegal read access on the stack,
   which the authors do not believe is exploitable but should
   nonetheless be fixed.  The following patch around line 137 of silk/
   NLSF_stabilize.c prevents the problem:

   <CODE BEGINS>
              /* Keep delta_min distance between the NLSFs */
            for( i = 1; i < L; i++ )
   -            NLSF_Q15[i] = silk_max_int( NLSF_Q15[i], \
   NLSF_Q15[i-1] + NDeltaMin_Q15[i] );
   +            NLSF_Q15[i] = silk_max_int( NLSF_Q15[i], \
   silk_ADD_SAT16( NLSF_Q15[i-1], NDeltaMin_Q15[i] ) );

            /* Last NLSF should be no higher than 1 - NDeltaMin[L] */
   <CODE ENDS>

8.  Cap on Band Energy



   On extreme bitstreams, it is possible for log-domain band energy
   levels to exceed the maximum single-precision floating point value
   once converted to a linear scale.  This would later cause the decoded
   values to be NaN (not a number), possibly causing problems in the
   software using the PCM values.  This can be avoided with the
   following patch around line 552 of celt/quant_bands.c:

   <CODE BEGINS>
          {
             opus_val16 lg = ADD16(oldEBands[i+c*m->nbEBands],
                             SHL16((opus_val16)eMeans[i],6));
   +         lg = MIN32(QCONST32(32.f, 16), lg);
             eBands[i+c*m->nbEBands] = PSHR32(celt_exp2(lg),4);
          }
          for (;i<m->nbEBands;i++)
   <CODE ENDS>












Valin & Vos                  Standards Track                    [Page 7]

RFC 8251                       Opus Update                  October 2017


9.  Hybrid Folding



   When encoding in hybrid mode at low bitrate, we sometimes only have
   enough bits to code a single Constrained-Energy Lapped Transform
   (CELT) band (8 - 9.6 kHz).  When that happens, the second band (CELT
   band 18, from 9.6 - 12 kHz) cannot use folding because it is wider
   than the amount already coded and falls back to white noise.  Because
   it can also happen on transients (e.g., stops), it can cause audible
   pre-echo.

   To address the issue, we change the folding behavior so that it is
   never forced to fall back to Linear Congruential Generator (LCG) due
   to the first band not containing enough coefficients to fold onto the
   second band.  This is achieved by simply repeating part of the first
   band in the folding of the second band.  This changes the code in
   celt/bands.c around line 1237:

  <CODE BEGINS>
            b = 0;
         }

  -      if (resynth && M*eBands[i]-N >= M*eBands[start] && \
  (update_lowband || lowband_offset==0))
  +      if (resynth && (M*eBands[i]-N >= M*eBands[start] || \
  i==start+1) && (update_lowband || lowband_offset==0))
               lowband_offset = i;

  +      if (i == start+1)
  +      {
  +         int n1, n2;
  +         int offset;
  +         n1 = M*(eBands[start+1]-eBands[start]);
  +         n2 = M*(eBands[start+2]-eBands[start+1]);
  +         offset = M*eBands[start];
  +         /* Duplicate enough of the first band folding data to \
  be able to fold the second band.
  +            Copies no data for CELT-only mode. */
  +         OPUS_COPY(&norm[offset+n1], &norm[offset+2*n1 - n2], n2-n1);
  +         if (C==2)
  +            OPUS_COPY(&norm2[offset+n1], &norm2[offset+2*n1 - n2], \
  n2-n1);
  +      }
  +
         tf_change = tf_res[i];
         if (i>=m->effEBands)
         {
  <CODE ENDS>




Valin & Vos                  Standards Track                    [Page 8]

RFC 8251                       Opus Update                  October 2017


   as well as around line 1260:

   <CODE BEGINS>
             fold_start = lowband_offset;
             while(M*eBands[--fold_start] > effective_lowband);
             fold_end = lowband_offset-1;
   -         while(M*eBands[++fold_end] < effective_lowband+N);
   +         while(++fold_end < i && M*eBands[fold_end] < \
   effective_lowband+N);
             x_cm = y_cm = 0;
             fold_i = fold_start; do {
               x_cm |= collapse_masks[fold_i*C+0];

   <CODE ENDS>

   The fix does not impact compatibility, because the improvement does
   not depend on the encoder doing anything special.  There is also no
   reasonable way for an encoder to use the original behavior to improve
   quality over the proposed change.

10.  Downmix to Mono



   The last issue is not strictly a bug, but it is an issue that has
   been reported when downmixing an Opus decoded stream to mono, whether
   this is done inside the decoder or as a post-processing step on the
   stereo decoder output.  Opus intensity stereo allows optionally
   coding the two channels 180 degrees out of phase on a per-band basis.
   This provides better stereo quality than forcing the two channels to
   be in phase, but when the output is downmixed to mono, the energy in
   the affected bands is canceled, sometimes resulting in audible
   artifacts.

   As a work-around for this issue, the decoder MAY choose not to apply
   the 180-degree phase shift.  This can be useful when downmixing to
   mono inside or outside of the decoder (e.g., requested explicitly
   from an API).

11.  New Test Vectors



   Changes in Sections 9 and 10 have sufficient impact on the test
   vectors to make them fail.  For this reason, this document also
   updates the Opus test vectors.  The new test vectors now include two
   decoded outputs for the same bitstream.  The outputs with suffix 'm'
   do not apply the CELT 180-degree phase shift as allowed in
   Section 10, while the outputs without the suffix do.  An
   implementation is compliant as long as it passes either set of
   vectors.




Valin & Vos                  Standards Track                    [Page 9]

RFC 8251                       Opus Update                  October 2017


   Any Opus implementation that passes either the original test vectors
   from RFC 6716 [RFC6716] or one of the new sets of test vectors is
   compliant with the Opus specification.  However, newer
   implementations SHOULD be based on the new test vectors rather than
   the old ones.

   The new test vectors are located at
   <https://www.ietf.org/proceedings/98/slides/materials-98-codec-opus-
   newvectors-00.tar.gz>.  The SHA-1 hashes of the test vectors are:

   e49b2862ceec7324790ed8019eb9744596d5be01  testvector01.bit
   b809795ae1bcd606049d76de4ad24236257135e0  testvector02.bit
   e0c4ecaeab44d35a2f5b6575cd996848e5ee2acc  testvector03.bit
   a0f870cbe14ebb71fa9066ef3ee96e59c9a75187  testvector04.bit
   9b3d92b48b965dfe9edf7b8a85edd4309f8cf7c8  testvector05.bit
   28e66769ab17e17f72875283c14b19690cbc4e57  testvector06.bit
   bacf467be3215fc7ec288f29e2477de1192947a6  testvector07.bit
   ddbe08b688bbf934071f3893cd0030ce48dba12f  testvector08.bit
   3932d9d61944dab1201645b8eeaad595d5705ecb  testvector09.bit
   521eb2a1e0cc9c31b8b740673307c2d3b10c1900  testvector10.bit
   6bc8f3146fcb96450c901b16c3d464ccdf4d5d96  testvector11.bit
   338c3f1b4b97226bc60bc41038becbc6de06b28f  testvector12.bit
   f5ef93884da6a814d311027918e9afc6f2e5c2c8  testvector01.dec
   48ac1ff1995250a756e1e17bd32acefa8cd2b820  testvector02.dec
   d15567e919db2d0e818727092c0af8dd9df23c95  testvector03.dec
   1249dd28f5bd1e39a66fd6d99449dca7a8316342  testvector04.dec
   b85675d81deef84a112c466cdff3b7aaa1d2fc76  testvector05.dec
   55f0b191e90bfa6f98b50d01a64b44255cb4813e  testvector06.dec
   61e8b357ab090b1801eeb578a28a6ae935e25b7b  testvector07.dec
   a58539ee5321453b2ddf4c0f2500e856b3966862  testvector08.dec
   bb96aad2cde188555862b7bbb3af6133851ef8f4  testvector09.dec
   1b6cdf0413ac9965b16184b1bea129b5c0b2a37a  testvector10.dec
   b1fff72b74666e3027801b29dbc48b31f80dee0d  testvector11.dec
   98e09bbafed329e341c3b4052e9c4ba5fc83f9b1  testvector12.dec
   1e7d984ea3fbb16ba998aea761f4893fbdb30157  testvector01m.dec
   48ac1ff1995250a756e1e17bd32acefa8cd2b820  testvector02m.dec
   d15567e919db2d0e818727092c0af8dd9df23c95  testvector03m.dec
   1249dd28f5bd1e39a66fd6d99449dca7a8316342  testvector04m.dec
   d70b0bad431e7d463bc3da49bd2d49f1c6d0a530  testvector05m.dec
   6ac1648c3174c95fada565161a6c78bdbe59c77d  testvector06m.dec
   fc5e2f709693738324fb4c8bdc0dad6dda04e713  testvector07m.dec
   aad2ba397bf1b6a18e8e09b50e4b19627d479f00  testvector08m.dec
   6feb7a7b9d7cdc1383baf8d5739e2a514bd0ba08  testvector09m.dec
   1b6cdf0413ac9965b16184b1bea129b5c0b2a37a  testvector10m.dec
   fd3d3a7b0dfbdab98d37ed9aa04b659b9fefbd18  testvector11m.dec
   98e09bbafed329e341c3b4052e9c4ba5fc83f9b1  testvector12m.dec

   Note that the decoder input bitstream files (.bit) are unchanged.



Valin & Vos                  Standards Track                   [Page 10]

RFC 8251                       Opus Update                  October 2017


12.  Security Considerations



   This document fixes two security issues reported on Opus that affect
   the reference implementation in RFC 6716 [RFC6716]: CVE-2013-0899
   <https://nvd.nist.gov/vuln/detail/CVE-2013-0899> and CVE-2017-0381
   <https://nvd.nist.gov/vuln/detail/CVE-2017-0381>.  CVE-2013-0899
   theoretically could have caused an information leak.  The leaked
   information would have gone through the decoder process before being
   accessible to the attacker.  The update in Section 4 fixes this.
   CVE-2017-0381 could have resulted in a 16-bit out-of-bounds read from
   a fixed location.  The update in Section 7 fixes this.  Beyond the
   two fixed Common Vulnerabilities and Exposures (CVEs), this document
   adds no new security considerations beyond those in RFC 6716
   [RFC6716].

13.  IANA Considerations



   This document does not require any IANA actions.

14.  Normative References



   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <https://www.rfc-editor.org/info/rfc2119>.

   [RFC6716]  Valin, JM., Vos, K., and T. Terriberry, "Definition of the
              Opus Audio Codec", RFC 6716, DOI 10.17487/RFC6716,
              September 2012, <https://www.rfc-editor.org/info/rfc6716>.

   [RFC8174]  Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
              2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
              May 2017, <https://www.rfc-editor.org/info/rfc8174>.

Acknowledgements



   We would like to thank Juri Aedla for reporting the issue with the
   parsing of the Opus padding.  Thanks to Felicia Lim for reporting the
   LSF integer overflow issue.  Also, thanks to Tina le Grand, Jonathan
   Lennox, and Mark Harris for their feedback on this document.











Valin & Vos                  Standards Track                   [Page 11]

RFC 8251                       Opus Update                  October 2017


Authors' Addresses



   Jean-Marc Valin
   Mozilla Corporation
   331 E. Evelyn Avenue
   Mountain View, CA  94041
   United States of America

   Phone: +1 650 903-0800
   Email: jmvalin@jmvalin.ca


   Koen Vos
   vocTone

   Email: koenvos74@gmail.com



































Valin & Vos                  Standards Track                   [Page 12]