idnits 2.17.1
draft-ietf-codec-opus-update-10.txt:
Checking boilerplate required by RFC 5378 and the IETF Trust (see
https://trustee.ietf.org/license-info):
----------------------------------------------------------------------------
No issues found here.
Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt:
----------------------------------------------------------------------------
No issues found here.
Checking nits according to https://www.ietf.org/id-info/checklist :
----------------------------------------------------------------------------
== There are 1 instance of lines with non-RFC2606-compliant FQDNs in the
document.
-- The draft header indicates that this document updates RFC6716, but the
abstract doesn't seem to directly say this. It does mention RFC6716
though, so this could be OK.
Miscellaneous warnings:
----------------------------------------------------------------------------
== The copyright year in the IETF Trust and authors Copyright Line does not
match the current year
== Line 411 has weird spacing: '...5ee2acc testv...'
== Line 417 has weird spacing: '...5705ecb testv...'
-- The document date (August 24, 2017) is 2435 days in the past. Is this
intentional?
Checking references for intended status: Proposed Standard
----------------------------------------------------------------------------
(See RFCs 3967 and 4897 for information about using normative references
to lower-maturity documents in RFCs)
No issues found here.
Summary: 0 errors (**), 0 flaws (~~), 4 warnings (==), 2 comments (--).
Run idnits with the --verbose option for more detailed information about
the items above.
--------------------------------------------------------------------------------
2 Network Working Group JM. Valin
3 Internet-Draft Mozilla Corporation
4 Updates: 6716 (if approved) K. Vos
5 Intended status: Standards Track vocTone
6 Expires: February 25, 2018 August 24, 2017
8 Updates to the Opus Audio Codec
9 draft-ietf-codec-opus-update-10
11 Abstract
13 This document addresses minor issues that were found in the
14 specification of the Opus audio codec in RFC 6716. It updates the
15 normative decoder implementation included in the appendix of RFC
16 6716. The changes fixes real and potential security-related issues,
17 as well minor quality-related issues.
19 Status of This Memo
21 This Internet-Draft is submitted in full conformance with the
22 provisions of BCP 78 and BCP 79.
24 Internet-Drafts are working documents of the Internet Engineering
25 Task Force (IETF). Note that other groups may also distribute
26 working documents as Internet-Drafts. The list of current Internet-
27 Drafts is at http://datatracker.ietf.org/drafts/current/.
29 Internet-Drafts are draft documents valid for a maximum of six months
30 and may be updated, replaced, or obsoleted by other documents at any
31 time. It is inappropriate to use Internet-Drafts as reference
32 material or to cite them other than as "work in progress."
34 This Internet-Draft will expire on February 25, 2018.
36 Copyright Notice
38 Copyright (c) 2017 IETF Trust and the persons identified as the
39 document authors. All rights reserved.
41 This document is subject to BCP 78 and the IETF Trust's Legal
42 Provisions Relating to IETF Documents
43 (http://trustee.ietf.org/license-info) in effect on the date of
44 publication of this document. Please review these documents
45 carefully, as they describe your rights and restrictions with respect
46 to this document. Code Components extracted from this document must
47 include Simplified BSD License text as described in Section 4.e of
48 the Trust Legal Provisions and are provided without warranty as
49 described in the Simplified BSD License.
51 Table of Contents
53 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
54 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3
55 3. Stereo State Reset in SILK . . . . . . . . . . . . . . . . . 3
56 4. Parsing of the Opus Packet Padding . . . . . . . . . . . . . 3
57 5. Resampler buffer . . . . . . . . . . . . . . . . . . . . . . 4
58 6. Integer wrap-around in inverse gain computation . . . . . . . 6
59 7. Integer wrap-around in LSF decoding . . . . . . . . . . . . . 6
60 8. Cap on Band Energy . . . . . . . . . . . . . . . . . . . . . 7
61 9. Hybrid Folding . . . . . . . . . . . . . . . . . . . . . . . 7
62 10. Downmix to Mono . . . . . . . . . . . . . . . . . . . . . . . 9
63 11. New Test Vectors . . . . . . . . . . . . . . . . . . . . . . 9
64 12. Security Considerations . . . . . . . . . . . . . . . . . . . 10
65 13. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 11
66 14. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 11
67 15. Normative References . . . . . . . . . . . . . . . . . . . . 11
68 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 11
70 1. Introduction
72 This document addresses minor issues that were discovered in the
73 reference implementation of the Opus codec. Unlike most IETF
74 specifications, Opus is defined in RFC 6716 [RFC6716] in terms of a
75 normative reference decoder implementation rather than from the
76 associated text description. That RFC includes the reference decoder
77 implementation as Appendix A. That's why only issues affecting the
78 decoder are listed here. An up-to-date implementation of the Opus
79 encoder can be found at .
81 Some of the changes in this document update normative behaviour in a
82 way that requires new test vectors. The English text of the
83 specification is unaffected, only the C implementation is. The
84 updated specification remains fully compatible with the original
85 specification.
87 Note: due to RFC formatting conventions, lines exceeding the column
88 width in the patch are split using a backslash character. The
89 backslashes at the end of a line and the white space at the beginning
90 of the following line are not part of the patch. A properly
91 formatted patch including all changes is available at
92 and has a SHA-1 hash of
94 029e3aa88fc342c91e67a21e7bfbc9458661cd5f.
96 2. Terminology
98 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
99 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
100 document are to be interpreted as described in RFC 2119 [RFC2119].
102 3. Stereo State Reset in SILK
104 The reference implementation does not reinitialize the stereo state
105 during a mode switch. The old stereo memory can produce a brief
106 impulse (i.e. single sample) in the decoded audio. This can be fixed
107 by changing silk/dec_API.c at line 72:
109
110 for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) {
111 ret = silk_init_decoder( &channel_state[ n ] );
112 }
113 + silk_memset(&((silk_decoder *)decState)->sStereo, 0,
114 + sizeof(((silk_decoder *)decState)->sStereo));
115 + /* Not strictly needed, but it's cleaner that way */
116 + ((silk_decoder *)decState)->prev_decode_only_middle = 0;
118 return ret;
119 }
120
122 This change affects the normative output of the decoder, but the
123 amount of change is within the tolerance and too small to make the
124 testvector check fail.
126 4. Parsing of the Opus Packet Padding
128 It was discovered that some invalid packets of very large size could
129 trigger an out-of-bounds read in the Opus packet parsing code
130 responsible for padding. This is due to an integer overflow if the
131 signaled padding exceeds 2^31-1 bytes (the actual packet may be
132 smaller). The code can be fixed by decrementing the (signed) len
133 value, instead of incrementing a separate padding counter. This is
134 done by applying the following changes at line 596 of src/
135 opus_decoder.c:
137
138 /* Padding flag is bit 6 */
139 if (ch&0x40)
140 {
141 - int padding=0;
142 int p;
143 do {
144 if (len<=0)
145 return OPUS_INVALID_PACKET;
146 p = *data++;
147 len--;
148 - padding += p==255 ? 254: p;
149 + len -= p==255 ? 254: p;
150 } while (p==255);
151 - len -= padding;
152 }
153
155 This packet parsing issue is limited to reading memory up to about 60
156 kB beyond the compressed buffer. This can only be triggered by a
157 compressed packet more than about 16 MB long, so it's not a problem
158 for RTP. In theory, it could crash a file decoder (e.g. Opus in
159 Ogg) if the memory just after the incoming packet is out-of-range,
160 but our attempts to trigger such a crash in a production application
161 built using an affected version of the Opus decoder failed.
163 5. Resampler buffer
165 The SILK resampler had the following issues:
167 1. The calls to memcpy() were using sizeof(opus_int32), but the type
168 of the local buffer was opus_int16.
170 2. Because the size was wrong, this potentially allowed the source
171 and destination regions of the memcpy() to overlap on the copy
172 from "buf" to "buf". We believe that nSamplesIn (number of input
173 samples) is at least fs_in_khZ (sampling rate in kHz), which is
174 at least 8. Since RESAMPLER_ORDER_FIR_12 is only 8, that should
175 not be a problem once the type size is fixed.
177 3. The size of the buffer used RESAMPLER_MAX_BATCH_SIZE_IN, but the
178 data stored in it was actually twice the input batch size
179 (nSamplesIn<<1).
181 The code can be fixed by applying the following changes to line 78 of
182 silk/resampler_private_IIR_FIR.c:
184
185 )
186 {
187 silk_resampler_state_struct *S = \
188 (silk_resampler_state_struct *)SS;
189 opus_int32 nSamplesIn;
190 opus_int32 max_index_Q16, index_increment_Q16;
191 - opus_int16 buf[ RESAMPLER_MAX_BATCH_SIZE_IN + \
192 RESAMPLER_ORDER_FIR_12 ];
193 + opus_int16 buf[ 2*RESAMPLER_MAX_BATCH_SIZE_IN + \
194 RESAMPLER_ORDER_FIR_12 ];
196 /* Copy buffered samples to start of buffer */
197 - silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 \
198 * sizeof( opus_int32 ) );
199 + silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 \
200 * sizeof( opus_int16 ) );
202 /* Iterate over blocks of frameSizeIn input samples */
203 index_increment_Q16 = S->invRatio_Q16;
204 while( 1 ) {
205 nSamplesIn = silk_min( inLen, S->batchSize );
207 /* Upsample 2x */
208 silk_resampler_private_up2_HQ( S->sIIR, &buf[ \
209 RESAMPLER_ORDER_FIR_12 ], in, nSamplesIn );
211 max_index_Q16 = silk_LSHIFT32( nSamplesIn, 16 + 1 \
212 ); /* + 1 because 2x upsampling */
213 out = silk_resampler_private_IIR_FIR_INTERPOL( out, \
214 buf, max_index_Q16, index_increment_Q16 );
215 in += nSamplesIn;
216 inLen -= nSamplesIn;
218 if( inLen > 0 ) {
219 /* More iterations to do; copy last part of \
220 filtered signal to beginning of buffer */
221 - silk_memcpy( buf, &buf[ nSamplesIn << 1 ], \
222 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) );
223 + silk_memmove( buf, &buf[ nSamplesIn << 1 ], \
224 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) );
225 } else {
226 break;
227 }
228 }
230 /* Copy last part of filtered signal to the state for \
231 the next call */
232 - silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], \
233 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) );
234 + silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], \
235 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) );
236 }
237
239 6. Integer wrap-around in inverse gain computation
241 It was discovered through decoder fuzzing that some bitstreams could
242 produce integer values exceeding 32-bits in
243 LPC_inverse_pred_gain_QA(), causing a wrap-around. The C standard
244 considers this behavior as undefined. The following patch to line 87
245 of silk/LPC_inv_pred_gain.c detects values that do not fit in a
246 32-bit integer and considers the corresponding filters unstable:
248
249 /* Update AR coefficient */
250 for( n = 0; n < k; n++ ) {
251 - tmp_QA = Aold_QA[ n ] - MUL32_FRAC_Q( \
252 Aold_QA[ k - n - 1 ], rc_Q31, 31 );
253 - Anew_QA[ n ] = MUL32_FRAC_Q( tmp_QA, rc_mult2 , mult2Q );
254 + opus_int64 tmp64;
255 + tmp_QA = silk_SUB_SAT32( Aold_QA[ n ], MUL32_FRAC_Q( \
256 Aold_QA[ k - n - 1 ], rc_Q31, 31 ) );
257 + tmp64 = silk_RSHIFT_ROUND64( silk_SMULL( tmp_QA, \
258 rc_mult2 ), mult2Q);
259 + if( tmp64 > silk_int32_MAX || tmp64 < silk_int32_MIN ) {
260 + return 0;
261 + }
262 + Anew_QA[ n ] = ( opus_int32 )tmp64;
263 }
264
266 7. Integer wrap-around in LSF decoding
268 It was discovered -- also from decoder fuzzing -- that an integer
269 wrap-around could occur when decoding bitstreams with extremely large
270 values for the high LSF parameters. The end result of the wrap-
271 around is an illegal read access on the stack, which the authors do
272 not believe is exploitable but should nonetheless be fixed. The
273 following patch to line 137 of silk/NLSF_stabilize.c prevents the
274 problem:
276
277 /* Keep delta_min distance between the NLSFs */
278 for( i = 1; i < L; i++ )
279 - NLSF_Q15[i] = silk_max_int( NLSF_Q15[i], \
280 NLSF_Q15[i-1] + NDeltaMin_Q15[i] );
281 + NLSF_Q15[i] = silk_max_int( NLSF_Q15[i], \
282 silk_ADD_SAT16( NLSF_Q15[i-1], NDeltaMin_Q15[i] ) );
284 /* Last NLSF should be no higher than 1 - NDeltaMin[L] */
285
287 8. Cap on Band Energy
289 On extreme bit-streams, it is possible for log-domain band energy
290 levels to exceed the maximum single-precision floating point value
291 once converted to a linear scale. This would later cause the decoded
292 values to be NaN (not a number), possibly causing problems in the
293 software using the PCM values. This can be avoided with the
294 following patch to line 552 of celt/quant_bands.c:
296
297 {
298 opus_val16 lg = ADD16(oldEBands[i+c*m->nbEBands],
299 SHL16((opus_val16)eMeans[i],6));
300 + lg = MIN32(QCONST32(32.f, 16), lg);
301 eBands[i+c*m->nbEBands] = PSHR32(celt_exp2(lg),4);
302 }
303 for (;inbEBands;i++)
304
306 9. Hybrid Folding
308 When encoding in hybrid mode at low bitrate, we sometimes only have
309 enough bits to code a single CELT band (8 - 9.6 kHz). When that
310 happens, the second band (CELT band 18, from 9.6 to 12 kHz) cannot
311 use folding because it is wider than the amount already coded, and
312 falls back to white noise. Because it can also happen on transients
313 (e.g. stops), it can cause audible pre-echo.
315 To address the issue, we change the folding behavior so that it is
316 never forced to fall back to LCG due to the first band not containing
317 enough coefficients to fold onto the second band. This is achieved
318 by simply repeating part of the first band in the folding of the
319 second band. This changes the code in celt/bands.c around line 1237:
321
322 b = 0;
323 }
325 - if (resynth && M*eBands[i]-N >= M*eBands[start] && \
326 (update_lowband || lowband_offset==0))
327 + if (resynth && (M*eBands[i]-N >= M*eBands[start] || \
328 i==start+1) && (update_lowband || lowband_offset==0))
329 lowband_offset = i;
331 + if (i == start+1)
332 + {
333 + int n1, n2;
334 + int offset;
335 + n1 = M*(eBands[start+1]-eBands[start]);
336 + n2 = M*(eBands[start+2]-eBands[start+1]);
337 + offset = M*eBands[start];
338 + /* Duplicate enough of the first band folding data to \
339 be able to fold the second band.
340 + Copies no data for CELT-only mode. */
341 + OPUS_COPY(&norm[offset+n1], &norm[offset+2*n1 - n2], n2-n1);
342 + if (C==2)
343 + OPUS_COPY(&norm2[offset+n1], &norm2[offset+2*n1 - n2], \
344 n2-n1);
345 + }
346 +
347 tf_change = tf_res[i];
348 if (i>=m->effEBands)
349 {
350
352 as well as line 1260:
354
355 fold_start = lowband_offset;
356 while(M*eBands[--fold_start] > effective_lowband);
357 fold_end = lowband_offset-1;
358 - while(M*eBands[++fold_end] < effective_lowband+N);
359 + while(++fold_end < i && M*eBands[fold_end] < \
360 effective_lowband+N);
361 x_cm = y_cm = 0;
362 fold_i = fold_start; do {
363 x_cm |= collapse_masks[fold_i*C+0];
365
367 The fix does not impact compatibility, because the improvement does
368 not depend on the encoder doing anything special. There is also no
369 reasonable way for an encoder to use the original behavior to improve
370 quality over the proposed change.
372 10. Downmix to Mono
374 The last issue is not strictly a bug, but it is an issue that has
375 been reported when downmixing an Opus decoded stream to mono, whether
376 this is done inside the decoder or as a post-processing step on the
377 stereo decoder output. Opus intensity stereo allows optionally
378 coding the two channels 180-degrees out of phase on a per-band basis.
379 This provides better stereo quality than forcing the two channels to
380 be in phase, but when the output is downmixed to mono, the energy in
381 the affected bands is cancelled sometimes resulting in audible
382 artifacts.
384 As a work-around for this issue, the decoder MAY choose not to apply
385 the 180-degree phase shift. This can be useful when downmixing to
386 mono inside or outside of the decoder (e.g. user-controllable).
388 11. New Test Vectors
390 Changes in Section 9 and Section 10 have sufficient impact on the
391 testvectors to make them fail. For this reason, this document also
392 updates the Opus test vectors. The new test vectors now include two
393 decoded outputs for the same bitstream. The outputs with suffix 'm'
394 do not apply the CELT 180-degree phase shift as allowed in
395 Section 10, while the outputs without the suffix do. An
396 implementation is compliant as long as it passes either set of
397 vectors.
399 Any Opus implementation that passes either the original test vectors
400 from RFC 6716 [RFC6716] or one of the new sets of test vectors is
401 compliant with the Opus specification. However, newer
402 implementations SHOULD be based on the new test vectors rather than
403 the old ones.
405 The new test vectors are located at
406 . The SHA-1 hashes of the test vectors are:
409 e49b2862ceec7324790ed8019eb9744596d5be01 testvector01.bit
410 b809795ae1bcd606049d76de4ad24236257135e0 testvector02.bit
411 e0c4ecaeab44d35a2f5b6575cd996848e5ee2acc testvector03.bit
412 a0f870cbe14ebb71fa9066ef3ee96e59c9a75187 testvector04.bit
413 9b3d92b48b965dfe9edf7b8a85edd4309f8cf7c8 testvector05.bit
414 28e66769ab17e17f72875283c14b19690cbc4e57 testvector06.bit
415 bacf467be3215fc7ec288f29e2477de1192947a6 testvector07.bit
416 ddbe08b688bbf934071f3893cd0030ce48dba12f testvector08.bit
417 3932d9d61944dab1201645b8eeaad595d5705ecb testvector09.bit
418 521eb2a1e0cc9c31b8b740673307c2d3b10c1900 testvector10.bit
419 6bc8f3146fcb96450c901b16c3d464ccdf4d5d96 testvector11.bit
420 338c3f1b4b97226bc60bc41038becbc6de06b28f testvector12.bit
421 f5ef93884da6a814d311027918e9afc6f2e5c2c8 testvector01.dec
422 48ac1ff1995250a756e1e17bd32acefa8cd2b820 testvector02.dec
423 d15567e919db2d0e818727092c0af8dd9df23c95 testvector03.dec
424 1249dd28f5bd1e39a66fd6d99449dca7a8316342 testvector04.dec
425 b85675d81deef84a112c466cdff3b7aaa1d2fc76 testvector05.dec
426 55f0b191e90bfa6f98b50d01a64b44255cb4813e testvector06.dec
427 61e8b357ab090b1801eeb578a28a6ae935e25b7b testvector07.dec
428 a58539ee5321453b2ddf4c0f2500e856b3966862 testvector08.dec
429 bb96aad2cde188555862b7bbb3af6133851ef8f4 testvector09.dec
430 1b6cdf0413ac9965b16184b1bea129b5c0b2a37a testvector10.dec
431 b1fff72b74666e3027801b29dbc48b31f80dee0d testvector11.dec
432 98e09bbafed329e341c3b4052e9c4ba5fc83f9b1 testvector12.dec
433 1e7d984ea3fbb16ba998aea761f4893fbdb30157 testvector01m.dec
434 48ac1ff1995250a756e1e17bd32acefa8cd2b820 testvector02m.dec
435 d15567e919db2d0e818727092c0af8dd9df23c95 testvector03m.dec
436 1249dd28f5bd1e39a66fd6d99449dca7a8316342 testvector04m.dec
437 d70b0bad431e7d463bc3da49bd2d49f1c6d0a530 testvector05m.dec
438 6ac1648c3174c95fada565161a6c78bdbe59c77d testvector06m.dec
439 fc5e2f709693738324fb4c8bdc0dad6dda04e713 testvector07m.dec
440 aad2ba397bf1b6a18e8e09b50e4b19627d479f00 testvector08m.dec
441 6feb7a7b9d7cdc1383baf8d5739e2a514bd0ba08 testvector09m.dec
442 1b6cdf0413ac9965b16184b1bea129b5c0b2a37a testvector10m.dec
443 fd3d3a7b0dfbdab98d37ed9aa04b659b9fefbd18 testvector11m.dec
444 98e09bbafed329e341c3b4052e9c4ba5fc83f9b1 testvector12m.dec
446 Note that the decoder input bitstream files (.bit) are unchanged.
448 12. Security Considerations
450 This document fixes two security issues reported on Opus and that
451 affect the reference implementation in RFC 6716 [RFC6716]: CVE-
452 2013-0899 and CVE-
453 2017-0381 . CVE-
454 2013-0899 theoretically could have caused an information leak. The
455 leaked information would have gone through the decoder process before
456 being accessible to the attacker. It is fixed by Section 4. CVE-
457 2017-0381 could have resulted in a 16-bit out-of-bounds read from a
458 fixed location. It is fixed in Section 7. Beyond the two fixed
459 CVEs, this document adds no new security considerations on top of RFC
460 6716 [RFC6716].
462 13. IANA Considerations
464 This document makes no request of IANA.
466 Note to RFC Editor: this section may be removed on publication as an
467 RFC.
469 14. Acknowledgements
471 We would like to thank Juri Aedla for reporting the issue with the
472 parsing of the Opus padding. Thanks to Felicia Lim for reporting the
473 LSF integer overflow issue. Also, thanks to Tina le Grand, Jonathan
474 Lennox, and Mark Harris for their feedback on this document.
476 15. Normative References
478 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
479 Requirement Levels", BCP 14, RFC 2119,
480 DOI 10.17487/RFC2119, March 1997, .
483 [RFC6716] Valin, JM., Vos, K., and T. Terriberry, "Definition of the
484 Opus Audio Codec", RFC 6716, DOI 10.17487/RFC6716,
485 September 2012, .
487 Authors' Addresses
489 Jean-Marc Valin
490 Mozilla Corporation
491 331 E. Evelyn Avenue
492 Mountain View, CA 94041
493 USA
495 Phone: +1 650 903-0800
496 Email: jmvalin@jmvalin.ca
498 Koen Vos
499 vocTone
501 Email: koenvos74@gmail.com