Add SIPREC recording support (RFC 7865/7866) to SBC by hecko · Pull Request #266 · sems-server/sems

hecko · 2026-02-11T14:04:24Z

Implement a SIPREC Session Recording Client (cc_siprec) as an SBC call_control module and a minimal Session Recording Server (siprec_srs) as a standalone SEMS application.

cc_siprec hooks into the SBC's RTP relay path to fork media streams to an SRS via SIPREC signaling. It manages the full SIPREC dialog lifecycle: INVITE with multipart SDP + RFC 7865 metadata XML, hold/resume updates, and BYE on call teardown.

siprec_srs receives recording INVITEs, negotiates dual RTP streams (one per call leg), decodes G.711 A-law/u-law based on the RTP payload type, and writes each leg to a separate 16-bit PCM WAV file alongside the SIPREC metadata XML.

Copilot

Pull request overview

This PR introduces SIPREC (RFC 7865/7866) session recording to the SEMS SBC by adding a new SBC call_control module (cc_siprec) that forks RTP to an SRS, plus a minimal SIPREC Session Recording Server application (siprec_srs) that receives the recording INVITEs and writes per-leg WAV + metadata XML.

Changes:

Add cc_siprec call_control module with SIPREC dialog handling, metadata generation, and RTP forking to an SRS.
Add siprec_srs SEMS application to accept SIPREC INVITEs, receive dual RTP streams, and write WAV/XML artifacts.
Extend SBC RTP relay with an ExtendedCCInterface::onAfterRTPRelay hook; adjust config installation behavior and improve Max-Forwards/registrar logging.

Reviewed changes

Copilot reviewed 28 out of 29 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
core/CMakeLists.txt	Changes config install behavior to preserve existing configs and install `.dist` references.
cmake/config.rules.txt	Updates shared config-install macro to install `.dist` and conditionally install real configs.
core/AmB2BSession.cpp	Adds WARN log when rejecting relayed requests with Max-Forwards=0.
apps/sbc/SBC.cpp	Adds WARN log when rejecting OoD requests with Max-Forwards=0.
apps/sbc/SBCCallLeg.cpp	Adds WARN log for Max-Forwards check; invokes new RTP relay hook for extended CC modules.
apps/sbc/ExtendedCCInterface.h	Introduces `onAfterRTPRelay` hook (RTP-thread callback).
apps/sbc/call_control/CMakeLists.txt	Adds `siprec` call_control subdirectory.
apps/sbc/call_control/siprec/*	Implements SIPREC client: SIP dialog, metadata XML builder, RTP forwarder, config + docs, and build rules.
apps/siprec_srs/*	Implements minimal SIPREC SRS: SDP negotiation for two legs, RTP receive/decoding, WAV writer, config + docs, and build rules.
apps/sbc/call_control/registrar/Registrar.cpp	Improves log messages for retargeting/404 and REGISTER handling.
apps/CMakeLists.txt	Adds `siprec_srs` application to build.
README.md	Documents `cc_siprec` under SBC call_control modules and adds `siprec_srs` under Recording.
.gitignore	Adds `.claude` to ignored files.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-11T14:18:43Z

apps/sbc/call_control/siprec/RtpForwarder.cpp

+    // Don't flood logs - only log occasionally
+    static int err_count = 0;
+    if (++err_count % 1000 == 1) {
+      ERROR("SIPREC RtpForwarder: sendto() failed: %s (error #%d)\n",
+            strerror(errno), err_count);
+    }


err_count is a function-local static int updated from sendPacket(), which is intended to be called from RTP threads. This introduces a data race (and couples rate limiting across all instances). Use an atomic counter, or move the counter to a per-instance member guarded appropriately.

Copilot · 2026-02-11T14:18:43Z

apps/sbc/call_control/siprec/CCSiprec.cpp

+    int ret = state->dialog->startRecording(srs_uri, caller_uri, callee_uri,
+                                            cs_sdp, call_id);
+    if (ret != 0) {
+      ERROR("SIPREC: failed to start recording for call %s\n", ltag.c_str());
+      removeState(ltag);
+    }


The PR description mentions recording_mandatory should reject/terminate the call if the SRS is unreachable, but recording_mandatory_ is never used and failures to start recording only log and remove state. If mandatory recording is required, this code should trigger a concrete call action (e.g., immediately hang up/refuse the call) when startRecording() fails or when the SRS rejects the INVITE.

Copilot · 2026-02-11T14:18:44Z

apps/sbc/call_control/siprec/SiprecDialog.cpp

+  if (req.method == SIP_METH_INVITE) {
+    if (reply.code >= 200 && reply.code < 300) {
+      // 200 OK - parse SDP answer to get SRS RTP ports
+      if (reply.body.getLen()) {


if (reply.body.getLen()) will be false for multipart bodies because AmMimeBody::getLen() returns 0 for multipart even when parts exist. This can cause SDP parsing to be skipped even though an application/sdp part is present. Use !reply.body.empty() (or just attempt hasContentType() unconditionally) instead of checking getLen().

Suggested change

if (reply.body.getLen()) {

if (!reply.body.empty()) {

Copilot · 2026-02-11T14:18:44Z

apps/siprec_srs/SiprecSrs.cpp

+         leg + 1, bound_port, codec_name.c_str(), pt, wav_path.c_str());
+
+    leg++;
+    if (leg >= 2) break;  // only accept first 2 audio m-lines


The SDP answer must include the same number of m-lines as the offer (RFC 3264). Breaking out of the loop after 2 legs can produce an answer with fewer m-lines than the offer if the INVITE contains additional media sections. Instead, continue iterating and set remaining m-lines to port=0 (rejected/disabled) to preserve ordering/count.

Suggested change

if (leg >= 2) break; // only accept first 2 audio m-lines

// Do not break here: per RFC 3264, the SDP answer must preserve the

// number and ordering of m-lines from the offer, so we continue

// iterating over all media sections.

Copilot · 2026-02-11T14:18:44Z

apps/siprec_srs/RtpReceiver.cpp

+    // Read payload type from RTP header (byte 1, bits 0-6)
+    uint8_t pt = buf[1] & 0x7F;
+
+    // Skip non-audio payload types (CN=13, telephone-event=96-127)
+    if (pt == 13 || pt >= 96) continue;
+
+    int payload_len = len - hdr_len;
+    uint8_t* payload = buf + hdr_len;
+
+    // Decode G.711 payload to 16-bit PCM based on actual RTP payload type
+    int16_t pcm_buf[RTP_MAX_PACKET];
+    for (int i = 0; i < payload_len; i++) {
+      if (pt == 0)
+        pcm_buf[i] = decodePCMU(payload[i]);
+      else
+        pcm_buf[i] = decodePCMA(payload[i]);
+    }


RTP decoding currently ignores the negotiated/expected codec (codec_ is set in init() but never used) and decodes any payload type < 96 as PCMA (except pt=0). This can produce corrupted WAV output (and is unsafe if unexpected PTs arrive). Consider validating pt against the expected PT from SDP and dropping mismatches; if dynamic PTs are possible, the SDP negotiation should enforce/translate to 0/8 or the receiver should accept the negotiated dynamic PT.

Copilot · 2026-02-11T14:18:45Z

apps/sbc/call_control/siprec/CCSiprec.cpp

+    string ltag = call->getLocalTag();
+    SiprecCallState* state = getState(ltag);
+    if (state && state->dialog) {
+      state->dialog->stopRecording();
+    }


getState() returns a raw pointer obtained under sessions_mut_, but it is used after the lock is released. If another thread calls removeState() concurrently, state/state->dialog can become dangling before stopRecording() is invoked. Consider returning a ref-counted handle (or otherwise guaranteeing lifetime) for all getState() call sites.

Copilot · 2026-02-11T14:18:45Z

apps/sbc/call_control/siprec/CCSiprec.cpp

+    if (it != sessions_.end()) state = it->second;
+  }
+
+  if (state && state->dialog) {
+    state->dialog->forwardRtp(p, is_aleg);


state is looked up under sessions_mut_ but then used after the lock is released. Since removeState() can delete the SiprecCallState concurrently (e.g., from onDestroyLeg() while the RTP thread is running), this can become a use-after-free when calling state->dialog->forwardRtp(). Consider keeping the object alive via ref-counting (e.g., shared_ptr), or copying out a stable pointer (e.g., SiprecDialog*) under the lock with a lifetime guarantee, or deferring deletion until RTP forwarding is stopped.

Suggested change

if (it != sessions_.end()) state = it->second;

}

if (state && state->dialog) {

state->dialog->forwardRtp(p, is_aleg);

if (it != sessions_.end())

state = it->second;

if (state && state->dialog) {

// Call forwardRtp while holding sessions_mut_ so that removeState()

// cannot delete the SiprecCallState concurrently.

state->dialog->forwardRtp(p, is_aleg);

}

Copilot · 2026-02-11T14:18:45Z

apps/sbc/call_control/siprec/CCSiprec.cpp

+      // For B-leg, search by b_leg_ltag
+      for (it = sessions_.begin(); it != sessions_.end(); ++it) {
+        if (it->second->b_leg_ltag == ltag) break;
+      }


For B-leg packets this performs a linear scan over all active sessions on every RTP packet (for (it = sessions_.begin(); ...)). This is O(N) per packet and will not scale with concurrent calls. Consider maintaining a secondary map keyed by b_leg_ltag (or storing the recording state pointer directly on the call leg) so lookup is O(1).

Copilot · 2026-02-11T14:18:45Z

apps/sbc/call_control/siprec/SiprecDialog.cpp

+  SdpMedia m1;
+  m1.type = MT_AUDIO;
+  m1.port = 0;         // SRS will tell us where to send (we use our own UDP socket)
+  m1.transport = TP_RTPAVP;
+  m1.send = true;
+  m1.recv = false;     // sendonly
+


Setting the offered m-line port to 0 makes the stream "disabled" per RFC 3264. Some SIPREC SRS implementations will reject/ignore such m-lines, which will break interoperability. If the SRC intends to send RTP, it should advertise a real local RTP port (and ideally send from that port by binding the forwarder socket accordingly).

Copilot · 2026-02-11T14:18:46Z

apps/siprec_srs/RtpReceiver.h

+#include "AmThread.h"
+
+#include <string>
+#include <atomic>


This header uses fixed-width integer types (uint32_t, uint8_t, int16_t) but doesn't include <cstdint> / <stdint.h>. Relying on indirect includes is non-portable and can break builds on some platforms/toolchains.

Suggested change

#include <atomic>

#include <atomic>

#include <cstdint>

Implement a SIPREC Session Recording Client (SRC) as an SBC call_control module (cc_siprec) that forks RTP to a Session Recording Server (SRS), plus a minimal built-in SRS (siprec_srs) for testing. SIP signaling (RFC 7866): - INVITE with Require: siprec and +sip.src Contact feature tag - multipart/mixed body: SDP offer + metadata XML - Content-Disposition: recording-session on metadata parts - Separate sendonly m-lines per CS direction with a=label - Symmetric RTP with allocated local port pairs (RFC 7866 8.1.8) - RTCP port allocation on port+1 (RFC 7866 8.1.1) - Configurable RTP profile: RTP/AVP, RTP/SAVP, RTP/AVPF, RTP/SAVPF - CANCEL for pending INVITE on call teardown - re-INVITE for hold/resume with updated SDP + metadata - BYE with final metadata and stop-time Metadata XML (RFC 7865): - Namespace urn:ietf:params:xml:ns:recording:1 - session, participant, stream elements with base64 IDs - sessionrecordingassoc and participantsessionassoc elements - participantstreamassoc with send/recv text-content stream refs - ISO 8601 timestamps, partial updates on hold Recording indication (RFC 7866 6.1.2): - a=record:on injected into B-leg INVITE SDP CS SDP codec extraction: - Parses initial INVITE SDP to auto-detect codec for RS offer - Falls back to configured codec (PCMA/PCMU) when unavailable Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Copilot AI review requested due to automatic review settings February 11, 2026 14:04

Copilot started reviewing on behalf of hecko February 11, 2026 14:04 View session

Copilot AI reviewed Feb 11, 2026

View reviewed changes

hecko force-pushed the feature/siprec branch from 1598442 to 28b11dc Compare February 11, 2026 14:33

hecko merged commit 22043a7 into master Feb 11, 2026
11 checks passed

hecko deleted the feature/siprec branch February 11, 2026 17:24

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add SIPREC recording support (RFC 7865/7866) to SBC#266

Add SIPREC recording support (RFC 7865/7866) to SBC#266
hecko merged 1 commit intomasterfrom
feature/siprec

hecko commented Feb 11, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

-    if (leg >= 2) break;  // only accept first 2 audio m-lines
+    // Do not break here: per RFC 3264, the SDP answer must preserve the
+    // number and ordering of m-lines from the offer, so we continue
+    // iterating over all media sections.

Conversation

hecko commented Feb 11, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant