BFD hardware offload for BGP sessions by baorliu · Pull Request #4253 · sonic-net/sonic-swss

baorliu · 2026-02-23T22:37:51Z

Original PR 3267 was messed up when trying to fix DCO error. this one is the replacement for that PR for code change reference.

For detailed description and commit history, please refer the original PR:
#3267

What I did
Provide an option to use hardware offloaded BFD to replace software BFD (frr/bfdd) for BGP neighbor reachability detection.

Why I did it
Hardware offloaded BFD usually provide more bfd sessions support, and has shorter detection time comparing to software bfd. and also lower CPU load because that BFD sessions are offloaded to hardware/ASIC.

How I verified it

unit test
The design was built into a sonic image, and verified in a hardware testbed. This design has been in multiple production image for more than one year.

Details if related
This program is built under sonic-swss, but running in bgp container.

HLD in PR review:
sonic-net/SONiC#1599

[notes] some commit history are lost after a force push after fixing a DCO error.
git log before the force push:

commit 42c0b3ef874d3f58c10e03b29cdce29876cb497a
Author: Baorong Liu <96146196+baorliu@users.noreply.github.com>
Date:   Mon Feb 23 11:51:51 2026 -0800

    fix makefile
    
    Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>
commit 84d4c1bac48e93e3c6fcf80ddda75d4325e61484
Author: Baorong Liu <96146196+baorliu@users.noreply.github.com>
Date:   Tue Apr 29 13:07:18 2025 -0700

    add ut for bfdsyncd
    
    Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>
commit 32db91f472501b67f42b74dc16f3a5ece9c49c23
Author: Baorong Liu <96146196+baorliu@users.noreply.github.com>
Date:   Wed Feb 12 12:03:23 2025 -0800

    handle bgp restart case
    
    Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

commit d9d71e8da3c2b3e7ae14557a01d3abed1d0eaa06
Author: Baorong Liu <96146196+baorliu@users.noreply.github.com>
Date:   Mon Nov 4 15:16:34 2024 -0800

    remove some debugging code
    
    Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

commit 4e608ca1d163454dd0d70d1fa50e5d1a53839ad6
Author: Baorong Liu <baorliu@cisco.com>
Date:   Mon Aug 26 21:14:19 2024 -0700

    bgp bfd hardware offload initial commit
    
    Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

mssonicbld · 2026-02-23T22:37:59Z

/azp run

azure-pipelines · 2026-02-23T22:38:08Z

Azure Pipelines successfully started running 1 pipeline(s).

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

mssonicbld · 2026-02-24T00:20:23Z

/azp run

azure-pipelines · 2026-02-24T00:20:33Z

Azure Pipelines successfully started running 1 pipeline(s).

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

mssonicbld · 2026-02-24T00:56:31Z

/azp run

azure-pipelines · 2026-02-24T00:56:41Z

Azure Pipelines successfully started running 1 pipeline(s).

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

mssonicbld · 2026-02-24T02:53:38Z

/azp run

azure-pipelines · 2026-02-24T02:53:48Z

Azure Pipelines successfully started running 1 pipeline(s).

Signed-off-by: Baorong Liu <baorliu@cisco.com>

mssonicbld · 2026-04-15T17:52:10Z

/azp run

azure-pipelines · 2026-04-15T17:52:17Z

Azure Pipelines will not run the associated pipelines, because the pull request was updated after the run command was issued. Review the pull request again and issue a new run command.

mssonicbld · 2026-04-15T18:25:15Z

/azp run

azure-pipelines · 2026-04-15T18:25:26Z

Azure Pipelines successfully started running 1 pipeline(s).

Signed-off-by: Baorong Liu <baorliu@cisco.com>

mssonicbld · 2026-04-15T18:28:40Z

/azp run

azure-pipelines · 2026-04-15T18:28:51Z

Azure Pipelines successfully started running 1 pipeline(s).

vaishnav-nexthop · 2026-05-20T13:14:48Z

+std::string BfdLink::exec(const char* cmd) {
+    std::array<char, 128> buffer;
+    std::string result;
+    std::unique_ptr<FILE, decltype(&pclose)> pipe(popen(cmd, "r"), pclose);


This line breaks the swss .deb build on Debian Trixie / GCC 14.2 with -Werror=ignored-attributes:

bfdsyncd/bfdlink.cpp:128:44: error: ignoring attributes on template argument 'int (*)(FILE*)' [-Werror=ignored-attributes]

pclose is declared in glibc with __attribute__((nonnull)), and that attribute is stripped when decltype(&pclose) is used as the Deleter template parameter to std::unique_ptr. GCC 13 (Bookworm) was silent about it, anyone moving to a Trixie-based slave image (sonic-slave-trixie) hits a hard build failure since swss builds -Werror.

vaishnav-nexthop · 2026-05-20T13:19:25Z

+        /* Remote discriminator. */
+        if (field == "remote_discriminator")
+        {
+            uint32_t rid = (uint32_t)strtoll(string(value).c_str(), NULL, 10);


All four strtoll(value.c_str(), NULL, 10) calls in handleBfdStateUpdate silently return 0 on garbage input. A malformed remote_discriminator / remote_min_rx / remote_min_tx / remote_multiplier field in STATE_DB will be turned into a zero, packed into msg.data.state.*, and sent over the BFDDP socket to bfdd — the operator gets no signal that the state notification is corrupt.

Worth replacing with swss::to_uint<uint32_t>(value) inside a try/catch (std::exception&) so a bad row logs the offending field name and the function returns false before sending a partial frame.

vaishnav-nexthop · 2026-05-22T10:07:25Z

+                m_bfdTable.del(bfdkey);
+                m_key2bfd.erase(bfdkey_map);
+                /* the symptom observed that redis eliminates consecutive del and add transaction sometime, get wrong result. need to wait to make sure deletion done */
+                usleep(100000);


Do the timer fields actually need a delete/recreate? Per saibfd.h, MIN_TX, MIN_RX, and MULTIPLIER are CREATE_AND_SET, so they should be modifiable in place via set_bfd_session_attribute. A delete/recreate bounces the session to Down momentarily — which can flap the BGP sessions BFD is protecting, defeating the point.

For the fields that are CREATE_ONLY (MULTIHOP, CBIT, SRC_IP_ADDRESS, DST_IP_ADDRESS, LOCAL_DISCRIMINATOR), delete/recreate is unavoidable

The reason to use delete/recreate approach is to make it compatible with existing systems/vendors:
1, current orchagent/bfdorch does not support attribute update:

In create_bfd_session: if (bfd_session_map.find(key) != bfd_session_map.end()) { SWSS_LOG_ERROR("BFD session for %s already exists", key.c_str()); return true; }

2, with the delete/recreate approach, parameter update from frr/bfdd still works even some vendor does not support BFD attribute update in SAI. so it is safe for all the vendors who already support BFD HW offload, even that vendor does not support BFD (HW offload) attribute update.

Comment from Baorong Liu

Copilot

Pull request overview

This PR adds support infrastructure for using hardware-offloaded BFD sessions (created/managed via SAI in orchagent) to feed BFD session updates to BGP/bfdd via a new bfdsyncd daemon, enabling hardware BFD to replace software BFD for BGP neighbor reachability detection.

Changes:

Add new bfdsyncd daemon that accepts FRR/bfdd dataplane protocol messages, programs APP_BFD_SESSION_TABLE, and forwards STATE_DB BFD state updates back to the client.
Extend BfdOrch state-change handling to persist additional remote session parameters in STATE_DB, and allow src_mac input when creating sessions with HW_LOOKUP_VALID=false.
Add a new mock unit test target for bfdsyncd and wire bfdsyncd into the build system.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
`Makefile.am`	Adds `bfdsyncd` to the top-level build subdirectories.
`configure.ac`	Registers `bfdsyncd/Makefile` for autotools generation.
`orchagent/bfdorch.cpp`	Persists BFD remote parameters on state change; supports optional `src_mac` during session creation when HW lookup is invalid.
`bfdsyncd/Makefile.am`	Adds build rules for the new `bfdsyncd` binary.
`bfdsyncd/bfdsyncd.cpp`	Implements the main loop for the new daemon (accept client + select loop with STATE_DB subscription).
`bfdsyncd/bfdlink.h`	Declares the `BfdLink` selectable handling the bfdd dataplane protocol and DB interactions.
`bfdsyncd/bfdlink.cpp`	Implements bfdd dataplane message parsing, DB programming, and state/counter message responses.
`bfdsyncd/bfdd/bfddp_packet.h`	Adds the bfdd dataplane protocol structures/constants used by `bfdsyncd`.
`tests/mock_tests/Makefile.am`	Adds `tests_bfdsyncd` target to the mock test suite build.
`tests/mock_tests/bfdsyncd/test_bfdlink.cpp`	Adds unit tests covering basic session add/state-update/counter-request/delete flows.

+    /* Message header. */
+    msg.header.version = BFD_DP_VERSION;
+    msg.header.length = ntohs(msglen);
+    msg.header.type = ntohs(BFD_STATE_CHANGE);
+


+BfdLink::~BfdLink()
+{
+    delete[] m_messageBuffer;
+    if (m_connected)
+        close(m_connection_socket);
+    if (m_server_up)
+        close(m_server_socket);
+}


+    m_connection_socket = ::accept(m_server_socket, (struct sockaddr *)&client_addr,
+                                   &client_len);
+    if (m_connection_socket < 0)
+        throw system_error(errno, system_category());
+
+    SWSS_LOG_WARN("New connection accepted from: %s\n", inet_ntoa(client_addr.sin_addr));
+}


+        /* check link local ip address 169.254.0.0/16 0xa9fe0000 */
+        if ((inet_pton(AF_INET, dst_addr, &v4) == 1) && ((v4.s_addr & 0x0000ffff) == 0x0000fea9)) {
+            is_linklocal = true;
+            SWSS_LOG_INFO("dst_addr %s is a link local ip address", dst_addr);
+        }


+    while ((c = getopt (argc, argv, "hdp:")) != -1)
+        switch (c)
+            {
+            case 'h':
+                cout << "Usage: bfdsyncd -d -p <tcp port number>" << endl;
+                break;
+            case 'd':


+#ifndef __BFDLINK__
+#define __BFDLINK__


+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the ?Software?), to
+ * deal in the Software without restriction, including without limitation the


+ * THE SOFTWARE IS PROVIDED ?AS IS?, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,


prsunny · 2026-06-12T01:40:51Z

@baorliu , please address copilot comments and resolve conflict.

Merge PR sonic-net#4253 and address review feedback: fix GCC 14 pclose deleter build failure, validate STATE_DB numeric fields, correct BFDDP header byte order, link-local detection, resource cleanup, and master merge conflicts in mock_tests Makefile. Use ephemeral port 0 in unit tests to avoid CI bind conflicts on the default BFDDP port. Signed-off-by: Sridhar talari <stalarir@cisco.com>

prsunny · 2026-06-30T22:50:28Z

Closing in favor of #4676

bfdsyncd, recover pr 3267

87c6166

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

baorliu requested a review from prsunny as a code owner February 23, 2026 22:37

fix tests

fa9d78d

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

fix tests

9152e81

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

fix tests

c9c67ac

Signed-off-by: Baorong Liu <96146196+baorliu@users.noreply.github.com>

baorliu mentioned this pull request Feb 27, 2026

[swss]: BFD hardware offload for BGP sessions #3267

Open

rajshekhar-nexthop mentioned this pull request Mar 26, 2026

[bfd] BFD HW offload for BGP sessions sonic-net/SONiC#1599

Merged

resolve conflicts in mock_test Makefile.am

cb9bca0

Signed-off-by: Baorong Liu <baorliu@cisco.com>

Merge branch 'master' into baorliu_bgp_bfd_new

7eac609

Signed-off-by: Baorong Liu <baorliu@cisco.com>

baorliu force-pushed the baorliu_bgp_bfd_new branch from a09e3a4 to 7eac609 Compare April 15, 2026 18:28

baorliu changed the title ~~[swss] bfdsyncd, recreate PR for messed-up pr-3267~~ [swss] BFD hardware offload for BGP sessions Apr 16, 2026

baorliu changed the title ~~[swss] BFD hardware offload for BGP sessions~~ BFD hardware offload for BGP sessions Apr 16, 2026

vaishnav-nexthop reviewed May 22, 2026

View reviewed changes

balajib-cisco mentioned this pull request May 24, 2026

[bfdsyncd]: BFD hardware offload for BGP sessions #4588

Draft

prsunny requested a review from Copilot June 12, 2026 01:33

Copilot started reviewing on behalf of prsunny June 12, 2026 01:33 View session

Copilot AI reviewed Jun 12, 2026

View reviewed changes

sridhartalari mentioned this pull request Jun 15, 2026

[bfdsyncd] BFD hardware offload for BGP sessions (patched #4253) #4676

Open

3 tasks

prsunny closed this Jun 30, 2026

		* THE SOFTWARE IS PROVIDED ?AS IS?, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
		* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

		#ifndef __BFDLINK__
		#define __BFDLINK__

Uh oh!

Conversation

baorliu commented Feb 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mssonicbld commented Feb 23, 2026

Uh oh!

azure-pipelines Bot commented Feb 23, 2026

Uh oh!

mssonicbld commented Feb 24, 2026

Uh oh!

azure-pipelines Bot commented Feb 24, 2026

Uh oh!

mssonicbld commented Feb 24, 2026

Uh oh!

azure-pipelines Bot commented Feb 24, 2026

Uh oh!

mssonicbld commented Feb 24, 2026

Uh oh!

azure-pipelines Bot commented Feb 24, 2026

Uh oh!

mssonicbld commented Apr 15, 2026

Uh oh!

azure-pipelines Bot commented Apr 15, 2026

Uh oh!

mssonicbld commented Apr 15, 2026

Uh oh!

azure-pipelines Bot commented Apr 15, 2026

Uh oh!

mssonicbld commented Apr 15, 2026

Uh oh!

azure-pipelines Bot commented Apr 15, 2026

Uh oh!

vaishnav-nexthop May 20, 2026

Choose a reason for hiding this comment

Uh oh!

vaishnav-nexthop May 20, 2026

Choose a reason for hiding this comment

Uh oh!

vaishnav-nexthop May 22, 2026

Choose a reason for hiding this comment

Uh oh!

liu14187 Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

prsunny commented Jun 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

prsunny commented Jun 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

baorliu commented Feb 23, 2026 •

edited

Loading

prsunny commented Jun 12, 2026 •

edited

Loading