[202405][FRR] Fixing FRR to make route node lock (sonic-net#20990)

To fix a statistical issue. The original fix was done in FRRouting/frr#17297. However to accommodate 8.5.4 the patch in the PR was added.

[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `/usr/lib/frr/zebra -A 127.0.0.1 -s 90000000 -M dplane_fpm_nl -M snmp'.
Program terminated with signal SIGABRT, Aborted.
#0  0x00007fccd7351e2c in ?? () from /lib/x86_64-linux-gnu/libc.so.6
[Current thread is 1 (Thread 0x7fccd6faf7c0 (LWP 36))]
(gdb) bt
#0  0x00007fccd7351e2c in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#1  0x00007fccd7302fb2 in raise () from /lib/x86_64-linux-gnu/libc.so.6
#2  0x00007fccd72ed472 in abort () from /lib/x86_64-linux-gnu/libc.so.6
#3  0x00007fccd75bb3a9 in _zlog_assert_failed (xref=xref@entry=0x7fccd7652380 <_xref.16>, extra=extra@entry=0x0) at ../lib/zlog.c:678
#4  0x00007fccd759b2fe in route_node_delete (node=<optimized out>) at ../lib/table.c:352
#5  0x00007fccd759b445 in route_unlock_node (node=0x0) at ../lib/table.h:258
#6  route_next (node=<optimized out>) at ../lib/table.c:436
#7  route_next (node=node@entry=0x56029d89e560) at ../lib/table.c:410
#8  0x000056029b6b6b7a in if_lookup_by_name_per_ns (ns=ns@entry=0x56029d873d90, ifname=ifname@entry=0x7fccc0029340 "PortChannel1020")
    at ../zebra/interface.c:312
#9  0x000056029b6b8b36 in zebra_if_dplane_ifp_handling (ctx=0x7fccc0029310) at ../zebra/interface.c:1867
#10 zebra_if_dplane_result (ctx=0x7fccc0029310) at ../zebra/interface.c:2221
#11 0x000056029b7137a9 in rib_process_dplane_results (thread=<optimized out>) at ../zebra/zebra_rib.c:4810
#12 0x00007fccd75a0e0d in thread_call (thread=thread@entry=0x7ffe8e553cc0) at ../lib/thread.c:1990
#13 0x00007fccd7559368 in frr_run (master=0x56029d65a040) at ../lib/libfrr.c:1198
#14 0x000056029b6ac317 in main (argc=9, argv=0x7ffe8e5540d8) at ../zebra/main.c:478

src/sonic-frr/patch/0054-make-route-node-lock-atomic.patch

@@ -0,0 +1,95 @@
+From 402665769dc5563c246003720e75c3d6244a88eb Mon Sep 17 00:00:00 2001
+From: "Barry Friedman (friedman)" <friedman@cisco.com>
+Date: Wed, 9 Oct 2024 18:01:00 +0000
+Subject: [PATCH 1/2] lib: make route_node lock atomic
+
+Multiple threads walking a route table perform read-modify-write
+operations on the lock variable. This can result in a wrong lock count
+which wrongly triggers a route_node_delete(). Making the access to the
+lock variable atomic should ensure the lock is correct.
+
+Signed-off-by: Barry Friedman (friedman) <friedman@cisco.com>
+---
+ lib/table.h | 6 +++---
+ 1 file changed, 3 insertions(+), 3 deletions(-)
+
+diff --git a/lib/table.h b/lib/table.h
+index 5dec69ee7e..89993dd279 100644
+--- a/lib/table.h
++++ b/lib/table.h
+@@ -129,7 +129,7 @@ struct route_table {
+ 	struct route_node *table_rdonly(link[2]);                              \
+                                                                                \
+ 	/* Lock of this radix */                                               \
+-	unsigned int table_rdonly(lock);                                       \
++	_Atomic unsigned int table_rdonly(lock);			       \
+                                                                                \
+ 	struct rn_hash_node_item nodehash;                                     \
+ 	/* Each node of route. */                                              \
+@@ -244,7 +244,7 @@ extern void route_table_iter_cleanup(route_table_iter_t *iter);
+ /* Lock node. */
+ static inline struct route_node *route_lock_node(struct route_node *node)
+ {
+-	(*(unsigned *)&node->lock)++;
++	atomic_fetch_add_explicit(&node->lock, 1, memory_order_relaxed);
+ 	return node;
+ }
+
+@@ -252,7 +252,7 @@ static inline struct route_node *route_lock_node(struct route_node *node)
+ static inline void route_unlock_node(struct route_node *node)
+ {
+ 	assert(node->lock > 0);
+-	(*(unsigned *)&node->lock)--;
++	atomic_fetch_sub_explicit(&node->lock, 1, memory_order_relaxed);
+
+ 	if (node->lock == 0)
+ 		route_node_delete(node);
+-- 
+2.43.2
+
+
+From bf96a5a334ea1e953bf98a92f6a942335f49700b Mon Sep 17 00:00:00 2001
+From: "Barry Friedman (friedman)" <friedman@cisco.com>
+Date: Fri, 11 Oct 2024 22:02:51 +0000
+Subject: [PATCH 2/2] lib: Fix atomic node lock warnings with enable-dev-build
+
+The route_node lock variable type adds a const qualifier when
+--enable-dev-build is configured. This is to generate warnings
+in case client code tries to modify this variable. As a result
+the functions route_lock_node() and route_unlock_node() need to
+cast away the const qualifier to prevent the compile warnings.
+When they _Atomic qualifier was added I wrongly removed the
+cast. This change adds it back to prevent the warnings.
+
+Signed-off-by: Barry Friedman (friedman) <friedman@cisco.com>
+---
+ lib/table.h | 6 ++++--
+ 1 file changed, 4 insertions(+), 2 deletions(-)
+
+diff --git a/lib/table.h b/lib/table.h
+index 89993dd279..8023f8aa30 100644
+--- a/lib/table.h
++++ b/lib/table.h
+@@ -244,7 +244,8 @@ extern void route_table_iter_cleanup(route_table_iter_t *iter);
+ /* Lock node. */
+ static inline struct route_node *route_lock_node(struct route_node *node)
+ {
+-	atomic_fetch_add_explicit(&node->lock, 1, memory_order_relaxed);
++	atomic_fetch_add_explicit((_Atomic unsigned *)&node->lock, 1,
++				  memory_order_relaxed);
+ 	return node;
+ }
+
+@@ -252,7 +253,8 @@ static inline struct route_node *route_lock_node(struct route_node *node)
+ static inline void route_unlock_node(struct route_node *node)
+ {
+ 	assert(node->lock > 0);
+-	atomic_fetch_sub_explicit(&node->lock, 1, memory_order_relaxed);
++	atomic_fetch_sub_explicit((_Atomic unsigned *)&node->lock, 1,
++				  memory_order_relaxed);
+
+ 	if (node->lock == 0)
+ 		route_node_delete(node);
+-- 
+2.43.2
+

src/sonic-frr/patch/series

@@ -51,3 +51,4 @@
 0051-bgpd-backpressure-fix-ret-value-evpn_route_select_in.patch
 0052-bgpd-backpressure-log-error-for-evpn-when-route-inst.patch
 0053-Patch-to-send-tag-value-associated-with-route-via-ne.patch
+0054-make-route-node-lock-atomic.patch