chore(sync): align codebase with upstream open version

Synchronize 51 files across bpf, core, integration, internal,
pkg and vendor modules. Key changes:

- add xfs procfs parser and metrics integration tests
- add dev procfs parser for block device resolution
- extend watch and netutil test coverage
- refine dropwatch, hungtask and iolatency bpf programs
- update grafana dashboards and vendor procfs/xfs dependency

Signed-off-by: huhong-web <2420524547@qq.com>

Author: huhong-web

bpf/dropwatch.c

@@ -21,11 +21,6 @@
 #define SK_FL_TYPE_SHIFT 16
 #define SK_FL_TYPE_MASK 0xffff0000
 
-/* Reserved for future kernel skb_drop_reason (SKB_DROP_REASON_NOT_SPECIFIED,
- * ...) passthrough; (u32)-1 is out of band: kernel reason values grow upward
- * from 0 (low 16 bits code, high 16 bits subsystem since v6.4) and can never
- * reach it. */
-#define SKB_DROP_REASON_UNSUPPORT ((u32)-1)
 #define PKT_RAW_LEN 120
 
 struct packet_meta {
@@ -38,11 +33,10 @@ struct packet_meta {
 	u32 dev_flags;           /* 4  */
 	u32 queue_mapping;       /* 4  */
 	u32 drop_reason;         /* 4  */
-	u32 type;                /* 4  */
 	u32 net_inum;             /* 4  */
 	u8  dev_name[IFNAMSIZ];  /* 16 */
 	u8  comm[COMPAT_TASK_COMM_LEN]; /* 16 */
-};                           /* total: 96 bytes */
+};                           /* 92 bytes + 4 tail pad = 96 */
 
 struct packet_raw {
 	u16 eth_proto;    /* 2  */
@@ -86,6 +80,31 @@ char __license[] SEC("license") = "Dual MIT/GPL";
 static const struct drop_packet_event zero_data = {};
 static const u32 stackmap_key = 0;
 
+/* kfree_skb gained an skb drop reason field in v5.17, absent from the BTF this
+ * object is compiled against. Carry it in a CO-RE flavor relocated at load time:
+ * the anonymous enum field matches the kernel's enum skb_drop_reason by the name
+ * "reason". No reason constants are hardcoded (their values shift across kernels);
+ * names are resolved at runtime from kernel BTF in userspace (loadDropReasonNames).
+ *
+ * SKB_DROP_REASON_UNSUPPORT = -1 satisfies C's "enum needs >=1 enumerator" rule
+ * and is the out-of-band fallback for kernels predating the field: as (u32)-1 it
+ * can never collide with real reasons (which grow from 0). */
+struct trace_event_raw_kfree_skb___reason {
+	enum { SKB_DROP_REASON_UNSUPPORT = -1 } reason;
+} __attribute__((preserve_access_index));
+
+/* Return the kernel skb drop reason when the running kernel supports it,
+ * otherwise the out-of-band SKB_DROP_REASON_UNSUPPORT sentinel. */
+static inline u32 skb_get_drop_reason(struct trace_event_raw_kfree_skb *ctx)
+{
+	struct trace_event_raw_kfree_skb___reason *ctx_reason = (void *)ctx;
+
+	if (bpf_core_field_exists(ctx_reason->reason))
+		return BPF_CORE_READ(ctx_reason, reason);
+
+	return SKB_DROP_REASON_UNSUPPORT;
+}
+
 struct sock___5_10 {
 	u16 sk_type;
 	u16 sk_protocol;
@@ -187,7 +206,7 @@ int bpf_kfree_skb_prog(struct trace_event_raw_kfree_skb *ctx)
 		return 0;
 
 	/* pcap filter via bpf_pcap_stub.h: pass-through stub patched at load
-	 * time by internal/pcapinject with the compiled tcpdump expression.
+	 * time by internal/pcapfilter with the compiled tcpdump expression.
 	 */
 	if (!PCAP_STUB_PASS_SKB(skb))
 		return 0;
@@ -205,16 +224,11 @@ int bpf_kfree_skb_prog(struct trace_event_raw_kfree_skb *ctx)
 
 	/* meta */
 	data->meta.ktime_ns = bpf_ktime_get_ns();
-	/* tgid_pid and comm in SoftIRQ Context do not accurately reflect
-	 * the process where the drop occurred, only for reference.
-	 */
 	data->meta.tgid_pid = bpf_get_current_pid_tgid();
 	bpf_get_current_comm(&data->meta.comm, sizeof(data->meta.comm));
-
 	data->meta.kfree_skb_addr = (u64)(unsigned long)ctx->location;
 	data->meta.queue_mapping = BPF_CORE_READ(skb, queue_mapping);
-	data->meta.drop_reason = SKB_DROP_REASON_UNSUPPORT;
-	data->meta.type = 0;
+	data->meta.drop_reason = skb_get_drop_reason(ctx);
 
 	data->pkt_hdr.pkt_len = BPF_CORE_READ(skb, len);
 

bpf/include/bpf_pcap_stub.h

@@ -9,7 +9,7 @@
 
 /*
  * pcap_stub_l{2,3}: reserved-NOP stub functions patched at load time by
- * internal/pcapinject with compiled tcpdump filter bytecode. The three
+ * internal/pcapfilter with compiled tcpdump filter bytecode. The three
  * ctx parameters reserve R1/R2/R3; the patched bytecode overwrites them.
  * Unpatched body is a pass-through: data != data_end && three-way ctx
  * equality both hold, so the stub always returns true.
@@ -20,7 +20,7 @@
 static __noinline bool pcap_stub_l3(void *_ctx, void *__ctx, void *___ctx,
 				    void *data, void *data_end)
 {
-	/* 512 × 8-byte NOP insns; must equal nopAreaSize in internal/pcapinject/elfpatch.go */
+	/* 512 × 8-byte NOP insns; must equal stubReservedInsns in internal/pcapfilter/elfpatch.go */
 	asm volatile(".rept 512\n\t"
 		     ".byte 0xb7, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00\n\t"
 		     ".endr\n\t");
@@ -30,7 +30,7 @@ static __noinline bool pcap_stub_l3(void *_ctx, void *__ctx, void *___ctx,
 static __noinline bool pcap_stub_l2(void *_ctx, void *__ctx, void *___ctx,
 				    void *data, void *data_end)
 {
-	/* 512 × 8-byte NOP insns; must equal nopAreaSize in internal/pcapinject/elfpatch.go */
+	/* 512 × 8-byte NOP insns; must equal stubReservedInsns in internal/pcapfilter/elfpatch.go */
 	asm volatile(".rept 512\n\t"
 		     ".byte 0xb7, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00\n\t"
 		     ".endr\n\t");

bpf/include/vmlinux_net.h

@@ -17,6 +17,12 @@
 #define IP_MF           0x2000          /* Flag: "More Fragments"       */
 #define IP_OFFSET       0x1FFF          /* "Fragment Offset" part       */
 
+// skb_mac_header - return the MAC header pointer from sk_buff
+static inline unsigned char *skb_mac_header(struct sk_buff *skb)
+{
+    return BPF_CORE_READ(skb, head) + BPF_CORE_READ(skb, mac_header);
+}
+
 // skb_network_header - get the network header from sk_buff
 static inline unsigned char *skb_network_header(struct sk_buff *skb)
 {
@@ -28,10 +34,4 @@ static inline unsigned char *skb_transport_header(struct sk_buff *skb)
 {
     return BPF_CORE_READ(skb, head) + BPF_CORE_READ(skb, transport_header);
 }
-
-// skb_mac_header - return the MAC header pointer from sk_buff
-static inline unsigned char *skb_mac_header(struct sk_buff *skb)
-{
-    return BPF_CORE_READ(skb, head) + BPF_CORE_READ(skb, mac_header);
-}
 #endif

build/docker/grafana/dashboards/metrics-container.json

@@ -3952,4 +3952,4 @@
   "title": "Metric 大盘 - 容器视图",
   "uid": "metrics-container",
   "version": 1
-}
+}

cmd/huatuo-bamai/config/config_test.go

@@ -26,8 +26,7 @@ func writeConfigFile(t *testing.T, dir, name, content string) string {
 
 	path := filepath.Join(dir, name)
 	if err := os.WriteFile(path, []byte(content), 0o600); err != nil {
-		t.Errorf("write config file %s: %v", path, err)
-		return ""
+		t.Fatalf("write config file %s: %v", path, err)
 	}
 
 	return path
@@ -61,8 +60,7 @@ ExcludedOnContainer = "writeback"
 	}
 
 	if err := Load(path); err != nil {
-		t.Errorf("Load returned error: %v", err)
-		return
+		t.Fatalf("Load returned error: %v", err)
 	}
 
 	if len(Get().BlackList) != 2 {
@@ -110,8 +108,7 @@ ExcludedOnContainer = "writeback"
 	}
 
 	if err := Load(path); err != nil {
-		t.Errorf("Load returned error: %v", err)
-		return
+		t.Fatalf("Load returned error: %v", err)
 	}
 
 	Set("BlackList", []string{"netdev_hw", "metax_gpu"})
@@ -121,13 +118,11 @@ ExcludedOnContainer = "writeback"
 	Set("MetricCollector.Vmstat.IncludedOnContainer", "workingset_refault_file")
 
 	if err := Sync(); err != nil {
-		t.Errorf("Sync returned error: %v", err)
-		return
+		t.Fatalf("Sync returned error: %v", err)
 	}
 
 	if err := Load(path); err != nil {
-		t.Errorf("Load after Sync returned error: %v", err)
-		return
+		t.Fatalf("Load after Sync returned error: %v", err)
 	}
 
 	if len(Get().BlackList) != 2 {
@@ -148,8 +143,7 @@ ExcludedOnContainer = "writeback"
 
 	raw, err := os.ReadFile(path)
 	if err != nil {
-		t.Errorf("read synced config: %v", err)
-		return
+		t.Fatalf("read synced config: %v", err)
 	}
 	if !strings.Contains(string(raw), "[AutoTracing]") || !strings.Contains(string(raw), "IssuesList = [[\"cpuidle\", \"perf\"]]") {
 		t.Errorf("synced config should persist AutoTracing.IssuesList, got %s", string(raw))

core/events/memory_reclaim_events.go

@@ -26,14 +26,15 @@ import (
 	"huatuo-bamai/pkg/tracing"
 )
 
-type memoryReclaimTracing struct{}
-
-type memoryReclaimPerfEvent struct {
-	Comm      [bpf.TaskCommLen]byte
-	Deltatime uint64
-	CSS       uint64
-	Pid       uint64
-}
+type (
+	memoryReclaimTracing   struct{}
+	memoryReclaimPerfEvent struct {
+		Comm      [bpf.TaskCommLen]byte
+		Deltatime uint64
+		CSS       uint64
+		Pid       uint64
+	}
+)
 
 // MemoryReclaimTracingData is the full data structure.
 type MemoryReclaimTracingData struct {
@@ -56,9 +57,9 @@ func newMemoryReclaim() (*tracing.EventTracingAttr, error) {
 
 const cssCacheTTL = 5 * time.Second
 
-//go:generate $BPF_COMPILE $BPF_INCLUDE -s $BPF_DIR/memory_reclaim_events.c -o $BPF_DIR/memory_reclaim_events.o
-
 // Start detect work, load bpf and wait data form perfevent
+//
+//go:generate $BPF_COMPILE $BPF_INCLUDE -s $BPF_DIR/memory_reclaim_events.c -o $BPF_DIR/memory_reclaim_events.o
 func (c *memoryReclaimTracing) Start(ctx context.Context) error {
 	b, err := bpf.LoadBpf(bpf.ThisBpfOBJ(), map[string]any{
 		"deltath": cfg.MemoryReclaim.BlockedThreshold,

core/events/net_rx_latency.go

@@ -124,7 +124,9 @@ func (c *netRecvLatTracing) Start(ctx context.Context) error {
 
 	log.Debugf("net_rx_latency offset of mono to walltime: %v ns", monoWallOffset)
 
-	// Enable skb software RX timestamps before starting the tracer.
+	// for tracing 'net_rx_latency' keep the skb timestamp enabled,
+	// kernel func net_enable_timestamp() is system wide, can enable by set SOF_TIMESTAMPING_RX_SOFTWARE,
+	// ref: https://www.kernel.org/doc/html/latest/networking/timestamping.html.
 	tsfd, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, 0)
 	if err != nil {
 		return fmt.Errorf("create timestamp socket: %w", err)
@@ -234,6 +236,7 @@ func (c *netRecvLatTracing) Start(ctx context.Context) error {
 			}
 			log.Debugf("net_rx_latency tracerData: %+v", tracerData)
 
+			// save storage
 			if err := tracing.Save(&tracing.WriteRequest{
 				TracerName:  "net_rx_latency",
 				ContainerID: containerID,
@@ -251,7 +254,7 @@ func ignore(pid uint64, comm string, hostNetnsInode uint64) (containerID string,
 	dstInode, err := netutil.NetNSInodeByPid(int(pid))
 	if err != nil {
 		// ignore the missing program
-		if errors.Is(err, syscall.ENOENT) {
+		if errors.Is(err, syscall.ENOENT) || errors.Is(err, unix.EACCES) || errors.Is(err, unix.ESRCH) {
 			return "", true, nil
 		}
 		return "", skip, fmt.Errorf("get netns inode of pid %v failed: %w", pid, err)