After using the atomic_increment() method of BPF_HASH, the value obtained by lookup() does not start from 0, but 8

[xiaonuoz]

code:

const (
	file_read_c = `
	#include <asm/ptrace.h>
    #include <linux/fs.h>
    #include <linux/sched.h>

	BPF_HASH(read_remove_map, u64, u64, 256);

	struct readfile_event_t {
	};

	BPF_PERF_OUTPUT(watch_readfile_events);

	int trace_vfs_read(struct pt_regs *ctx, struct file *file, char __user *buf, size_t count, loff_t *pos)
	{
		u64 id = bpf_get_current_pid_tgid();
	
		read_remove_map.atomic_increment(id);

		char file_name[256];
		bpf_probe_read_kernel_str(&file_name, sizeof(file_name), (void *)file->f_path.dentry->d_name.name);
		if (file_name[0] != 'a' || file_name[1] != 'b'){
			return 0;
		}

		u64* ret = NULL;
		ret = read_remove_map.lookup(&id);
		if ( ret ){
			bpf_trace_printk("file_name:%s    count:%d\n",file_name, *ret);
		}
		return 0;
	}

	`
)

func WatchReadFile(ctx context.Context) {
	BpfModule := bpf.NewModule(file_read_c, []string{})

	traceVfsRead, err := BpfModule.LoadKprobe("trace_vfs_read")
	if err != nil {
		fmt.Println("traceVfsWrite err:", err)
		return
	}

	err = BpfModule.AttachKprobe("vfs_read", traceVfsRead, -1)
	if err != nil {
		fmt.Println("AttachKprobe err:", err)
		return
	}

	table := bpf.NewTable(BpfModule.TableId("watch_readfile_events"), BpfModule)

	channel := make(chan []byte, 2048)

	perfMap, err := bpf.InitPerfMap(table, channel, nil)
	if err != nil {
		fmt.Println("InitPerfMap err:", err)
		return
	}

	perfMap.Start()
	defer perfMap.Stop()

	for range channel {
		select {
		case <-ctx.Done():
			return
		default:
		}
	}
}

result:

[xiaonuoz]

Sorry, this issue has been resolved, because bpf_get_current_pid_tgid may appear duplicate