samples/bpf/sampleip_user.c

   1 /*
   2  * sampleip: sample instruction pointer and frequency count in a BPF map.
   3  *
   4  * Copyright 2016 Netflix, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of version 2 of the GNU General Public
   8  * License as published by the Free Software Foundation.
   9  */
  10 #include <stdio.h>
  11 #include <stdlib.h>
  12 #include <stdio.h>
  13 #include <unistd.h>
  14 #include <errno.h>
  15 #include <signal.h>
  16 #include <string.h>
  17 #include <assert.h>
  18 #include <linux/perf_event.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/bpf.h>
  21 #include <sys/ioctl.h>
  22 #include "libbpf.h"
  23 #include "bpf_load.h"
  24 #include "perf-sys.h"
  25
  26 #define DEFAULT_FREQ    99
  27 #define DEFAULT_SECS    5
  28 #define MAX_IPS         8192
  29 #define PAGE_OFFSET     0xffff880000000000
  30
  31 static int nr_cpus;
  32
  33 static void usage(void)
  34 {
  35         printf("USAGE: sampleip [-F freq] [duration]\n");
  36         printf("       -F freq    # sample frequency (Hertz), default 99\n");
  37         printf("       duration   # sampling duration (seconds), default 5\n");
  38 }
  39
  40 static int sampling_start(int *pmu_fd, int freq)
  41 {
  42         int i;
  43
  44         struct perf_event_attr pe_sample_attr = {
  45                 .type = PERF_TYPE_SOFTWARE,
  46                 .freq = 1,
  47                 .sample_period = freq,
  48                 .config = PERF_COUNT_SW_CPU_CLOCK,
  49                 .inherit = 1,
  50         };
  51
  52         for (i = 0; i < nr_cpus; i++) {
  53                 pmu_fd[i] = sys_perf_event_open(&pe_sample_attr, -1 /* pid */, i,
  54                                             -1 /* group_fd */, 0 /* flags */);
  55                 if (pmu_fd[i] < 0) {
  56                         fprintf(stderr, "ERROR: Initializing perf sampling\n");
  57                         return 1;
  58                 }
  59                 assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_SET_BPF,
  60                              prog_fd[0]) == 0);
  61                 assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE, 0) == 0);
  62         }
  63
  64         return 0;
  65 }
  66
  67 static void sampling_end(int *pmu_fd)
  68 {
  69         int i;
  70
  71         for (i = 0; i < nr_cpus; i++)
  72                 close(pmu_fd[i]);
  73 }
  74
  75 struct ipcount {
  76         __u64 ip;
  77         __u32 count;
  78 };
  79
  80 /* used for sorting */
  81 struct ipcount counts[MAX_IPS];
  82
  83 static int count_cmp(const void *p1, const void *p2)
  84 {
  85         return ((struct ipcount *)p1)->count - ((struct ipcount *)p2)->count;
  86 }
  87
  88 static void print_ip_map(int fd)
  89 {
  90         struct ksym *sym;
  91         __u64 key, next_key;
  92         __u32 value;
  93         int i, max;
  94
  95         printf("%-19s %-32s %s\n", "ADDR", "KSYM", "COUNT");
  96
  97         /* fetch IPs and counts */
  98         key = 0, i = 0;
  99         while (bpf_map_get_next_key(fd, &key, &next_key) == 0) {
 100                 bpf_map_lookup_elem(fd, &next_key, &value);
 101                 counts[i].ip = next_key;
 102                 counts[i++].count = value;
 103                 key = next_key;
 104         }
 105         max = i;
 106
 107         /* sort and print */
 108         qsort(counts, max, sizeof(struct ipcount), count_cmp);
 109         for (i = 0; i < max; i++) {
 110                 if (counts[i].ip > PAGE_OFFSET) {
 111                         sym = ksym_search(counts[i].ip);
 112                         printf("0x%-17llx %-32s %u\n", counts[i].ip, sym->name,
 113                                counts[i].count);
 114                 } else {
 115                         printf("0x%-17llx %-32s %u\n", counts[i].ip, "(user)",
 116                                counts[i].count);
 117                 }
 118         }
 119
 120         if (max == MAX_IPS) {
 121                 printf("WARNING: IP hash was full (max %d entries); ", max);
 122                 printf("may have dropped samples\n");
 123         }
 124 }
 125
 126 static void int_exit(int sig)
 127 {
 128         printf("\n");
 129         print_ip_map(map_fd[0]);
 130         exit(0);
 131 }
 132
 133 int main(int argc, char **argv)
 134 {
 135         char filename[256];
 136         int *pmu_fd, opt, freq = DEFAULT_FREQ, secs = DEFAULT_SECS;
 137
 138         /* process arguments */
 139         while ((opt = getopt(argc, argv, "F:h")) != -1) {
 140                 switch (opt) {
 141                 case 'F':
 142                         freq = atoi(optarg);
 143                         break;
 144                 case 'h':
 145                 default:
 146                         usage();
 147                         return 0;
 148                 }
 149         }
 150         if (argc - optind == 1)
 151                 secs = atoi(argv[optind]);
 152         if (freq == 0 || secs == 0) {
 153                 usage();
 154                 return 1;
 155         }
 156
 157         /* initialize kernel symbol translation */
 158         if (load_kallsyms()) {
 159                 fprintf(stderr, "ERROR: loading /proc/kallsyms\n");
 160                 return 2;
 161         }
 162
 163         /* create perf FDs for each CPU */
 164         nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
 165         pmu_fd = malloc(nr_cpus * sizeof(int));
 166         if (pmu_fd == NULL) {
 167                 fprintf(stderr, "ERROR: malloc of pmu_fd\n");
 168                 return 1;
 169         }
 170
 171         /* load BPF program */
 172         snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 173         if (load_bpf_file(filename)) {
 174                 fprintf(stderr, "ERROR: loading BPF program (errno %d):\n",
 175                         errno);
 176                 if (strcmp(bpf_log_buf, "") == 0)
 177                         fprintf(stderr, "Try: ulimit -l unlimited\n");
 178                 else
 179                         fprintf(stderr, "%s", bpf_log_buf);
 180                 return 1;
 181         }
 182         signal(SIGINT, int_exit);
 183         signal(SIGTERM, int_exit);
 184
 185         /* do sampling */
 186         printf("Sampling at %d Hertz for %d seconds. Ctrl-C also ends.\n",
 187                freq, secs);
 188         if (sampling_start(pmu_fd, freq) != 0)
 189                 return 1;
 190         sleep(secs);
 191         sampling_end(pmu_fd);
 192         free(pmu_fd);
 193
 194         /* output sample counts */
 195         print_ip_map(map_fd[0]);
 196
 197         return 0;
 198 }