bpf_load.c 6.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312
  1. #include <stdio.h>
  2. #include <sys/types.h>
  3. #include <sys/stat.h>
  4. #include <fcntl.h>
  5. #include <libelf.h>
  6. #include <gelf.h>
  7. #include <errno.h>
  8. #include <unistd.h>
  9. #include <string.h>
  10. #include <stdbool.h>
  11. #include <stdlib.h>
  12. #include <linux/bpf.h>
  13. #include <linux/filter.h>
  14. #include <linux/perf_event.h>
  15. #include <sys/syscall.h>
  16. #include <sys/ioctl.h>
  17. #include <sys/mman.h>
  18. #include <poll.h>
  19. #include "libbpf.h"
  20. #include "bpf_helpers.h"
  21. #include "bpf_load.h"
  22. #define DEBUGFS "/sys/kernel/debug/tracing/"
  23. static char license[128];
  24. static int kern_version;
  25. static bool processed_sec[128];
  26. int map_fd[MAX_MAPS];
  27. int prog_fd[MAX_PROGS];
  28. int event_fd[MAX_PROGS];
  29. int prog_cnt;
  30. static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
  31. {
  32. bool is_socket = strncmp(event, "socket", 6) == 0;
  33. bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
  34. bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
  35. enum bpf_prog_type prog_type;
  36. char buf[256];
  37. int fd, efd, err, id;
  38. struct perf_event_attr attr = {};
  39. attr.type = PERF_TYPE_TRACEPOINT;
  40. attr.sample_type = PERF_SAMPLE_RAW;
  41. attr.sample_period = 1;
  42. attr.wakeup_events = 1;
  43. if (is_socket) {
  44. prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
  45. } else if (is_kprobe || is_kretprobe) {
  46. prog_type = BPF_PROG_TYPE_KPROBE;
  47. } else {
  48. printf("Unknown event '%s'\n", event);
  49. return -1;
  50. }
  51. if (is_kprobe || is_kretprobe) {
  52. if (is_kprobe)
  53. event += 7;
  54. else
  55. event += 10;
  56. snprintf(buf, sizeof(buf),
  57. "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
  58. is_kprobe ? 'p' : 'r', event, event);
  59. err = system(buf);
  60. if (err < 0) {
  61. printf("failed to create kprobe '%s' error '%s'\n",
  62. event, strerror(errno));
  63. return -1;
  64. }
  65. }
  66. fd = bpf_prog_load(prog_type, prog, size, license, kern_version);
  67. if (fd < 0) {
  68. printf("bpf_prog_load() err=%d\n%s", errno, bpf_log_buf);
  69. return -1;
  70. }
  71. prog_fd[prog_cnt++] = fd;
  72. if (is_socket)
  73. return 0;
  74. strcpy(buf, DEBUGFS);
  75. strcat(buf, "events/kprobes/");
  76. strcat(buf, event);
  77. strcat(buf, "/id");
  78. efd = open(buf, O_RDONLY, 0);
  79. if (efd < 0) {
  80. printf("failed to open event %s\n", event);
  81. return -1;
  82. }
  83. err = read(efd, buf, sizeof(buf));
  84. if (err < 0 || err >= sizeof(buf)) {
  85. printf("read from '%s' failed '%s'\n", event, strerror(errno));
  86. return -1;
  87. }
  88. close(efd);
  89. buf[err] = 0;
  90. id = atoi(buf);
  91. attr.config = id;
  92. efd = perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
  93. if (efd < 0) {
  94. printf("event %d fd %d err %s\n", id, efd, strerror(errno));
  95. return -1;
  96. }
  97. event_fd[prog_cnt - 1] = efd;
  98. ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
  99. ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
  100. return 0;
  101. }
  102. static int load_maps(struct bpf_map_def *maps, int len)
  103. {
  104. int i;
  105. for (i = 0; i < len / sizeof(struct bpf_map_def); i++) {
  106. map_fd[i] = bpf_create_map(maps[i].type,
  107. maps[i].key_size,
  108. maps[i].value_size,
  109. maps[i].max_entries);
  110. if (map_fd[i] < 0)
  111. return 1;
  112. }
  113. return 0;
  114. }
  115. static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
  116. GElf_Shdr *shdr, Elf_Data **data)
  117. {
  118. Elf_Scn *scn;
  119. scn = elf_getscn(elf, i);
  120. if (!scn)
  121. return 1;
  122. if (gelf_getshdr(scn, shdr) != shdr)
  123. return 2;
  124. *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
  125. if (!*shname || !shdr->sh_size)
  126. return 3;
  127. *data = elf_getdata(scn, 0);
  128. if (!*data || elf_getdata(scn, *data) != NULL)
  129. return 4;
  130. return 0;
  131. }
  132. static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
  133. GElf_Shdr *shdr, struct bpf_insn *insn)
  134. {
  135. int i, nrels;
  136. nrels = shdr->sh_size / shdr->sh_entsize;
  137. for (i = 0; i < nrels; i++) {
  138. GElf_Sym sym;
  139. GElf_Rel rel;
  140. unsigned int insn_idx;
  141. gelf_getrel(data, i, &rel);
  142. insn_idx = rel.r_offset / sizeof(struct bpf_insn);
  143. gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
  144. if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
  145. printf("invalid relo for insn[%d].code 0x%x\n",
  146. insn_idx, insn[insn_idx].code);
  147. return 1;
  148. }
  149. insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
  150. insn[insn_idx].imm = map_fd[sym.st_value / sizeof(struct bpf_map_def)];
  151. }
  152. return 0;
  153. }
  154. int load_bpf_file(char *path)
  155. {
  156. int fd, i;
  157. Elf *elf;
  158. GElf_Ehdr ehdr;
  159. GElf_Shdr shdr, shdr_prog;
  160. Elf_Data *data, *data_prog, *symbols = NULL;
  161. char *shname, *shname_prog;
  162. if (elf_version(EV_CURRENT) == EV_NONE)
  163. return 1;
  164. fd = open(path, O_RDONLY, 0);
  165. if (fd < 0)
  166. return 1;
  167. elf = elf_begin(fd, ELF_C_READ, NULL);
  168. if (!elf)
  169. return 1;
  170. if (gelf_getehdr(elf, &ehdr) != &ehdr)
  171. return 1;
  172. /* clear all kprobes */
  173. i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events");
  174. /* scan over all elf sections to get license and map info */
  175. for (i = 1; i < ehdr.e_shnum; i++) {
  176. if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
  177. continue;
  178. if (0) /* helpful for llvm debugging */
  179. printf("section %d:%s data %p size %zd link %d flags %d\n",
  180. i, shname, data->d_buf, data->d_size,
  181. shdr.sh_link, (int) shdr.sh_flags);
  182. if (strcmp(shname, "license") == 0) {
  183. processed_sec[i] = true;
  184. memcpy(license, data->d_buf, data->d_size);
  185. } else if (strcmp(shname, "version") == 0) {
  186. processed_sec[i] = true;
  187. if (data->d_size != sizeof(int)) {
  188. printf("invalid size of version section %zd\n",
  189. data->d_size);
  190. return 1;
  191. }
  192. memcpy(&kern_version, data->d_buf, sizeof(int));
  193. } else if (strcmp(shname, "maps") == 0) {
  194. processed_sec[i] = true;
  195. if (load_maps(data->d_buf, data->d_size))
  196. return 1;
  197. } else if (shdr.sh_type == SHT_SYMTAB) {
  198. symbols = data;
  199. }
  200. }
  201. /* load programs that need map fixup (relocations) */
  202. for (i = 1; i < ehdr.e_shnum; i++) {
  203. if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
  204. continue;
  205. if (shdr.sh_type == SHT_REL) {
  206. struct bpf_insn *insns;
  207. if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
  208. &shdr_prog, &data_prog))
  209. continue;
  210. insns = (struct bpf_insn *) data_prog->d_buf;
  211. processed_sec[shdr.sh_info] = true;
  212. processed_sec[i] = true;
  213. if (parse_relo_and_apply(data, symbols, &shdr, insns))
  214. continue;
  215. if (memcmp(shname_prog, "kprobe/", 7) == 0 ||
  216. memcmp(shname_prog, "kretprobe/", 10) == 0 ||
  217. memcmp(shname_prog, "socket", 6) == 0)
  218. load_and_attach(shname_prog, insns, data_prog->d_size);
  219. }
  220. }
  221. /* load programs that don't use maps */
  222. for (i = 1; i < ehdr.e_shnum; i++) {
  223. if (processed_sec[i])
  224. continue;
  225. if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
  226. continue;
  227. if (memcmp(shname, "kprobe/", 7) == 0 ||
  228. memcmp(shname, "kretprobe/", 10) == 0 ||
  229. memcmp(shname, "socket", 6) == 0)
  230. load_and_attach(shname, data->d_buf, data->d_size);
  231. }
  232. close(fd);
  233. return 0;
  234. }
  235. void read_trace_pipe(void)
  236. {
  237. int trace_fd;
  238. trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
  239. if (trace_fd < 0)
  240. return;
  241. while (1) {
  242. static char buf[4096];
  243. ssize_t sz;
  244. sz = read(trace_fd, buf, sizeof(buf));
  245. if (sz > 0) {
  246. buf[sz] = 0;
  247. puts(buf);
  248. }
  249. }
  250. }