1 // SPDX-License-Identifier: GPL-2.0
13 #include <linux/bpf.h>
14 #include <linux/filter.h>
15 #include <linux/perf_event.h>
16 #include <linux/netlink.h>
17 #include <linux/rtnetlink.h>
18 #include <linux/types.h>
19 #include <sys/types.h>
20 #include <sys/socket.h>
21 #include <sys/syscall.h>
22 #include <sys/ioctl.h>
31 #define DEBUGFS "/sys/kernel/debug/tracing/"
33 static char license[128];
34 static int kern_version;
35 static bool processed_sec[128];
36 char bpf_log_buf[BPF_LOG_BUF_SIZE];
38 int prog_fd[MAX_PROGS];
39 int event_fd[MAX_PROGS];
41 int prog_array_fd = -1;
43 struct bpf_map_data map_data[MAX_MAPS];
44 int map_data_count = 0;
46 static int populate_prog_array(const char *event, int prog_fd)
48 int ind = atoi(event), err;
50 err = bpf_map_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY);
52 printf("failed to store prog_fd in prog_array\n");
58 static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
60 bool is_socket = strncmp(event, "socket", 6) == 0;
61 bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
62 bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
63 bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
64 bool is_raw_tracepoint = strncmp(event, "raw_tracepoint/", 15) == 0;
65 bool is_xdp = strncmp(event, "xdp", 3) == 0;
66 bool is_perf_event = strncmp(event, "perf_event", 10) == 0;
67 bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0;
68 bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0;
69 bool is_sockops = strncmp(event, "sockops", 7) == 0;
70 bool is_sk_skb = strncmp(event, "sk_skb", 6) == 0;
71 bool is_sk_msg = strncmp(event, "sk_msg", 6) == 0;
72 size_t insns_cnt = size / sizeof(struct bpf_insn);
73 enum bpf_prog_type prog_type;
76 struct perf_event_attr attr = {};
78 attr.type = PERF_TYPE_TRACEPOINT;
79 attr.sample_type = PERF_SAMPLE_RAW;
80 attr.sample_period = 1;
81 attr.wakeup_events = 1;
84 prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
85 } else if (is_kprobe || is_kretprobe) {
86 prog_type = BPF_PROG_TYPE_KPROBE;
87 } else if (is_tracepoint) {
88 prog_type = BPF_PROG_TYPE_TRACEPOINT;
89 } else if (is_raw_tracepoint) {
90 prog_type = BPF_PROG_TYPE_RAW_TRACEPOINT;
92 prog_type = BPF_PROG_TYPE_XDP;
93 } else if (is_perf_event) {
94 prog_type = BPF_PROG_TYPE_PERF_EVENT;
95 } else if (is_cgroup_skb) {
96 prog_type = BPF_PROG_TYPE_CGROUP_SKB;
97 } else if (is_cgroup_sk) {
98 prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
99 } else if (is_sockops) {
100 prog_type = BPF_PROG_TYPE_SOCK_OPS;
101 } else if (is_sk_skb) {
102 prog_type = BPF_PROG_TYPE_SK_SKB;
103 } else if (is_sk_msg) {
104 prog_type = BPF_PROG_TYPE_SK_MSG;
106 printf("Unknown event '%s'\n", event);
110 if (prog_cnt == MAX_PROGS)
113 fd = bpf_load_program(prog_type, prog, insns_cnt, license, kern_version,
114 bpf_log_buf, BPF_LOG_BUF_SIZE);
116 printf("bpf_load_program() err=%d\n%s", errno, bpf_log_buf);
120 prog_fd[prog_cnt++] = fd;
122 if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk)
125 if (is_socket || is_sockops || is_sk_skb || is_sk_msg) {
133 if (!isdigit(*event)) {
134 printf("invalid prog number\n");
137 return populate_prog_array(event, fd);
140 if (is_raw_tracepoint) {
141 efd = bpf_raw_tracepoint_open(event + 15, fd);
143 printf("tracepoint %s %s\n", event + 15, strerror(errno));
146 event_fd[prog_cnt - 1] = efd;
150 if (is_kprobe || is_kretprobe) {
151 bool need_normal_check = true;
152 const char *event_prefix = "";
160 printf("event name cannot be empty\n");
165 return populate_prog_array(event, fd);
168 if (strncmp(event, "sys_", 4) == 0) {
169 snprintf(buf, sizeof(buf),
170 "echo '%c:__x64_%s __x64_%s' >> /sys/kernel/debug/tracing/kprobe_events",
171 is_kprobe ? 'p' : 'r', event, event);
174 need_normal_check = false;
175 event_prefix = "__x64_";
179 if (need_normal_check) {
180 snprintf(buf, sizeof(buf),
181 "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
182 is_kprobe ? 'p' : 'r', event, event);
185 printf("failed to create kprobe '%s' error '%s'\n",
186 event, strerror(errno));
191 strcpy(buf, DEBUGFS);
192 strcat(buf, "events/kprobes/");
193 strcat(buf, event_prefix);
196 } else if (is_tracepoint) {
200 printf("event name cannot be empty\n");
203 strcpy(buf, DEBUGFS);
204 strcat(buf, "events/");
209 efd = open(buf, O_RDONLY, 0);
211 printf("failed to open event %s\n", event);
215 err = read(efd, buf, sizeof(buf));
216 if (err < 0 || err >= sizeof(buf)) {
217 printf("read from '%s' failed '%s'\n", event, strerror(errno));
227 efd = sys_perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
229 printf("event %d fd %d err %s\n", id, efd, strerror(errno));
232 event_fd[prog_cnt - 1] = efd;
233 err = ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
235 printf("ioctl PERF_EVENT_IOC_ENABLE failed err %s\n",
239 err = ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
241 printf("ioctl PERF_EVENT_IOC_SET_BPF failed err %s\n",
249 static int load_maps(struct bpf_map_data *maps, int nr_maps,
250 fixup_map_cb fixup_map)
254 for (i = 0; i < nr_maps; i++) {
256 fixup_map(&maps[i], i);
257 /* Allow userspace to assign map FD prior to creation */
258 if (maps[i].fd != -1) {
259 map_fd[i] = maps[i].fd;
264 numa_node = maps[i].def.map_flags & BPF_F_NUMA_NODE ?
265 maps[i].def.numa_node : -1;
267 if (maps[i].def.type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
268 maps[i].def.type == BPF_MAP_TYPE_HASH_OF_MAPS) {
269 int inner_map_fd = map_fd[maps[i].def.inner_map_idx];
271 map_fd[i] = bpf_create_map_in_map_node(maps[i].def.type,
273 maps[i].def.key_size,
275 maps[i].def.max_entries,
276 maps[i].def.map_flags,
279 map_fd[i] = bpf_create_map_node(maps[i].def.type,
281 maps[i].def.key_size,
282 maps[i].def.value_size,
283 maps[i].def.max_entries,
284 maps[i].def.map_flags,
288 printf("failed to create a map: %d %s\n",
289 errno, strerror(errno));
292 maps[i].fd = map_fd[i];
294 if (maps[i].def.type == BPF_MAP_TYPE_PROG_ARRAY)
295 prog_array_fd = map_fd[i];
300 static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
301 GElf_Shdr *shdr, Elf_Data **data)
305 scn = elf_getscn(elf, i);
309 if (gelf_getshdr(scn, shdr) != shdr)
312 *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
313 if (!*shname || !shdr->sh_size)
316 *data = elf_getdata(scn, 0);
317 if (!*data || elf_getdata(scn, *data) != NULL)
323 static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
324 GElf_Shdr *shdr, struct bpf_insn *insn,
325 struct bpf_map_data *maps, int nr_maps)
329 nrels = shdr->sh_size / shdr->sh_entsize;
331 for (i = 0; i < nrels; i++) {
334 unsigned int insn_idx;
338 gelf_getrel(data, i, &rel);
340 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
342 gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
344 if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
345 printf("invalid relo for insn[%d].code 0x%x\n",
346 insn_idx, insn[insn_idx].code);
349 insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
351 /* Match FD relocation against recorded map_data[] offset */
352 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
353 if (maps[map_idx].elf_offset == sym.st_value) {
359 insn[insn_idx].imm = maps[map_idx].fd;
361 printf("invalid relo for insn[%d] no map_data match\n",
370 static int cmp_symbols(const void *l, const void *r)
372 const GElf_Sym *lsym = (const GElf_Sym *)l;
373 const GElf_Sym *rsym = (const GElf_Sym *)r;
375 if (lsym->st_value < rsym->st_value)
377 else if (lsym->st_value > rsym->st_value)
383 static int load_elf_maps_section(struct bpf_map_data *maps, int maps_shndx,
384 Elf *elf, Elf_Data *symbols, int strtabidx)
386 int map_sz_elf, map_sz_copy;
387 bool validate_zero = false;
399 /* Get data for maps section via elf index */
400 scn = elf_getscn(elf, maps_shndx);
402 data_maps = elf_getdata(scn, NULL);
403 if (!scn || !data_maps) {
404 printf("Failed to get Elf_Data from maps section %d\n",
409 /* For each map get corrosponding symbol table entry */
410 sym = calloc(MAX_MAPS+1, sizeof(GElf_Sym));
411 for (i = 0, nr_maps = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
412 assert(nr_maps < MAX_MAPS+1);
413 if (!gelf_getsym(symbols, i, &sym[nr_maps]))
415 if (sym[nr_maps].st_shndx != maps_shndx)
417 /* Only increment iif maps section */
421 /* Align to map_fd[] order, via sort on offset in sym.st_value */
422 qsort(sym, nr_maps, sizeof(GElf_Sym), cmp_symbols);
424 /* Keeping compatible with ELF maps section changes
425 * ------------------------------------------------
426 * The program size of struct bpf_load_map_def is known by loader
427 * code, but struct stored in ELF file can be different.
429 * Unfortunately sym[i].st_size is zero. To calculate the
430 * struct size stored in the ELF file, assume all struct have
431 * the same size, and simply divide with number of map
434 map_sz_elf = data_maps->d_size / nr_maps;
435 map_sz_copy = sizeof(struct bpf_load_map_def);
436 if (map_sz_elf < map_sz_copy) {
438 * Backward compat, loading older ELF file with
439 * smaller struct, keeping remaining bytes zero.
441 map_sz_copy = map_sz_elf;
442 } else if (map_sz_elf > map_sz_copy) {
444 * Forward compat, loading newer ELF file with larger
445 * struct with unknown features. Assume zero means
446 * feature not used. Thus, validate rest of struct
449 validate_zero = true;
452 /* Memcpy relevant part of ELF maps data to loader maps */
453 for (i = 0; i < nr_maps; i++) {
454 struct bpf_load_map_def *def;
455 unsigned char *addr, *end;
456 const char *map_name;
459 map_name = elf_strptr(elf, strtabidx, sym[i].st_name);
460 maps[i].name = strdup(map_name);
462 printf("strdup(%s): %s(%d)\n", map_name,
463 strerror(errno), errno);
468 /* Symbol value is offset into ELF maps section data area */
469 offset = sym[i].st_value;
470 def = (struct bpf_load_map_def *)(data_maps->d_buf + offset);
471 maps[i].elf_offset = offset;
472 memset(&maps[i].def, 0, sizeof(struct bpf_load_map_def));
473 memcpy(&maps[i].def, def, map_sz_copy);
475 /* Verify no newer features were requested */
477 addr = (unsigned char*) def + map_sz_copy;
478 end = (unsigned char*) def + map_sz_elf;
479 for (; addr < end; addr++) {
492 static int do_load_bpf_file(const char *path, fixup_map_cb fixup_map)
494 int fd, i, ret, maps_shndx = -1, strtabidx = -1;
497 GElf_Shdr shdr, shdr_prog;
498 Elf_Data *data, *data_prog, *data_maps = NULL, *symbols = NULL;
499 char *shname, *shname_prog;
502 /* reset global variables */
504 memset(license, 0, sizeof(license));
505 memset(processed_sec, 0, sizeof(processed_sec));
507 if (elf_version(EV_CURRENT) == EV_NONE)
510 fd = open(path, O_RDONLY, 0);
514 elf = elf_begin(fd, ELF_C_READ, NULL);
519 if (gelf_getehdr(elf, &ehdr) != &ehdr)
522 /* clear all kprobes */
523 i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events");
525 /* scan over all elf sections to get license and map info */
526 for (i = 1; i < ehdr.e_shnum; i++) {
528 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
531 if (0) /* helpful for llvm debugging */
532 printf("section %d:%s data %p size %zd link %d flags %d\n",
533 i, shname, data->d_buf, data->d_size,
534 shdr.sh_link, (int) shdr.sh_flags);
536 if (strcmp(shname, "license") == 0) {
537 processed_sec[i] = true;
538 memcpy(license, data->d_buf, data->d_size);
539 } else if (strcmp(shname, "version") == 0) {
540 processed_sec[i] = true;
541 if (data->d_size != sizeof(int)) {
542 printf("invalid size of version section %zd\n",
546 memcpy(&kern_version, data->d_buf, sizeof(int));
547 } else if (strcmp(shname, "maps") == 0) {
552 for (j = 0; j < MAX_MAPS; j++)
554 } else if (shdr.sh_type == SHT_SYMTAB) {
555 strtabidx = shdr.sh_link;
563 printf("missing SHT_SYMTAB section\n");
568 nr_maps = load_elf_maps_section(map_data, maps_shndx,
569 elf, symbols, strtabidx);
571 printf("Error: Failed loading ELF maps (errno:%d):%s\n",
572 nr_maps, strerror(-nr_maps));
575 if (load_maps(map_data, nr_maps, fixup_map))
577 map_data_count = nr_maps;
579 processed_sec[maps_shndx] = true;
582 /* process all relo sections, and rewrite bpf insns for maps */
583 for (i = 1; i < ehdr.e_shnum; i++) {
584 if (processed_sec[i])
587 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
590 if (shdr.sh_type == SHT_REL) {
591 struct bpf_insn *insns;
593 /* locate prog sec that need map fixup (relocations) */
594 if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
595 &shdr_prog, &data_prog))
598 if (shdr_prog.sh_type != SHT_PROGBITS ||
599 !(shdr_prog.sh_flags & SHF_EXECINSTR))
602 insns = (struct bpf_insn *) data_prog->d_buf;
603 processed_sec[i] = true; /* relo section */
605 if (parse_relo_and_apply(data, symbols, &shdr, insns,
612 for (i = 1; i < ehdr.e_shnum; i++) {
614 if (processed_sec[i])
617 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
620 if (memcmp(shname, "kprobe/", 7) == 0 ||
621 memcmp(shname, "kretprobe/", 10) == 0 ||
622 memcmp(shname, "tracepoint/", 11) == 0 ||
623 memcmp(shname, "raw_tracepoint/", 15) == 0 ||
624 memcmp(shname, "xdp", 3) == 0 ||
625 memcmp(shname, "perf_event", 10) == 0 ||
626 memcmp(shname, "socket", 6) == 0 ||
627 memcmp(shname, "cgroup/", 7) == 0 ||
628 memcmp(shname, "sockops", 7) == 0 ||
629 memcmp(shname, "sk_skb", 6) == 0 ||
630 memcmp(shname, "sk_msg", 6) == 0) {
631 ret = load_and_attach(shname, data->d_buf,
643 int load_bpf_file(char *path)
645 return do_load_bpf_file(path, NULL);
648 int load_bpf_file_fixup_map(const char *path, fixup_map_cb fixup_map)
650 return do_load_bpf_file(path, fixup_map);
653 void read_trace_pipe(void)
657 trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
662 static char buf[4096];
665 sz = read(trace_fd, buf, sizeof(buf));