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>
21 #include "bpf_helpers.h"
24 #define DEBUGFS "/sys/kernel/debug/tracing/"
26 static char license[128];
27 static int kern_version;
28 static bool processed_sec[128];
30 int prog_fd[MAX_PROGS];
31 int event_fd[MAX_PROGS];
33 int prog_array_fd = -1;
35 static int populate_prog_array(const char *event, int prog_fd)
37 int ind = atoi(event), err;
39 err = bpf_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY);
41 printf("failed to store prog_fd in prog_array\n");
47 static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
49 bool is_socket = strncmp(event, "socket", 6) == 0;
50 bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
51 bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
52 bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
53 bool is_xdp = strncmp(event, "xdp", 3) == 0;
54 enum bpf_prog_type prog_type;
57 struct perf_event_attr attr = {};
59 attr.type = PERF_TYPE_TRACEPOINT;
60 attr.sample_type = PERF_SAMPLE_RAW;
61 attr.sample_period = 1;
62 attr.wakeup_events = 1;
65 prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
66 } else if (is_kprobe || is_kretprobe) {
67 prog_type = BPF_PROG_TYPE_KPROBE;
68 } else if (is_tracepoint) {
69 prog_type = BPF_PROG_TYPE_TRACEPOINT;
71 prog_type = BPF_PROG_TYPE_XDP;
73 printf("Unknown event '%s'\n", event);
77 fd = bpf_prog_load(prog_type, prog, size, license, kern_version);
79 printf("bpf_prog_load() err=%d\n%s", errno, bpf_log_buf);
83 prog_fd[prog_cnt++] = fd;
93 if (!isdigit(*event)) {
94 printf("invalid prog number\n");
97 return populate_prog_array(event, fd);
100 if (is_kprobe || is_kretprobe) {
107 printf("event name cannot be empty\n");
112 return populate_prog_array(event, fd);
114 snprintf(buf, sizeof(buf),
115 "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
116 is_kprobe ? 'p' : 'r', event, event);
119 printf("failed to create kprobe '%s' error '%s'\n",
120 event, strerror(errno));
124 strcpy(buf, DEBUGFS);
125 strcat(buf, "events/kprobes/");
128 } else if (is_tracepoint) {
132 printf("event name cannot be empty\n");
135 strcpy(buf, DEBUGFS);
136 strcat(buf, "events/");
141 efd = open(buf, O_RDONLY, 0);
143 printf("failed to open event %s\n", event);
147 err = read(efd, buf, sizeof(buf));
148 if (err < 0 || err >= sizeof(buf)) {
149 printf("read from '%s' failed '%s'\n", event, strerror(errno));
159 efd = perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
161 printf("event %d fd %d err %s\n", id, efd, strerror(errno));
164 event_fd[prog_cnt - 1] = efd;
165 ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
166 ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
171 static int load_maps(struct bpf_map_def *maps, int len)
175 for (i = 0; i < len / sizeof(struct bpf_map_def); i++) {
177 map_fd[i] = bpf_create_map(maps[i].type,
183 printf("failed to create a map: %d %s\n",
184 errno, strerror(errno));
188 if (maps[i].type == BPF_MAP_TYPE_PROG_ARRAY)
189 prog_array_fd = map_fd[i];
194 static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
195 GElf_Shdr *shdr, Elf_Data **data)
199 scn = elf_getscn(elf, i);
203 if (gelf_getshdr(scn, shdr) != shdr)
206 *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
207 if (!*shname || !shdr->sh_size)
210 *data = elf_getdata(scn, 0);
211 if (!*data || elf_getdata(scn, *data) != NULL)
217 static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
218 GElf_Shdr *shdr, struct bpf_insn *insn)
222 nrels = shdr->sh_size / shdr->sh_entsize;
224 for (i = 0; i < nrels; i++) {
227 unsigned int insn_idx;
229 gelf_getrel(data, i, &rel);
231 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
233 gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
235 if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
236 printf("invalid relo for insn[%d].code 0x%x\n",
237 insn_idx, insn[insn_idx].code);
240 insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
241 insn[insn_idx].imm = map_fd[sym.st_value / sizeof(struct bpf_map_def)];
247 int load_bpf_file(char *path)
252 GElf_Shdr shdr, shdr_prog;
253 Elf_Data *data, *data_prog, *symbols = NULL;
254 char *shname, *shname_prog;
256 if (elf_version(EV_CURRENT) == EV_NONE)
259 fd = open(path, O_RDONLY, 0);
263 elf = elf_begin(fd, ELF_C_READ, NULL);
268 if (gelf_getehdr(elf, &ehdr) != &ehdr)
271 /* clear all kprobes */
272 i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events");
274 /* scan over all elf sections to get license and map info */
275 for (i = 1; i < ehdr.e_shnum; i++) {
277 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
280 if (0) /* helpful for llvm debugging */
281 printf("section %d:%s data %p size %zd link %d flags %d\n",
282 i, shname, data->d_buf, data->d_size,
283 shdr.sh_link, (int) shdr.sh_flags);
285 if (strcmp(shname, "license") == 0) {
286 processed_sec[i] = true;
287 memcpy(license, data->d_buf, data->d_size);
288 } else if (strcmp(shname, "version") == 0) {
289 processed_sec[i] = true;
290 if (data->d_size != sizeof(int)) {
291 printf("invalid size of version section %zd\n",
295 memcpy(&kern_version, data->d_buf, sizeof(int));
296 } else if (strcmp(shname, "maps") == 0) {
297 processed_sec[i] = true;
298 if (load_maps(data->d_buf, data->d_size))
300 } else if (shdr.sh_type == SHT_SYMTAB) {
305 /* load programs that need map fixup (relocations) */
306 for (i = 1; i < ehdr.e_shnum; i++) {
308 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
310 if (shdr.sh_type == SHT_REL) {
311 struct bpf_insn *insns;
313 if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
314 &shdr_prog, &data_prog))
317 insns = (struct bpf_insn *) data_prog->d_buf;
319 processed_sec[shdr.sh_info] = true;
320 processed_sec[i] = true;
322 if (parse_relo_and_apply(data, symbols, &shdr, insns))
325 if (memcmp(shname_prog, "kprobe/", 7) == 0 ||
326 memcmp(shname_prog, "kretprobe/", 10) == 0 ||
327 memcmp(shname_prog, "tracepoint/", 11) == 0 ||
328 memcmp(shname_prog, "xdp", 3) == 0 ||
329 memcmp(shname_prog, "socket", 6) == 0)
330 load_and_attach(shname_prog, insns, data_prog->d_size);
334 /* load programs that don't use maps */
335 for (i = 1; i < ehdr.e_shnum; i++) {
337 if (processed_sec[i])
340 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
343 if (memcmp(shname, "kprobe/", 7) == 0 ||
344 memcmp(shname, "kretprobe/", 10) == 0 ||
345 memcmp(shname, "tracepoint/", 11) == 0 ||
346 memcmp(shname, "xdp", 3) == 0 ||
347 memcmp(shname, "socket", 6) == 0)
348 load_and_attach(shname, data->d_buf, data->d_size);
355 void read_trace_pipe(void)
359 trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
364 static char buf[4096];
367 sz = read(trace_fd, buf, sizeof(buf));
375 #define MAX_SYMS 300000
376 static struct ksym syms[MAX_SYMS];
379 static int ksym_cmp(const void *p1, const void *p2)
381 return ((struct ksym *)p1)->addr - ((struct ksym *)p2)->addr;
384 int load_kallsyms(void)
386 FILE *f = fopen("/proc/kallsyms", "r");
387 char func[256], buf[256];
396 if (!fgets(buf, sizeof(buf), f))
398 if (sscanf(buf, "%p %c %s", &addr, &symbol, func) != 3)
402 syms[i].addr = (long) addr;
403 syms[i].name = strdup(func);
407 qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp);
411 struct ksym *ksym_search(long key)
413 int start = 0, end = sym_cnt;
416 while (start < end) {
417 size_t mid = start + (end - start) / 2;
419 result = key - syms[mid].addr;
428 if (start >= 1 && syms[start - 1].addr < key &&
429 key < syms[start].addr)
431 return &syms[start - 1];
433 /* out of range. return _stext */