12 #include <linux/bpf.h>
13 #include <linux/filter.h>
14 #include <linux/perf_event.h>
15 #include <linux/netlink.h>
16 #include <linux/rtnetlink.h>
17 #include <sys/types.h>
18 #include <sys/socket.h>
19 #include <sys/syscall.h>
20 #include <sys/ioctl.h>
25 #include "bpf_helpers.h"
28 #define DEBUGFS "/sys/kernel/debug/tracing/"
30 static char license[128];
31 static int kern_version;
32 static bool processed_sec[128];
34 int prog_fd[MAX_PROGS];
35 int event_fd[MAX_PROGS];
37 int prog_array_fd = -1;
39 static int populate_prog_array(const char *event, int prog_fd)
41 int ind = atoi(event), err;
43 err = bpf_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY);
45 printf("failed to store prog_fd in prog_array\n");
51 static int load_and_attach(const char *event, struct bpf_insn *prog, int size)
53 bool is_socket = strncmp(event, "socket", 6) == 0;
54 bool is_kprobe = strncmp(event, "kprobe/", 7) == 0;
55 bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0;
56 bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0;
57 bool is_xdp = strncmp(event, "xdp", 3) == 0;
58 bool is_perf_event = strncmp(event, "perf_event", 10) == 0;
59 bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0;
60 bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0;
61 enum bpf_prog_type prog_type;
64 struct perf_event_attr attr = {};
66 attr.type = PERF_TYPE_TRACEPOINT;
67 attr.sample_type = PERF_SAMPLE_RAW;
68 attr.sample_period = 1;
69 attr.wakeup_events = 1;
72 prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
73 } else if (is_kprobe || is_kretprobe) {
74 prog_type = BPF_PROG_TYPE_KPROBE;
75 } else if (is_tracepoint) {
76 prog_type = BPF_PROG_TYPE_TRACEPOINT;
78 prog_type = BPF_PROG_TYPE_XDP;
79 } else if (is_perf_event) {
80 prog_type = BPF_PROG_TYPE_PERF_EVENT;
81 } else if (is_cgroup_skb) {
82 prog_type = BPF_PROG_TYPE_CGROUP_SKB;
83 } else if (is_cgroup_sk) {
84 prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
86 printf("Unknown event '%s'\n", event);
90 fd = bpf_prog_load(prog_type, prog, size, license, kern_version);
92 printf("bpf_prog_load() err=%d\n%s", errno, bpf_log_buf);
96 prog_fd[prog_cnt++] = fd;
98 if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk)
106 if (!isdigit(*event)) {
107 printf("invalid prog number\n");
110 return populate_prog_array(event, fd);
113 if (is_kprobe || is_kretprobe) {
120 printf("event name cannot be empty\n");
125 return populate_prog_array(event, fd);
127 snprintf(buf, sizeof(buf),
128 "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events",
129 is_kprobe ? 'p' : 'r', event, event);
132 printf("failed to create kprobe '%s' error '%s'\n",
133 event, strerror(errno));
137 strcpy(buf, DEBUGFS);
138 strcat(buf, "events/kprobes/");
141 } else if (is_tracepoint) {
145 printf("event name cannot be empty\n");
148 strcpy(buf, DEBUGFS);
149 strcat(buf, "events/");
154 efd = open(buf, O_RDONLY, 0);
156 printf("failed to open event %s\n", event);
160 err = read(efd, buf, sizeof(buf));
161 if (err < 0 || err >= sizeof(buf)) {
162 printf("read from '%s' failed '%s'\n", event, strerror(errno));
172 efd = perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
174 printf("event %d fd %d err %s\n", id, efd, strerror(errno));
177 event_fd[prog_cnt - 1] = efd;
178 ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
179 ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
184 static int load_maps(struct bpf_map_def *maps, int len)
188 for (i = 0; i < len / sizeof(struct bpf_map_def); i++) {
190 map_fd[i] = bpf_create_map(maps[i].type,
196 printf("failed to create a map: %d %s\n",
197 errno, strerror(errno));
201 if (maps[i].type == BPF_MAP_TYPE_PROG_ARRAY)
202 prog_array_fd = map_fd[i];
207 static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname,
208 GElf_Shdr *shdr, Elf_Data **data)
212 scn = elf_getscn(elf, i);
216 if (gelf_getshdr(scn, shdr) != shdr)
219 *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name);
220 if (!*shname || !shdr->sh_size)
223 *data = elf_getdata(scn, 0);
224 if (!*data || elf_getdata(scn, *data) != NULL)
230 static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols,
231 GElf_Shdr *shdr, struct bpf_insn *insn)
235 nrels = shdr->sh_size / shdr->sh_entsize;
237 for (i = 0; i < nrels; i++) {
240 unsigned int insn_idx;
242 gelf_getrel(data, i, &rel);
244 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
246 gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym);
248 if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
249 printf("invalid relo for insn[%d].code 0x%x\n",
250 insn_idx, insn[insn_idx].code);
253 insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
254 insn[insn_idx].imm = map_fd[sym.st_value / sizeof(struct bpf_map_def)];
260 int load_bpf_file(char *path)
265 GElf_Shdr shdr, shdr_prog;
266 Elf_Data *data, *data_prog, *symbols = NULL;
267 char *shname, *shname_prog;
269 if (elf_version(EV_CURRENT) == EV_NONE)
272 fd = open(path, O_RDONLY, 0);
276 elf = elf_begin(fd, ELF_C_READ, NULL);
281 if (gelf_getehdr(elf, &ehdr) != &ehdr)
284 /* clear all kprobes */
285 i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events");
287 /* scan over all elf sections to get license and map info */
288 for (i = 1; i < ehdr.e_shnum; i++) {
290 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
293 if (0) /* helpful for llvm debugging */
294 printf("section %d:%s data %p size %zd link %d flags %d\n",
295 i, shname, data->d_buf, data->d_size,
296 shdr.sh_link, (int) shdr.sh_flags);
298 if (strcmp(shname, "license") == 0) {
299 processed_sec[i] = true;
300 memcpy(license, data->d_buf, data->d_size);
301 } else if (strcmp(shname, "version") == 0) {
302 processed_sec[i] = true;
303 if (data->d_size != sizeof(int)) {
304 printf("invalid size of version section %zd\n",
308 memcpy(&kern_version, data->d_buf, sizeof(int));
309 } else if (strcmp(shname, "maps") == 0) {
310 processed_sec[i] = true;
311 if (load_maps(data->d_buf, data->d_size))
313 } else if (shdr.sh_type == SHT_SYMTAB) {
318 /* load programs that need map fixup (relocations) */
319 for (i = 1; i < ehdr.e_shnum; i++) {
321 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
323 if (shdr.sh_type == SHT_REL) {
324 struct bpf_insn *insns;
326 if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog,
327 &shdr_prog, &data_prog))
330 if (shdr_prog.sh_type != SHT_PROGBITS ||
331 !(shdr_prog.sh_flags & SHF_EXECINSTR))
334 insns = (struct bpf_insn *) data_prog->d_buf;
336 processed_sec[shdr.sh_info] = true;
337 processed_sec[i] = true;
339 if (parse_relo_and_apply(data, symbols, &shdr, insns))
342 if (memcmp(shname_prog, "kprobe/", 7) == 0 ||
343 memcmp(shname_prog, "kretprobe/", 10) == 0 ||
344 memcmp(shname_prog, "tracepoint/", 11) == 0 ||
345 memcmp(shname_prog, "xdp", 3) == 0 ||
346 memcmp(shname_prog, "perf_event", 10) == 0 ||
347 memcmp(shname_prog, "socket", 6) == 0 ||
348 memcmp(shname_prog, "cgroup/", 7) == 0)
349 load_and_attach(shname_prog, insns, data_prog->d_size);
353 /* load programs that don't use maps */
354 for (i = 1; i < ehdr.e_shnum; i++) {
356 if (processed_sec[i])
359 if (get_sec(elf, i, &ehdr, &shname, &shdr, &data))
362 if (memcmp(shname, "kprobe/", 7) == 0 ||
363 memcmp(shname, "kretprobe/", 10) == 0 ||
364 memcmp(shname, "tracepoint/", 11) == 0 ||
365 memcmp(shname, "xdp", 3) == 0 ||
366 memcmp(shname, "perf_event", 10) == 0 ||
367 memcmp(shname, "socket", 6) == 0 ||
368 memcmp(shname, "cgroup/", 7) == 0)
369 load_and_attach(shname, data->d_buf, data->d_size);
376 void read_trace_pipe(void)
380 trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
385 static char buf[4096];
388 sz = read(trace_fd, buf, sizeof(buf));
396 #define MAX_SYMS 300000
397 static struct ksym syms[MAX_SYMS];
400 static int ksym_cmp(const void *p1, const void *p2)
402 return ((struct ksym *)p1)->addr - ((struct ksym *)p2)->addr;
405 int load_kallsyms(void)
407 FILE *f = fopen("/proc/kallsyms", "r");
408 char func[256], buf[256];
417 if (!fgets(buf, sizeof(buf), f))
419 if (sscanf(buf, "%p %c %s", &addr, &symbol, func) != 3)
423 syms[i].addr = (long) addr;
424 syms[i].name = strdup(func);
428 qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp);
432 struct ksym *ksym_search(long key)
434 int start = 0, end = sym_cnt;
437 while (start < end) {
438 size_t mid = start + (end - start) / 2;
440 result = key - syms[mid].addr;
449 if (start >= 1 && syms[start - 1].addr < key &&
450 key < syms[start].addr)
452 return &syms[start - 1];
454 /* out of range. return _stext */
458 int set_link_xdp_fd(int ifindex, int fd)
460 struct sockaddr_nl sa;
461 int sock, seq = 0, len, ret = -1;
463 struct nlattr *nla, *nla_xdp;
466 struct ifinfomsg ifinfo;
470 struct nlmsgerr *err;
472 memset(&sa, 0, sizeof(sa));
473 sa.nl_family = AF_NETLINK;
475 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
477 printf("open netlink socket: %s\n", strerror(errno));
481 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
482 printf("bind to netlink: %s\n", strerror(errno));
486 memset(&req, 0, sizeof(req));
487 req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg));
488 req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
489 req.nh.nlmsg_type = RTM_SETLINK;
490 req.nh.nlmsg_pid = 0;
491 req.nh.nlmsg_seq = ++seq;
492 req.ifinfo.ifi_family = AF_UNSPEC;
493 req.ifinfo.ifi_index = ifindex;
494 nla = (struct nlattr *)(((char *)&req)
495 + NLMSG_ALIGN(req.nh.nlmsg_len));
496 nla->nla_type = NLA_F_NESTED | 43/*IFLA_XDP*/;
498 nla_xdp = (struct nlattr *)((char *)nla + NLA_HDRLEN);
499 nla_xdp->nla_type = 1/*IFLA_XDP_FD*/;
500 nla_xdp->nla_len = NLA_HDRLEN + sizeof(int);
501 memcpy((char *)nla_xdp + NLA_HDRLEN, &fd, sizeof(fd));
502 nla->nla_len = NLA_HDRLEN + nla_xdp->nla_len;
504 req.nh.nlmsg_len += NLA_ALIGN(nla->nla_len);
506 if (send(sock, &req, req.nh.nlmsg_len, 0) < 0) {
507 printf("send to netlink: %s\n", strerror(errno));
511 len = recv(sock, buf, sizeof(buf), 0);
513 printf("recv from netlink: %s\n", strerror(errno));
517 for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, len);
518 nh = NLMSG_NEXT(nh, len)) {
519 if (nh->nlmsg_pid != getpid()) {
520 printf("Wrong pid %d, expected %d\n",
521 nh->nlmsg_pid, getpid());
524 if (nh->nlmsg_seq != seq) {
525 printf("Wrong seq %d, expected %d\n",
529 switch (nh->nlmsg_type) {
531 err = (struct nlmsgerr *)NLMSG_DATA(nh);
534 printf("nlmsg error %s\n", strerror(-err->error));