]>
Commit | Line | Data |
---|---|---|
249b812d AS |
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> | |
b896c4f9 | 11 | #include <stdlib.h> |
249b812d AS |
12 | #include <linux/bpf.h> |
13 | #include <linux/filter.h> | |
b896c4f9 | 14 | #include <linux/perf_event.h> |
12d8bb64 MKL |
15 | #include <linux/netlink.h> |
16 | #include <linux/rtnetlink.h> | |
6387d011 | 17 | #include <linux/types.h> |
12d8bb64 MKL |
18 | #include <sys/types.h> |
19 | #include <sys/socket.h> | |
b896c4f9 AS |
20 | #include <sys/syscall.h> |
21 | #include <sys/ioctl.h> | |
22 | #include <sys/mman.h> | |
23 | #include <poll.h> | |
5bacd780 | 24 | #include <ctype.h> |
9fd63d05 | 25 | #include <assert.h> |
249b812d | 26 | #include "libbpf.h" |
249b812d | 27 | #include "bpf_load.h" |
205c8ada | 28 | #include "perf-sys.h" |
249b812d | 29 | |
b896c4f9 AS |
30 | #define DEBUGFS "/sys/kernel/debug/tracing/" |
31 | ||
249b812d | 32 | static char license[128]; |
b896c4f9 | 33 | static int kern_version; |
249b812d | 34 | static bool processed_sec[128]; |
d40fc181 | 35 | char bpf_log_buf[BPF_LOG_BUF_SIZE]; |
249b812d AS |
36 | int map_fd[MAX_MAPS]; |
37 | int prog_fd[MAX_PROGS]; | |
b896c4f9 | 38 | int event_fd[MAX_PROGS]; |
249b812d | 39 | int prog_cnt; |
5bacd780 AS |
40 | int prog_array_fd = -1; |
41 | ||
156450d9 JDB |
42 | /* Keeping relevant info on maps */ |
43 | struct bpf_map_data { | |
44 | int fd; | |
45 | char *name; | |
46 | size_t elf_offset; | |
47 | struct bpf_map_def def; | |
48 | }; | |
49 | struct bpf_map_data map_data[MAX_MAPS]; | |
50 | int map_data_count = 0; | |
51 | ||
5bacd780 AS |
52 | static int populate_prog_array(const char *event, int prog_fd) |
53 | { | |
54 | int ind = atoi(event), err; | |
55 | ||
d40fc181 | 56 | err = bpf_map_update_elem(prog_array_fd, &ind, &prog_fd, BPF_ANY); |
5bacd780 AS |
57 | if (err < 0) { |
58 | printf("failed to store prog_fd in prog_array\n"); | |
59 | return -1; | |
60 | } | |
61 | return 0; | |
62 | } | |
249b812d AS |
63 | |
64 | static int load_and_attach(const char *event, struct bpf_insn *prog, int size) | |
65 | { | |
249b812d | 66 | bool is_socket = strncmp(event, "socket", 6) == 0; |
b896c4f9 AS |
67 | bool is_kprobe = strncmp(event, "kprobe/", 7) == 0; |
68 | bool is_kretprobe = strncmp(event, "kretprobe/", 10) == 0; | |
c0766040 | 69 | bool is_tracepoint = strncmp(event, "tracepoint/", 11) == 0; |
86af8b41 | 70 | bool is_xdp = strncmp(event, "xdp", 3) == 0; |
1c47910e | 71 | bool is_perf_event = strncmp(event, "perf_event", 10) == 0; |
4f2e7ae5 DA |
72 | bool is_cgroup_skb = strncmp(event, "cgroup/skb", 10) == 0; |
73 | bool is_cgroup_sk = strncmp(event, "cgroup/sock", 11) == 0; | |
43371c83 | 74 | size_t insns_cnt = size / sizeof(struct bpf_insn); |
b896c4f9 AS |
75 | enum bpf_prog_type prog_type; |
76 | char buf[256]; | |
77 | int fd, efd, err, id; | |
78 | struct perf_event_attr attr = {}; | |
79 | ||
80 | attr.type = PERF_TYPE_TRACEPOINT; | |
81 | attr.sample_type = PERF_SAMPLE_RAW; | |
82 | attr.sample_period = 1; | |
83 | attr.wakeup_events = 1; | |
84 | ||
85 | if (is_socket) { | |
86 | prog_type = BPF_PROG_TYPE_SOCKET_FILTER; | |
87 | } else if (is_kprobe || is_kretprobe) { | |
88 | prog_type = BPF_PROG_TYPE_KPROBE; | |
c0766040 AS |
89 | } else if (is_tracepoint) { |
90 | prog_type = BPF_PROG_TYPE_TRACEPOINT; | |
86af8b41 BB |
91 | } else if (is_xdp) { |
92 | prog_type = BPF_PROG_TYPE_XDP; | |
1c47910e AS |
93 | } else if (is_perf_event) { |
94 | prog_type = BPF_PROG_TYPE_PERF_EVENT; | |
4f2e7ae5 DA |
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; | |
b896c4f9 AS |
99 | } else { |
100 | printf("Unknown event '%s'\n", event); | |
249b812d | 101 | return -1; |
b896c4f9 AS |
102 | } |
103 | ||
43371c83 | 104 | fd = bpf_load_program(prog_type, prog, insns_cnt, license, kern_version, |
d40fc181 | 105 | bpf_log_buf, BPF_LOG_BUF_SIZE); |
5bacd780 | 106 | if (fd < 0) { |
d40fc181 | 107 | printf("bpf_load_program() err=%d\n%s", errno, bpf_log_buf); |
5bacd780 AS |
108 | return -1; |
109 | } | |
110 | ||
111 | prog_fd[prog_cnt++] = fd; | |
112 | ||
4f2e7ae5 | 113 | if (is_xdp || is_perf_event || is_cgroup_skb || is_cgroup_sk) |
86af8b41 BB |
114 | return 0; |
115 | ||
5bacd780 AS |
116 | if (is_socket) { |
117 | event += 6; | |
118 | if (*event != '/') | |
119 | return 0; | |
120 | event++; | |
121 | if (!isdigit(*event)) { | |
122 | printf("invalid prog number\n"); | |
123 | return -1; | |
124 | } | |
125 | return populate_prog_array(event, fd); | |
126 | } | |
127 | ||
b896c4f9 AS |
128 | if (is_kprobe || is_kretprobe) { |
129 | if (is_kprobe) | |
130 | event += 7; | |
131 | else | |
132 | event += 10; | |
133 | ||
5bacd780 AS |
134 | if (*event == 0) { |
135 | printf("event name cannot be empty\n"); | |
136 | return -1; | |
137 | } | |
138 | ||
139 | if (isdigit(*event)) | |
140 | return populate_prog_array(event, fd); | |
141 | ||
b896c4f9 AS |
142 | snprintf(buf, sizeof(buf), |
143 | "echo '%c:%s %s' >> /sys/kernel/debug/tracing/kprobe_events", | |
144 | is_kprobe ? 'p' : 'r', event, event); | |
145 | err = system(buf); | |
146 | if (err < 0) { | |
147 | printf("failed to create kprobe '%s' error '%s'\n", | |
148 | event, strerror(errno)); | |
149 | return -1; | |
150 | } | |
249b812d | 151 | |
c0766040 AS |
152 | strcpy(buf, DEBUGFS); |
153 | strcat(buf, "events/kprobes/"); | |
154 | strcat(buf, event); | |
155 | strcat(buf, "/id"); | |
156 | } else if (is_tracepoint) { | |
157 | event += 11; | |
158 | ||
159 | if (*event == 0) { | |
160 | printf("event name cannot be empty\n"); | |
161 | return -1; | |
162 | } | |
163 | strcpy(buf, DEBUGFS); | |
164 | strcat(buf, "events/"); | |
165 | strcat(buf, event); | |
166 | strcat(buf, "/id"); | |
167 | } | |
b896c4f9 AS |
168 | |
169 | efd = open(buf, O_RDONLY, 0); | |
170 | if (efd < 0) { | |
171 | printf("failed to open event %s\n", event); | |
172 | return -1; | |
173 | } | |
174 | ||
175 | err = read(efd, buf, sizeof(buf)); | |
176 | if (err < 0 || err >= sizeof(buf)) { | |
177 | printf("read from '%s' failed '%s'\n", event, strerror(errno)); | |
178 | return -1; | |
179 | } | |
180 | ||
181 | close(efd); | |
182 | ||
183 | buf[err] = 0; | |
184 | id = atoi(buf); | |
185 | attr.config = id; | |
186 | ||
205c8ada | 187 | efd = sys_perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0); |
b896c4f9 AS |
188 | if (efd < 0) { |
189 | printf("event %d fd %d err %s\n", id, efd, strerror(errno)); | |
190 | return -1; | |
191 | } | |
192 | event_fd[prog_cnt - 1] = efd; | |
193 | ioctl(efd, PERF_EVENT_IOC_ENABLE, 0); | |
194 | ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd); | |
195 | ||
249b812d AS |
196 | return 0; |
197 | } | |
198 | ||
156450d9 JDB |
199 | static int load_maps(struct bpf_map_data *maps, int nr_maps, |
200 | fixup_map_cb fixup_map) | |
249b812d AS |
201 | { |
202 | int i; | |
156450d9 | 203 | |
5010e948 | 204 | for (i = 0; i < nr_maps; i++) { |
9fd63d05 | 205 | if (fixup_map) |
156450d9 | 206 | fixup_map(&maps[i].def, maps[i].name, i); |
249b812d | 207 | |
156450d9 JDB |
208 | if (maps[i].def.type == BPF_MAP_TYPE_ARRAY_OF_MAPS || |
209 | maps[i].def.type == BPF_MAP_TYPE_HASH_OF_MAPS) { | |
210 | int inner_map_fd = map_fd[maps[i].def.inner_map_idx]; | |
fb30d4b7 | 211 | |
156450d9 JDB |
212 | map_fd[i] = bpf_create_map_in_map(maps[i].def.type, |
213 | maps[i].def.key_size, | |
214 | inner_map_fd, | |
215 | maps[i].def.max_entries, | |
216 | maps[i].def.map_flags); | |
fb30d4b7 | 217 | } else { |
156450d9 JDB |
218 | map_fd[i] = bpf_create_map(maps[i].def.type, |
219 | maps[i].def.key_size, | |
220 | maps[i].def.value_size, | |
221 | maps[i].def.max_entries, | |
222 | maps[i].def.map_flags); | |
fb30d4b7 | 223 | } |
618ec9a7 AS |
224 | if (map_fd[i] < 0) { |
225 | printf("failed to create a map: %d %s\n", | |
226 | errno, strerror(errno)); | |
249b812d | 227 | return 1; |
618ec9a7 | 228 | } |
156450d9 | 229 | maps[i].fd = map_fd[i]; |
5bacd780 | 230 | |
156450d9 | 231 | if (maps[i].def.type == BPF_MAP_TYPE_PROG_ARRAY) |
5bacd780 | 232 | prog_array_fd = map_fd[i]; |
249b812d AS |
233 | } |
234 | return 0; | |
235 | } | |
236 | ||
237 | static int get_sec(Elf *elf, int i, GElf_Ehdr *ehdr, char **shname, | |
238 | GElf_Shdr *shdr, Elf_Data **data) | |
239 | { | |
240 | Elf_Scn *scn; | |
241 | ||
242 | scn = elf_getscn(elf, i); | |
243 | if (!scn) | |
244 | return 1; | |
245 | ||
246 | if (gelf_getshdr(scn, shdr) != shdr) | |
247 | return 2; | |
248 | ||
249 | *shname = elf_strptr(elf, ehdr->e_shstrndx, shdr->sh_name); | |
250 | if (!*shname || !shdr->sh_size) | |
251 | return 3; | |
252 | ||
253 | *data = elf_getdata(scn, 0); | |
254 | if (!*data || elf_getdata(scn, *data) != NULL) | |
255 | return 4; | |
256 | ||
257 | return 0; | |
258 | } | |
259 | ||
260 | static int parse_relo_and_apply(Elf_Data *data, Elf_Data *symbols, | |
156450d9 JDB |
261 | GElf_Shdr *shdr, struct bpf_insn *insn, |
262 | struct bpf_map_data *maps, int nr_maps) | |
249b812d AS |
263 | { |
264 | int i, nrels; | |
265 | ||
266 | nrels = shdr->sh_size / shdr->sh_entsize; | |
267 | ||
268 | for (i = 0; i < nrels; i++) { | |
269 | GElf_Sym sym; | |
270 | GElf_Rel rel; | |
271 | unsigned int insn_idx; | |
156450d9 JDB |
272 | bool match = false; |
273 | int j, map_idx; | |
249b812d AS |
274 | |
275 | gelf_getrel(data, i, &rel); | |
276 | ||
277 | insn_idx = rel.r_offset / sizeof(struct bpf_insn); | |
278 | ||
279 | gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym); | |
280 | ||
281 | if (insn[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) { | |
282 | printf("invalid relo for insn[%d].code 0x%x\n", | |
283 | insn_idx, insn[insn_idx].code); | |
284 | return 1; | |
285 | } | |
286 | insn[insn_idx].src_reg = BPF_PSEUDO_MAP_FD; | |
156450d9 JDB |
287 | |
288 | /* Match FD relocation against recorded map_data[] offset */ | |
289 | for (map_idx = 0; map_idx < nr_maps; map_idx++) { | |
290 | if (maps[map_idx].elf_offset == sym.st_value) { | |
291 | match = true; | |
292 | break; | |
293 | } | |
294 | } | |
295 | if (match) { | |
296 | insn[insn_idx].imm = maps[map_idx].fd; | |
297 | } else { | |
298 | printf("invalid relo for insn[%d] no map_data match\n", | |
299 | insn_idx); | |
300 | return 1; | |
301 | } | |
249b812d AS |
302 | } |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
9fd63d05 MKL |
307 | static int cmp_symbols(const void *l, const void *r) |
308 | { | |
309 | const GElf_Sym *lsym = (const GElf_Sym *)l; | |
310 | const GElf_Sym *rsym = (const GElf_Sym *)r; | |
311 | ||
312 | if (lsym->st_value < rsym->st_value) | |
313 | return -1; | |
314 | else if (lsym->st_value > rsym->st_value) | |
315 | return 1; | |
316 | else | |
317 | return 0; | |
318 | } | |
319 | ||
156450d9 JDB |
320 | static int load_elf_maps_section(struct bpf_map_data *maps, int maps_shndx, |
321 | Elf *elf, Elf_Data *symbols, int strtabidx) | |
9fd63d05 | 322 | { |
156450d9 JDB |
323 | int map_sz_elf, map_sz_copy; |
324 | bool validate_zero = false; | |
325 | Elf_Data *data_maps; | |
326 | int i, nr_maps; | |
327 | GElf_Sym *sym; | |
328 | Elf_Scn *scn; | |
329 | int copy_sz; | |
330 | ||
331 | if (maps_shndx < 0) | |
332 | return -EINVAL; | |
333 | if (!symbols) | |
334 | return -EINVAL; | |
335 | ||
336 | /* Get data for maps section via elf index */ | |
337 | scn = elf_getscn(elf, maps_shndx); | |
338 | if (scn) | |
339 | data_maps = elf_getdata(scn, NULL); | |
340 | if (!scn || !data_maps) { | |
341 | printf("Failed to get Elf_Data from maps section %d\n", | |
342 | maps_shndx); | |
343 | return -EINVAL; | |
344 | } | |
9fd63d05 | 345 | |
156450d9 JDB |
346 | /* For each map get corrosponding symbol table entry */ |
347 | sym = calloc(MAX_MAPS+1, sizeof(GElf_Sym)); | |
348 | for (i = 0, nr_maps = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) { | |
349 | assert(nr_maps < MAX_MAPS+1); | |
350 | if (!gelf_getsym(symbols, i, &sym[nr_maps])) | |
9fd63d05 | 351 | continue; |
156450d9 | 352 | if (sym[nr_maps].st_shndx != maps_shndx) |
9fd63d05 | 353 | continue; |
156450d9 | 354 | /* Only increment iif maps section */ |
9fd63d05 MKL |
355 | nr_maps++; |
356 | } | |
357 | ||
156450d9 JDB |
358 | /* Align to map_fd[] order, via sort on offset in sym.st_value */ |
359 | qsort(sym, nr_maps, sizeof(GElf_Sym), cmp_symbols); | |
360 | ||
361 | /* Keeping compatible with ELF maps section changes | |
362 | * ------------------------------------------------ | |
363 | * The program size of struct bpf_map_def is known by loader | |
364 | * code, but struct stored in ELF file can be different. | |
365 | * | |
366 | * Unfortunately sym[i].st_size is zero. To calculate the | |
367 | * struct size stored in the ELF file, assume all struct have | |
368 | * the same size, and simply divide with number of map | |
369 | * symbols. | |
370 | */ | |
371 | map_sz_elf = data_maps->d_size / nr_maps; | |
372 | map_sz_copy = sizeof(struct bpf_map_def); | |
373 | if (map_sz_elf < map_sz_copy) { | |
374 | /* | |
375 | * Backward compat, loading older ELF file with | |
376 | * smaller struct, keeping remaining bytes zero. | |
377 | */ | |
378 | map_sz_copy = map_sz_elf; | |
379 | } else if (map_sz_elf > map_sz_copy) { | |
380 | /* | |
381 | * Forward compat, loading newer ELF file with larger | |
382 | * struct with unknown features. Assume zero means | |
383 | * feature not used. Thus, validate rest of struct | |
384 | * data is zero. | |
385 | */ | |
386 | validate_zero = true; | |
387 | } | |
9fd63d05 | 388 | |
156450d9 | 389 | /* Memcpy relevant part of ELF maps data to loader maps */ |
9fd63d05 | 390 | for (i = 0; i < nr_maps; i++) { |
156450d9 JDB |
391 | unsigned char *addr, *end; |
392 | struct bpf_map_def *def; | |
393 | const char *map_name; | |
394 | size_t offset; | |
395 | ||
396 | map_name = elf_strptr(elf, strtabidx, sym[i].st_name); | |
397 | maps[i].name = strdup(map_name); | |
398 | if (!maps[i].name) { | |
9fd63d05 MKL |
399 | printf("strdup(%s): %s(%d)\n", map_name, |
400 | strerror(errno), errno); | |
156450d9 JDB |
401 | free(sym); |
402 | return -errno; | |
403 | } | |
404 | ||
405 | /* Symbol value is offset into ELF maps section data area */ | |
406 | offset = sym[i].st_value; | |
407 | def = (struct bpf_map_def *)(data_maps->d_buf + offset); | |
408 | maps[i].elf_offset = offset; | |
409 | memset(&maps[i].def, 0, sizeof(struct bpf_map_def)); | |
410 | memcpy(&maps[i].def, def, map_sz_copy); | |
411 | ||
412 | /* Verify no newer features were requested */ | |
413 | if (validate_zero) { | |
414 | addr = (unsigned char*) def + map_sz_copy; | |
415 | end = (unsigned char*) def + map_sz_elf; | |
416 | for (; addr < end; addr++) { | |
417 | if (*addr != 0) { | |
418 | free(sym); | |
419 | return -EFBIG; | |
420 | } | |
421 | } | |
9fd63d05 MKL |
422 | } |
423 | } | |
424 | ||
156450d9 | 425 | free(sym); |
5010e948 | 426 | return nr_maps; |
9fd63d05 MKL |
427 | } |
428 | ||
429 | static int do_load_bpf_file(const char *path, fixup_map_cb fixup_map) | |
249b812d | 430 | { |
9fd63d05 | 431 | int fd, i, ret, maps_shndx = -1, strtabidx = -1; |
249b812d AS |
432 | Elf *elf; |
433 | GElf_Ehdr ehdr; | |
434 | GElf_Shdr shdr, shdr_prog; | |
9fd63d05 | 435 | Elf_Data *data, *data_prog, *data_maps = NULL, *symbols = NULL; |
156450d9 JDB |
436 | char *shname, *shname_prog; |
437 | int nr_maps = 0; | |
249b812d | 438 | |
a734fb5d MS |
439 | /* reset global variables */ |
440 | kern_version = 0; | |
441 | memset(license, 0, sizeof(license)); | |
442 | memset(processed_sec, 0, sizeof(processed_sec)); | |
443 | ||
249b812d AS |
444 | if (elf_version(EV_CURRENT) == EV_NONE) |
445 | return 1; | |
446 | ||
447 | fd = open(path, O_RDONLY, 0); | |
448 | if (fd < 0) | |
449 | return 1; | |
450 | ||
451 | elf = elf_begin(fd, ELF_C_READ, NULL); | |
452 | ||
453 | if (!elf) | |
454 | return 1; | |
455 | ||
456 | if (gelf_getehdr(elf, &ehdr) != &ehdr) | |
457 | return 1; | |
458 | ||
b896c4f9 AS |
459 | /* clear all kprobes */ |
460 | i = system("echo \"\" > /sys/kernel/debug/tracing/kprobe_events"); | |
461 | ||
249b812d AS |
462 | /* scan over all elf sections to get license and map info */ |
463 | for (i = 1; i < ehdr.e_shnum; i++) { | |
464 | ||
465 | if (get_sec(elf, i, &ehdr, &shname, &shdr, &data)) | |
466 | continue; | |
467 | ||
468 | if (0) /* helpful for llvm debugging */ | |
469 | printf("section %d:%s data %p size %zd link %d flags %d\n", | |
470 | i, shname, data->d_buf, data->d_size, | |
471 | shdr.sh_link, (int) shdr.sh_flags); | |
472 | ||
473 | if (strcmp(shname, "license") == 0) { | |
474 | processed_sec[i] = true; | |
475 | memcpy(license, data->d_buf, data->d_size); | |
b896c4f9 AS |
476 | } else if (strcmp(shname, "version") == 0) { |
477 | processed_sec[i] = true; | |
478 | if (data->d_size != sizeof(int)) { | |
479 | printf("invalid size of version section %zd\n", | |
480 | data->d_size); | |
481 | return 1; | |
482 | } | |
483 | memcpy(&kern_version, data->d_buf, sizeof(int)); | |
249b812d | 484 | } else if (strcmp(shname, "maps") == 0) { |
156450d9 JDB |
485 | int j; |
486 | ||
9fd63d05 MKL |
487 | maps_shndx = i; |
488 | data_maps = data; | |
156450d9 JDB |
489 | for (j = 0; j < MAX_MAPS; j++) |
490 | map_data[j].fd = -1; | |
249b812d | 491 | } else if (shdr.sh_type == SHT_SYMTAB) { |
9fd63d05 | 492 | strtabidx = shdr.sh_link; |
249b812d AS |
493 | symbols = data; |
494 | } | |
495 | } | |
496 | ||
9fd63d05 MKL |
497 | ret = 1; |
498 | ||
499 | if (!symbols) { | |
500 | printf("missing SHT_SYMTAB section\n"); | |
501 | goto done; | |
502 | } | |
503 | ||
504 | if (data_maps) { | |
156450d9 JDB |
505 | nr_maps = load_elf_maps_section(map_data, maps_shndx, |
506 | elf, symbols, strtabidx); | |
507 | if (nr_maps < 0) { | |
508 | printf("Error: Failed loading ELF maps (errno:%d):%s\n", | |
509 | nr_maps, strerror(-nr_maps)); | |
5010e948 JDB |
510 | ret = 1; |
511 | goto done; | |
512 | } | |
156450d9 | 513 | if (load_maps(map_data, nr_maps, fixup_map)) |
9fd63d05 | 514 | goto done; |
156450d9 | 515 | map_data_count = nr_maps; |
9fd63d05 MKL |
516 | |
517 | processed_sec[maps_shndx] = true; | |
518 | } | |
519 | ||
249b812d AS |
520 | /* load programs that need map fixup (relocations) */ |
521 | for (i = 1; i < ehdr.e_shnum; i++) { | |
16ad1329 MS |
522 | if (processed_sec[i]) |
523 | continue; | |
249b812d AS |
524 | |
525 | if (get_sec(elf, i, &ehdr, &shname, &shdr, &data)) | |
526 | continue; | |
527 | if (shdr.sh_type == SHT_REL) { | |
528 | struct bpf_insn *insns; | |
529 | ||
530 | if (get_sec(elf, shdr.sh_info, &ehdr, &shname_prog, | |
531 | &shdr_prog, &data_prog)) | |
532 | continue; | |
533 | ||
db6a71dd AS |
534 | if (shdr_prog.sh_type != SHT_PROGBITS || |
535 | !(shdr_prog.sh_flags & SHF_EXECINSTR)) | |
536 | continue; | |
537 | ||
249b812d AS |
538 | insns = (struct bpf_insn *) data_prog->d_buf; |
539 | ||
540 | processed_sec[shdr.sh_info] = true; | |
541 | processed_sec[i] = true; | |
542 | ||
156450d9 JDB |
543 | if (parse_relo_and_apply(data, symbols, &shdr, insns, |
544 | map_data, nr_maps)) | |
249b812d AS |
545 | continue; |
546 | ||
b896c4f9 AS |
547 | if (memcmp(shname_prog, "kprobe/", 7) == 0 || |
548 | memcmp(shname_prog, "kretprobe/", 10) == 0 || | |
c0766040 | 549 | memcmp(shname_prog, "tracepoint/", 11) == 0 || |
86af8b41 | 550 | memcmp(shname_prog, "xdp", 3) == 0 || |
1c47910e | 551 | memcmp(shname_prog, "perf_event", 10) == 0 || |
4f2e7ae5 DA |
552 | memcmp(shname_prog, "socket", 6) == 0 || |
553 | memcmp(shname_prog, "cgroup/", 7) == 0) | |
249b812d AS |
554 | load_and_attach(shname_prog, insns, data_prog->d_size); |
555 | } | |
556 | } | |
557 | ||
558 | /* load programs that don't use maps */ | |
559 | for (i = 1; i < ehdr.e_shnum; i++) { | |
560 | ||
561 | if (processed_sec[i]) | |
562 | continue; | |
563 | ||
564 | if (get_sec(elf, i, &ehdr, &shname, &shdr, &data)) | |
565 | continue; | |
566 | ||
b896c4f9 AS |
567 | if (memcmp(shname, "kprobe/", 7) == 0 || |
568 | memcmp(shname, "kretprobe/", 10) == 0 || | |
c0766040 | 569 | memcmp(shname, "tracepoint/", 11) == 0 || |
86af8b41 | 570 | memcmp(shname, "xdp", 3) == 0 || |
1c47910e | 571 | memcmp(shname, "perf_event", 10) == 0 || |
4f2e7ae5 DA |
572 | memcmp(shname, "socket", 6) == 0 || |
573 | memcmp(shname, "cgroup/", 7) == 0) | |
249b812d AS |
574 | load_and_attach(shname, data->d_buf, data->d_size); |
575 | } | |
576 | ||
9fd63d05 MKL |
577 | ret = 0; |
578 | done: | |
249b812d | 579 | close(fd); |
9fd63d05 MKL |
580 | return ret; |
581 | } | |
582 | ||
583 | int load_bpf_file(char *path) | |
584 | { | |
585 | return do_load_bpf_file(path, NULL); | |
586 | } | |
587 | ||
588 | int load_bpf_file_fixup_map(const char *path, fixup_map_cb fixup_map) | |
589 | { | |
590 | return do_load_bpf_file(path, fixup_map); | |
249b812d | 591 | } |
b896c4f9 AS |
592 | |
593 | void read_trace_pipe(void) | |
594 | { | |
595 | int trace_fd; | |
596 | ||
597 | trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0); | |
598 | if (trace_fd < 0) | |
599 | return; | |
600 | ||
601 | while (1) { | |
602 | static char buf[4096]; | |
603 | ssize_t sz; | |
604 | ||
605 | sz = read(trace_fd, buf, sizeof(buf)); | |
606 | if (sz > 0) { | |
607 | buf[sz] = 0; | |
608 | puts(buf); | |
609 | } | |
610 | } | |
611 | } | |
3622e7e4 AS |
612 | |
613 | #define MAX_SYMS 300000 | |
614 | static struct ksym syms[MAX_SYMS]; | |
615 | static int sym_cnt; | |
616 | ||
617 | static int ksym_cmp(const void *p1, const void *p2) | |
618 | { | |
619 | return ((struct ksym *)p1)->addr - ((struct ksym *)p2)->addr; | |
620 | } | |
621 | ||
622 | int load_kallsyms(void) | |
623 | { | |
624 | FILE *f = fopen("/proc/kallsyms", "r"); | |
625 | char func[256], buf[256]; | |
626 | char symbol; | |
627 | void *addr; | |
628 | int i = 0; | |
629 | ||
630 | if (!f) | |
631 | return -ENOENT; | |
632 | ||
633 | while (!feof(f)) { | |
634 | if (!fgets(buf, sizeof(buf), f)) | |
635 | break; | |
636 | if (sscanf(buf, "%p %c %s", &addr, &symbol, func) != 3) | |
637 | break; | |
638 | if (!addr) | |
639 | continue; | |
640 | syms[i].addr = (long) addr; | |
641 | syms[i].name = strdup(func); | |
642 | i++; | |
643 | } | |
644 | sym_cnt = i; | |
645 | qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp); | |
646 | return 0; | |
647 | } | |
648 | ||
649 | struct ksym *ksym_search(long key) | |
650 | { | |
651 | int start = 0, end = sym_cnt; | |
652 | int result; | |
653 | ||
654 | while (start < end) { | |
655 | size_t mid = start + (end - start) / 2; | |
656 | ||
657 | result = key - syms[mid].addr; | |
658 | if (result < 0) | |
659 | end = mid; | |
660 | else if (result > 0) | |
661 | start = mid + 1; | |
662 | else | |
663 | return &syms[mid]; | |
664 | } | |
665 | ||
666 | if (start >= 1 && syms[start - 1].addr < key && | |
667 | key < syms[start].addr) | |
668 | /* valid ksym */ | |
669 | return &syms[start - 1]; | |
670 | ||
671 | /* out of range. return _stext */ | |
672 | return &syms[0]; | |
673 | } | |
12d8bb64 | 674 | |
6387d011 | 675 | int set_link_xdp_fd(int ifindex, int fd, __u32 flags) |
12d8bb64 MKL |
676 | { |
677 | struct sockaddr_nl sa; | |
678 | int sock, seq = 0, len, ret = -1; | |
679 | char buf[4096]; | |
680 | struct nlattr *nla, *nla_xdp; | |
681 | struct { | |
682 | struct nlmsghdr nh; | |
683 | struct ifinfomsg ifinfo; | |
684 | char attrbuf[64]; | |
685 | } req; | |
686 | struct nlmsghdr *nh; | |
687 | struct nlmsgerr *err; | |
688 | ||
689 | memset(&sa, 0, sizeof(sa)); | |
690 | sa.nl_family = AF_NETLINK; | |
691 | ||
692 | sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE); | |
693 | if (sock < 0) { | |
694 | printf("open netlink socket: %s\n", strerror(errno)); | |
695 | return -1; | |
696 | } | |
697 | ||
698 | if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) { | |
699 | printf("bind to netlink: %s\n", strerror(errno)); | |
700 | goto cleanup; | |
701 | } | |
702 | ||
703 | memset(&req, 0, sizeof(req)); | |
704 | req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)); | |
705 | req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK; | |
706 | req.nh.nlmsg_type = RTM_SETLINK; | |
707 | req.nh.nlmsg_pid = 0; | |
708 | req.nh.nlmsg_seq = ++seq; | |
709 | req.ifinfo.ifi_family = AF_UNSPEC; | |
710 | req.ifinfo.ifi_index = ifindex; | |
3993f2cb DA |
711 | |
712 | /* started nested attribute for XDP */ | |
12d8bb64 MKL |
713 | nla = (struct nlattr *)(((char *)&req) |
714 | + NLMSG_ALIGN(req.nh.nlmsg_len)); | |
715 | nla->nla_type = NLA_F_NESTED | 43/*IFLA_XDP*/; | |
3993f2cb | 716 | nla->nla_len = NLA_HDRLEN; |
12d8bb64 | 717 | |
3993f2cb DA |
718 | /* add XDP fd */ |
719 | nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len); | |
12d8bb64 MKL |
720 | nla_xdp->nla_type = 1/*IFLA_XDP_FD*/; |
721 | nla_xdp->nla_len = NLA_HDRLEN + sizeof(int); | |
722 | memcpy((char *)nla_xdp + NLA_HDRLEN, &fd, sizeof(fd)); | |
3993f2cb DA |
723 | nla->nla_len += nla_xdp->nla_len; |
724 | ||
725 | /* if user passed in any flags, add those too */ | |
726 | if (flags) { | |
727 | nla_xdp = (struct nlattr *)((char *)nla + nla->nla_len); | |
728 | nla_xdp->nla_type = 3/*IFLA_XDP_FLAGS*/; | |
729 | nla_xdp->nla_len = NLA_HDRLEN + sizeof(flags); | |
730 | memcpy((char *)nla_xdp + NLA_HDRLEN, &flags, sizeof(flags)); | |
731 | nla->nla_len += nla_xdp->nla_len; | |
732 | } | |
12d8bb64 MKL |
733 | |
734 | req.nh.nlmsg_len += NLA_ALIGN(nla->nla_len); | |
735 | ||
736 | if (send(sock, &req, req.nh.nlmsg_len, 0) < 0) { | |
737 | printf("send to netlink: %s\n", strerror(errno)); | |
738 | goto cleanup; | |
739 | } | |
740 | ||
741 | len = recv(sock, buf, sizeof(buf), 0); | |
742 | if (len < 0) { | |
743 | printf("recv from netlink: %s\n", strerror(errno)); | |
744 | goto cleanup; | |
745 | } | |
746 | ||
747 | for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, len); | |
748 | nh = NLMSG_NEXT(nh, len)) { | |
749 | if (nh->nlmsg_pid != getpid()) { | |
750 | printf("Wrong pid %d, expected %d\n", | |
751 | nh->nlmsg_pid, getpid()); | |
752 | goto cleanup; | |
753 | } | |
754 | if (nh->nlmsg_seq != seq) { | |
755 | printf("Wrong seq %d, expected %d\n", | |
756 | nh->nlmsg_seq, seq); | |
757 | goto cleanup; | |
758 | } | |
759 | switch (nh->nlmsg_type) { | |
760 | case NLMSG_ERROR: | |
761 | err = (struct nlmsgerr *)NLMSG_DATA(nh); | |
762 | if (!err->error) | |
763 | continue; | |
764 | printf("nlmsg error %s\n", strerror(-err->error)); | |
765 | goto cleanup; | |
766 | case NLMSG_DONE: | |
767 | break; | |
768 | } | |
769 | } | |
770 | ||
771 | ret = 0; | |
772 | ||
773 | cleanup: | |
774 | close(sock); | |
775 | return ret; | |
776 | } |