]>
Commit | Line | Data |
---|---|---|
1d129d19 | 1 | /* |
e4225669 | 2 | * bpf.c BPF common code |
1d129d19 JP |
3 | * |
4 | * This program is free software; you can distribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
e4225669 | 9 | * Authors: Daniel Borkmann <daniel@iogearbox.net> |
1d129d19 | 10 | * Jiri Pirko <jiri@resnulli.us> |
e4225669 | 11 | * Alexei Starovoitov <ast@kernel.org> |
1d129d19 JP |
12 | */ |
13 | ||
14 | #include <stdio.h> | |
15 | #include <stdlib.h> | |
16 | #include <unistd.h> | |
17 | #include <string.h> | |
18 | #include <stdbool.h> | |
473d7840 | 19 | #include <stdint.h> |
1d129d19 | 20 | #include <errno.h> |
11c39b5e DB |
21 | #include <fcntl.h> |
22 | #include <stdarg.h> | |
5c5a0f3d | 23 | #include <limits.h> |
e4225669 | 24 | #include <assert.h> |
1d129d19 | 25 | |
11c39b5e DB |
26 | #ifdef HAVE_ELF |
27 | #include <libelf.h> | |
28 | #include <gelf.h> | |
29 | #endif | |
30 | ||
32e93fb7 DB |
31 | #include <sys/types.h> |
32 | #include <sys/stat.h> | |
33 | #include <sys/un.h> | |
34 | #include <sys/vfs.h> | |
35 | #include <sys/mount.h> | |
36 | #include <sys/syscall.h> | |
37 | #include <sys/sendfile.h> | |
38 | #include <sys/resource.h> | |
39 | ||
8187b012 DB |
40 | #include <arpa/inet.h> |
41 | ||
1d129d19 | 42 | #include "utils.h" |
6256f8c9 | 43 | |
e4225669 | 44 | #include "bpf_util.h" |
6256f8c9 DB |
45 | #include "bpf_elf.h" |
46 | #include "bpf_scm.h" | |
47 | ||
e4225669 DB |
48 | struct bpf_prog_meta { |
49 | const char *type; | |
50 | const char *subdir; | |
51 | const char *section; | |
52 | bool may_uds_export; | |
53 | }; | |
1d129d19 | 54 | |
e4225669 DB |
55 | static const enum bpf_prog_type __bpf_types[] = { |
56 | BPF_PROG_TYPE_SCHED_CLS, | |
57 | BPF_PROG_TYPE_SCHED_ACT, | |
c7272ca7 | 58 | BPF_PROG_TYPE_XDP, |
b15f440e TG |
59 | BPF_PROG_TYPE_LWT_IN, |
60 | BPF_PROG_TYPE_LWT_OUT, | |
61 | BPF_PROG_TYPE_LWT_XMIT, | |
e4225669 | 62 | }; |
67584e3a | 63 | |
e4225669 DB |
64 | static const struct bpf_prog_meta __bpf_prog_meta[] = { |
65 | [BPF_PROG_TYPE_SCHED_CLS] = { | |
66 | .type = "cls", | |
67 | .subdir = "tc", | |
68 | .section = ELF_SECTION_CLASSIFIER, | |
69 | .may_uds_export = true, | |
70 | }, | |
71 | [BPF_PROG_TYPE_SCHED_ACT] = { | |
72 | .type = "act", | |
73 | .subdir = "tc", | |
74 | .section = ELF_SECTION_ACTION, | |
75 | .may_uds_export = true, | |
76 | }, | |
c7272ca7 DB |
77 | [BPF_PROG_TYPE_XDP] = { |
78 | .type = "xdp", | |
79 | .subdir = "xdp", | |
80 | .section = ELF_SECTION_PROG, | |
81 | }, | |
b15f440e TG |
82 | [BPF_PROG_TYPE_LWT_IN] = { |
83 | .type = "lwt_in", | |
84 | .subdir = "ip", | |
85 | .section = ELF_SECTION_PROG, | |
86 | }, | |
87 | [BPF_PROG_TYPE_LWT_OUT] = { | |
88 | .type = "lwt_out", | |
89 | .subdir = "ip", | |
90 | .section = ELF_SECTION_PROG, | |
91 | }, | |
92 | [BPF_PROG_TYPE_LWT_XMIT] = { | |
93 | .type = "lwt_xmit", | |
94 | .subdir = "ip", | |
95 | .section = ELF_SECTION_PROG, | |
96 | }, | |
e4225669 DB |
97 | }; |
98 | ||
99 | static const char *bpf_prog_to_subdir(enum bpf_prog_type type) | |
100 | { | |
101 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
102 | __bpf_prog_meta[type].subdir); | |
103 | return __bpf_prog_meta[type].subdir; | |
104 | } | |
105 | ||
106 | const char *bpf_prog_to_default_section(enum bpf_prog_type type) | |
107 | { | |
108 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
109 | __bpf_prog_meta[type].section); | |
110 | return __bpf_prog_meta[type].section; | |
111 | } | |
e77fa41d | 112 | |
32e93fb7 DB |
113 | #ifdef HAVE_ELF |
114 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
115 | const char *sec, bool verbose); | |
116 | #else | |
117 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
118 | const char *sec, bool verbose) | |
119 | { | |
120 | fprintf(stderr, "No ELF library support compiled in.\n"); | |
121 | errno = ENOSYS; | |
122 | return -1; | |
123 | } | |
124 | #endif | |
125 | ||
126 | static inline __u64 bpf_ptr_to_u64(const void *ptr) | |
127 | { | |
128 | return (__u64)(unsigned long)ptr; | |
129 | } | |
130 | ||
131 | static int bpf(int cmd, union bpf_attr *attr, unsigned int size) | |
132 | { | |
133 | #ifdef __NR_bpf | |
134 | return syscall(__NR_bpf, cmd, attr, size); | |
135 | #else | |
136 | fprintf(stderr, "No bpf syscall, kernel headers too old?\n"); | |
137 | errno = ENOSYS; | |
138 | return -1; | |
139 | #endif | |
140 | } | |
141 | ||
91d88eeb DB |
142 | static int bpf_map_update(int fd, const void *key, const void *value, |
143 | uint64_t flags) | |
32e93fb7 | 144 | { |
d17b136f | 145 | union bpf_attr attr = {}; |
67584e3a | 146 | |
67584e3a ND |
147 | attr.map_fd = fd; |
148 | attr.key = bpf_ptr_to_u64(key); | |
149 | attr.value = bpf_ptr_to_u64(value); | |
150 | attr.flags = flags; | |
32e93fb7 | 151 | |
91d88eeb | 152 | return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
32e93fb7 DB |
153 | } |
154 | ||
779525cd DB |
155 | static int bpf_prog_fd_by_id(uint32_t id) |
156 | { | |
157 | union bpf_attr attr = {}; | |
158 | ||
159 | attr.prog_id = id; | |
160 | ||
161 | return bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); | |
162 | } | |
163 | ||
164 | static int bpf_prog_info_by_fd(int fd, struct bpf_prog_info *info, | |
165 | uint32_t *info_len) | |
166 | { | |
167 | union bpf_attr attr = {}; | |
168 | int ret; | |
169 | ||
170 | attr.info.bpf_fd = fd; | |
171 | attr.info.info = bpf_ptr_to_u64(info); | |
172 | attr.info.info_len = *info_len; | |
173 | ||
174 | *info_len = 0; | |
175 | ret = bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); | |
176 | if (!ret) | |
177 | *info_len = attr.info.info_len; | |
178 | ||
179 | return ret; | |
180 | } | |
181 | ||
182 | void bpf_dump_prog_info(FILE *f, uint32_t id) | |
183 | { | |
184 | struct bpf_prog_info info = {}; | |
185 | uint32_t len = sizeof(info); | |
186 | int fd, ret; | |
187 | ||
188 | fprintf(f, "id %u ", id); | |
189 | ||
190 | fd = bpf_prog_fd_by_id(id); | |
191 | if (fd < 0) | |
192 | return; | |
193 | ||
194 | ret = bpf_prog_info_by_fd(fd, &info, &len); | |
195 | if (!ret && len) { | |
196 | if (info.jited_prog_len) | |
197 | fprintf(f, "jited "); | |
198 | } | |
199 | ||
200 | close(fd); | |
201 | } | |
202 | ||
32e93fb7 DB |
203 | static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len, |
204 | char **bpf_string, bool *need_release, | |
205 | const char separator) | |
1d129d19 JP |
206 | { |
207 | char sp; | |
208 | ||
209 | if (from_file) { | |
210 | size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,"); | |
3da3ebfc | 211 | char *tmp_string, *pos, c, c_prev = ' '; |
1d129d19 JP |
212 | FILE *fp; |
213 | ||
214 | tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len; | |
3da3ebfc | 215 | tmp_string = pos = calloc(1, tmp_len); |
1d129d19 JP |
216 | if (tmp_string == NULL) |
217 | return -ENOMEM; | |
218 | ||
1d129d19 JP |
219 | fp = fopen(arg, "r"); |
220 | if (fp == NULL) { | |
221 | perror("Cannot fopen"); | |
222 | free(tmp_string); | |
223 | return -ENOENT; | |
224 | } | |
225 | ||
3da3ebfc PS |
226 | while ((c = fgetc(fp)) != EOF) { |
227 | switch (c) { | |
228 | case '\n': | |
229 | if (c_prev != ',') | |
230 | *(pos++) = ','; | |
231 | break; | |
232 | case ' ': | |
233 | case '\t': | |
234 | if (c_prev != ' ') | |
235 | *(pos++) = c; | |
236 | break; | |
237 | default: | |
238 | *(pos++) = c; | |
239 | } | |
240 | if (pos - tmp_string == tmp_len) | |
241 | break; | |
242 | c_prev = c; | |
243 | } | |
244 | ||
245 | if (!feof(fp)) { | |
1d129d19 JP |
246 | free(tmp_string); |
247 | fclose(fp); | |
3da3ebfc | 248 | return -E2BIG; |
1d129d19 JP |
249 | } |
250 | ||
251 | fclose(fp); | |
3da3ebfc | 252 | *pos = 0; |
e4225669 | 253 | |
1d129d19 JP |
254 | *need_release = true; |
255 | *bpf_string = tmp_string; | |
256 | } else { | |
257 | *need_release = false; | |
258 | *bpf_string = arg; | |
259 | } | |
260 | ||
261 | if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 || | |
262 | sp != separator) { | |
263 | if (*need_release) | |
264 | free(*bpf_string); | |
265 | return -EINVAL; | |
266 | } | |
267 | ||
268 | return 0; | |
269 | } | |
270 | ||
32e93fb7 DB |
271 | static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops, |
272 | bool from_file) | |
1d129d19 JP |
273 | { |
274 | char *bpf_string, *token, separator = ','; | |
275 | int ret = 0, i = 0; | |
276 | bool need_release; | |
277 | __u16 bpf_len = 0; | |
278 | ||
279 | if (argc < 1) | |
280 | return -EINVAL; | |
281 | if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string, | |
282 | &need_release, separator)) | |
283 | return -EINVAL; | |
284 | if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) { | |
285 | ret = -EINVAL; | |
286 | goto out; | |
287 | } | |
288 | ||
289 | token = bpf_string; | |
290 | while ((token = strchr(token, separator)) && (++token)[0]) { | |
291 | if (i >= bpf_len) { | |
32a121cb | 292 | fprintf(stderr, "Real program length exceeds encoded length parameter!\n"); |
1d129d19 JP |
293 | ret = -EINVAL; |
294 | goto out; | |
295 | } | |
296 | ||
297 | if (sscanf(token, "%hu %hhu %hhu %u,", | |
298 | &bpf_ops[i].code, &bpf_ops[i].jt, | |
299 | &bpf_ops[i].jf, &bpf_ops[i].k) != 4) { | |
300 | fprintf(stderr, "Error at instruction %d!\n", i); | |
301 | ret = -EINVAL; | |
302 | goto out; | |
303 | } | |
304 | ||
305 | i++; | |
306 | } | |
307 | ||
308 | if (i != bpf_len) { | |
afc1a200 | 309 | fprintf(stderr, "Parsed program length is less than encoded length parameter!\n"); |
1d129d19 JP |
310 | ret = -EINVAL; |
311 | goto out; | |
312 | } | |
313 | ret = bpf_len; | |
1d129d19 JP |
314 | out: |
315 | if (need_release) | |
316 | free(bpf_string); | |
317 | ||
318 | return ret; | |
319 | } | |
320 | ||
321 | void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len) | |
322 | { | |
d896797c | 323 | struct sock_filter *ops = RTA_DATA(bpf_ops); |
1d129d19 JP |
324 | int i; |
325 | ||
326 | if (len == 0) | |
327 | return; | |
328 | ||
329 | fprintf(f, "bytecode \'%u,", len); | |
330 | ||
331 | for (i = 0; i < len - 1; i++) | |
332 | fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt, | |
333 | ops[i].jf, ops[i].k); | |
334 | ||
6256f8c9 | 335 | fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt, |
1d129d19 JP |
336 | ops[i].jf, ops[i].k); |
337 | } | |
11c39b5e | 338 | |
afc1a200 DB |
339 | static void bpf_map_pin_report(const struct bpf_elf_map *pin, |
340 | const struct bpf_elf_map *obj) | |
341 | { | |
342 | fprintf(stderr, "Map specification differs from pinned file!\n"); | |
343 | ||
344 | if (obj->type != pin->type) | |
345 | fprintf(stderr, " - Type: %u (obj) != %u (pin)\n", | |
346 | obj->type, pin->type); | |
347 | if (obj->size_key != pin->size_key) | |
348 | fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n", | |
349 | obj->size_key, pin->size_key); | |
350 | if (obj->size_value != pin->size_value) | |
351 | fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n", | |
352 | obj->size_value, pin->size_value); | |
353 | if (obj->max_elem != pin->max_elem) | |
354 | fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n", | |
355 | obj->max_elem, pin->max_elem); | |
4dd3f50a DB |
356 | if (obj->flags != pin->flags) |
357 | fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n", | |
358 | obj->flags, pin->flags); | |
afc1a200 DB |
359 | |
360 | fprintf(stderr, "\n"); | |
361 | } | |
362 | ||
ecb05c0f DB |
363 | struct bpf_prog_data { |
364 | unsigned int type; | |
365 | unsigned int jited; | |
366 | }; | |
367 | ||
368 | struct bpf_map_ext { | |
369 | struct bpf_prog_data owner; | |
370 | }; | |
371 | ||
372 | static int bpf_derive_elf_map_from_fdinfo(int fd, struct bpf_elf_map *map, | |
373 | struct bpf_map_ext *ext) | |
9e607f2e | 374 | { |
ecb05c0f | 375 | unsigned int val, owner_type = 0, owner_jited = 0; |
9e607f2e | 376 | char file[PATH_MAX], buff[4096]; |
9e607f2e DB |
377 | FILE *fp; |
378 | ||
379 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
ecb05c0f | 380 | memset(map, 0, sizeof(*map)); |
9e607f2e DB |
381 | |
382 | fp = fopen(file, "r"); | |
383 | if (!fp) { | |
384 | fprintf(stderr, "No procfs support?!\n"); | |
385 | return -EIO; | |
386 | } | |
387 | ||
9e607f2e DB |
388 | while (fgets(buff, sizeof(buff), fp)) { |
389 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
ecb05c0f | 390 | map->type = val; |
9e607f2e | 391 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) |
ecb05c0f | 392 | map->size_key = val; |
9e607f2e | 393 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) |
ecb05c0f | 394 | map->size_value = val; |
9e607f2e | 395 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) |
ecb05c0f | 396 | map->max_elem = val; |
4dd3f50a | 397 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) |
ecb05c0f | 398 | map->flags = val; |
fb24802b DB |
399 | else if (sscanf(buff, "owner_prog_type:\t%i", &val) == 1) |
400 | owner_type = val; | |
ecb05c0f DB |
401 | else if (sscanf(buff, "owner_jited:\t%i", &val) == 1) |
402 | owner_jited = val; | |
9e607f2e DB |
403 | } |
404 | ||
405 | fclose(fp); | |
ecb05c0f DB |
406 | if (ext) { |
407 | memset(ext, 0, sizeof(*ext)); | |
408 | ext->owner.type = owner_type; | |
409 | ext->owner.jited = owner_jited; | |
410 | } | |
411 | ||
412 | return 0; | |
413 | } | |
414 | ||
415 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, | |
416 | struct bpf_map_ext *ext, int length, | |
417 | enum bpf_prog_type type) | |
418 | { | |
419 | struct bpf_elf_map tmp, zero = {}; | |
420 | int ret; | |
421 | ||
422 | ret = bpf_derive_elf_map_from_fdinfo(fd, &tmp, ext); | |
423 | if (ret < 0) | |
424 | return ret; | |
9e607f2e | 425 | |
fb24802b DB |
426 | /* The decision to reject this is on kernel side eventually, but |
427 | * at least give the user a chance to know what's wrong. | |
428 | */ | |
ecb05c0f | 429 | if (ext->owner.type && ext->owner.type != type) |
fb24802b | 430 | fprintf(stderr, "Program array map owner types differ: %u (obj) != %u (pin)\n", |
ecb05c0f | 431 | type, ext->owner.type); |
fb24802b | 432 | |
91d88eeb | 433 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
434 | return 0; |
435 | } else { | |
9e607f2e DB |
436 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, |
437 | * so just accept it. We know we do have an eBPF fd and in this | |
438 | * case, everything is 0. It is guaranteed that no such map exists | |
439 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
440 | */ | |
91d88eeb | 441 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
442 | return 0; |
443 | ||
afc1a200 | 444 | bpf_map_pin_report(&tmp, map); |
9e607f2e DB |
445 | return -EINVAL; |
446 | } | |
447 | } | |
448 | ||
91d88eeb DB |
449 | static int bpf_mnt_fs(const char *target) |
450 | { | |
451 | bool bind_done = false; | |
452 | ||
453 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
454 | if (errno != EINVAL || bind_done) { | |
455 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
456 | target, strerror(errno)); | |
457 | return -1; | |
458 | } | |
459 | ||
460 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
461 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
462 | target, target, strerror(errno)); | |
463 | return -1; | |
464 | } | |
465 | ||
466 | bind_done = true; | |
467 | } | |
468 | ||
e4225669 | 469 | if (mount("bpf", target, "bpf", 0, "mode=0700")) { |
91d88eeb DB |
470 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", |
471 | target, strerror(errno)); | |
472 | return -1; | |
473 | } | |
474 | ||
475 | return 0; | |
476 | } | |
477 | ||
95ae9a48 DB |
478 | static int bpf_mnt_check_target(const char *target) |
479 | { | |
480 | struct stat sb = {}; | |
481 | int ret; | |
482 | ||
483 | ret = stat(target, &sb); | |
484 | if (ret) { | |
485 | ret = mkdir(target, S_IRWXU); | |
486 | if (ret) { | |
487 | fprintf(stderr, "mkdir %s failed: %s\n", target, | |
488 | strerror(errno)); | |
489 | return ret; | |
490 | } | |
491 | } | |
492 | ||
493 | return 0; | |
494 | } | |
495 | ||
32e93fb7 DB |
496 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
497 | { | |
498 | struct statfs st_fs; | |
499 | ||
500 | if (statfs(mnt, &st_fs) < 0) | |
501 | return -ENOENT; | |
502 | if ((unsigned long)st_fs.f_type != magic) | |
503 | return -ENOENT; | |
504 | ||
505 | return 0; | |
506 | } | |
507 | ||
95ae9a48 DB |
508 | static const char *bpf_find_mntpt_single(unsigned long magic, char *mnt, |
509 | int len, const char *mntpt) | |
510 | { | |
511 | int ret; | |
512 | ||
513 | ret = bpf_valid_mntpt(mntpt, magic); | |
514 | if (!ret) { | |
515 | strncpy(mnt, mntpt, len - 1); | |
516 | mnt[len - 1] = 0; | |
517 | return mnt; | |
518 | } | |
519 | ||
520 | return NULL; | |
521 | } | |
522 | ||
32e93fb7 DB |
523 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, |
524 | char *mnt, int len, | |
525 | const char * const *known_mnts) | |
526 | { | |
527 | const char * const *ptr; | |
528 | char type[100]; | |
529 | FILE *fp; | |
530 | ||
531 | if (known_mnts) { | |
532 | ptr = known_mnts; | |
533 | while (*ptr) { | |
95ae9a48 | 534 | if (bpf_find_mntpt_single(magic, mnt, len, *ptr)) |
32e93fb7 | 535 | return mnt; |
32e93fb7 DB |
536 | ptr++; |
537 | } | |
538 | } | |
539 | ||
c3724e4b PS |
540 | if (len != PATH_MAX) |
541 | return NULL; | |
542 | ||
32e93fb7 | 543 | fp = fopen("/proc/mounts", "r"); |
c3724e4b | 544 | if (fp == NULL) |
32e93fb7 DB |
545 | return NULL; |
546 | ||
547 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
548 | mnt, type) == 2) { | |
549 | if (strcmp(type, fstype) == 0) | |
550 | break; | |
551 | } | |
552 | ||
553 | fclose(fp); | |
554 | if (strcmp(type, fstype) != 0) | |
555 | return NULL; | |
556 | ||
557 | return mnt; | |
558 | } | |
559 | ||
560 | int bpf_trace_pipe(void) | |
561 | { | |
562 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
563 | static const char * const tracefs_known_mnts[] = { | |
564 | TRACE_DIR_MNT, | |
565 | "/sys/kernel/debug/tracing", | |
566 | "/tracing", | |
567 | "/trace", | |
568 | 0, | |
569 | }; | |
570 | char tpipe[PATH_MAX]; | |
571 | const char *mnt; | |
572 | int fd; | |
573 | ||
574 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
575 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
576 | if (!mnt) { | |
577 | fprintf(stderr, "tracefs not mounted?\n"); | |
578 | return -1; | |
579 | } | |
580 | ||
581 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
582 | ||
583 | fd = open(tpipe, O_RDONLY); | |
584 | if (fd < 0) | |
585 | return -1; | |
586 | ||
587 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
588 | while (1) { | |
589 | static char buff[4096]; | |
590 | ssize_t ret; | |
591 | ||
592 | ret = read(fd, buff, sizeof(buff) - 1); | |
593 | if (ret > 0) { | |
b5c78e1b PS |
594 | if (write(STDERR_FILENO, buff, ret) != ret) |
595 | return -1; | |
32e93fb7 DB |
596 | fflush(stderr); |
597 | } | |
598 | } | |
599 | ||
600 | return 0; | |
601 | } | |
602 | ||
e4225669 | 603 | static int bpf_gen_global(const char *bpf_sub_dir) |
91d88eeb | 604 | { |
e4225669 DB |
605 | char bpf_glo_dir[PATH_MAX]; |
606 | int ret; | |
607 | ||
608 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/", | |
609 | bpf_sub_dir, BPF_DIR_GLOBALS); | |
610 | ||
611 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
612 | if (ret && errno != EEXIST) { | |
613 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
614 | strerror(errno)); | |
615 | return ret; | |
616 | } | |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
621 | static int bpf_gen_master(const char *base, const char *name) | |
622 | { | |
623 | char bpf_sub_dir[PATH_MAX]; | |
624 | int ret; | |
625 | ||
626 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name); | |
627 | ||
628 | ret = mkdir(bpf_sub_dir, S_IRWXU); | |
629 | if (ret && errno != EEXIST) { | |
630 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir, | |
631 | strerror(errno)); | |
632 | return ret; | |
633 | } | |
634 | ||
635 | return bpf_gen_global(bpf_sub_dir); | |
636 | } | |
637 | ||
638 | static int bpf_slave_via_bind_mnt(const char *full_name, | |
639 | const char *full_link) | |
640 | { | |
641 | int ret; | |
642 | ||
643 | ret = mkdir(full_name, S_IRWXU); | |
644 | if (ret) { | |
645 | assert(errno != EEXIST); | |
646 | fprintf(stderr, "mkdir %s failed: %s\n", full_name, | |
647 | strerror(errno)); | |
648 | return ret; | |
649 | } | |
650 | ||
651 | ret = mount(full_link, full_name, "none", MS_BIND, NULL); | |
652 | if (ret) { | |
653 | rmdir(full_name); | |
654 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
655 | full_link, full_name, strerror(errno)); | |
656 | } | |
657 | ||
658 | return ret; | |
659 | } | |
660 | ||
661 | static int bpf_gen_slave(const char *base, const char *name, | |
662 | const char *link) | |
663 | { | |
664 | char bpf_lnk_dir[PATH_MAX]; | |
665 | char bpf_sub_dir[PATH_MAX]; | |
666 | struct stat sb = {}; | |
667 | int ret; | |
668 | ||
669 | snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link); | |
670 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name); | |
671 | ||
672 | ret = symlink(bpf_lnk_dir, bpf_sub_dir); | |
673 | if (ret) { | |
674 | if (errno != EEXIST) { | |
675 | if (errno != EPERM) { | |
676 | fprintf(stderr, "symlink %s failed: %s\n", | |
677 | bpf_sub_dir, strerror(errno)); | |
678 | return ret; | |
679 | } | |
680 | ||
681 | return bpf_slave_via_bind_mnt(bpf_sub_dir, | |
682 | bpf_lnk_dir); | |
683 | } | |
684 | ||
685 | ret = lstat(bpf_sub_dir, &sb); | |
686 | if (ret) { | |
687 | fprintf(stderr, "lstat %s failed: %s\n", | |
688 | bpf_sub_dir, strerror(errno)); | |
689 | return ret; | |
690 | } | |
691 | ||
692 | if ((sb.st_mode & S_IFMT) != S_IFLNK) | |
693 | return bpf_gen_global(bpf_sub_dir); | |
694 | } | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
699 | static int bpf_gen_hierarchy(const char *base) | |
700 | { | |
701 | int ret, i; | |
702 | ||
703 | ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0])); | |
704 | for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++) | |
705 | ret = bpf_gen_slave(base, | |
706 | bpf_prog_to_subdir(__bpf_types[i]), | |
707 | bpf_prog_to_subdir(__bpf_types[0])); | |
708 | return ret; | |
709 | } | |
710 | ||
711 | static const char *bpf_get_work_dir(enum bpf_prog_type type) | |
712 | { | |
713 | static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT; | |
714 | static char bpf_wrk_dir[PATH_MAX]; | |
91d88eeb | 715 | static const char *mnt; |
e4225669 | 716 | static bool bpf_mnt_cached; |
95ae9a48 | 717 | const char *mnt_env = getenv(BPF_ENV_MNT); |
91d88eeb DB |
718 | static const char * const bpf_known_mnts[] = { |
719 | BPF_DIR_MNT, | |
e4225669 | 720 | "/bpf", |
91d88eeb DB |
721 | 0, |
722 | }; | |
91d88eeb DB |
723 | int ret; |
724 | ||
e4225669 DB |
725 | if (bpf_mnt_cached) { |
726 | const char *out = mnt; | |
727 | ||
51361a9f | 728 | if (out && type) { |
e4225669 DB |
729 | snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/", |
730 | out, bpf_prog_to_subdir(type)); | |
731 | out = bpf_tmp; | |
732 | } | |
733 | return out; | |
734 | } | |
91d88eeb | 735 | |
95ae9a48 DB |
736 | if (mnt_env) |
737 | mnt = bpf_find_mntpt_single(BPF_FS_MAGIC, bpf_tmp, | |
738 | sizeof(bpf_tmp), mnt_env); | |
739 | else | |
740 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp, | |
741 | sizeof(bpf_tmp), bpf_known_mnts); | |
91d88eeb | 742 | if (!mnt) { |
95ae9a48 DB |
743 | mnt = mnt_env ? : BPF_DIR_MNT; |
744 | ret = bpf_mnt_check_target(mnt); | |
745 | if (!ret) | |
746 | ret = bpf_mnt_fs(mnt); | |
91d88eeb DB |
747 | if (ret) { |
748 | mnt = NULL; | |
749 | goto out; | |
750 | } | |
751 | } | |
752 | ||
e4225669 | 753 | snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt); |
91d88eeb | 754 | |
e4225669 DB |
755 | ret = bpf_gen_hierarchy(bpf_wrk_dir); |
756 | if (ret) { | |
91d88eeb DB |
757 | mnt = NULL; |
758 | goto out; | |
759 | } | |
760 | ||
e4225669 | 761 | mnt = bpf_wrk_dir; |
91d88eeb DB |
762 | out: |
763 | bpf_mnt_cached = true; | |
91d88eeb DB |
764 | return mnt; |
765 | } | |
766 | ||
e4225669 | 767 | static int bpf_obj_get(const char *pathname, enum bpf_prog_type type) |
91d88eeb | 768 | { |
d17b136f | 769 | union bpf_attr attr = {}; |
91d88eeb DB |
770 | char tmp[PATH_MAX]; |
771 | ||
772 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
e4225669 | 773 | pathname[1] == ':' && bpf_get_work_dir(type)) { |
91d88eeb | 774 | snprintf(tmp, sizeof(tmp), "%s/%s", |
e4225669 | 775 | bpf_get_work_dir(type), pathname + 2); |
91d88eeb DB |
776 | pathname = tmp; |
777 | } | |
778 | ||
91d88eeb DB |
779 | attr.pathname = bpf_ptr_to_u64(pathname); |
780 | ||
781 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
782 | } | |
783 | ||
21856018 DB |
784 | static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type) |
785 | { | |
786 | int prog_fd = bpf_obj_get(pathname, type); | |
787 | ||
788 | if (prog_fd < 0) | |
789 | fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n", | |
790 | pathname, strerror(errno)); | |
791 | return prog_fd; | |
792 | } | |
793 | ||
91d88eeb | 794 | enum bpf_mode { |
e4225669 | 795 | CBPF_BYTECODE, |
91d88eeb DB |
796 | CBPF_FILE, |
797 | EBPF_OBJECT, | |
798 | EBPF_PINNED, | |
e4225669 | 799 | BPF_MODE_MAX, |
91d88eeb DB |
800 | }; |
801 | ||
e4225669 DB |
802 | static int bpf_parse(enum bpf_prog_type *type, enum bpf_mode *mode, |
803 | struct bpf_cfg_in *cfg, const bool *opt_tbl) | |
32e93fb7 | 804 | { |
32e93fb7 | 805 | const char *file, *section, *uds_name; |
32e93fb7 | 806 | bool verbose = false; |
e4225669 | 807 | int i, ret, argc; |
91d88eeb DB |
808 | char **argv; |
809 | ||
e4225669 DB |
810 | argv = cfg->argv; |
811 | argc = cfg->argc; | |
91d88eeb DB |
812 | |
813 | if (opt_tbl[CBPF_BYTECODE] && | |
814 | (matches(*argv, "bytecode") == 0 || | |
815 | strcmp(*argv, "bc") == 0)) { | |
816 | *mode = CBPF_BYTECODE; | |
817 | } else if (opt_tbl[CBPF_FILE] && | |
818 | (matches(*argv, "bytecode-file") == 0 || | |
819 | strcmp(*argv, "bcf") == 0)) { | |
820 | *mode = CBPF_FILE; | |
821 | } else if (opt_tbl[EBPF_OBJECT] && | |
822 | (matches(*argv, "object-file") == 0 || | |
823 | strcmp(*argv, "obj") == 0)) { | |
824 | *mode = EBPF_OBJECT; | |
825 | } else if (opt_tbl[EBPF_PINNED] && | |
826 | (matches(*argv, "object-pinned") == 0 || | |
827 | matches(*argv, "pinned") == 0 || | |
828 | matches(*argv, "fd") == 0)) { | |
829 | *mode = EBPF_PINNED; | |
32e93fb7 DB |
830 | } else { |
831 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
832 | return -1; | |
833 | } | |
834 | ||
835 | NEXT_ARG(); | |
836 | file = section = uds_name = NULL; | |
91d88eeb | 837 | if (*mode == EBPF_OBJECT || *mode == EBPF_PINNED) { |
32e93fb7 DB |
838 | file = *argv; |
839 | NEXT_ARG_FWD(); | |
840 | ||
91d88eeb DB |
841 | if (*type == BPF_PROG_TYPE_UNSPEC) { |
842 | if (argc > 0 && matches(*argv, "type") == 0) { | |
843 | NEXT_ARG(); | |
e4225669 DB |
844 | for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta); |
845 | i++) { | |
846 | if (!__bpf_prog_meta[i].type) | |
847 | continue; | |
848 | if (!matches(*argv, | |
849 | __bpf_prog_meta[i].type)) { | |
850 | *type = i; | |
851 | break; | |
852 | } | |
853 | } | |
854 | ||
855 | if (*type == BPF_PROG_TYPE_UNSPEC) { | |
91d88eeb DB |
856 | fprintf(stderr, "What type is \"%s\"?\n", |
857 | *argv); | |
858 | return -1; | |
859 | } | |
860 | NEXT_ARG_FWD(); | |
861 | } else { | |
862 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
863 | } | |
864 | } | |
865 | ||
e4225669 | 866 | section = bpf_prog_to_default_section(*type); |
32e93fb7 DB |
867 | if (argc > 0 && matches(*argv, "section") == 0) { |
868 | NEXT_ARG(); | |
869 | section = *argv; | |
870 | NEXT_ARG_FWD(); | |
871 | } | |
872 | ||
e4225669 DB |
873 | if (__bpf_prog_meta[*type].may_uds_export) { |
874 | uds_name = getenv(BPF_ENV_UDS); | |
875 | if (argc > 0 && !uds_name && | |
876 | matches(*argv, "export") == 0) { | |
877 | NEXT_ARG(); | |
878 | uds_name = *argv; | |
879 | NEXT_ARG_FWD(); | |
880 | } | |
32e93fb7 DB |
881 | } |
882 | ||
883 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
884 | verbose = true; | |
885 | NEXT_ARG_FWD(); | |
886 | } | |
887 | ||
888 | PREV_ARG(); | |
889 | } | |
890 | ||
91d88eeb | 891 | if (*mode == CBPF_BYTECODE || *mode == CBPF_FILE) |
e4225669 | 892 | ret = bpf_ops_parse(argc, argv, cfg->ops, *mode == CBPF_FILE); |
91d88eeb DB |
893 | else if (*mode == EBPF_OBJECT) |
894 | ret = bpf_obj_open(file, *type, section, verbose); | |
895 | else if (*mode == EBPF_PINNED) | |
21856018 | 896 | ret = bpf_obj_pinned(file, *type); |
91d88eeb | 897 | else |
32e93fb7 DB |
898 | return -1; |
899 | ||
e4225669 DB |
900 | cfg->object = file; |
901 | cfg->section = section; | |
902 | cfg->uds = uds_name; | |
903 | cfg->argc = argc; | |
904 | cfg->argv = argv; | |
91d88eeb DB |
905 | |
906 | return ret; | |
907 | } | |
908 | ||
e4225669 DB |
909 | static int bpf_parse_opt_tbl(enum bpf_prog_type type, struct bpf_cfg_in *cfg, |
910 | const struct bpf_cfg_ops *ops, void *nl, | |
911 | const bool *opt_tbl) | |
91d88eeb DB |
912 | { |
913 | struct sock_filter opcodes[BPF_MAXINSNS]; | |
91d88eeb | 914 | char annotation[256]; |
91d88eeb DB |
915 | enum bpf_mode mode; |
916 | int ret; | |
917 | ||
e4225669 DB |
918 | cfg->ops = opcodes; |
919 | ret = bpf_parse(&type, &mode, cfg, opt_tbl); | |
920 | cfg->ops = NULL; | |
91d88eeb DB |
921 | if (ret < 0) |
922 | return ret; | |
923 | ||
e4225669 DB |
924 | if (mode == CBPF_BYTECODE || mode == CBPF_FILE) |
925 | ops->cbpf_cb(nl, opcodes, ret); | |
91d88eeb | 926 | if (mode == EBPF_OBJECT || mode == EBPF_PINNED) { |
32e93fb7 | 927 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
e4225669 DB |
928 | basename(cfg->object), mode == EBPF_PINNED ? |
929 | "*fsobj" : cfg->section); | |
930 | ops->ebpf_cb(nl, ret, annotation); | |
32e93fb7 DB |
931 | } |
932 | ||
91d88eeb DB |
933 | return 0; |
934 | } | |
32e93fb7 | 935 | |
e4225669 DB |
936 | int bpf_parse_common(enum bpf_prog_type type, struct bpf_cfg_in *cfg, |
937 | const struct bpf_cfg_ops *ops, void *nl) | |
938 | { | |
939 | bool opt_tbl[BPF_MODE_MAX] = {}; | |
940 | ||
941 | if (ops->cbpf_cb) { | |
942 | opt_tbl[CBPF_BYTECODE] = true; | |
943 | opt_tbl[CBPF_FILE] = true; | |
944 | } | |
945 | ||
946 | if (ops->ebpf_cb) { | |
947 | opt_tbl[EBPF_OBJECT] = true; | |
948 | opt_tbl[EBPF_PINNED] = true; | |
949 | } | |
950 | ||
951 | return bpf_parse_opt_tbl(type, cfg, ops, nl, opt_tbl); | |
952 | } | |
953 | ||
91d88eeb DB |
954 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
955 | { | |
956 | enum bpf_prog_type type = BPF_PROG_TYPE_UNSPEC; | |
957 | const bool opt_tbl[BPF_MODE_MAX] = { | |
91d88eeb DB |
958 | [EBPF_OBJECT] = true, |
959 | [EBPF_PINNED] = true, | |
960 | }; | |
961 | const struct bpf_elf_map test = { | |
962 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
963 | .size_key = sizeof(int), | |
964 | .size_value = sizeof(int), | |
965 | }; | |
e4225669 DB |
966 | struct bpf_cfg_in cfg = { |
967 | .argc = argc, | |
968 | .argv = argv, | |
969 | }; | |
ecb05c0f | 970 | struct bpf_map_ext ext = {}; |
91d88eeb | 971 | int ret, prog_fd, map_fd; |
91d88eeb DB |
972 | enum bpf_mode mode; |
973 | uint32_t map_key; | |
974 | ||
e4225669 | 975 | prog_fd = bpf_parse(&type, &mode, &cfg, opt_tbl); |
91d88eeb DB |
976 | if (prog_fd < 0) |
977 | return prog_fd; | |
978 | if (key) { | |
979 | map_key = *key; | |
980 | } else { | |
e4225669 | 981 | ret = sscanf(cfg.section, "%*i/%i", &map_key); |
91d88eeb | 982 | if (ret != 1) { |
32a121cb | 983 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
984 | ret = -EINVAL; |
985 | goto out_prog; | |
986 | } | |
987 | } | |
32e93fb7 | 988 | |
e4225669 | 989 | map_fd = bpf_obj_get(map_path, type); |
91d88eeb DB |
990 | if (map_fd < 0) { |
991 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
992 | map_path, strerror(errno)); | |
993 | ret = map_fd; | |
994 | goto out_prog; | |
995 | } | |
996 | ||
ecb05c0f | 997 | ret = bpf_map_selfcheck_pinned(map_fd, &test, &ext, |
fb24802b DB |
998 | offsetof(struct bpf_elf_map, max_elem), |
999 | type); | |
91d88eeb DB |
1000 | if (ret < 0) { |
1001 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
1002 | goto out_map; | |
1003 | } | |
1004 | ||
1005 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
1006 | if (ret < 0) | |
1007 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
1008 | out_map: | |
1009 | close(map_fd); | |
1010 | out_prog: | |
1011 | close(prog_fd); | |
1012 | return ret; | |
32e93fb7 DB |
1013 | } |
1014 | ||
fc4ccce0 DA |
1015 | int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type) |
1016 | { | |
1017 | union bpf_attr attr = {}; | |
1018 | ||
1019 | attr.target_fd = target_fd; | |
1020 | attr.attach_bpf_fd = prog_fd; | |
1021 | attr.attach_type = type; | |
1022 | ||
1023 | return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
1024 | } | |
1025 | ||
1026 | int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type) | |
1027 | { | |
1028 | union bpf_attr attr = {}; | |
1029 | ||
1030 | attr.target_fd = target_fd; | |
1031 | attr.attach_type = type; | |
1032 | ||
1033 | return bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); | |
1034 | } | |
1035 | ||
869d889e DA |
1036 | int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, |
1037 | size_t size_insns, const char *license, char *log, | |
1038 | size_t size_log) | |
1039 | { | |
1040 | union bpf_attr attr = {}; | |
1041 | ||
1042 | attr.prog_type = type; | |
1043 | attr.insns = bpf_ptr_to_u64(insns); | |
1044 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); | |
1045 | attr.license = bpf_ptr_to_u64(license); | |
1046 | ||
1047 | if (size_log > 0) { | |
1048 | attr.log_buf = bpf_ptr_to_u64(log); | |
1049 | attr.log_size = size_log; | |
1050 | attr.log_level = 1; | |
1051 | } | |
1052 | ||
1053 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); | |
1054 | } | |
1055 | ||
6256f8c9 | 1056 | #ifdef HAVE_ELF |
32e93fb7 DB |
1057 | struct bpf_elf_prog { |
1058 | enum bpf_prog_type type; | |
1059 | const struct bpf_insn *insns; | |
1060 | size_t size; | |
1061 | const char *license; | |
1062 | }; | |
1063 | ||
f6793eec DB |
1064 | struct bpf_hash_entry { |
1065 | unsigned int pinning; | |
1066 | const char *subpath; | |
1067 | struct bpf_hash_entry *next; | |
1068 | }; | |
1069 | ||
ecb05c0f DB |
1070 | struct bpf_config { |
1071 | unsigned int jit_enabled; | |
1072 | }; | |
1073 | ||
32e93fb7 | 1074 | struct bpf_elf_ctx { |
ecb05c0f | 1075 | struct bpf_config cfg; |
32e93fb7 DB |
1076 | Elf *elf_fd; |
1077 | GElf_Ehdr elf_hdr; | |
1078 | Elf_Data *sym_tab; | |
1079 | Elf_Data *str_tab; | |
1080 | int obj_fd; | |
1081 | int map_fds[ELF_MAX_MAPS]; | |
1082 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
ecb05c0f | 1083 | struct bpf_map_ext maps_ext[ELF_MAX_MAPS]; |
32e93fb7 DB |
1084 | int sym_num; |
1085 | int map_num; | |
e4225669 | 1086 | int map_len; |
32e93fb7 DB |
1087 | bool *sec_done; |
1088 | int sec_maps; | |
1089 | char license[ELF_MAX_LICENSE_LEN]; | |
1090 | enum bpf_prog_type type; | |
1091 | bool verbose; | |
1092 | struct bpf_elf_st stat; | |
f6793eec | 1093 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
1094 | char *log; |
1095 | size_t log_size; | |
32e93fb7 DB |
1096 | }; |
1097 | ||
6256f8c9 | 1098 | struct bpf_elf_sec_data { |
32e93fb7 DB |
1099 | GElf_Shdr sec_hdr; |
1100 | Elf_Data *sec_data; | |
1101 | const char *sec_name; | |
6256f8c9 DB |
1102 | }; |
1103 | ||
1104 | struct bpf_map_data { | |
32e93fb7 DB |
1105 | int *fds; |
1106 | const char *obj; | |
1107 | struct bpf_elf_st *st; | |
1108 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
1109 | }; |
1110 | ||
f31645d1 DB |
1111 | static __check_format_string(2, 3) void |
1112 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
1113 | { |
1114 | va_list vl; | |
1115 | ||
1116 | va_start(vl, format); | |
1117 | vfprintf(stderr, format, vl); | |
1118 | va_end(vl); | |
1119 | ||
f31645d1 | 1120 | if (ctx->log && ctx->log[0]) { |
afc1a200 DB |
1121 | if (ctx->verbose) { |
1122 | fprintf(stderr, "%s\n", ctx->log); | |
1123 | } else { | |
1124 | unsigned int off = 0, len = strlen(ctx->log); | |
1125 | ||
1126 | if (len > BPF_MAX_LOG) { | |
1127 | off = len - BPF_MAX_LOG; | |
1128 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
1129 | off); | |
1130 | } | |
1131 | fprintf(stderr, "%s\n", ctx->log + off); | |
1132 | } | |
1133 | ||
f31645d1 DB |
1134 | memset(ctx->log, 0, ctx->log_size); |
1135 | } | |
1136 | } | |
1137 | ||
1138 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
1139 | { | |
0f74d0f3 | 1140 | const size_t log_max = UINT_MAX >> 8; |
f31645d1 DB |
1141 | size_t log_size = ctx->log_size; |
1142 | void *ptr; | |
1143 | ||
1144 | if (!ctx->log) { | |
1145 | log_size = 65536; | |
0f74d0f3 | 1146 | } else if (log_size < log_max) { |
f31645d1 | 1147 | log_size <<= 1; |
0f74d0f3 TG |
1148 | if (log_size > log_max) |
1149 | log_size = log_max; | |
1150 | } else { | |
1151 | return -EINVAL; | |
d937a74b | 1152 | } |
f31645d1 DB |
1153 | |
1154 | ptr = realloc(ctx->log, log_size); | |
1155 | if (!ptr) | |
1156 | return -ENOMEM; | |
1157 | ||
1158 | ctx->log = ptr; | |
1159 | ctx->log_size = log_size; | |
1160 | ||
1161 | return 0; | |
11c39b5e DB |
1162 | } |
1163 | ||
4dd3f50a DB |
1164 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
1165 | uint32_t size_value, uint32_t max_elem, | |
612ff099 | 1166 | uint32_t flags, int inner_fd) |
11c39b5e | 1167 | { |
d17b136f | 1168 | union bpf_attr attr = {}; |
67584e3a | 1169 | |
67584e3a ND |
1170 | attr.map_type = type; |
1171 | attr.key_size = size_key; | |
612ff099 | 1172 | attr.value_size = inner_fd ? sizeof(int) : size_value; |
67584e3a | 1173 | attr.max_entries = max_elem; |
4dd3f50a | 1174 | attr.map_flags = flags; |
612ff099 | 1175 | attr.inner_map_fd = inner_fd; |
11c39b5e DB |
1176 | |
1177 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
1178 | } | |
11c39b5e | 1179 | |
32e93fb7 | 1180 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 1181 | { |
d17b136f | 1182 | union bpf_attr attr = {}; |
67584e3a | 1183 | |
67584e3a ND |
1184 | attr.pathname = bpf_ptr_to_u64(pathname); |
1185 | attr.bpf_fd = fd; | |
32e93fb7 DB |
1186 | |
1187 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
1188 | } | |
11c39b5e | 1189 | |
32e93fb7 DB |
1190 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
1191 | { | |
1192 | struct sockaddr_alg alg = { | |
1193 | .salg_family = AF_ALG, | |
1194 | .salg_type = "hash", | |
1195 | .salg_name = "sha1", | |
1196 | }; | |
1197 | int ret, cfd, ofd, ffd; | |
1198 | struct stat stbuff; | |
1199 | ssize_t size; | |
1200 | ||
1201 | if (!object || len != 20) | |
1202 | return -EINVAL; | |
1203 | ||
1204 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
1205 | if (cfd < 0) { | |
1206 | fprintf(stderr, "Cannot get AF_ALG socket: %s\n", | |
1207 | strerror(errno)); | |
1208 | return cfd; | |
1209 | } | |
1210 | ||
1211 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
1212 | if (ret < 0) { | |
1213 | fprintf(stderr, "Error binding socket: %s\n", strerror(errno)); | |
1214 | goto out_cfd; | |
1215 | } | |
1216 | ||
1217 | ofd = accept(cfd, NULL, 0); | |
1218 | if (ofd < 0) { | |
1219 | fprintf(stderr, "Error accepting socket: %s\n", | |
1220 | strerror(errno)); | |
1221 | ret = ofd; | |
1222 | goto out_cfd; | |
1223 | } | |
1224 | ||
1225 | ffd = open(object, O_RDONLY); | |
1226 | if (ffd < 0) { | |
1227 | fprintf(stderr, "Error opening object %s: %s\n", | |
1228 | object, strerror(errno)); | |
1229 | ret = ffd; | |
1230 | goto out_ofd; | |
1231 | } | |
1232 | ||
32a121cb | 1233 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
1234 | if (ret < 0) { |
1235 | fprintf(stderr, "Error doing fstat: %s\n", | |
1236 | strerror(errno)); | |
1237 | goto out_ffd; | |
d937a74b | 1238 | } |
11c39b5e | 1239 | |
32e93fb7 DB |
1240 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
1241 | if (size != stbuff.st_size) { | |
1242 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
1243 | size, stbuff.st_size, strerror(errno)); | |
1244 | ret = -1; | |
1245 | goto out_ffd; | |
1246 | } | |
1247 | ||
1248 | size = read(ofd, out, len); | |
1249 | if (size != len) { | |
1250 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
1251 | size, len, strerror(errno)); | |
1252 | ret = -1; | |
1253 | } else { | |
1254 | ret = 0; | |
1255 | } | |
1256 | out_ffd: | |
1257 | close(ffd); | |
1258 | out_ofd: | |
1259 | close(ofd); | |
1260 | out_cfd: | |
1261 | close(cfd); | |
1262 | return ret; | |
11c39b5e DB |
1263 | } |
1264 | ||
32e93fb7 | 1265 | static const char *bpf_get_obj_uid(const char *pathname) |
11c39b5e | 1266 | { |
32a121cb | 1267 | static bool bpf_uid_cached; |
32e93fb7 DB |
1268 | static char bpf_uid[64]; |
1269 | uint8_t tmp[20]; | |
1270 | int ret; | |
11c39b5e | 1271 | |
32e93fb7 DB |
1272 | if (bpf_uid_cached) |
1273 | goto done; | |
11c39b5e | 1274 | |
32e93fb7 DB |
1275 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); |
1276 | if (ret) { | |
1277 | fprintf(stderr, "Object hashing failed!\n"); | |
1278 | return NULL; | |
1279 | } | |
1280 | ||
1281 | hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid)); | |
1282 | bpf_uid_cached = true; | |
1283 | done: | |
1284 | return bpf_uid; | |
11c39b5e DB |
1285 | } |
1286 | ||
32e93fb7 DB |
1287 | static int bpf_init_env(const char *pathname) |
1288 | { | |
1289 | struct rlimit limit = { | |
1290 | .rlim_cur = RLIM_INFINITY, | |
1291 | .rlim_max = RLIM_INFINITY, | |
1292 | }; | |
1293 | ||
1294 | /* Don't bother in case we fail! */ | |
1295 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
1296 | ||
e4225669 | 1297 | if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) { |
32a121cb | 1298 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
32e93fb7 DB |
1299 | return 0; |
1300 | } | |
1301 | ||
1302 | if (!bpf_get_obj_uid(pathname)) | |
1303 | return -1; | |
1304 | ||
1305 | return 0; | |
6256f8c9 DB |
1306 | } |
1307 | ||
f6793eec DB |
1308 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
1309 | uint32_t pinning) | |
1310 | { | |
1311 | struct bpf_hash_entry *entry; | |
1312 | ||
1313 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1314 | while (entry && entry->pinning != pinning) | |
1315 | entry = entry->next; | |
1316 | ||
1317 | return entry ? entry->subpath : NULL; | |
1318 | } | |
1319 | ||
1320 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
1321 | uint32_t pinning) | |
11c39b5e | 1322 | { |
32e93fb7 DB |
1323 | switch (pinning) { |
1324 | case PIN_OBJECT_NS: | |
1325 | case PIN_GLOBAL_NS: | |
1326 | return false; | |
1327 | case PIN_NONE: | |
32e93fb7 | 1328 | return true; |
f6793eec DB |
1329 | default: |
1330 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
1331 | } |
1332 | } | |
1333 | ||
1334 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 1335 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
1336 | { |
1337 | switch (pinning) { | |
1338 | case PIN_OBJECT_NS: | |
e4225669 DB |
1339 | snprintf(pathname, len, "%s/%s/%s", |
1340 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1341 | bpf_get_obj_uid(NULL), name); |
1342 | break; | |
1343 | case PIN_GLOBAL_NS: | |
e4225669 DB |
1344 | snprintf(pathname, len, "%s/%s/%s", |
1345 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1346 | BPF_DIR_GLOBALS, name); |
1347 | break; | |
f6793eec | 1348 | default: |
e4225669 DB |
1349 | snprintf(pathname, len, "%s/../%s/%s", |
1350 | bpf_get_work_dir(ctx->type), | |
f6793eec DB |
1351 | bpf_custom_pinning(ctx, pinning), name); |
1352 | break; | |
32e93fb7 DB |
1353 | } |
1354 | } | |
1355 | ||
f6793eec DB |
1356 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1357 | uint32_t pinning) | |
32e93fb7 DB |
1358 | { |
1359 | char pathname[PATH_MAX]; | |
1360 | ||
e4225669 | 1361 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
32e93fb7 DB |
1362 | return 0; |
1363 | ||
f6793eec | 1364 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
e4225669 | 1365 | return bpf_obj_get(pathname, ctx->type); |
32e93fb7 DB |
1366 | } |
1367 | ||
e4225669 | 1368 | static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx) |
32e93fb7 | 1369 | { |
f6793eec | 1370 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1371 | int ret; |
1372 | ||
e4225669 | 1373 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type), |
f6793eec DB |
1374 | bpf_get_obj_uid(NULL)); |
1375 | ||
1376 | ret = mkdir(tmp, S_IRWXU); | |
1377 | if (ret && errno != EEXIST) { | |
1378 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1379 | return ret; | |
1380 | } | |
1381 | ||
1382 | return 0; | |
1383 | } | |
1384 | ||
e4225669 DB |
1385 | static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx, |
1386 | const char *todo) | |
f6793eec DB |
1387 | { |
1388 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1389 | int ret; | |
1390 | ||
e4225669 | 1391 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type)); |
f6793eec DB |
1392 | snprintf(rem, sizeof(rem), "%s/", todo); |
1393 | sub = strtok(rem, "/"); | |
32e93fb7 | 1394 | |
f6793eec DB |
1395 | while (sub) { |
1396 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1397 | return -EINVAL; | |
1398 | ||
1399 | strcat(tmp, sub); | |
1400 | strcat(tmp, "/"); | |
32e93fb7 | 1401 | |
f6793eec | 1402 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1403 | if (ret && errno != EEXIST) { |
f6793eec | 1404 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1405 | strerror(errno)); |
1406 | return ret; | |
1407 | } | |
f6793eec DB |
1408 | |
1409 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1410 | } |
1411 | ||
f6793eec DB |
1412 | return 0; |
1413 | } | |
1414 | ||
1415 | static int bpf_place_pinned(int fd, const char *name, | |
1416 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1417 | { | |
1418 | char pathname[PATH_MAX]; | |
1419 | const char *tmp; | |
1420 | int ret = 0; | |
1421 | ||
e4225669 | 1422 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
f6793eec DB |
1423 | return 0; |
1424 | ||
1425 | if (pinning == PIN_OBJECT_NS) | |
e4225669 | 1426 | ret = bpf_make_obj_path(ctx); |
f6793eec | 1427 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) |
e4225669 | 1428 | ret = bpf_make_custom_path(ctx, tmp); |
f6793eec DB |
1429 | if (ret < 0) |
1430 | return ret; | |
1431 | ||
1432 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1433 | return bpf_obj_pin(fd, pathname); |
1434 | } | |
1435 | ||
f31645d1 DB |
1436 | static void bpf_prog_report(int fd, const char *section, |
1437 | const struct bpf_elf_prog *prog, | |
1438 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1439 | { |
afc1a200 DB |
1440 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1441 | ||
1442 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1443 | fd < 0 ? "rejected: " : "loaded", |
1444 | fd < 0 ? strerror(errno) : "", | |
1445 | fd < 0 ? errno : fd); | |
1446 | ||
1447 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1448 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1449 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1450 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1451 | ||
1452 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1453 | } | |
32e93fb7 | 1454 | |
f31645d1 DB |
1455 | static int bpf_prog_attach(const char *section, |
1456 | const struct bpf_elf_prog *prog, | |
1457 | struct bpf_elf_ctx *ctx) | |
1458 | { | |
1459 | int tries = 0, fd; | |
1460 | retry: | |
32e93fb7 DB |
1461 | errno = 0; |
1462 | fd = bpf_prog_load(prog->type, prog->insns, prog->size, | |
f31645d1 DB |
1463 | prog->license, ctx->log, ctx->log_size); |
1464 | if (fd < 0 || ctx->verbose) { | |
1465 | /* The verifier log is pretty chatty, sometimes so chatty | |
1466 | * on larger programs, that we could fail to dump everything | |
1467 | * into our buffer. Still, try to give a debuggable error | |
1468 | * log for the user, so enlarge it and re-fail. | |
1469 | */ | |
1470 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
0f74d0f3 | 1471 | if (tries++ < 10 && !bpf_log_realloc(ctx)) |
f31645d1 DB |
1472 | goto retry; |
1473 | ||
32a121cb | 1474 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1475 | ctx->log_size, tries); |
1476 | return fd; | |
1477 | } | |
1478 | ||
1479 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1480 | } |
1481 | ||
1482 | return fd; | |
1483 | } | |
1484 | ||
f31645d1 DB |
1485 | static void bpf_map_report(int fd, const char *name, |
1486 | const struct bpf_elf_map *map, | |
612ff099 | 1487 | struct bpf_elf_ctx *ctx, int inner_fd) |
f31645d1 DB |
1488 | { |
1489 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1490 | fd < 0 ? "rejected: " : "loaded", | |
1491 | fd < 0 ? strerror(errno) : "", | |
1492 | fd < 0 ? errno : fd); | |
1493 | ||
1494 | fprintf(stderr, " - Type: %u\n", map->type); | |
1495 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1496 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1497 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
612ff099 DB |
1498 | fprintf(stderr, " - Size value: %u\n", |
1499 | inner_fd ? (int)sizeof(int) : map->size_value); | |
4dd3f50a DB |
1500 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1501 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1502 | } |
1503 | ||
612ff099 DB |
1504 | static int bpf_find_map_id(const struct bpf_elf_ctx *ctx, uint32_t id) |
1505 | { | |
1506 | int i; | |
1507 | ||
1508 | for (i = 0; i < ctx->map_num; i++) { | |
1509 | if (ctx->maps[i].id != id) | |
1510 | continue; | |
1511 | if (ctx->map_fds[i] < 0) | |
1512 | return -EINVAL; | |
1513 | ||
1514 | return ctx->map_fds[i]; | |
1515 | } | |
1516 | ||
1517 | return -ENOENT; | |
1518 | } | |
1519 | ||
0b5eadc5 | 1520 | static void bpf_report_map_in_map(int outer_fd, uint32_t idx) |
612ff099 DB |
1521 | { |
1522 | struct bpf_elf_map outer_map; | |
1523 | int ret; | |
1524 | ||
1525 | fprintf(stderr, "Cannot insert map into map! "); | |
1526 | ||
ecb05c0f | 1527 | ret = bpf_derive_elf_map_from_fdinfo(outer_fd, &outer_map, NULL); |
612ff099 DB |
1528 | if (!ret) { |
1529 | if (idx >= outer_map.max_elem && | |
1530 | outer_map.type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
1531 | fprintf(stderr, "Outer map has %u elements, index %u is invalid!\n", | |
1532 | outer_map.max_elem, idx); | |
1533 | return; | |
1534 | } | |
1535 | } | |
1536 | ||
1537 | fprintf(stderr, "Different map specs used for outer and inner map?\n"); | |
1538 | } | |
1539 | ||
1540 | static bool bpf_is_map_in_map_type(const struct bpf_elf_map *map) | |
1541 | { | |
1542 | return map->type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1543 | map->type == BPF_MAP_TYPE_HASH_OF_MAPS; | |
1544 | } | |
1545 | ||
ecb05c0f DB |
1546 | static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx, |
1547 | const struct bpf_elf_map *map, struct bpf_map_ext *ext, | |
1548 | int *have_map_in_map) | |
32e93fb7 | 1549 | { |
612ff099 | 1550 | int fd, ret, map_inner_fd = 0; |
32e93fb7 | 1551 | |
f6793eec | 1552 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1553 | if (fd > 0) { |
ecb05c0f | 1554 | ret = bpf_map_selfcheck_pinned(fd, map, ext, |
91d88eeb | 1555 | offsetof(struct bpf_elf_map, |
fb24802b | 1556 | id), ctx->type); |
9e607f2e DB |
1557 | if (ret < 0) { |
1558 | close(fd); | |
1559 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1560 | name); | |
1561 | return ret; | |
1562 | } | |
f31645d1 | 1563 | if (ctx->verbose) |
32e93fb7 DB |
1564 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1565 | name); | |
1566 | return fd; | |
1567 | } | |
1568 | ||
612ff099 DB |
1569 | if (have_map_in_map && bpf_is_map_in_map_type(map)) { |
1570 | (*have_map_in_map)++; | |
1571 | if (map->inner_id) | |
1572 | return 0; | |
1573 | fprintf(stderr, "Map \'%s\' cannot be created since no inner map ID defined!\n", | |
1574 | name); | |
1575 | return -EINVAL; | |
1576 | } | |
1577 | ||
1578 | if (!have_map_in_map && bpf_is_map_in_map_type(map)) { | |
1579 | map_inner_fd = bpf_find_map_id(ctx, map->inner_id); | |
1580 | if (map_inner_fd < 0) { | |
1581 | fprintf(stderr, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n", | |
1582 | name, map->inner_id); | |
1583 | return -EINVAL; | |
1584 | } | |
1585 | } | |
1586 | ||
32e93fb7 DB |
1587 | errno = 0; |
1588 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
612ff099 | 1589 | map->max_elem, map->flags, map_inner_fd); |
f31645d1 | 1590 | if (fd < 0 || ctx->verbose) { |
612ff099 | 1591 | bpf_map_report(fd, name, map, ctx, map_inner_fd); |
32e93fb7 DB |
1592 | if (fd < 0) |
1593 | return fd; | |
1594 | } | |
1595 | ||
f6793eec | 1596 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1597 | if (ret < 0 && errno != EEXIST) { |
1598 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1599 | strerror(errno)); | |
1600 | close(fd); | |
1601 | return ret; | |
1602 | } | |
1603 | ||
1604 | return fd; | |
1605 | } | |
1606 | ||
32e93fb7 DB |
1607 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1608 | const GElf_Sym *sym) | |
1609 | { | |
1610 | return ctx->str_tab->d_buf + sym->st_name; | |
1611 | } | |
1612 | ||
1613 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) | |
1614 | { | |
1615 | GElf_Sym sym; | |
11c39b5e DB |
1616 | int i; |
1617 | ||
32e93fb7 DB |
1618 | for (i = 0; i < ctx->sym_num; i++) { |
1619 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1620 | continue; | |
1621 | ||
5230a2ed DB |
1622 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
1623 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
32e93fb7 | 1624 | sym.st_shndx != ctx->sec_maps || |
e4225669 | 1625 | sym.st_value / ctx->map_len != which) |
32e93fb7 DB |
1626 | continue; |
1627 | ||
1628 | return bpf_str_tab_name(ctx, &sym); | |
11c39b5e | 1629 | } |
32e93fb7 DB |
1630 | |
1631 | return NULL; | |
11c39b5e DB |
1632 | } |
1633 | ||
32e93fb7 | 1634 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1635 | { |
612ff099 | 1636 | int i, j, ret, fd, inner_fd, inner_idx, have_map_in_map = 0; |
32e93fb7 | 1637 | const char *map_name; |
11c39b5e | 1638 | |
32e93fb7 DB |
1639 | for (i = 0; i < ctx->map_num; i++) { |
1640 | map_name = bpf_map_fetch_name(ctx, i); | |
1641 | if (!map_name) | |
1642 | return -EIO; | |
11c39b5e | 1643 | |
ecb05c0f DB |
1644 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1645 | &ctx->maps_ext[i], &have_map_in_map); | |
612ff099 DB |
1646 | if (fd < 0) |
1647 | return fd; | |
1648 | ||
1649 | ctx->map_fds[i] = !fd ? -1 : fd; | |
1650 | } | |
1651 | ||
1652 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { | |
1653 | if (ctx->map_fds[i] >= 0) | |
1654 | continue; | |
1655 | ||
1656 | map_name = bpf_map_fetch_name(ctx, i); | |
1657 | if (!map_name) | |
1658 | return -EIO; | |
1659 | ||
ecb05c0f DB |
1660 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1661 | &ctx->maps_ext[i], NULL); | |
32e93fb7 DB |
1662 | if (fd < 0) |
1663 | return fd; | |
11c39b5e | 1664 | |
32e93fb7 | 1665 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1666 | } |
1667 | ||
612ff099 DB |
1668 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { |
1669 | if (!ctx->maps[i].id || | |
1670 | ctx->maps[i].inner_id || | |
1671 | ctx->maps[i].inner_idx == -1) | |
1672 | continue; | |
1673 | ||
1674 | inner_fd = ctx->map_fds[i]; | |
1675 | inner_idx = ctx->maps[i].inner_idx; | |
1676 | ||
1677 | for (j = 0; j < ctx->map_num; j++) { | |
1678 | if (!bpf_is_map_in_map_type(&ctx->maps[j])) | |
1679 | continue; | |
1680 | if (ctx->maps[j].inner_id != ctx->maps[i].id) | |
1681 | continue; | |
1682 | ||
1683 | ret = bpf_map_update(ctx->map_fds[j], &inner_idx, | |
1684 | &inner_fd, BPF_ANY); | |
1685 | if (ret < 0) { | |
1686 | bpf_report_map_in_map(ctx->map_fds[j], | |
0b5eadc5 | 1687 | inner_idx); |
612ff099 DB |
1688 | return ret; |
1689 | } | |
1690 | } | |
1691 | } | |
1692 | ||
11c39b5e | 1693 | return 0; |
11c39b5e DB |
1694 | } |
1695 | ||
e4225669 DB |
1696 | static int bpf_map_num_sym(struct bpf_elf_ctx *ctx) |
1697 | { | |
1698 | int i, num = 0; | |
1699 | GElf_Sym sym; | |
1700 | ||
1701 | for (i = 0; i < ctx->sym_num; i++) { | |
1702 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1703 | continue; | |
1704 | ||
1705 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1706 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1707 | sym.st_shndx != ctx->sec_maps) | |
1708 | continue; | |
1709 | num++; | |
1710 | } | |
1711 | ||
1712 | return num; | |
1713 | } | |
1714 | ||
32e93fb7 DB |
1715 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1716 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1717 | { |
32e93fb7 | 1718 | Elf_Data *sec_edata; |
11c39b5e DB |
1719 | GElf_Shdr sec_hdr; |
1720 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1721 | char *sec_name; |
1722 | ||
32e93fb7 | 1723 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1724 | |
32e93fb7 | 1725 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1726 | if (!sec_fd) |
1727 | return -EINVAL; | |
11c39b5e DB |
1728 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1729 | return -EIO; | |
1730 | ||
32e93fb7 | 1731 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1732 | sec_hdr.sh_name); |
1733 | if (!sec_name || !sec_hdr.sh_size) | |
1734 | return -ENOENT; | |
1735 | ||
1736 | sec_edata = elf_getdata(sec_fd, NULL); | |
1737 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1738 | return -EIO; | |
1739 | ||
32e93fb7 | 1740 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1741 | |
32e93fb7 DB |
1742 | data->sec_name = sec_name; |
1743 | data->sec_data = sec_edata; | |
11c39b5e DB |
1744 | return 0; |
1745 | } | |
1746 | ||
e4225669 DB |
1747 | struct bpf_elf_map_min { |
1748 | __u32 type; | |
1749 | __u32 size_key; | |
1750 | __u32 size_value; | |
1751 | __u32 max_elem; | |
1752 | }; | |
11c39b5e | 1753 | |
e4225669 DB |
1754 | static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section, |
1755 | struct bpf_elf_sec_data *data) | |
1756 | { | |
1757 | ctx->map_num = data->sec_data->d_size; | |
32e93fb7 DB |
1758 | ctx->sec_maps = section; |
1759 | ctx->sec_done[section] = true; | |
11c39b5e | 1760 | |
e4225669 | 1761 | if (ctx->map_num > sizeof(ctx->maps)) { |
32e93fb7 DB |
1762 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); |
1763 | return -ENOMEM; | |
1764 | } | |
11c39b5e | 1765 | |
e4225669 DB |
1766 | memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num); |
1767 | return 0; | |
1768 | } | |
1769 | ||
1770 | static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end) | |
1771 | { | |
1772 | GElf_Sym sym; | |
1773 | int off, i; | |
1774 | ||
1775 | for (off = 0; off < end; off += ctx->map_len) { | |
1776 | /* Order doesn't need to be linear here, hence we walk | |
1777 | * the table again. | |
1778 | */ | |
1779 | for (i = 0; i < ctx->sym_num; i++) { | |
1780 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1781 | continue; | |
1782 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1783 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1784 | sym.st_shndx != ctx->sec_maps) | |
1785 | continue; | |
1786 | if (sym.st_value == off) | |
1787 | break; | |
1788 | if (i == ctx->sym_num - 1) | |
1789 | return -1; | |
1790 | } | |
1791 | } | |
1792 | ||
1793 | return off == end ? 0 : -1; | |
1794 | } | |
1795 | ||
1796 | static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx) | |
1797 | { | |
1798 | struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {}; | |
1799 | int i, sym_num = bpf_map_num_sym(ctx); | |
1800 | __u8 *buff; | |
1801 | ||
1802 | if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) { | |
1803 | fprintf(stderr, "%u maps not supported in current map section!\n", | |
1804 | sym_num); | |
1805 | return -EINVAL; | |
1806 | } | |
1807 | ||
1808 | if (ctx->map_num % sym_num != 0 || | |
1809 | ctx->map_num % sizeof(__u32) != 0) { | |
1810 | fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n"); | |
1811 | return -EINVAL; | |
1812 | } | |
1813 | ||
1814 | ctx->map_len = ctx->map_num / sym_num; | |
1815 | if (bpf_map_verify_all_offs(ctx, ctx->map_num)) { | |
1816 | fprintf(stderr, "Different struct bpf_elf_map in use!\n"); | |
1817 | return -EINVAL; | |
1818 | } | |
1819 | ||
1820 | if (ctx->map_len == sizeof(struct bpf_elf_map)) { | |
1821 | ctx->map_num = sym_num; | |
1822 | return 0; | |
1823 | } else if (ctx->map_len > sizeof(struct bpf_elf_map)) { | |
1824 | fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n"); | |
1825 | return -EINVAL; | |
1826 | } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) { | |
1827 | fprintf(stderr, "struct bpf_elf_map too small, not supported!\n"); | |
1828 | return -EINVAL; | |
1829 | } | |
1830 | ||
1831 | ctx->map_num = sym_num; | |
1832 | for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num; | |
1833 | i++, buff += ctx->map_len) { | |
1834 | /* The fixup leaves the rest of the members as zero, which | |
1835 | * is fine currently, but option exist to set some other | |
1836 | * default value as well when needed in future. | |
1837 | */ | |
1838 | memcpy(&fixup[i], buff, ctx->map_len); | |
1839 | } | |
1840 | ||
1841 | memcpy(ctx->maps, fixup, sizeof(fixup)); | |
1842 | ||
1843 | printf("Note: %zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n", | |
1844 | sizeof(struct bpf_elf_map) - ctx->map_len); | |
32e93fb7 DB |
1845 | return 0; |
1846 | } | |
11c39b5e | 1847 | |
32e93fb7 DB |
1848 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
1849 | struct bpf_elf_sec_data *data) | |
1850 | { | |
1851 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
1852 | return -ENOMEM; | |
11c39b5e | 1853 | |
32e93fb7 DB |
1854 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
1855 | ctx->sec_done[section] = true; | |
1856 | return 0; | |
1857 | } | |
11c39b5e | 1858 | |
32e93fb7 DB |
1859 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
1860 | struct bpf_elf_sec_data *data) | |
1861 | { | |
1862 | ctx->sym_tab = data->sec_data; | |
1863 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
1864 | ctx->sec_done[section] = true; | |
11c39b5e DB |
1865 | return 0; |
1866 | } | |
1867 | ||
32e93fb7 DB |
1868 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
1869 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1870 | { |
32e93fb7 DB |
1871 | ctx->str_tab = data->sec_data; |
1872 | ctx->sec_done[section] = true; | |
1873 | return 0; | |
1874 | } | |
11c39b5e | 1875 | |
afc1a200 DB |
1876 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
1877 | { | |
1878 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
1879 | } | |
1880 | ||
32e93fb7 DB |
1881 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx) |
1882 | { | |
1883 | struct bpf_elf_sec_data data; | |
1884 | int i, ret = -1; | |
11c39b5e | 1885 | |
32e93fb7 DB |
1886 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1887 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
1888 | if (ret < 0) |
1889 | continue; | |
1890 | ||
cce3d466 DB |
1891 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1892 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
e4225669 | 1893 | ret = bpf_fetch_maps_begin(ctx, i, &data); |
cce3d466 DB |
1894 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1895 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 1896 | ret = bpf_fetch_license(ctx, i, &data); |
cce3d466 DB |
1897 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
1898 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
1899 | ret = bpf_fetch_symtab(ctx, i, &data); |
1900 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 1901 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 DB |
1902 | ret = bpf_fetch_strtab(ctx, i, &data); |
1903 | if (ret < 0) { | |
afc1a200 | 1904 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 1905 | i); |
e4225669 | 1906 | return ret; |
11c39b5e | 1907 | } |
32e93fb7 DB |
1908 | } |
1909 | ||
afc1a200 | 1910 | if (bpf_has_map_data(ctx)) { |
e4225669 DB |
1911 | ret = bpf_fetch_maps_end(ctx); |
1912 | if (ret < 0) { | |
1913 | fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n"); | |
1914 | return ret; | |
1915 | } | |
1916 | ||
32e93fb7 DB |
1917 | ret = bpf_maps_attach_all(ctx); |
1918 | if (ret < 0) { | |
1919 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
1920 | return ret; | |
11c39b5e DB |
1921 | } |
1922 | } | |
1923 | ||
1924 | return ret; | |
1925 | } | |
1926 | ||
e4225669 DB |
1927 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section, |
1928 | bool *sseen) | |
11c39b5e | 1929 | { |
32e93fb7 DB |
1930 | struct bpf_elf_sec_data data; |
1931 | struct bpf_elf_prog prog; | |
1932 | int ret, i, fd = -1; | |
11c39b5e | 1933 | |
32e93fb7 DB |
1934 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1935 | if (ctx->sec_done[i]) | |
11c39b5e DB |
1936 | continue; |
1937 | ||
32e93fb7 | 1938 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
1939 | if (ret < 0 || |
1940 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
1941 | data.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1942 | !strcmp(data.sec_name, section))) | |
11c39b5e DB |
1943 | continue; |
1944 | ||
e4225669 DB |
1945 | *sseen = true; |
1946 | ||
32e93fb7 DB |
1947 | memset(&prog, 0, sizeof(prog)); |
1948 | prog.type = ctx->type; | |
1949 | prog.insns = data.sec_data->d_buf; | |
1950 | prog.size = data.sec_data->d_size; | |
1951 | prog.license = ctx->license; | |
11c39b5e | 1952 | |
f31645d1 | 1953 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1954 | if (fd < 0) |
e4225669 | 1955 | return fd; |
11c39b5e | 1956 | |
32e93fb7 | 1957 | ctx->sec_done[i] = true; |
11c39b5e DB |
1958 | break; |
1959 | } | |
1960 | ||
32e93fb7 | 1961 | return fd; |
11c39b5e DB |
1962 | } |
1963 | ||
ecb05c0f DB |
1964 | struct bpf_tail_call_props { |
1965 | unsigned int total; | |
1966 | unsigned int jited; | |
1967 | }; | |
1968 | ||
32e93fb7 DB |
1969 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
1970 | struct bpf_elf_sec_data *data_relo, | |
ecb05c0f DB |
1971 | struct bpf_elf_sec_data *data_insn, |
1972 | struct bpf_tail_call_props *props) | |
11c39b5e | 1973 | { |
32e93fb7 DB |
1974 | Elf_Data *idata = data_insn->sec_data; |
1975 | GElf_Shdr *rhdr = &data_relo->sec_hdr; | |
1976 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
1977 | struct bpf_insn *insns = idata->d_buf; | |
1978 | unsigned int num_insns = idata->d_size / sizeof(*insns); | |
11c39b5e | 1979 | |
32e93fb7 DB |
1980 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
1981 | unsigned int ioff, rmap; | |
1982 | GElf_Rel relo; | |
1983 | GElf_Sym sym; | |
1984 | ||
1985 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
1986 | return -EIO; | |
1987 | ||
1988 | ioff = relo.r_offset / sizeof(struct bpf_insn); | |
1989 | if (ioff >= num_insns || | |
a576c6b9 | 1990 | insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) { |
32a121cb | 1991 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", |
a576c6b9 DB |
1992 | ioff); |
1993 | if (ioff < num_insns && | |
1994 | insns[ioff].code == (BPF_JMP | BPF_CALL)) | |
32a121cb | 1995 | fprintf(stderr, " - Try to annotate functions with always_inline attribute!\n"); |
32e93fb7 | 1996 | return -EINVAL; |
a576c6b9 | 1997 | } |
32e93fb7 DB |
1998 | |
1999 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) | |
2000 | return -EIO; | |
2486337a | 2001 | if (sym.st_shndx != ctx->sec_maps) { |
32a121cb | 2002 | fprintf(stderr, "ELF contains non-map related relo data in entry %u pointing to section %u! Compiler bug?!\n", |
2486337a DB |
2003 | relo_ent, sym.st_shndx); |
2004 | return -EIO; | |
2005 | } | |
32e93fb7 | 2006 | |
e4225669 | 2007 | rmap = sym.st_value / ctx->map_len; |
32e93fb7 DB |
2008 | if (rmap >= ARRAY_SIZE(ctx->map_fds)) |
2009 | return -EINVAL; | |
2010 | if (!ctx->map_fds[rmap]) | |
2011 | return -EINVAL; | |
ecb05c0f DB |
2012 | if (ctx->maps[rmap].type == BPF_MAP_TYPE_PROG_ARRAY) { |
2013 | props->total++; | |
2014 | if (ctx->maps_ext[rmap].owner.jited || | |
2015 | (ctx->maps_ext[rmap].owner.type == 0 && | |
2016 | ctx->cfg.jit_enabled)) | |
2017 | props->jited++; | |
2018 | } | |
32e93fb7 DB |
2019 | |
2020 | if (ctx->verbose) | |
32a121cb | 2021 | fprintf(stderr, "Map \'%s\' (%d) injected into prog section \'%s\' at offset %u!\n", |
32e93fb7 DB |
2022 | bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap], |
2023 | data_insn->sec_name, ioff); | |
11c39b5e | 2024 | |
32e93fb7 DB |
2025 | insns[ioff].src_reg = BPF_PSEUDO_MAP_FD; |
2026 | insns[ioff].imm = ctx->map_fds[rmap]; | |
2027 | } | |
2028 | ||
2029 | return 0; | |
2030 | } | |
2031 | ||
afc1a200 | 2032 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
e4225669 | 2033 | bool *lderr, bool *sseen) |
32e93fb7 DB |
2034 | { |
2035 | struct bpf_elf_sec_data data_relo, data_insn; | |
2036 | struct bpf_elf_prog prog; | |
2037 | int ret, idx, i, fd = -1; | |
2038 | ||
2039 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
ecb05c0f DB |
2040 | struct bpf_tail_call_props props = {}; |
2041 | ||
32e93fb7 DB |
2042 | ret = bpf_fill_section_data(ctx, i, &data_relo); |
2043 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
2044 | continue; |
2045 | ||
32e93fb7 | 2046 | idx = data_relo.sec_hdr.sh_info; |
e4225669 | 2047 | |
32e93fb7 | 2048 | ret = bpf_fill_section_data(ctx, idx, &data_insn); |
cce3d466 DB |
2049 | if (ret < 0 || |
2050 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
2051 | data_insn.sec_hdr.sh_flags & SHF_EXECINSTR && | |
2052 | !strcmp(data_insn.sec_name, section))) | |
11c39b5e | 2053 | continue; |
32e93fb7 | 2054 | |
e4225669 DB |
2055 | *sseen = true; |
2056 | ||
ecb05c0f | 2057 | ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn, &props); |
c9c3720d DB |
2058 | if (ret < 0) { |
2059 | *lderr = true; | |
e4225669 | 2060 | return ret; |
c9c3720d | 2061 | } |
11c39b5e | 2062 | |
32e93fb7 DB |
2063 | memset(&prog, 0, sizeof(prog)); |
2064 | prog.type = ctx->type; | |
2065 | prog.insns = data_insn.sec_data->d_buf; | |
2066 | prog.size = data_insn.sec_data->d_size; | |
2067 | prog.license = ctx->license; | |
2068 | ||
f31645d1 | 2069 | fd = bpf_prog_attach(section, &prog, ctx); |
afc1a200 DB |
2070 | if (fd < 0) { |
2071 | *lderr = true; | |
ecb05c0f DB |
2072 | if (props.total) { |
2073 | if (ctx->cfg.jit_enabled && | |
2074 | props.total != props.jited) | |
2075 | fprintf(stderr, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n", | |
2076 | props.jited, props.total); | |
2077 | if (!ctx->cfg.jit_enabled && | |
2078 | props.jited) | |
2079 | fprintf(stderr, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n", | |
2080 | props.jited, props.total); | |
2081 | } | |
e4225669 | 2082 | return fd; |
afc1a200 | 2083 | } |
11c39b5e | 2084 | |
32e93fb7 DB |
2085 | ctx->sec_done[i] = true; |
2086 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
2087 | break; |
2088 | } | |
2089 | ||
32e93fb7 | 2090 | return fd; |
11c39b5e DB |
2091 | } |
2092 | ||
32e93fb7 | 2093 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 2094 | { |
e4225669 | 2095 | bool lderr = false, sseen = false; |
473d7840 DB |
2096 | int ret = -1; |
2097 | ||
afc1a200 | 2098 | if (bpf_has_map_data(ctx)) |
e4225669 | 2099 | ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen); |
afc1a200 | 2100 | if (ret < 0 && !lderr) |
e4225669 DB |
2101 | ret = bpf_fetch_prog(ctx, section, &sseen); |
2102 | if (ret < 0 && !sseen) | |
2103 | fprintf(stderr, "Program section \'%s\' not found in ELF file!\n", | |
2104 | section); | |
473d7840 DB |
2105 | return ret; |
2106 | } | |
2107 | ||
910b543d DB |
2108 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
2109 | { | |
2110 | int i; | |
2111 | ||
2112 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
2113 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
2114 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
2115 | return i; | |
2116 | return -1; | |
2117 | } | |
2118 | ||
ecb05c0f DB |
2119 | struct bpf_jited_aux { |
2120 | int prog_fd; | |
2121 | int map_fd; | |
2122 | struct bpf_prog_data prog; | |
2123 | struct bpf_map_ext map; | |
2124 | }; | |
2125 | ||
2126 | static int bpf_derive_prog_from_fdinfo(int fd, struct bpf_prog_data *prog) | |
2127 | { | |
2128 | char file[PATH_MAX], buff[4096]; | |
2129 | unsigned int val; | |
2130 | FILE *fp; | |
2131 | ||
2132 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
2133 | memset(prog, 0, sizeof(*prog)); | |
2134 | ||
2135 | fp = fopen(file, "r"); | |
2136 | if (!fp) { | |
2137 | fprintf(stderr, "No procfs support?!\n"); | |
2138 | return -EIO; | |
2139 | } | |
2140 | ||
2141 | while (fgets(buff, sizeof(buff), fp)) { | |
2142 | if (sscanf(buff, "prog_type:\t%u", &val) == 1) | |
2143 | prog->type = val; | |
2144 | else if (sscanf(buff, "prog_jited:\t%u", &val) == 1) | |
2145 | prog->jited = val; | |
2146 | } | |
2147 | ||
2148 | fclose(fp); | |
2149 | return 0; | |
2150 | } | |
2151 | ||
2152 | static int bpf_tail_call_get_aux(struct bpf_jited_aux *aux) | |
2153 | { | |
2154 | struct bpf_elf_map tmp; | |
2155 | int ret; | |
2156 | ||
2157 | ret = bpf_derive_elf_map_from_fdinfo(aux->map_fd, &tmp, &aux->map); | |
2158 | if (!ret) | |
2159 | ret = bpf_derive_prog_from_fdinfo(aux->prog_fd, &aux->prog); | |
2160 | ||
2161 | return ret; | |
2162 | } | |
2163 | ||
32e93fb7 | 2164 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 2165 | { |
32e93fb7 DB |
2166 | struct bpf_elf_sec_data data; |
2167 | uint32_t map_id, key_id; | |
910b543d | 2168 | int fd, i, ret, idx; |
473d7840 | 2169 | |
32e93fb7 DB |
2170 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2171 | if (ctx->sec_done[i]) | |
473d7840 DB |
2172 | continue; |
2173 | ||
32e93fb7 | 2174 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
2175 | if (ret < 0) |
2176 | continue; | |
2177 | ||
910b543d DB |
2178 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
2179 | if (ret != 2) | |
32e93fb7 | 2180 | continue; |
910b543d DB |
2181 | |
2182 | idx = bpf_find_map_by_id(ctx, map_id); | |
2183 | if (idx < 0) | |
473d7840 DB |
2184 | continue; |
2185 | ||
32e93fb7 DB |
2186 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
2187 | if (fd < 0) | |
473d7840 DB |
2188 | return -EIO; |
2189 | ||
910b543d DB |
2190 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
2191 | &fd, BPF_ANY); | |
afc1a200 | 2192 | if (ret < 0) { |
ecb05c0f DB |
2193 | struct bpf_jited_aux aux = {}; |
2194 | ||
2195 | ret = -errno; | |
2196 | if (errno == E2BIG) { | |
afc1a200 DB |
2197 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", |
2198 | key_id, map_id); | |
ecb05c0f DB |
2199 | return ret; |
2200 | } | |
2201 | ||
2202 | aux.map_fd = ctx->map_fds[idx]; | |
2203 | aux.prog_fd = fd; | |
2204 | ||
2205 | if (bpf_tail_call_get_aux(&aux)) | |
2206 | return ret; | |
2207 | if (!aux.map.owner.type) | |
2208 | return ret; | |
2209 | ||
2210 | if (aux.prog.type != aux.map.owner.type) | |
2211 | fprintf(stderr, "Tail call map owned by prog type %u, but prog type is %u!\n", | |
2212 | aux.map.owner.type, aux.prog.type); | |
2213 | if (aux.prog.jited != aux.map.owner.jited) | |
2214 | fprintf(stderr, "Tail call map %s jited, but prog %s!\n", | |
2215 | aux.map.owner.jited ? "is" : "not", | |
2216 | aux.prog.jited ? "is" : "not"); | |
2217 | return ret; | |
afc1a200 | 2218 | } |
473d7840 | 2219 | |
32e93fb7 | 2220 | ctx->sec_done[i] = true; |
473d7840 DB |
2221 | } |
2222 | ||
2223 | return 0; | |
2224 | } | |
2225 | ||
32e93fb7 | 2226 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 2227 | { |
32e93fb7 DB |
2228 | struct stat st; |
2229 | int ret; | |
11c39b5e | 2230 | |
32e93fb7 | 2231 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 2232 | |
32e93fb7 DB |
2233 | ret = fstat(ctx->obj_fd, &st); |
2234 | if (ret < 0) { | |
2235 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
2236 | strerror(errno)); | |
2237 | return; | |
2238 | } | |
11c39b5e | 2239 | |
32e93fb7 DB |
2240 | ctx->stat.st_dev = st.st_dev; |
2241 | ctx->stat.st_ino = st.st_ino; | |
2242 | } | |
2243 | ||
f6793eec DB |
2244 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
2245 | { | |
2246 | char buff[PATH_MAX]; | |
2247 | ||
2248 | while (fgets(buff, sizeof(buff), fp)) { | |
2249 | char *ptr = buff; | |
2250 | ||
2251 | while (*ptr == ' ' || *ptr == '\t') | |
2252 | ptr++; | |
2253 | ||
2254 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
2255 | continue; | |
2256 | ||
2257 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
2258 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
2259 | strcpy(path, ptr); | |
2260 | return -1; | |
2261 | } | |
2262 | ||
2263 | return 1; | |
2264 | } | |
2265 | ||
2266 | return 0; | |
2267 | } | |
2268 | ||
2269 | static bool bpf_pinning_reserved(uint32_t pinning) | |
2270 | { | |
2271 | switch (pinning) { | |
2272 | case PIN_NONE: | |
2273 | case PIN_OBJECT_NS: | |
2274 | case PIN_GLOBAL_NS: | |
2275 | return true; | |
2276 | default: | |
2277 | return false; | |
2278 | } | |
2279 | } | |
2280 | ||
2281 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
2282 | { | |
2283 | struct bpf_hash_entry *entry; | |
d17b136f | 2284 | char subpath[PATH_MAX] = {}; |
f6793eec DB |
2285 | uint32_t pinning; |
2286 | FILE *fp; | |
2287 | int ret; | |
2288 | ||
2289 | fp = fopen(db_file, "r"); | |
2290 | if (!fp) | |
2291 | return; | |
2292 | ||
f6793eec DB |
2293 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { |
2294 | if (ret == -1) { | |
2295 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
2296 | db_file, subpath); | |
2297 | fclose(fp); | |
2298 | return; | |
2299 | } | |
2300 | ||
2301 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
2302 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
2303 | db_file, pinning); | |
f6793eec DB |
2304 | continue; |
2305 | } | |
2306 | ||
2307 | entry = malloc(sizeof(*entry)); | |
2308 | if (!entry) { | |
2309 | fprintf(stderr, "No memory left for db entry!\n"); | |
2310 | continue; | |
2311 | } | |
2312 | ||
2313 | entry->pinning = pinning; | |
2314 | entry->subpath = strdup(subpath); | |
2315 | if (!entry->subpath) { | |
2316 | fprintf(stderr, "No memory left for db entry!\n"); | |
2317 | free(entry); | |
2318 | continue; | |
2319 | } | |
2320 | ||
2321 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
2322 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
2323 | } | |
2324 | ||
2325 | fclose(fp); | |
2326 | } | |
2327 | ||
2328 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
2329 | { | |
2330 | struct bpf_hash_entry *entry; | |
2331 | int i; | |
2332 | ||
2333 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
2334 | while ((entry = ctx->ht[i]) != NULL) { | |
2335 | ctx->ht[i] = entry->next; | |
2336 | free((char *)entry->subpath); | |
2337 | free(entry); | |
2338 | } | |
2339 | } | |
2340 | } | |
2341 | ||
8187b012 DB |
2342 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
2343 | { | |
2344 | if (ctx->elf_hdr.e_type != ET_REL || | |
e77fa41d DB |
2345 | (ctx->elf_hdr.e_machine != EM_NONE && |
2346 | ctx->elf_hdr.e_machine != EM_BPF) || | |
8187b012 DB |
2347 | ctx->elf_hdr.e_version != EV_CURRENT) { |
2348 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
2349 | return -EINVAL; | |
2350 | } | |
2351 | ||
2352 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
2353 | default: | |
2354 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
2355 | return -EINVAL; | |
2356 | case ELFDATA2LSB: | |
2357 | if (htons(1) == 1) { | |
2358 | fprintf(stderr, | |
2359 | "We are big endian, eBPF object is little endian!\n"); | |
2360 | return -EIO; | |
2361 | } | |
2362 | break; | |
2363 | case ELFDATA2MSB: | |
2364 | if (htons(1) != 1) { | |
2365 | fprintf(stderr, | |
2366 | "We are little endian, eBPF object is big endian!\n"); | |
2367 | return -EIO; | |
2368 | } | |
2369 | break; | |
2370 | } | |
2371 | ||
2372 | return 0; | |
2373 | } | |
2374 | ||
ecb05c0f DB |
2375 | static void bpf_get_cfg(struct bpf_elf_ctx *ctx) |
2376 | { | |
2377 | static const char *path_jit = "/proc/sys/net/core/bpf_jit_enable"; | |
2378 | int fd; | |
2379 | ||
2380 | fd = open(path_jit, O_RDONLY); | |
2381 | if (fd > 0) { | |
2382 | char tmp[16] = {}; | |
2383 | ||
2384 | if (read(fd, tmp, sizeof(tmp)) > 0) | |
2385 | ctx->cfg.jit_enabled = atoi(tmp); | |
2386 | close(fd); | |
2387 | } | |
2388 | } | |
2389 | ||
32e93fb7 DB |
2390 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
2391 | enum bpf_prog_type type, bool verbose) | |
2392 | { | |
2393 | int ret = -EINVAL; | |
2394 | ||
2395 | if (elf_version(EV_CURRENT) == EV_NONE || | |
2396 | bpf_init_env(pathname)) | |
2397 | return ret; | |
2398 | ||
2399 | memset(ctx, 0, sizeof(*ctx)); | |
ecb05c0f | 2400 | bpf_get_cfg(ctx); |
32e93fb7 DB |
2401 | ctx->verbose = verbose; |
2402 | ctx->type = type; | |
2403 | ||
2404 | ctx->obj_fd = open(pathname, O_RDONLY); | |
2405 | if (ctx->obj_fd < 0) | |
2406 | return ctx->obj_fd; | |
2407 | ||
2408 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
2409 | if (!ctx->elf_fd) { | |
11c39b5e | 2410 | ret = -EINVAL; |
32e93fb7 | 2411 | goto out_fd; |
11c39b5e DB |
2412 | } |
2413 | ||
8187b012 DB |
2414 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
2415 | ret = -EINVAL; | |
2416 | goto out_fd; | |
2417 | } | |
2418 | ||
32e93fb7 DB |
2419 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
2420 | &ctx->elf_hdr) { | |
11c39b5e DB |
2421 | ret = -EIO; |
2422 | goto out_elf; | |
2423 | } | |
2424 | ||
8187b012 DB |
2425 | ret = bpf_elf_check_ehdr(ctx); |
2426 | if (ret < 0) | |
2427 | goto out_elf; | |
2428 | ||
32e93fb7 DB |
2429 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
2430 | sizeof(*(ctx->sec_done))); | |
2431 | if (!ctx->sec_done) { | |
11c39b5e DB |
2432 | ret = -ENOMEM; |
2433 | goto out_elf; | |
2434 | } | |
2435 | ||
f31645d1 DB |
2436 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
2437 | ret = -ENOMEM; | |
2438 | goto out_free; | |
2439 | } | |
2440 | ||
32e93fb7 | 2441 | bpf_save_finfo(ctx); |
f6793eec DB |
2442 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
2443 | ||
32e93fb7 | 2444 | return 0; |
f31645d1 DB |
2445 | out_free: |
2446 | free(ctx->sec_done); | |
32e93fb7 DB |
2447 | out_elf: |
2448 | elf_end(ctx->elf_fd); | |
2449 | out_fd: | |
2450 | close(ctx->obj_fd); | |
2451 | return ret; | |
2452 | } | |
d937a74b | 2453 | |
32e93fb7 DB |
2454 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
2455 | { | |
2456 | int i, count = 0; | |
11c39b5e | 2457 | |
32e93fb7 DB |
2458 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
2459 | if (!ctx->map_fds[i]) | |
2460 | break; | |
2461 | count++; | |
2462 | } | |
473d7840 | 2463 | |
32e93fb7 DB |
2464 | return count; |
2465 | } | |
6256f8c9 | 2466 | |
32e93fb7 DB |
2467 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
2468 | { | |
2469 | int i; | |
2470 | ||
2471 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
2472 | if (ctx->map_fds[i]) | |
2473 | close(ctx->map_fds[i]); | |
473d7840 | 2474 | } |
32e93fb7 DB |
2475 | } |
2476 | ||
2477 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
2478 | { | |
2479 | if (failure) | |
2480 | bpf_maps_teardown(ctx); | |
473d7840 | 2481 | |
f6793eec | 2482 | bpf_hash_destroy(ctx); |
f31645d1 | 2483 | |
32e93fb7 | 2484 | free(ctx->sec_done); |
f31645d1 DB |
2485 | free(ctx->log); |
2486 | ||
32e93fb7 DB |
2487 | elf_end(ctx->elf_fd); |
2488 | close(ctx->obj_fd); | |
2489 | } | |
6256f8c9 | 2490 | |
32e93fb7 | 2491 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 2492 | |
32e93fb7 DB |
2493 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
2494 | const char *section, bool verbose) | |
2495 | { | |
2496 | struct bpf_elf_ctx *ctx = &__ctx; | |
2497 | int fd = 0, ret; | |
6256f8c9 | 2498 | |
32e93fb7 DB |
2499 | ret = bpf_elf_ctx_init(ctx, pathname, type, verbose); |
2500 | if (ret < 0) { | |
2501 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
2502 | return ret; | |
2503 | } | |
6256f8c9 | 2504 | |
32e93fb7 DB |
2505 | ret = bpf_fetch_ancillary(ctx); |
2506 | if (ret < 0) { | |
2507 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
2508 | goto out; | |
2509 | } | |
2510 | ||
2511 | fd = bpf_fetch_prog_sec(ctx, section); | |
2512 | if (fd < 0) { | |
2513 | fprintf(stderr, "Error fetching program/map!\n"); | |
2514 | ret = fd; | |
2515 | goto out; | |
2516 | } | |
2517 | ||
2518 | ret = bpf_fill_prog_arrays(ctx); | |
2519 | if (ret < 0) | |
2520 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 2521 | out: |
32e93fb7 DB |
2522 | bpf_elf_ctx_destroy(ctx, ret < 0); |
2523 | if (ret < 0) { | |
2524 | if (fd) | |
2525 | close(fd); | |
2526 | return ret; | |
2527 | } | |
2528 | ||
2529 | return fd; | |
6256f8c9 | 2530 | } |
11c39b5e | 2531 | |
6256f8c9 | 2532 | static int |
4bd62446 DB |
2533 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
2534 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 | 2535 | { |
d17b136f PS |
2536 | struct bpf_map_set_msg msg = { |
2537 | .aux.uds_ver = BPF_SCM_AUX_VER, | |
2538 | .aux.num_ent = entries, | |
2539 | }; | |
6256f8c9 DB |
2540 | int *cmsg_buf, min_fd; |
2541 | char *amsg_buf; | |
2542 | int i; | |
2543 | ||
6256f8c9 DB |
2544 | strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); |
2545 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); | |
2546 | ||
2547 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
2548 | amsg_buf = (char *)msg.aux.ent; | |
2549 | ||
4bd62446 | 2550 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
2551 | int ret; |
2552 | ||
4bd62446 | 2553 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
2554 | bpf_map_set_init_single(&msg, min_fd); |
2555 | ||
2556 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
2557 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
2558 | ||
2559 | ret = sendmsg(fd, &msg.hdr, 0); | |
2560 | if (ret <= 0) | |
2561 | return ret ? : -1; | |
2562 | } | |
2563 | ||
2564 | return 0; | |
11c39b5e DB |
2565 | } |
2566 | ||
4bd62446 DB |
2567 | static int |
2568 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
2569 | unsigned int entries) | |
2570 | { | |
2571 | struct bpf_map_set_msg msg; | |
2572 | int *cmsg_buf, min_fd; | |
2573 | char *amsg_buf, *mmsg_buf; | |
2574 | unsigned int needed = 1; | |
2575 | int i; | |
2576 | ||
2577 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
2578 | amsg_buf = (char *)msg.aux.ent; | |
2579 | mmsg_buf = (char *)&msg.aux; | |
2580 | ||
2581 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
2582 | struct cmsghdr *cmsg; | |
2583 | int ret; | |
2584 | ||
2585 | min_fd = min(entries, entries - i); | |
2586 | bpf_map_set_init_single(&msg, min_fd); | |
2587 | ||
2588 | ret = recvmsg(fd, &msg.hdr, 0); | |
2589 | if (ret <= 0) | |
2590 | return ret ? : -1; | |
2591 | ||
2592 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
2593 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
2594 | return -EINVAL; | |
2595 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
2596 | return -EIO; | |
2597 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
2598 | return -ENOSYS; | |
2599 | ||
2600 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
2601 | if (min_fd > entries || min_fd <= 0) | |
2602 | return -EINVAL; | |
2603 | ||
2604 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
2605 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
2606 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
2607 | ||
2608 | needed = aux->num_ent; | |
2609 | } | |
2610 | ||
2611 | return 0; | |
2612 | } | |
2613 | ||
2614 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 2615 | { |
32e93fb7 | 2616 | struct bpf_elf_ctx *ctx = &__ctx; |
d17b136f PS |
2617 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
2618 | struct bpf_map_data bpf_aux = { | |
2619 | .fds = ctx->map_fds, | |
2620 | .ent = ctx->maps, | |
2621 | .st = &ctx->stat, | |
2622 | .obj = obj, | |
2623 | }; | |
6256f8c9 DB |
2624 | int fd, ret; |
2625 | ||
2626 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2627 | if (fd < 0) { | |
2628 | fprintf(stderr, "Cannot open socket: %s\n", | |
2629 | strerror(errno)); | |
2630 | return -1; | |
2631 | } | |
2632 | ||
6256f8c9 DB |
2633 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2634 | ||
2635 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2636 | if (ret < 0) { | |
2637 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
2638 | path, strerror(errno)); | |
2639 | return -1; | |
2640 | } | |
2641 | ||
4bd62446 | 2642 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 2643 | bpf_maps_count(ctx)); |
6256f8c9 | 2644 | if (ret < 0) |
4bd62446 DB |
2645 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
2646 | path, strerror(errno)); | |
2647 | ||
32e93fb7 | 2648 | bpf_maps_teardown(ctx); |
4bd62446 DB |
2649 | close(fd); |
2650 | return ret; | |
2651 | } | |
2652 | ||
2653 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
2654 | unsigned int entries) | |
2655 | { | |
d17b136f | 2656 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
4bd62446 DB |
2657 | int fd, ret; |
2658 | ||
2659 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2660 | if (fd < 0) { | |
2661 | fprintf(stderr, "Cannot open socket: %s\n", | |
2662 | strerror(errno)); | |
2663 | return -1; | |
2664 | } | |
2665 | ||
4bd62446 DB |
2666 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2667 | ||
2668 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2669 | if (ret < 0) { | |
2670 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
2671 | strerror(errno)); | |
2672 | return -1; | |
2673 | } | |
2674 | ||
2675 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
2676 | if (ret < 0) | |
2677 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
2678 | path, strerror(errno)); |
2679 | ||
4bd62446 | 2680 | unlink(addr.sun_path); |
6256f8c9 DB |
2681 | close(fd); |
2682 | return ret; | |
2683 | } | |
11c39b5e | 2684 | #endif /* HAVE_ELF */ |