]>
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) { | |
18f156bf | 515 | strlcpy(mnt, mntpt, len); |
95ae9a48 DB |
516 | return mnt; |
517 | } | |
518 | ||
519 | return NULL; | |
520 | } | |
521 | ||
32e93fb7 DB |
522 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, |
523 | char *mnt, int len, | |
524 | const char * const *known_mnts) | |
525 | { | |
526 | const char * const *ptr; | |
527 | char type[100]; | |
528 | FILE *fp; | |
529 | ||
530 | if (known_mnts) { | |
531 | ptr = known_mnts; | |
532 | while (*ptr) { | |
95ae9a48 | 533 | if (bpf_find_mntpt_single(magic, mnt, len, *ptr)) |
32e93fb7 | 534 | return mnt; |
32e93fb7 DB |
535 | ptr++; |
536 | } | |
537 | } | |
538 | ||
c3724e4b PS |
539 | if (len != PATH_MAX) |
540 | return NULL; | |
541 | ||
32e93fb7 | 542 | fp = fopen("/proc/mounts", "r"); |
c3724e4b | 543 | if (fp == NULL) |
32e93fb7 DB |
544 | return NULL; |
545 | ||
546 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
547 | mnt, type) == 2) { | |
548 | if (strcmp(type, fstype) == 0) | |
549 | break; | |
550 | } | |
551 | ||
552 | fclose(fp); | |
553 | if (strcmp(type, fstype) != 0) | |
554 | return NULL; | |
555 | ||
556 | return mnt; | |
557 | } | |
558 | ||
559 | int bpf_trace_pipe(void) | |
560 | { | |
561 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
562 | static const char * const tracefs_known_mnts[] = { | |
563 | TRACE_DIR_MNT, | |
564 | "/sys/kernel/debug/tracing", | |
565 | "/tracing", | |
566 | "/trace", | |
567 | 0, | |
568 | }; | |
569 | char tpipe[PATH_MAX]; | |
570 | const char *mnt; | |
571 | int fd; | |
572 | ||
573 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
574 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
575 | if (!mnt) { | |
576 | fprintf(stderr, "tracefs not mounted?\n"); | |
577 | return -1; | |
578 | } | |
579 | ||
580 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
581 | ||
582 | fd = open(tpipe, O_RDONLY); | |
583 | if (fd < 0) | |
584 | return -1; | |
585 | ||
586 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
587 | while (1) { | |
588 | static char buff[4096]; | |
589 | ssize_t ret; | |
590 | ||
591 | ret = read(fd, buff, sizeof(buff) - 1); | |
592 | if (ret > 0) { | |
b5c78e1b PS |
593 | if (write(STDERR_FILENO, buff, ret) != ret) |
594 | return -1; | |
32e93fb7 DB |
595 | fflush(stderr); |
596 | } | |
597 | } | |
598 | ||
599 | return 0; | |
600 | } | |
601 | ||
e4225669 | 602 | static int bpf_gen_global(const char *bpf_sub_dir) |
91d88eeb | 603 | { |
e4225669 DB |
604 | char bpf_glo_dir[PATH_MAX]; |
605 | int ret; | |
606 | ||
607 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/", | |
608 | bpf_sub_dir, BPF_DIR_GLOBALS); | |
609 | ||
610 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
611 | if (ret && errno != EEXIST) { | |
612 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
613 | strerror(errno)); | |
614 | return ret; | |
615 | } | |
616 | ||
617 | return 0; | |
618 | } | |
619 | ||
620 | static int bpf_gen_master(const char *base, const char *name) | |
621 | { | |
622 | char bpf_sub_dir[PATH_MAX]; | |
623 | int ret; | |
624 | ||
625 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name); | |
626 | ||
627 | ret = mkdir(bpf_sub_dir, S_IRWXU); | |
628 | if (ret && errno != EEXIST) { | |
629 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir, | |
630 | strerror(errno)); | |
631 | return ret; | |
632 | } | |
633 | ||
634 | return bpf_gen_global(bpf_sub_dir); | |
635 | } | |
636 | ||
637 | static int bpf_slave_via_bind_mnt(const char *full_name, | |
638 | const char *full_link) | |
639 | { | |
640 | int ret; | |
641 | ||
642 | ret = mkdir(full_name, S_IRWXU); | |
643 | if (ret) { | |
644 | assert(errno != EEXIST); | |
645 | fprintf(stderr, "mkdir %s failed: %s\n", full_name, | |
646 | strerror(errno)); | |
647 | return ret; | |
648 | } | |
649 | ||
650 | ret = mount(full_link, full_name, "none", MS_BIND, NULL); | |
651 | if (ret) { | |
652 | rmdir(full_name); | |
653 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
654 | full_link, full_name, strerror(errno)); | |
655 | } | |
656 | ||
657 | return ret; | |
658 | } | |
659 | ||
660 | static int bpf_gen_slave(const char *base, const char *name, | |
661 | const char *link) | |
662 | { | |
663 | char bpf_lnk_dir[PATH_MAX]; | |
664 | char bpf_sub_dir[PATH_MAX]; | |
665 | struct stat sb = {}; | |
666 | int ret; | |
667 | ||
668 | snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link); | |
669 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name); | |
670 | ||
671 | ret = symlink(bpf_lnk_dir, bpf_sub_dir); | |
672 | if (ret) { | |
673 | if (errno != EEXIST) { | |
674 | if (errno != EPERM) { | |
675 | fprintf(stderr, "symlink %s failed: %s\n", | |
676 | bpf_sub_dir, strerror(errno)); | |
677 | return ret; | |
678 | } | |
679 | ||
680 | return bpf_slave_via_bind_mnt(bpf_sub_dir, | |
681 | bpf_lnk_dir); | |
682 | } | |
683 | ||
684 | ret = lstat(bpf_sub_dir, &sb); | |
685 | if (ret) { | |
686 | fprintf(stderr, "lstat %s failed: %s\n", | |
687 | bpf_sub_dir, strerror(errno)); | |
688 | return ret; | |
689 | } | |
690 | ||
691 | if ((sb.st_mode & S_IFMT) != S_IFLNK) | |
692 | return bpf_gen_global(bpf_sub_dir); | |
693 | } | |
694 | ||
695 | return 0; | |
696 | } | |
697 | ||
698 | static int bpf_gen_hierarchy(const char *base) | |
699 | { | |
700 | int ret, i; | |
701 | ||
702 | ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0])); | |
703 | for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++) | |
704 | ret = bpf_gen_slave(base, | |
705 | bpf_prog_to_subdir(__bpf_types[i]), | |
706 | bpf_prog_to_subdir(__bpf_types[0])); | |
707 | return ret; | |
708 | } | |
709 | ||
710 | static const char *bpf_get_work_dir(enum bpf_prog_type type) | |
711 | { | |
712 | static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT; | |
713 | static char bpf_wrk_dir[PATH_MAX]; | |
91d88eeb | 714 | static const char *mnt; |
e4225669 | 715 | static bool bpf_mnt_cached; |
95ae9a48 | 716 | const char *mnt_env = getenv(BPF_ENV_MNT); |
91d88eeb DB |
717 | static const char * const bpf_known_mnts[] = { |
718 | BPF_DIR_MNT, | |
e4225669 | 719 | "/bpf", |
91d88eeb DB |
720 | 0, |
721 | }; | |
91d88eeb DB |
722 | int ret; |
723 | ||
e4225669 DB |
724 | if (bpf_mnt_cached) { |
725 | const char *out = mnt; | |
726 | ||
51361a9f | 727 | if (out && type) { |
e4225669 DB |
728 | snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/", |
729 | out, bpf_prog_to_subdir(type)); | |
730 | out = bpf_tmp; | |
731 | } | |
732 | return out; | |
733 | } | |
91d88eeb | 734 | |
95ae9a48 DB |
735 | if (mnt_env) |
736 | mnt = bpf_find_mntpt_single(BPF_FS_MAGIC, bpf_tmp, | |
737 | sizeof(bpf_tmp), mnt_env); | |
738 | else | |
739 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp, | |
740 | sizeof(bpf_tmp), bpf_known_mnts); | |
91d88eeb | 741 | if (!mnt) { |
95ae9a48 DB |
742 | mnt = mnt_env ? : BPF_DIR_MNT; |
743 | ret = bpf_mnt_check_target(mnt); | |
744 | if (!ret) | |
745 | ret = bpf_mnt_fs(mnt); | |
91d88eeb DB |
746 | if (ret) { |
747 | mnt = NULL; | |
748 | goto out; | |
749 | } | |
750 | } | |
751 | ||
e4225669 | 752 | snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt); |
91d88eeb | 753 | |
e4225669 DB |
754 | ret = bpf_gen_hierarchy(bpf_wrk_dir); |
755 | if (ret) { | |
91d88eeb DB |
756 | mnt = NULL; |
757 | goto out; | |
758 | } | |
759 | ||
e4225669 | 760 | mnt = bpf_wrk_dir; |
91d88eeb DB |
761 | out: |
762 | bpf_mnt_cached = true; | |
91d88eeb DB |
763 | return mnt; |
764 | } | |
765 | ||
e4225669 | 766 | static int bpf_obj_get(const char *pathname, enum bpf_prog_type type) |
91d88eeb | 767 | { |
d17b136f | 768 | union bpf_attr attr = {}; |
91d88eeb DB |
769 | char tmp[PATH_MAX]; |
770 | ||
771 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
e4225669 | 772 | pathname[1] == ':' && bpf_get_work_dir(type)) { |
91d88eeb | 773 | snprintf(tmp, sizeof(tmp), "%s/%s", |
e4225669 | 774 | bpf_get_work_dir(type), pathname + 2); |
91d88eeb DB |
775 | pathname = tmp; |
776 | } | |
777 | ||
91d88eeb DB |
778 | attr.pathname = bpf_ptr_to_u64(pathname); |
779 | ||
780 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
781 | } | |
782 | ||
21856018 DB |
783 | static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type) |
784 | { | |
785 | int prog_fd = bpf_obj_get(pathname, type); | |
786 | ||
787 | if (prog_fd < 0) | |
788 | fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n", | |
789 | pathname, strerror(errno)); | |
790 | return prog_fd; | |
791 | } | |
792 | ||
91d88eeb | 793 | enum bpf_mode { |
e4225669 | 794 | CBPF_BYTECODE, |
91d88eeb DB |
795 | CBPF_FILE, |
796 | EBPF_OBJECT, | |
797 | EBPF_PINNED, | |
e4225669 | 798 | BPF_MODE_MAX, |
91d88eeb DB |
799 | }; |
800 | ||
e4225669 DB |
801 | static int bpf_parse(enum bpf_prog_type *type, enum bpf_mode *mode, |
802 | struct bpf_cfg_in *cfg, const bool *opt_tbl) | |
32e93fb7 | 803 | { |
32e93fb7 | 804 | const char *file, *section, *uds_name; |
32e93fb7 | 805 | bool verbose = false; |
e4225669 | 806 | int i, ret, argc; |
91d88eeb DB |
807 | char **argv; |
808 | ||
e4225669 DB |
809 | argv = cfg->argv; |
810 | argc = cfg->argc; | |
91d88eeb DB |
811 | |
812 | if (opt_tbl[CBPF_BYTECODE] && | |
813 | (matches(*argv, "bytecode") == 0 || | |
814 | strcmp(*argv, "bc") == 0)) { | |
815 | *mode = CBPF_BYTECODE; | |
816 | } else if (opt_tbl[CBPF_FILE] && | |
817 | (matches(*argv, "bytecode-file") == 0 || | |
818 | strcmp(*argv, "bcf") == 0)) { | |
819 | *mode = CBPF_FILE; | |
820 | } else if (opt_tbl[EBPF_OBJECT] && | |
821 | (matches(*argv, "object-file") == 0 || | |
822 | strcmp(*argv, "obj") == 0)) { | |
823 | *mode = EBPF_OBJECT; | |
824 | } else if (opt_tbl[EBPF_PINNED] && | |
825 | (matches(*argv, "object-pinned") == 0 || | |
826 | matches(*argv, "pinned") == 0 || | |
827 | matches(*argv, "fd") == 0)) { | |
828 | *mode = EBPF_PINNED; | |
32e93fb7 DB |
829 | } else { |
830 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
831 | return -1; | |
832 | } | |
833 | ||
834 | NEXT_ARG(); | |
835 | file = section = uds_name = NULL; | |
91d88eeb | 836 | if (*mode == EBPF_OBJECT || *mode == EBPF_PINNED) { |
32e93fb7 DB |
837 | file = *argv; |
838 | NEXT_ARG_FWD(); | |
839 | ||
91d88eeb DB |
840 | if (*type == BPF_PROG_TYPE_UNSPEC) { |
841 | if (argc > 0 && matches(*argv, "type") == 0) { | |
842 | NEXT_ARG(); | |
e4225669 DB |
843 | for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta); |
844 | i++) { | |
845 | if (!__bpf_prog_meta[i].type) | |
846 | continue; | |
847 | if (!matches(*argv, | |
848 | __bpf_prog_meta[i].type)) { | |
849 | *type = i; | |
850 | break; | |
851 | } | |
852 | } | |
853 | ||
854 | if (*type == BPF_PROG_TYPE_UNSPEC) { | |
91d88eeb DB |
855 | fprintf(stderr, "What type is \"%s\"?\n", |
856 | *argv); | |
857 | return -1; | |
858 | } | |
859 | NEXT_ARG_FWD(); | |
860 | } else { | |
861 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
862 | } | |
863 | } | |
864 | ||
e4225669 | 865 | section = bpf_prog_to_default_section(*type); |
32e93fb7 DB |
866 | if (argc > 0 && matches(*argv, "section") == 0) { |
867 | NEXT_ARG(); | |
868 | section = *argv; | |
869 | NEXT_ARG_FWD(); | |
870 | } | |
871 | ||
e4225669 DB |
872 | if (__bpf_prog_meta[*type].may_uds_export) { |
873 | uds_name = getenv(BPF_ENV_UDS); | |
874 | if (argc > 0 && !uds_name && | |
875 | matches(*argv, "export") == 0) { | |
876 | NEXT_ARG(); | |
877 | uds_name = *argv; | |
878 | NEXT_ARG_FWD(); | |
879 | } | |
32e93fb7 DB |
880 | } |
881 | ||
882 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
883 | verbose = true; | |
884 | NEXT_ARG_FWD(); | |
885 | } | |
886 | ||
887 | PREV_ARG(); | |
888 | } | |
889 | ||
91d88eeb | 890 | if (*mode == CBPF_BYTECODE || *mode == CBPF_FILE) |
e4225669 | 891 | ret = bpf_ops_parse(argc, argv, cfg->ops, *mode == CBPF_FILE); |
91d88eeb DB |
892 | else if (*mode == EBPF_OBJECT) |
893 | ret = bpf_obj_open(file, *type, section, verbose); | |
894 | else if (*mode == EBPF_PINNED) | |
21856018 | 895 | ret = bpf_obj_pinned(file, *type); |
91d88eeb | 896 | else |
32e93fb7 DB |
897 | return -1; |
898 | ||
e4225669 DB |
899 | cfg->object = file; |
900 | cfg->section = section; | |
901 | cfg->uds = uds_name; | |
902 | cfg->argc = argc; | |
903 | cfg->argv = argv; | |
91d88eeb DB |
904 | |
905 | return ret; | |
906 | } | |
907 | ||
e4225669 DB |
908 | static int bpf_parse_opt_tbl(enum bpf_prog_type type, struct bpf_cfg_in *cfg, |
909 | const struct bpf_cfg_ops *ops, void *nl, | |
910 | const bool *opt_tbl) | |
91d88eeb DB |
911 | { |
912 | struct sock_filter opcodes[BPF_MAXINSNS]; | |
91d88eeb | 913 | char annotation[256]; |
91d88eeb DB |
914 | enum bpf_mode mode; |
915 | int ret; | |
916 | ||
e4225669 DB |
917 | cfg->ops = opcodes; |
918 | ret = bpf_parse(&type, &mode, cfg, opt_tbl); | |
919 | cfg->ops = NULL; | |
91d88eeb DB |
920 | if (ret < 0) |
921 | return ret; | |
922 | ||
e4225669 DB |
923 | if (mode == CBPF_BYTECODE || mode == CBPF_FILE) |
924 | ops->cbpf_cb(nl, opcodes, ret); | |
91d88eeb | 925 | if (mode == EBPF_OBJECT || mode == EBPF_PINNED) { |
32e93fb7 | 926 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
e4225669 DB |
927 | basename(cfg->object), mode == EBPF_PINNED ? |
928 | "*fsobj" : cfg->section); | |
929 | ops->ebpf_cb(nl, ret, annotation); | |
32e93fb7 DB |
930 | } |
931 | ||
91d88eeb DB |
932 | return 0; |
933 | } | |
32e93fb7 | 934 | |
e4225669 DB |
935 | int bpf_parse_common(enum bpf_prog_type type, struct bpf_cfg_in *cfg, |
936 | const struct bpf_cfg_ops *ops, void *nl) | |
937 | { | |
938 | bool opt_tbl[BPF_MODE_MAX] = {}; | |
939 | ||
940 | if (ops->cbpf_cb) { | |
941 | opt_tbl[CBPF_BYTECODE] = true; | |
942 | opt_tbl[CBPF_FILE] = true; | |
943 | } | |
944 | ||
945 | if (ops->ebpf_cb) { | |
946 | opt_tbl[EBPF_OBJECT] = true; | |
947 | opt_tbl[EBPF_PINNED] = true; | |
948 | } | |
949 | ||
950 | return bpf_parse_opt_tbl(type, cfg, ops, nl, opt_tbl); | |
951 | } | |
952 | ||
91d88eeb DB |
953 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
954 | { | |
955 | enum bpf_prog_type type = BPF_PROG_TYPE_UNSPEC; | |
956 | const bool opt_tbl[BPF_MODE_MAX] = { | |
91d88eeb DB |
957 | [EBPF_OBJECT] = true, |
958 | [EBPF_PINNED] = true, | |
959 | }; | |
960 | const struct bpf_elf_map test = { | |
961 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
962 | .size_key = sizeof(int), | |
963 | .size_value = sizeof(int), | |
964 | }; | |
e4225669 DB |
965 | struct bpf_cfg_in cfg = { |
966 | .argc = argc, | |
967 | .argv = argv, | |
968 | }; | |
ecb05c0f | 969 | struct bpf_map_ext ext = {}; |
91d88eeb | 970 | int ret, prog_fd, map_fd; |
91d88eeb DB |
971 | enum bpf_mode mode; |
972 | uint32_t map_key; | |
973 | ||
e4225669 | 974 | prog_fd = bpf_parse(&type, &mode, &cfg, opt_tbl); |
91d88eeb DB |
975 | if (prog_fd < 0) |
976 | return prog_fd; | |
977 | if (key) { | |
978 | map_key = *key; | |
979 | } else { | |
e4225669 | 980 | ret = sscanf(cfg.section, "%*i/%i", &map_key); |
91d88eeb | 981 | if (ret != 1) { |
32a121cb | 982 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
983 | ret = -EINVAL; |
984 | goto out_prog; | |
985 | } | |
986 | } | |
32e93fb7 | 987 | |
e4225669 | 988 | map_fd = bpf_obj_get(map_path, type); |
91d88eeb DB |
989 | if (map_fd < 0) { |
990 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
991 | map_path, strerror(errno)); | |
992 | ret = map_fd; | |
993 | goto out_prog; | |
994 | } | |
995 | ||
ecb05c0f | 996 | ret = bpf_map_selfcheck_pinned(map_fd, &test, &ext, |
fb24802b DB |
997 | offsetof(struct bpf_elf_map, max_elem), |
998 | type); | |
91d88eeb DB |
999 | if (ret < 0) { |
1000 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
1001 | goto out_map; | |
1002 | } | |
1003 | ||
1004 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
1005 | if (ret < 0) | |
1006 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
1007 | out_map: | |
1008 | close(map_fd); | |
1009 | out_prog: | |
1010 | close(prog_fd); | |
1011 | return ret; | |
32e93fb7 DB |
1012 | } |
1013 | ||
fc4ccce0 DA |
1014 | int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type) |
1015 | { | |
1016 | union bpf_attr attr = {}; | |
1017 | ||
1018 | attr.target_fd = target_fd; | |
1019 | attr.attach_bpf_fd = prog_fd; | |
1020 | attr.attach_type = type; | |
1021 | ||
1022 | return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
1023 | } | |
1024 | ||
1025 | int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type) | |
1026 | { | |
1027 | union bpf_attr attr = {}; | |
1028 | ||
1029 | attr.target_fd = target_fd; | |
1030 | attr.attach_type = type; | |
1031 | ||
1032 | return bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); | |
1033 | } | |
1034 | ||
869d889e DA |
1035 | int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, |
1036 | size_t size_insns, const char *license, char *log, | |
1037 | size_t size_log) | |
1038 | { | |
1039 | union bpf_attr attr = {}; | |
1040 | ||
1041 | attr.prog_type = type; | |
1042 | attr.insns = bpf_ptr_to_u64(insns); | |
1043 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); | |
1044 | attr.license = bpf_ptr_to_u64(license); | |
1045 | ||
1046 | if (size_log > 0) { | |
1047 | attr.log_buf = bpf_ptr_to_u64(log); | |
1048 | attr.log_size = size_log; | |
1049 | attr.log_level = 1; | |
1050 | } | |
1051 | ||
1052 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); | |
1053 | } | |
1054 | ||
6256f8c9 | 1055 | #ifdef HAVE_ELF |
32e93fb7 DB |
1056 | struct bpf_elf_prog { |
1057 | enum bpf_prog_type type; | |
1058 | const struct bpf_insn *insns; | |
1059 | size_t size; | |
1060 | const char *license; | |
1061 | }; | |
1062 | ||
f6793eec DB |
1063 | struct bpf_hash_entry { |
1064 | unsigned int pinning; | |
1065 | const char *subpath; | |
1066 | struct bpf_hash_entry *next; | |
1067 | }; | |
1068 | ||
ecb05c0f DB |
1069 | struct bpf_config { |
1070 | unsigned int jit_enabled; | |
1071 | }; | |
1072 | ||
32e93fb7 | 1073 | struct bpf_elf_ctx { |
ecb05c0f | 1074 | struct bpf_config cfg; |
32e93fb7 DB |
1075 | Elf *elf_fd; |
1076 | GElf_Ehdr elf_hdr; | |
1077 | Elf_Data *sym_tab; | |
1078 | Elf_Data *str_tab; | |
1079 | int obj_fd; | |
1080 | int map_fds[ELF_MAX_MAPS]; | |
1081 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
ecb05c0f | 1082 | struct bpf_map_ext maps_ext[ELF_MAX_MAPS]; |
32e93fb7 DB |
1083 | int sym_num; |
1084 | int map_num; | |
e4225669 | 1085 | int map_len; |
32e93fb7 DB |
1086 | bool *sec_done; |
1087 | int sec_maps; | |
1088 | char license[ELF_MAX_LICENSE_LEN]; | |
1089 | enum bpf_prog_type type; | |
1090 | bool verbose; | |
1091 | struct bpf_elf_st stat; | |
f6793eec | 1092 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
1093 | char *log; |
1094 | size_t log_size; | |
32e93fb7 DB |
1095 | }; |
1096 | ||
6256f8c9 | 1097 | struct bpf_elf_sec_data { |
32e93fb7 DB |
1098 | GElf_Shdr sec_hdr; |
1099 | Elf_Data *sec_data; | |
1100 | const char *sec_name; | |
6256f8c9 DB |
1101 | }; |
1102 | ||
1103 | struct bpf_map_data { | |
32e93fb7 DB |
1104 | int *fds; |
1105 | const char *obj; | |
1106 | struct bpf_elf_st *st; | |
1107 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
1108 | }; |
1109 | ||
f31645d1 DB |
1110 | static __check_format_string(2, 3) void |
1111 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
1112 | { |
1113 | va_list vl; | |
1114 | ||
1115 | va_start(vl, format); | |
1116 | vfprintf(stderr, format, vl); | |
1117 | va_end(vl); | |
1118 | ||
f31645d1 | 1119 | if (ctx->log && ctx->log[0]) { |
afc1a200 DB |
1120 | if (ctx->verbose) { |
1121 | fprintf(stderr, "%s\n", ctx->log); | |
1122 | } else { | |
1123 | unsigned int off = 0, len = strlen(ctx->log); | |
1124 | ||
1125 | if (len > BPF_MAX_LOG) { | |
1126 | off = len - BPF_MAX_LOG; | |
1127 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
1128 | off); | |
1129 | } | |
1130 | fprintf(stderr, "%s\n", ctx->log + off); | |
1131 | } | |
1132 | ||
f31645d1 DB |
1133 | memset(ctx->log, 0, ctx->log_size); |
1134 | } | |
1135 | } | |
1136 | ||
1137 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
1138 | { | |
0f74d0f3 | 1139 | const size_t log_max = UINT_MAX >> 8; |
f31645d1 DB |
1140 | size_t log_size = ctx->log_size; |
1141 | void *ptr; | |
1142 | ||
1143 | if (!ctx->log) { | |
1144 | log_size = 65536; | |
0f74d0f3 | 1145 | } else if (log_size < log_max) { |
f31645d1 | 1146 | log_size <<= 1; |
0f74d0f3 TG |
1147 | if (log_size > log_max) |
1148 | log_size = log_max; | |
1149 | } else { | |
1150 | return -EINVAL; | |
d937a74b | 1151 | } |
f31645d1 DB |
1152 | |
1153 | ptr = realloc(ctx->log, log_size); | |
1154 | if (!ptr) | |
1155 | return -ENOMEM; | |
1156 | ||
1157 | ctx->log = ptr; | |
1158 | ctx->log_size = log_size; | |
1159 | ||
1160 | return 0; | |
11c39b5e DB |
1161 | } |
1162 | ||
4dd3f50a DB |
1163 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
1164 | uint32_t size_value, uint32_t max_elem, | |
612ff099 | 1165 | uint32_t flags, int inner_fd) |
11c39b5e | 1166 | { |
d17b136f | 1167 | union bpf_attr attr = {}; |
67584e3a | 1168 | |
67584e3a ND |
1169 | attr.map_type = type; |
1170 | attr.key_size = size_key; | |
612ff099 | 1171 | attr.value_size = inner_fd ? sizeof(int) : size_value; |
67584e3a | 1172 | attr.max_entries = max_elem; |
4dd3f50a | 1173 | attr.map_flags = flags; |
612ff099 | 1174 | attr.inner_map_fd = inner_fd; |
11c39b5e DB |
1175 | |
1176 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
1177 | } | |
11c39b5e | 1178 | |
32e93fb7 | 1179 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 1180 | { |
d17b136f | 1181 | union bpf_attr attr = {}; |
67584e3a | 1182 | |
67584e3a ND |
1183 | attr.pathname = bpf_ptr_to_u64(pathname); |
1184 | attr.bpf_fd = fd; | |
32e93fb7 DB |
1185 | |
1186 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
1187 | } | |
11c39b5e | 1188 | |
32e93fb7 DB |
1189 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
1190 | { | |
1191 | struct sockaddr_alg alg = { | |
1192 | .salg_family = AF_ALG, | |
1193 | .salg_type = "hash", | |
1194 | .salg_name = "sha1", | |
1195 | }; | |
1196 | int ret, cfd, ofd, ffd; | |
1197 | struct stat stbuff; | |
1198 | ssize_t size; | |
1199 | ||
1200 | if (!object || len != 20) | |
1201 | return -EINVAL; | |
1202 | ||
1203 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
1204 | if (cfd < 0) { | |
1205 | fprintf(stderr, "Cannot get AF_ALG socket: %s\n", | |
1206 | strerror(errno)); | |
1207 | return cfd; | |
1208 | } | |
1209 | ||
1210 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
1211 | if (ret < 0) { | |
1212 | fprintf(stderr, "Error binding socket: %s\n", strerror(errno)); | |
1213 | goto out_cfd; | |
1214 | } | |
1215 | ||
1216 | ofd = accept(cfd, NULL, 0); | |
1217 | if (ofd < 0) { | |
1218 | fprintf(stderr, "Error accepting socket: %s\n", | |
1219 | strerror(errno)); | |
1220 | ret = ofd; | |
1221 | goto out_cfd; | |
1222 | } | |
1223 | ||
1224 | ffd = open(object, O_RDONLY); | |
1225 | if (ffd < 0) { | |
1226 | fprintf(stderr, "Error opening object %s: %s\n", | |
1227 | object, strerror(errno)); | |
1228 | ret = ffd; | |
1229 | goto out_ofd; | |
1230 | } | |
1231 | ||
32a121cb | 1232 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
1233 | if (ret < 0) { |
1234 | fprintf(stderr, "Error doing fstat: %s\n", | |
1235 | strerror(errno)); | |
1236 | goto out_ffd; | |
d937a74b | 1237 | } |
11c39b5e | 1238 | |
32e93fb7 DB |
1239 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
1240 | if (size != stbuff.st_size) { | |
1241 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
1242 | size, stbuff.st_size, strerror(errno)); | |
1243 | ret = -1; | |
1244 | goto out_ffd; | |
1245 | } | |
1246 | ||
1247 | size = read(ofd, out, len); | |
1248 | if (size != len) { | |
1249 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
1250 | size, len, strerror(errno)); | |
1251 | ret = -1; | |
1252 | } else { | |
1253 | ret = 0; | |
1254 | } | |
1255 | out_ffd: | |
1256 | close(ffd); | |
1257 | out_ofd: | |
1258 | close(ofd); | |
1259 | out_cfd: | |
1260 | close(cfd); | |
1261 | return ret; | |
11c39b5e DB |
1262 | } |
1263 | ||
32e93fb7 | 1264 | static const char *bpf_get_obj_uid(const char *pathname) |
11c39b5e | 1265 | { |
32a121cb | 1266 | static bool bpf_uid_cached; |
32e93fb7 DB |
1267 | static char bpf_uid[64]; |
1268 | uint8_t tmp[20]; | |
1269 | int ret; | |
11c39b5e | 1270 | |
32e93fb7 DB |
1271 | if (bpf_uid_cached) |
1272 | goto done; | |
11c39b5e | 1273 | |
32e93fb7 DB |
1274 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); |
1275 | if (ret) { | |
1276 | fprintf(stderr, "Object hashing failed!\n"); | |
1277 | return NULL; | |
1278 | } | |
1279 | ||
1280 | hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid)); | |
1281 | bpf_uid_cached = true; | |
1282 | done: | |
1283 | return bpf_uid; | |
11c39b5e DB |
1284 | } |
1285 | ||
32e93fb7 DB |
1286 | static int bpf_init_env(const char *pathname) |
1287 | { | |
1288 | struct rlimit limit = { | |
1289 | .rlim_cur = RLIM_INFINITY, | |
1290 | .rlim_max = RLIM_INFINITY, | |
1291 | }; | |
1292 | ||
1293 | /* Don't bother in case we fail! */ | |
1294 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
1295 | ||
e4225669 | 1296 | if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) { |
32a121cb | 1297 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
32e93fb7 DB |
1298 | return 0; |
1299 | } | |
1300 | ||
1301 | if (!bpf_get_obj_uid(pathname)) | |
1302 | return -1; | |
1303 | ||
1304 | return 0; | |
6256f8c9 DB |
1305 | } |
1306 | ||
f6793eec DB |
1307 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
1308 | uint32_t pinning) | |
1309 | { | |
1310 | struct bpf_hash_entry *entry; | |
1311 | ||
1312 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1313 | while (entry && entry->pinning != pinning) | |
1314 | entry = entry->next; | |
1315 | ||
1316 | return entry ? entry->subpath : NULL; | |
1317 | } | |
1318 | ||
1319 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
1320 | uint32_t pinning) | |
11c39b5e | 1321 | { |
32e93fb7 DB |
1322 | switch (pinning) { |
1323 | case PIN_OBJECT_NS: | |
1324 | case PIN_GLOBAL_NS: | |
1325 | return false; | |
1326 | case PIN_NONE: | |
32e93fb7 | 1327 | return true; |
f6793eec DB |
1328 | default: |
1329 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
1330 | } |
1331 | } | |
1332 | ||
1333 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 1334 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
1335 | { |
1336 | switch (pinning) { | |
1337 | case PIN_OBJECT_NS: | |
e4225669 DB |
1338 | snprintf(pathname, len, "%s/%s/%s", |
1339 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1340 | bpf_get_obj_uid(NULL), name); |
1341 | break; | |
1342 | case PIN_GLOBAL_NS: | |
e4225669 DB |
1343 | snprintf(pathname, len, "%s/%s/%s", |
1344 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1345 | BPF_DIR_GLOBALS, name); |
1346 | break; | |
f6793eec | 1347 | default: |
e4225669 DB |
1348 | snprintf(pathname, len, "%s/../%s/%s", |
1349 | bpf_get_work_dir(ctx->type), | |
f6793eec DB |
1350 | bpf_custom_pinning(ctx, pinning), name); |
1351 | break; | |
32e93fb7 DB |
1352 | } |
1353 | } | |
1354 | ||
f6793eec DB |
1355 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1356 | uint32_t pinning) | |
32e93fb7 DB |
1357 | { |
1358 | char pathname[PATH_MAX]; | |
1359 | ||
e4225669 | 1360 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
32e93fb7 DB |
1361 | return 0; |
1362 | ||
f6793eec | 1363 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
e4225669 | 1364 | return bpf_obj_get(pathname, ctx->type); |
32e93fb7 DB |
1365 | } |
1366 | ||
e4225669 | 1367 | static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx) |
32e93fb7 | 1368 | { |
f6793eec | 1369 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1370 | int ret; |
1371 | ||
e4225669 | 1372 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type), |
f6793eec DB |
1373 | bpf_get_obj_uid(NULL)); |
1374 | ||
1375 | ret = mkdir(tmp, S_IRWXU); | |
1376 | if (ret && errno != EEXIST) { | |
1377 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1378 | return ret; | |
1379 | } | |
1380 | ||
1381 | return 0; | |
1382 | } | |
1383 | ||
e4225669 DB |
1384 | static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx, |
1385 | const char *todo) | |
f6793eec DB |
1386 | { |
1387 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1388 | int ret; | |
1389 | ||
e4225669 | 1390 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type)); |
f6793eec DB |
1391 | snprintf(rem, sizeof(rem), "%s/", todo); |
1392 | sub = strtok(rem, "/"); | |
32e93fb7 | 1393 | |
f6793eec DB |
1394 | while (sub) { |
1395 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1396 | return -EINVAL; | |
1397 | ||
1398 | strcat(tmp, sub); | |
1399 | strcat(tmp, "/"); | |
32e93fb7 | 1400 | |
f6793eec | 1401 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1402 | if (ret && errno != EEXIST) { |
f6793eec | 1403 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1404 | strerror(errno)); |
1405 | return ret; | |
1406 | } | |
f6793eec DB |
1407 | |
1408 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1409 | } |
1410 | ||
f6793eec DB |
1411 | return 0; |
1412 | } | |
1413 | ||
1414 | static int bpf_place_pinned(int fd, const char *name, | |
1415 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1416 | { | |
1417 | char pathname[PATH_MAX]; | |
1418 | const char *tmp; | |
1419 | int ret = 0; | |
1420 | ||
e4225669 | 1421 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
f6793eec DB |
1422 | return 0; |
1423 | ||
1424 | if (pinning == PIN_OBJECT_NS) | |
e4225669 | 1425 | ret = bpf_make_obj_path(ctx); |
f6793eec | 1426 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) |
e4225669 | 1427 | ret = bpf_make_custom_path(ctx, tmp); |
f6793eec DB |
1428 | if (ret < 0) |
1429 | return ret; | |
1430 | ||
1431 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1432 | return bpf_obj_pin(fd, pathname); |
1433 | } | |
1434 | ||
f31645d1 DB |
1435 | static void bpf_prog_report(int fd, const char *section, |
1436 | const struct bpf_elf_prog *prog, | |
1437 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1438 | { |
afc1a200 DB |
1439 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1440 | ||
1441 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1442 | fd < 0 ? "rejected: " : "loaded", |
1443 | fd < 0 ? strerror(errno) : "", | |
1444 | fd < 0 ? errno : fd); | |
1445 | ||
1446 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1447 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1448 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1449 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1450 | ||
1451 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1452 | } | |
32e93fb7 | 1453 | |
f31645d1 DB |
1454 | static int bpf_prog_attach(const char *section, |
1455 | const struct bpf_elf_prog *prog, | |
1456 | struct bpf_elf_ctx *ctx) | |
1457 | { | |
1458 | int tries = 0, fd; | |
1459 | retry: | |
32e93fb7 DB |
1460 | errno = 0; |
1461 | fd = bpf_prog_load(prog->type, prog->insns, prog->size, | |
f31645d1 DB |
1462 | prog->license, ctx->log, ctx->log_size); |
1463 | if (fd < 0 || ctx->verbose) { | |
1464 | /* The verifier log is pretty chatty, sometimes so chatty | |
1465 | * on larger programs, that we could fail to dump everything | |
1466 | * into our buffer. Still, try to give a debuggable error | |
1467 | * log for the user, so enlarge it and re-fail. | |
1468 | */ | |
1469 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
0f74d0f3 | 1470 | if (tries++ < 10 && !bpf_log_realloc(ctx)) |
f31645d1 DB |
1471 | goto retry; |
1472 | ||
32a121cb | 1473 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1474 | ctx->log_size, tries); |
1475 | return fd; | |
1476 | } | |
1477 | ||
1478 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1479 | } |
1480 | ||
1481 | return fd; | |
1482 | } | |
1483 | ||
f31645d1 DB |
1484 | static void bpf_map_report(int fd, const char *name, |
1485 | const struct bpf_elf_map *map, | |
612ff099 | 1486 | struct bpf_elf_ctx *ctx, int inner_fd) |
f31645d1 DB |
1487 | { |
1488 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1489 | fd < 0 ? "rejected: " : "loaded", | |
1490 | fd < 0 ? strerror(errno) : "", | |
1491 | fd < 0 ? errno : fd); | |
1492 | ||
1493 | fprintf(stderr, " - Type: %u\n", map->type); | |
1494 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1495 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1496 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
612ff099 DB |
1497 | fprintf(stderr, " - Size value: %u\n", |
1498 | inner_fd ? (int)sizeof(int) : map->size_value); | |
4dd3f50a DB |
1499 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1500 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1501 | } |
1502 | ||
612ff099 DB |
1503 | static int bpf_find_map_id(const struct bpf_elf_ctx *ctx, uint32_t id) |
1504 | { | |
1505 | int i; | |
1506 | ||
1507 | for (i = 0; i < ctx->map_num; i++) { | |
1508 | if (ctx->maps[i].id != id) | |
1509 | continue; | |
1510 | if (ctx->map_fds[i] < 0) | |
1511 | return -EINVAL; | |
1512 | ||
1513 | return ctx->map_fds[i]; | |
1514 | } | |
1515 | ||
1516 | return -ENOENT; | |
1517 | } | |
1518 | ||
0b5eadc5 | 1519 | static void bpf_report_map_in_map(int outer_fd, uint32_t idx) |
612ff099 DB |
1520 | { |
1521 | struct bpf_elf_map outer_map; | |
1522 | int ret; | |
1523 | ||
1524 | fprintf(stderr, "Cannot insert map into map! "); | |
1525 | ||
ecb05c0f | 1526 | ret = bpf_derive_elf_map_from_fdinfo(outer_fd, &outer_map, NULL); |
612ff099 DB |
1527 | if (!ret) { |
1528 | if (idx >= outer_map.max_elem && | |
1529 | outer_map.type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
1530 | fprintf(stderr, "Outer map has %u elements, index %u is invalid!\n", | |
1531 | outer_map.max_elem, idx); | |
1532 | return; | |
1533 | } | |
1534 | } | |
1535 | ||
1536 | fprintf(stderr, "Different map specs used for outer and inner map?\n"); | |
1537 | } | |
1538 | ||
1539 | static bool bpf_is_map_in_map_type(const struct bpf_elf_map *map) | |
1540 | { | |
1541 | return map->type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1542 | map->type == BPF_MAP_TYPE_HASH_OF_MAPS; | |
1543 | } | |
1544 | ||
ecb05c0f DB |
1545 | static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx, |
1546 | const struct bpf_elf_map *map, struct bpf_map_ext *ext, | |
1547 | int *have_map_in_map) | |
32e93fb7 | 1548 | { |
612ff099 | 1549 | int fd, ret, map_inner_fd = 0; |
32e93fb7 | 1550 | |
f6793eec | 1551 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1552 | if (fd > 0) { |
ecb05c0f | 1553 | ret = bpf_map_selfcheck_pinned(fd, map, ext, |
91d88eeb | 1554 | offsetof(struct bpf_elf_map, |
fb24802b | 1555 | id), ctx->type); |
9e607f2e DB |
1556 | if (ret < 0) { |
1557 | close(fd); | |
1558 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1559 | name); | |
1560 | return ret; | |
1561 | } | |
f31645d1 | 1562 | if (ctx->verbose) |
32e93fb7 DB |
1563 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1564 | name); | |
1565 | return fd; | |
1566 | } | |
1567 | ||
612ff099 DB |
1568 | if (have_map_in_map && bpf_is_map_in_map_type(map)) { |
1569 | (*have_map_in_map)++; | |
1570 | if (map->inner_id) | |
1571 | return 0; | |
1572 | fprintf(stderr, "Map \'%s\' cannot be created since no inner map ID defined!\n", | |
1573 | name); | |
1574 | return -EINVAL; | |
1575 | } | |
1576 | ||
1577 | if (!have_map_in_map && bpf_is_map_in_map_type(map)) { | |
1578 | map_inner_fd = bpf_find_map_id(ctx, map->inner_id); | |
1579 | if (map_inner_fd < 0) { | |
1580 | fprintf(stderr, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n", | |
1581 | name, map->inner_id); | |
1582 | return -EINVAL; | |
1583 | } | |
1584 | } | |
1585 | ||
32e93fb7 DB |
1586 | errno = 0; |
1587 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
612ff099 | 1588 | map->max_elem, map->flags, map_inner_fd); |
f31645d1 | 1589 | if (fd < 0 || ctx->verbose) { |
612ff099 | 1590 | bpf_map_report(fd, name, map, ctx, map_inner_fd); |
32e93fb7 DB |
1591 | if (fd < 0) |
1592 | return fd; | |
1593 | } | |
1594 | ||
f6793eec | 1595 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1596 | if (ret < 0 && errno != EEXIST) { |
1597 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1598 | strerror(errno)); | |
1599 | close(fd); | |
1600 | return ret; | |
1601 | } | |
1602 | ||
1603 | return fd; | |
1604 | } | |
1605 | ||
32e93fb7 DB |
1606 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1607 | const GElf_Sym *sym) | |
1608 | { | |
1609 | return ctx->str_tab->d_buf + sym->st_name; | |
1610 | } | |
1611 | ||
1612 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) | |
1613 | { | |
1614 | GElf_Sym sym; | |
11c39b5e DB |
1615 | int i; |
1616 | ||
32e93fb7 DB |
1617 | for (i = 0; i < ctx->sym_num; i++) { |
1618 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1619 | continue; | |
1620 | ||
5230a2ed DB |
1621 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
1622 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
32e93fb7 | 1623 | sym.st_shndx != ctx->sec_maps || |
e4225669 | 1624 | sym.st_value / ctx->map_len != which) |
32e93fb7 DB |
1625 | continue; |
1626 | ||
1627 | return bpf_str_tab_name(ctx, &sym); | |
11c39b5e | 1628 | } |
32e93fb7 DB |
1629 | |
1630 | return NULL; | |
11c39b5e DB |
1631 | } |
1632 | ||
32e93fb7 | 1633 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1634 | { |
612ff099 | 1635 | int i, j, ret, fd, inner_fd, inner_idx, have_map_in_map = 0; |
32e93fb7 | 1636 | const char *map_name; |
11c39b5e | 1637 | |
32e93fb7 DB |
1638 | for (i = 0; i < ctx->map_num; i++) { |
1639 | map_name = bpf_map_fetch_name(ctx, i); | |
1640 | if (!map_name) | |
1641 | return -EIO; | |
11c39b5e | 1642 | |
ecb05c0f DB |
1643 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1644 | &ctx->maps_ext[i], &have_map_in_map); | |
612ff099 DB |
1645 | if (fd < 0) |
1646 | return fd; | |
1647 | ||
1648 | ctx->map_fds[i] = !fd ? -1 : fd; | |
1649 | } | |
1650 | ||
1651 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { | |
1652 | if (ctx->map_fds[i] >= 0) | |
1653 | continue; | |
1654 | ||
1655 | map_name = bpf_map_fetch_name(ctx, i); | |
1656 | if (!map_name) | |
1657 | return -EIO; | |
1658 | ||
ecb05c0f DB |
1659 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1660 | &ctx->maps_ext[i], NULL); | |
32e93fb7 DB |
1661 | if (fd < 0) |
1662 | return fd; | |
11c39b5e | 1663 | |
32e93fb7 | 1664 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1665 | } |
1666 | ||
612ff099 DB |
1667 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { |
1668 | if (!ctx->maps[i].id || | |
1669 | ctx->maps[i].inner_id || | |
1670 | ctx->maps[i].inner_idx == -1) | |
1671 | continue; | |
1672 | ||
1673 | inner_fd = ctx->map_fds[i]; | |
1674 | inner_idx = ctx->maps[i].inner_idx; | |
1675 | ||
1676 | for (j = 0; j < ctx->map_num; j++) { | |
1677 | if (!bpf_is_map_in_map_type(&ctx->maps[j])) | |
1678 | continue; | |
1679 | if (ctx->maps[j].inner_id != ctx->maps[i].id) | |
1680 | continue; | |
1681 | ||
1682 | ret = bpf_map_update(ctx->map_fds[j], &inner_idx, | |
1683 | &inner_fd, BPF_ANY); | |
1684 | if (ret < 0) { | |
1685 | bpf_report_map_in_map(ctx->map_fds[j], | |
0b5eadc5 | 1686 | inner_idx); |
612ff099 DB |
1687 | return ret; |
1688 | } | |
1689 | } | |
1690 | } | |
1691 | ||
11c39b5e | 1692 | return 0; |
11c39b5e DB |
1693 | } |
1694 | ||
e4225669 DB |
1695 | static int bpf_map_num_sym(struct bpf_elf_ctx *ctx) |
1696 | { | |
1697 | int i, num = 0; | |
1698 | GElf_Sym sym; | |
1699 | ||
1700 | for (i = 0; i < ctx->sym_num; i++) { | |
1701 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1702 | continue; | |
1703 | ||
1704 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1705 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1706 | sym.st_shndx != ctx->sec_maps) | |
1707 | continue; | |
1708 | num++; | |
1709 | } | |
1710 | ||
1711 | return num; | |
1712 | } | |
1713 | ||
32e93fb7 DB |
1714 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1715 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1716 | { |
32e93fb7 | 1717 | Elf_Data *sec_edata; |
11c39b5e DB |
1718 | GElf_Shdr sec_hdr; |
1719 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1720 | char *sec_name; |
1721 | ||
32e93fb7 | 1722 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1723 | |
32e93fb7 | 1724 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1725 | if (!sec_fd) |
1726 | return -EINVAL; | |
11c39b5e DB |
1727 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1728 | return -EIO; | |
1729 | ||
32e93fb7 | 1730 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1731 | sec_hdr.sh_name); |
1732 | if (!sec_name || !sec_hdr.sh_size) | |
1733 | return -ENOENT; | |
1734 | ||
1735 | sec_edata = elf_getdata(sec_fd, NULL); | |
1736 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1737 | return -EIO; | |
1738 | ||
32e93fb7 | 1739 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1740 | |
32e93fb7 DB |
1741 | data->sec_name = sec_name; |
1742 | data->sec_data = sec_edata; | |
11c39b5e DB |
1743 | return 0; |
1744 | } | |
1745 | ||
e4225669 DB |
1746 | struct bpf_elf_map_min { |
1747 | __u32 type; | |
1748 | __u32 size_key; | |
1749 | __u32 size_value; | |
1750 | __u32 max_elem; | |
1751 | }; | |
11c39b5e | 1752 | |
e4225669 DB |
1753 | static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section, |
1754 | struct bpf_elf_sec_data *data) | |
1755 | { | |
1756 | ctx->map_num = data->sec_data->d_size; | |
32e93fb7 DB |
1757 | ctx->sec_maps = section; |
1758 | ctx->sec_done[section] = true; | |
11c39b5e | 1759 | |
e4225669 | 1760 | if (ctx->map_num > sizeof(ctx->maps)) { |
32e93fb7 DB |
1761 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); |
1762 | return -ENOMEM; | |
1763 | } | |
11c39b5e | 1764 | |
e4225669 DB |
1765 | memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num); |
1766 | return 0; | |
1767 | } | |
1768 | ||
1769 | static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end) | |
1770 | { | |
1771 | GElf_Sym sym; | |
1772 | int off, i; | |
1773 | ||
1774 | for (off = 0; off < end; off += ctx->map_len) { | |
1775 | /* Order doesn't need to be linear here, hence we walk | |
1776 | * the table again. | |
1777 | */ | |
1778 | for (i = 0; i < ctx->sym_num; i++) { | |
1779 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1780 | continue; | |
1781 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1782 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1783 | sym.st_shndx != ctx->sec_maps) | |
1784 | continue; | |
1785 | if (sym.st_value == off) | |
1786 | break; | |
1787 | if (i == ctx->sym_num - 1) | |
1788 | return -1; | |
1789 | } | |
1790 | } | |
1791 | ||
1792 | return off == end ? 0 : -1; | |
1793 | } | |
1794 | ||
1795 | static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx) | |
1796 | { | |
1797 | struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {}; | |
1798 | int i, sym_num = bpf_map_num_sym(ctx); | |
1799 | __u8 *buff; | |
1800 | ||
1801 | if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) { | |
1802 | fprintf(stderr, "%u maps not supported in current map section!\n", | |
1803 | sym_num); | |
1804 | return -EINVAL; | |
1805 | } | |
1806 | ||
1807 | if (ctx->map_num % sym_num != 0 || | |
1808 | ctx->map_num % sizeof(__u32) != 0) { | |
1809 | fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n"); | |
1810 | return -EINVAL; | |
1811 | } | |
1812 | ||
1813 | ctx->map_len = ctx->map_num / sym_num; | |
1814 | if (bpf_map_verify_all_offs(ctx, ctx->map_num)) { | |
1815 | fprintf(stderr, "Different struct bpf_elf_map in use!\n"); | |
1816 | return -EINVAL; | |
1817 | } | |
1818 | ||
1819 | if (ctx->map_len == sizeof(struct bpf_elf_map)) { | |
1820 | ctx->map_num = sym_num; | |
1821 | return 0; | |
1822 | } else if (ctx->map_len > sizeof(struct bpf_elf_map)) { | |
1823 | fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n"); | |
1824 | return -EINVAL; | |
1825 | } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) { | |
1826 | fprintf(stderr, "struct bpf_elf_map too small, not supported!\n"); | |
1827 | return -EINVAL; | |
1828 | } | |
1829 | ||
1830 | ctx->map_num = sym_num; | |
1831 | for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num; | |
1832 | i++, buff += ctx->map_len) { | |
1833 | /* The fixup leaves the rest of the members as zero, which | |
1834 | * is fine currently, but option exist to set some other | |
1835 | * default value as well when needed in future. | |
1836 | */ | |
1837 | memcpy(&fixup[i], buff, ctx->map_len); | |
1838 | } | |
1839 | ||
1840 | memcpy(ctx->maps, fixup, sizeof(fixup)); | |
1841 | ||
1842 | printf("Note: %zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n", | |
1843 | sizeof(struct bpf_elf_map) - ctx->map_len); | |
32e93fb7 DB |
1844 | return 0; |
1845 | } | |
11c39b5e | 1846 | |
32e93fb7 DB |
1847 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
1848 | struct bpf_elf_sec_data *data) | |
1849 | { | |
1850 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
1851 | return -ENOMEM; | |
11c39b5e | 1852 | |
32e93fb7 DB |
1853 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
1854 | ctx->sec_done[section] = true; | |
1855 | return 0; | |
1856 | } | |
11c39b5e | 1857 | |
32e93fb7 DB |
1858 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
1859 | struct bpf_elf_sec_data *data) | |
1860 | { | |
1861 | ctx->sym_tab = data->sec_data; | |
1862 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
1863 | ctx->sec_done[section] = true; | |
11c39b5e DB |
1864 | return 0; |
1865 | } | |
1866 | ||
32e93fb7 DB |
1867 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
1868 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1869 | { |
32e93fb7 DB |
1870 | ctx->str_tab = data->sec_data; |
1871 | ctx->sec_done[section] = true; | |
1872 | return 0; | |
1873 | } | |
11c39b5e | 1874 | |
afc1a200 DB |
1875 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
1876 | { | |
1877 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
1878 | } | |
1879 | ||
32e93fb7 DB |
1880 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx) |
1881 | { | |
1882 | struct bpf_elf_sec_data data; | |
1883 | int i, ret = -1; | |
11c39b5e | 1884 | |
32e93fb7 DB |
1885 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1886 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
1887 | if (ret < 0) |
1888 | continue; | |
1889 | ||
cce3d466 DB |
1890 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1891 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
e4225669 | 1892 | ret = bpf_fetch_maps_begin(ctx, i, &data); |
cce3d466 DB |
1893 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1894 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 1895 | ret = bpf_fetch_license(ctx, i, &data); |
cce3d466 DB |
1896 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
1897 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
1898 | ret = bpf_fetch_symtab(ctx, i, &data); |
1899 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 1900 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 DB |
1901 | ret = bpf_fetch_strtab(ctx, i, &data); |
1902 | if (ret < 0) { | |
afc1a200 | 1903 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 1904 | i); |
e4225669 | 1905 | return ret; |
11c39b5e | 1906 | } |
32e93fb7 DB |
1907 | } |
1908 | ||
afc1a200 | 1909 | if (bpf_has_map_data(ctx)) { |
e4225669 DB |
1910 | ret = bpf_fetch_maps_end(ctx); |
1911 | if (ret < 0) { | |
1912 | fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n"); | |
1913 | return ret; | |
1914 | } | |
1915 | ||
32e93fb7 DB |
1916 | ret = bpf_maps_attach_all(ctx); |
1917 | if (ret < 0) { | |
1918 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
1919 | return ret; | |
11c39b5e DB |
1920 | } |
1921 | } | |
1922 | ||
1923 | return ret; | |
1924 | } | |
1925 | ||
e4225669 DB |
1926 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section, |
1927 | bool *sseen) | |
11c39b5e | 1928 | { |
32e93fb7 DB |
1929 | struct bpf_elf_sec_data data; |
1930 | struct bpf_elf_prog prog; | |
1931 | int ret, i, fd = -1; | |
11c39b5e | 1932 | |
32e93fb7 DB |
1933 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1934 | if (ctx->sec_done[i]) | |
11c39b5e DB |
1935 | continue; |
1936 | ||
32e93fb7 | 1937 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
1938 | if (ret < 0 || |
1939 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
1940 | data.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1941 | !strcmp(data.sec_name, section))) | |
11c39b5e DB |
1942 | continue; |
1943 | ||
e4225669 DB |
1944 | *sseen = true; |
1945 | ||
32e93fb7 DB |
1946 | memset(&prog, 0, sizeof(prog)); |
1947 | prog.type = ctx->type; | |
1948 | prog.insns = data.sec_data->d_buf; | |
1949 | prog.size = data.sec_data->d_size; | |
1950 | prog.license = ctx->license; | |
11c39b5e | 1951 | |
f31645d1 | 1952 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1953 | if (fd < 0) |
e4225669 | 1954 | return fd; |
11c39b5e | 1955 | |
32e93fb7 | 1956 | ctx->sec_done[i] = true; |
11c39b5e DB |
1957 | break; |
1958 | } | |
1959 | ||
32e93fb7 | 1960 | return fd; |
11c39b5e DB |
1961 | } |
1962 | ||
ecb05c0f DB |
1963 | struct bpf_tail_call_props { |
1964 | unsigned int total; | |
1965 | unsigned int jited; | |
1966 | }; | |
1967 | ||
32e93fb7 DB |
1968 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
1969 | struct bpf_elf_sec_data *data_relo, | |
ecb05c0f DB |
1970 | struct bpf_elf_sec_data *data_insn, |
1971 | struct bpf_tail_call_props *props) | |
11c39b5e | 1972 | { |
32e93fb7 DB |
1973 | Elf_Data *idata = data_insn->sec_data; |
1974 | GElf_Shdr *rhdr = &data_relo->sec_hdr; | |
1975 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
1976 | struct bpf_insn *insns = idata->d_buf; | |
1977 | unsigned int num_insns = idata->d_size / sizeof(*insns); | |
11c39b5e | 1978 | |
32e93fb7 DB |
1979 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
1980 | unsigned int ioff, rmap; | |
1981 | GElf_Rel relo; | |
1982 | GElf_Sym sym; | |
1983 | ||
1984 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
1985 | return -EIO; | |
1986 | ||
1987 | ioff = relo.r_offset / sizeof(struct bpf_insn); | |
1988 | if (ioff >= num_insns || | |
a576c6b9 | 1989 | insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) { |
32a121cb | 1990 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", |
a576c6b9 DB |
1991 | ioff); |
1992 | if (ioff < num_insns && | |
1993 | insns[ioff].code == (BPF_JMP | BPF_CALL)) | |
32a121cb | 1994 | fprintf(stderr, " - Try to annotate functions with always_inline attribute!\n"); |
32e93fb7 | 1995 | return -EINVAL; |
a576c6b9 | 1996 | } |
32e93fb7 DB |
1997 | |
1998 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) | |
1999 | return -EIO; | |
2486337a | 2000 | if (sym.st_shndx != ctx->sec_maps) { |
32a121cb | 2001 | fprintf(stderr, "ELF contains non-map related relo data in entry %u pointing to section %u! Compiler bug?!\n", |
2486337a DB |
2002 | relo_ent, sym.st_shndx); |
2003 | return -EIO; | |
2004 | } | |
32e93fb7 | 2005 | |
e4225669 | 2006 | rmap = sym.st_value / ctx->map_len; |
32e93fb7 DB |
2007 | if (rmap >= ARRAY_SIZE(ctx->map_fds)) |
2008 | return -EINVAL; | |
2009 | if (!ctx->map_fds[rmap]) | |
2010 | return -EINVAL; | |
ecb05c0f DB |
2011 | if (ctx->maps[rmap].type == BPF_MAP_TYPE_PROG_ARRAY) { |
2012 | props->total++; | |
2013 | if (ctx->maps_ext[rmap].owner.jited || | |
2014 | (ctx->maps_ext[rmap].owner.type == 0 && | |
2015 | ctx->cfg.jit_enabled)) | |
2016 | props->jited++; | |
2017 | } | |
32e93fb7 DB |
2018 | |
2019 | if (ctx->verbose) | |
32a121cb | 2020 | fprintf(stderr, "Map \'%s\' (%d) injected into prog section \'%s\' at offset %u!\n", |
32e93fb7 DB |
2021 | bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap], |
2022 | data_insn->sec_name, ioff); | |
11c39b5e | 2023 | |
32e93fb7 DB |
2024 | insns[ioff].src_reg = BPF_PSEUDO_MAP_FD; |
2025 | insns[ioff].imm = ctx->map_fds[rmap]; | |
2026 | } | |
2027 | ||
2028 | return 0; | |
2029 | } | |
2030 | ||
afc1a200 | 2031 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
e4225669 | 2032 | bool *lderr, bool *sseen) |
32e93fb7 DB |
2033 | { |
2034 | struct bpf_elf_sec_data data_relo, data_insn; | |
2035 | struct bpf_elf_prog prog; | |
2036 | int ret, idx, i, fd = -1; | |
2037 | ||
2038 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
ecb05c0f DB |
2039 | struct bpf_tail_call_props props = {}; |
2040 | ||
32e93fb7 DB |
2041 | ret = bpf_fill_section_data(ctx, i, &data_relo); |
2042 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
2043 | continue; |
2044 | ||
32e93fb7 | 2045 | idx = data_relo.sec_hdr.sh_info; |
e4225669 | 2046 | |
32e93fb7 | 2047 | ret = bpf_fill_section_data(ctx, idx, &data_insn); |
cce3d466 DB |
2048 | if (ret < 0 || |
2049 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
2050 | data_insn.sec_hdr.sh_flags & SHF_EXECINSTR && | |
2051 | !strcmp(data_insn.sec_name, section))) | |
11c39b5e | 2052 | continue; |
32e93fb7 | 2053 | |
e4225669 DB |
2054 | *sseen = true; |
2055 | ||
ecb05c0f | 2056 | ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn, &props); |
c9c3720d DB |
2057 | if (ret < 0) { |
2058 | *lderr = true; | |
e4225669 | 2059 | return ret; |
c9c3720d | 2060 | } |
11c39b5e | 2061 | |
32e93fb7 DB |
2062 | memset(&prog, 0, sizeof(prog)); |
2063 | prog.type = ctx->type; | |
2064 | prog.insns = data_insn.sec_data->d_buf; | |
2065 | prog.size = data_insn.sec_data->d_size; | |
2066 | prog.license = ctx->license; | |
2067 | ||
f31645d1 | 2068 | fd = bpf_prog_attach(section, &prog, ctx); |
afc1a200 DB |
2069 | if (fd < 0) { |
2070 | *lderr = true; | |
ecb05c0f DB |
2071 | if (props.total) { |
2072 | if (ctx->cfg.jit_enabled && | |
2073 | props.total != props.jited) | |
2074 | fprintf(stderr, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n", | |
2075 | props.jited, props.total); | |
2076 | if (!ctx->cfg.jit_enabled && | |
2077 | props.jited) | |
2078 | fprintf(stderr, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n", | |
2079 | props.jited, props.total); | |
2080 | } | |
e4225669 | 2081 | return fd; |
afc1a200 | 2082 | } |
11c39b5e | 2083 | |
32e93fb7 DB |
2084 | ctx->sec_done[i] = true; |
2085 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
2086 | break; |
2087 | } | |
2088 | ||
32e93fb7 | 2089 | return fd; |
11c39b5e DB |
2090 | } |
2091 | ||
32e93fb7 | 2092 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 2093 | { |
e4225669 | 2094 | bool lderr = false, sseen = false; |
473d7840 DB |
2095 | int ret = -1; |
2096 | ||
afc1a200 | 2097 | if (bpf_has_map_data(ctx)) |
e4225669 | 2098 | ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen); |
afc1a200 | 2099 | if (ret < 0 && !lderr) |
e4225669 DB |
2100 | ret = bpf_fetch_prog(ctx, section, &sseen); |
2101 | if (ret < 0 && !sseen) | |
2102 | fprintf(stderr, "Program section \'%s\' not found in ELF file!\n", | |
2103 | section); | |
473d7840 DB |
2104 | return ret; |
2105 | } | |
2106 | ||
910b543d DB |
2107 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
2108 | { | |
2109 | int i; | |
2110 | ||
2111 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
2112 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
2113 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
2114 | return i; | |
2115 | return -1; | |
2116 | } | |
2117 | ||
ecb05c0f DB |
2118 | struct bpf_jited_aux { |
2119 | int prog_fd; | |
2120 | int map_fd; | |
2121 | struct bpf_prog_data prog; | |
2122 | struct bpf_map_ext map; | |
2123 | }; | |
2124 | ||
2125 | static int bpf_derive_prog_from_fdinfo(int fd, struct bpf_prog_data *prog) | |
2126 | { | |
2127 | char file[PATH_MAX], buff[4096]; | |
2128 | unsigned int val; | |
2129 | FILE *fp; | |
2130 | ||
2131 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
2132 | memset(prog, 0, sizeof(*prog)); | |
2133 | ||
2134 | fp = fopen(file, "r"); | |
2135 | if (!fp) { | |
2136 | fprintf(stderr, "No procfs support?!\n"); | |
2137 | return -EIO; | |
2138 | } | |
2139 | ||
2140 | while (fgets(buff, sizeof(buff), fp)) { | |
2141 | if (sscanf(buff, "prog_type:\t%u", &val) == 1) | |
2142 | prog->type = val; | |
2143 | else if (sscanf(buff, "prog_jited:\t%u", &val) == 1) | |
2144 | prog->jited = val; | |
2145 | } | |
2146 | ||
2147 | fclose(fp); | |
2148 | return 0; | |
2149 | } | |
2150 | ||
2151 | static int bpf_tail_call_get_aux(struct bpf_jited_aux *aux) | |
2152 | { | |
2153 | struct bpf_elf_map tmp; | |
2154 | int ret; | |
2155 | ||
2156 | ret = bpf_derive_elf_map_from_fdinfo(aux->map_fd, &tmp, &aux->map); | |
2157 | if (!ret) | |
2158 | ret = bpf_derive_prog_from_fdinfo(aux->prog_fd, &aux->prog); | |
2159 | ||
2160 | return ret; | |
2161 | } | |
2162 | ||
32e93fb7 | 2163 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 2164 | { |
32e93fb7 DB |
2165 | struct bpf_elf_sec_data data; |
2166 | uint32_t map_id, key_id; | |
910b543d | 2167 | int fd, i, ret, idx; |
473d7840 | 2168 | |
32e93fb7 DB |
2169 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2170 | if (ctx->sec_done[i]) | |
473d7840 DB |
2171 | continue; |
2172 | ||
32e93fb7 | 2173 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
2174 | if (ret < 0) |
2175 | continue; | |
2176 | ||
910b543d DB |
2177 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
2178 | if (ret != 2) | |
32e93fb7 | 2179 | continue; |
910b543d DB |
2180 | |
2181 | idx = bpf_find_map_by_id(ctx, map_id); | |
2182 | if (idx < 0) | |
473d7840 DB |
2183 | continue; |
2184 | ||
32e93fb7 DB |
2185 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
2186 | if (fd < 0) | |
473d7840 DB |
2187 | return -EIO; |
2188 | ||
910b543d DB |
2189 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
2190 | &fd, BPF_ANY); | |
afc1a200 | 2191 | if (ret < 0) { |
ecb05c0f DB |
2192 | struct bpf_jited_aux aux = {}; |
2193 | ||
2194 | ret = -errno; | |
2195 | if (errno == E2BIG) { | |
afc1a200 DB |
2196 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", |
2197 | key_id, map_id); | |
ecb05c0f DB |
2198 | return ret; |
2199 | } | |
2200 | ||
2201 | aux.map_fd = ctx->map_fds[idx]; | |
2202 | aux.prog_fd = fd; | |
2203 | ||
2204 | if (bpf_tail_call_get_aux(&aux)) | |
2205 | return ret; | |
2206 | if (!aux.map.owner.type) | |
2207 | return ret; | |
2208 | ||
2209 | if (aux.prog.type != aux.map.owner.type) | |
2210 | fprintf(stderr, "Tail call map owned by prog type %u, but prog type is %u!\n", | |
2211 | aux.map.owner.type, aux.prog.type); | |
2212 | if (aux.prog.jited != aux.map.owner.jited) | |
2213 | fprintf(stderr, "Tail call map %s jited, but prog %s!\n", | |
2214 | aux.map.owner.jited ? "is" : "not", | |
2215 | aux.prog.jited ? "is" : "not"); | |
2216 | return ret; | |
afc1a200 | 2217 | } |
473d7840 | 2218 | |
32e93fb7 | 2219 | ctx->sec_done[i] = true; |
473d7840 DB |
2220 | } |
2221 | ||
2222 | return 0; | |
2223 | } | |
2224 | ||
32e93fb7 | 2225 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 2226 | { |
32e93fb7 DB |
2227 | struct stat st; |
2228 | int ret; | |
11c39b5e | 2229 | |
32e93fb7 | 2230 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 2231 | |
32e93fb7 DB |
2232 | ret = fstat(ctx->obj_fd, &st); |
2233 | if (ret < 0) { | |
2234 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
2235 | strerror(errno)); | |
2236 | return; | |
2237 | } | |
11c39b5e | 2238 | |
32e93fb7 DB |
2239 | ctx->stat.st_dev = st.st_dev; |
2240 | ctx->stat.st_ino = st.st_ino; | |
2241 | } | |
2242 | ||
f6793eec DB |
2243 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
2244 | { | |
2245 | char buff[PATH_MAX]; | |
2246 | ||
2247 | while (fgets(buff, sizeof(buff), fp)) { | |
2248 | char *ptr = buff; | |
2249 | ||
2250 | while (*ptr == ' ' || *ptr == '\t') | |
2251 | ptr++; | |
2252 | ||
2253 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
2254 | continue; | |
2255 | ||
2256 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
2257 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
2258 | strcpy(path, ptr); | |
2259 | return -1; | |
2260 | } | |
2261 | ||
2262 | return 1; | |
2263 | } | |
2264 | ||
2265 | return 0; | |
2266 | } | |
2267 | ||
2268 | static bool bpf_pinning_reserved(uint32_t pinning) | |
2269 | { | |
2270 | switch (pinning) { | |
2271 | case PIN_NONE: | |
2272 | case PIN_OBJECT_NS: | |
2273 | case PIN_GLOBAL_NS: | |
2274 | return true; | |
2275 | default: | |
2276 | return false; | |
2277 | } | |
2278 | } | |
2279 | ||
2280 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
2281 | { | |
2282 | struct bpf_hash_entry *entry; | |
d17b136f | 2283 | char subpath[PATH_MAX] = {}; |
f6793eec DB |
2284 | uint32_t pinning; |
2285 | FILE *fp; | |
2286 | int ret; | |
2287 | ||
2288 | fp = fopen(db_file, "r"); | |
2289 | if (!fp) | |
2290 | return; | |
2291 | ||
f6793eec DB |
2292 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { |
2293 | if (ret == -1) { | |
2294 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
2295 | db_file, subpath); | |
2296 | fclose(fp); | |
2297 | return; | |
2298 | } | |
2299 | ||
2300 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
2301 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
2302 | db_file, pinning); | |
f6793eec DB |
2303 | continue; |
2304 | } | |
2305 | ||
2306 | entry = malloc(sizeof(*entry)); | |
2307 | if (!entry) { | |
2308 | fprintf(stderr, "No memory left for db entry!\n"); | |
2309 | continue; | |
2310 | } | |
2311 | ||
2312 | entry->pinning = pinning; | |
2313 | entry->subpath = strdup(subpath); | |
2314 | if (!entry->subpath) { | |
2315 | fprintf(stderr, "No memory left for db entry!\n"); | |
2316 | free(entry); | |
2317 | continue; | |
2318 | } | |
2319 | ||
2320 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
2321 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
2322 | } | |
2323 | ||
2324 | fclose(fp); | |
2325 | } | |
2326 | ||
2327 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
2328 | { | |
2329 | struct bpf_hash_entry *entry; | |
2330 | int i; | |
2331 | ||
2332 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
2333 | while ((entry = ctx->ht[i]) != NULL) { | |
2334 | ctx->ht[i] = entry->next; | |
2335 | free((char *)entry->subpath); | |
2336 | free(entry); | |
2337 | } | |
2338 | } | |
2339 | } | |
2340 | ||
8187b012 DB |
2341 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
2342 | { | |
2343 | if (ctx->elf_hdr.e_type != ET_REL || | |
e77fa41d DB |
2344 | (ctx->elf_hdr.e_machine != EM_NONE && |
2345 | ctx->elf_hdr.e_machine != EM_BPF) || | |
8187b012 DB |
2346 | ctx->elf_hdr.e_version != EV_CURRENT) { |
2347 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
2348 | return -EINVAL; | |
2349 | } | |
2350 | ||
2351 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
2352 | default: | |
2353 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
2354 | return -EINVAL; | |
2355 | case ELFDATA2LSB: | |
2356 | if (htons(1) == 1) { | |
2357 | fprintf(stderr, | |
2358 | "We are big endian, eBPF object is little endian!\n"); | |
2359 | return -EIO; | |
2360 | } | |
2361 | break; | |
2362 | case ELFDATA2MSB: | |
2363 | if (htons(1) != 1) { | |
2364 | fprintf(stderr, | |
2365 | "We are little endian, eBPF object is big endian!\n"); | |
2366 | return -EIO; | |
2367 | } | |
2368 | break; | |
2369 | } | |
2370 | ||
2371 | return 0; | |
2372 | } | |
2373 | ||
ecb05c0f DB |
2374 | static void bpf_get_cfg(struct bpf_elf_ctx *ctx) |
2375 | { | |
2376 | static const char *path_jit = "/proc/sys/net/core/bpf_jit_enable"; | |
2377 | int fd; | |
2378 | ||
2379 | fd = open(path_jit, O_RDONLY); | |
2380 | if (fd > 0) { | |
2381 | char tmp[16] = {}; | |
2382 | ||
2383 | if (read(fd, tmp, sizeof(tmp)) > 0) | |
2384 | ctx->cfg.jit_enabled = atoi(tmp); | |
2385 | close(fd); | |
2386 | } | |
2387 | } | |
2388 | ||
32e93fb7 DB |
2389 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
2390 | enum bpf_prog_type type, bool verbose) | |
2391 | { | |
2392 | int ret = -EINVAL; | |
2393 | ||
2394 | if (elf_version(EV_CURRENT) == EV_NONE || | |
2395 | bpf_init_env(pathname)) | |
2396 | return ret; | |
2397 | ||
2398 | memset(ctx, 0, sizeof(*ctx)); | |
ecb05c0f | 2399 | bpf_get_cfg(ctx); |
32e93fb7 DB |
2400 | ctx->verbose = verbose; |
2401 | ctx->type = type; | |
2402 | ||
2403 | ctx->obj_fd = open(pathname, O_RDONLY); | |
2404 | if (ctx->obj_fd < 0) | |
2405 | return ctx->obj_fd; | |
2406 | ||
2407 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
2408 | if (!ctx->elf_fd) { | |
11c39b5e | 2409 | ret = -EINVAL; |
32e93fb7 | 2410 | goto out_fd; |
11c39b5e DB |
2411 | } |
2412 | ||
8187b012 DB |
2413 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
2414 | ret = -EINVAL; | |
2415 | goto out_fd; | |
2416 | } | |
2417 | ||
32e93fb7 DB |
2418 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
2419 | &ctx->elf_hdr) { | |
11c39b5e DB |
2420 | ret = -EIO; |
2421 | goto out_elf; | |
2422 | } | |
2423 | ||
8187b012 DB |
2424 | ret = bpf_elf_check_ehdr(ctx); |
2425 | if (ret < 0) | |
2426 | goto out_elf; | |
2427 | ||
32e93fb7 DB |
2428 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
2429 | sizeof(*(ctx->sec_done))); | |
2430 | if (!ctx->sec_done) { | |
11c39b5e DB |
2431 | ret = -ENOMEM; |
2432 | goto out_elf; | |
2433 | } | |
2434 | ||
f31645d1 DB |
2435 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
2436 | ret = -ENOMEM; | |
2437 | goto out_free; | |
2438 | } | |
2439 | ||
32e93fb7 | 2440 | bpf_save_finfo(ctx); |
f6793eec DB |
2441 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
2442 | ||
32e93fb7 | 2443 | return 0; |
f31645d1 DB |
2444 | out_free: |
2445 | free(ctx->sec_done); | |
32e93fb7 DB |
2446 | out_elf: |
2447 | elf_end(ctx->elf_fd); | |
2448 | out_fd: | |
2449 | close(ctx->obj_fd); | |
2450 | return ret; | |
2451 | } | |
d937a74b | 2452 | |
32e93fb7 DB |
2453 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
2454 | { | |
2455 | int i, count = 0; | |
11c39b5e | 2456 | |
32e93fb7 DB |
2457 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
2458 | if (!ctx->map_fds[i]) | |
2459 | break; | |
2460 | count++; | |
2461 | } | |
473d7840 | 2462 | |
32e93fb7 DB |
2463 | return count; |
2464 | } | |
6256f8c9 | 2465 | |
32e93fb7 DB |
2466 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
2467 | { | |
2468 | int i; | |
2469 | ||
2470 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
2471 | if (ctx->map_fds[i]) | |
2472 | close(ctx->map_fds[i]); | |
473d7840 | 2473 | } |
32e93fb7 DB |
2474 | } |
2475 | ||
2476 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
2477 | { | |
2478 | if (failure) | |
2479 | bpf_maps_teardown(ctx); | |
473d7840 | 2480 | |
f6793eec | 2481 | bpf_hash_destroy(ctx); |
f31645d1 | 2482 | |
32e93fb7 | 2483 | free(ctx->sec_done); |
f31645d1 DB |
2484 | free(ctx->log); |
2485 | ||
32e93fb7 DB |
2486 | elf_end(ctx->elf_fd); |
2487 | close(ctx->obj_fd); | |
2488 | } | |
6256f8c9 | 2489 | |
32e93fb7 | 2490 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 2491 | |
32e93fb7 DB |
2492 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
2493 | const char *section, bool verbose) | |
2494 | { | |
2495 | struct bpf_elf_ctx *ctx = &__ctx; | |
2496 | int fd = 0, ret; | |
6256f8c9 | 2497 | |
32e93fb7 DB |
2498 | ret = bpf_elf_ctx_init(ctx, pathname, type, verbose); |
2499 | if (ret < 0) { | |
2500 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
2501 | return ret; | |
2502 | } | |
6256f8c9 | 2503 | |
32e93fb7 DB |
2504 | ret = bpf_fetch_ancillary(ctx); |
2505 | if (ret < 0) { | |
2506 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
2507 | goto out; | |
2508 | } | |
2509 | ||
2510 | fd = bpf_fetch_prog_sec(ctx, section); | |
2511 | if (fd < 0) { | |
2512 | fprintf(stderr, "Error fetching program/map!\n"); | |
2513 | ret = fd; | |
2514 | goto out; | |
2515 | } | |
2516 | ||
2517 | ret = bpf_fill_prog_arrays(ctx); | |
2518 | if (ret < 0) | |
2519 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 2520 | out: |
32e93fb7 DB |
2521 | bpf_elf_ctx_destroy(ctx, ret < 0); |
2522 | if (ret < 0) { | |
2523 | if (fd) | |
2524 | close(fd); | |
2525 | return ret; | |
2526 | } | |
2527 | ||
2528 | return fd; | |
6256f8c9 | 2529 | } |
11c39b5e | 2530 | |
6256f8c9 | 2531 | static int |
4bd62446 DB |
2532 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
2533 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 | 2534 | { |
d17b136f PS |
2535 | struct bpf_map_set_msg msg = { |
2536 | .aux.uds_ver = BPF_SCM_AUX_VER, | |
2537 | .aux.num_ent = entries, | |
2538 | }; | |
6256f8c9 DB |
2539 | int *cmsg_buf, min_fd; |
2540 | char *amsg_buf; | |
2541 | int i; | |
2542 | ||
6256f8c9 DB |
2543 | strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); |
2544 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); | |
2545 | ||
2546 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
2547 | amsg_buf = (char *)msg.aux.ent; | |
2548 | ||
4bd62446 | 2549 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
2550 | int ret; |
2551 | ||
4bd62446 | 2552 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
2553 | bpf_map_set_init_single(&msg, min_fd); |
2554 | ||
2555 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
2556 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
2557 | ||
2558 | ret = sendmsg(fd, &msg.hdr, 0); | |
2559 | if (ret <= 0) | |
2560 | return ret ? : -1; | |
2561 | } | |
2562 | ||
2563 | return 0; | |
11c39b5e DB |
2564 | } |
2565 | ||
4bd62446 DB |
2566 | static int |
2567 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
2568 | unsigned int entries) | |
2569 | { | |
2570 | struct bpf_map_set_msg msg; | |
2571 | int *cmsg_buf, min_fd; | |
2572 | char *amsg_buf, *mmsg_buf; | |
2573 | unsigned int needed = 1; | |
2574 | int i; | |
2575 | ||
2576 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
2577 | amsg_buf = (char *)msg.aux.ent; | |
2578 | mmsg_buf = (char *)&msg.aux; | |
2579 | ||
2580 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
2581 | struct cmsghdr *cmsg; | |
2582 | int ret; | |
2583 | ||
2584 | min_fd = min(entries, entries - i); | |
2585 | bpf_map_set_init_single(&msg, min_fd); | |
2586 | ||
2587 | ret = recvmsg(fd, &msg.hdr, 0); | |
2588 | if (ret <= 0) | |
2589 | return ret ? : -1; | |
2590 | ||
2591 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
2592 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
2593 | return -EINVAL; | |
2594 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
2595 | return -EIO; | |
2596 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
2597 | return -ENOSYS; | |
2598 | ||
2599 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
2600 | if (min_fd > entries || min_fd <= 0) | |
2601 | return -EINVAL; | |
2602 | ||
2603 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
2604 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
2605 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
2606 | ||
2607 | needed = aux->num_ent; | |
2608 | } | |
2609 | ||
2610 | return 0; | |
2611 | } | |
2612 | ||
2613 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 2614 | { |
32e93fb7 | 2615 | struct bpf_elf_ctx *ctx = &__ctx; |
d17b136f PS |
2616 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
2617 | struct bpf_map_data bpf_aux = { | |
2618 | .fds = ctx->map_fds, | |
2619 | .ent = ctx->maps, | |
2620 | .st = &ctx->stat, | |
2621 | .obj = obj, | |
2622 | }; | |
6256f8c9 DB |
2623 | int fd, ret; |
2624 | ||
2625 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2626 | if (fd < 0) { | |
2627 | fprintf(stderr, "Cannot open socket: %s\n", | |
2628 | strerror(errno)); | |
2629 | return -1; | |
2630 | } | |
2631 | ||
6256f8c9 DB |
2632 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2633 | ||
2634 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2635 | if (ret < 0) { | |
2636 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
2637 | path, strerror(errno)); | |
2638 | return -1; | |
2639 | } | |
2640 | ||
4bd62446 | 2641 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 2642 | bpf_maps_count(ctx)); |
6256f8c9 | 2643 | if (ret < 0) |
4bd62446 DB |
2644 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
2645 | path, strerror(errno)); | |
2646 | ||
32e93fb7 | 2647 | bpf_maps_teardown(ctx); |
4bd62446 DB |
2648 | close(fd); |
2649 | return ret; | |
2650 | } | |
2651 | ||
2652 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
2653 | unsigned int entries) | |
2654 | { | |
d17b136f | 2655 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
4bd62446 DB |
2656 | int fd, ret; |
2657 | ||
2658 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2659 | if (fd < 0) { | |
2660 | fprintf(stderr, "Cannot open socket: %s\n", | |
2661 | strerror(errno)); | |
2662 | return -1; | |
2663 | } | |
2664 | ||
4bd62446 DB |
2665 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2666 | ||
2667 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2668 | if (ret < 0) { | |
2669 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
2670 | strerror(errno)); | |
2671 | return -1; | |
2672 | } | |
2673 | ||
2674 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
2675 | if (ret < 0) | |
2676 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
2677 | path, strerror(errno)); |
2678 | ||
4bd62446 | 2679 | unlink(addr.sun_path); |
6256f8c9 DB |
2680 | close(fd); |
2681 | return ret; | |
2682 | } | |
11c39b5e | 2683 | #endif /* HAVE_ELF */ |