]>
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" |
bc2d4d83 | 43 | #include "json_print.h" |
6256f8c9 | 44 | |
e4225669 | 45 | #include "bpf_util.h" |
6256f8c9 DB |
46 | #include "bpf_elf.h" |
47 | #include "bpf_scm.h" | |
48 | ||
e4225669 DB |
49 | struct bpf_prog_meta { |
50 | const char *type; | |
51 | const char *subdir; | |
52 | const char *section; | |
53 | bool may_uds_export; | |
54 | }; | |
1d129d19 | 55 | |
e4225669 DB |
56 | static const enum bpf_prog_type __bpf_types[] = { |
57 | BPF_PROG_TYPE_SCHED_CLS, | |
58 | BPF_PROG_TYPE_SCHED_ACT, | |
c7272ca7 | 59 | BPF_PROG_TYPE_XDP, |
b15f440e TG |
60 | BPF_PROG_TYPE_LWT_IN, |
61 | BPF_PROG_TYPE_LWT_OUT, | |
62 | BPF_PROG_TYPE_LWT_XMIT, | |
e4225669 | 63 | }; |
67584e3a | 64 | |
e4225669 DB |
65 | static const struct bpf_prog_meta __bpf_prog_meta[] = { |
66 | [BPF_PROG_TYPE_SCHED_CLS] = { | |
67 | .type = "cls", | |
68 | .subdir = "tc", | |
69 | .section = ELF_SECTION_CLASSIFIER, | |
70 | .may_uds_export = true, | |
71 | }, | |
72 | [BPF_PROG_TYPE_SCHED_ACT] = { | |
73 | .type = "act", | |
74 | .subdir = "tc", | |
75 | .section = ELF_SECTION_ACTION, | |
76 | .may_uds_export = true, | |
77 | }, | |
c7272ca7 DB |
78 | [BPF_PROG_TYPE_XDP] = { |
79 | .type = "xdp", | |
80 | .subdir = "xdp", | |
81 | .section = ELF_SECTION_PROG, | |
82 | }, | |
b15f440e TG |
83 | [BPF_PROG_TYPE_LWT_IN] = { |
84 | .type = "lwt_in", | |
85 | .subdir = "ip", | |
86 | .section = ELF_SECTION_PROG, | |
87 | }, | |
88 | [BPF_PROG_TYPE_LWT_OUT] = { | |
89 | .type = "lwt_out", | |
90 | .subdir = "ip", | |
91 | .section = ELF_SECTION_PROG, | |
92 | }, | |
93 | [BPF_PROG_TYPE_LWT_XMIT] = { | |
94 | .type = "lwt_xmit", | |
95 | .subdir = "ip", | |
96 | .section = ELF_SECTION_PROG, | |
97 | }, | |
04cb3c0d MX |
98 | [BPF_PROG_TYPE_LWT_SEG6LOCAL] = { |
99 | .type = "lwt_seg6local", | |
100 | .subdir = "ip", | |
101 | .section = ELF_SECTION_PROG, | |
102 | }, | |
e4225669 DB |
103 | }; |
104 | ||
105 | static const char *bpf_prog_to_subdir(enum bpf_prog_type type) | |
106 | { | |
107 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
108 | __bpf_prog_meta[type].subdir); | |
109 | return __bpf_prog_meta[type].subdir; | |
110 | } | |
111 | ||
112 | const char *bpf_prog_to_default_section(enum bpf_prog_type type) | |
113 | { | |
114 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
115 | __bpf_prog_meta[type].section); | |
116 | return __bpf_prog_meta[type].section; | |
117 | } | |
e77fa41d | 118 | |
32e93fb7 DB |
119 | #ifdef HAVE_ELF |
120 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
65fdae3d | 121 | const char *sec, __u32 ifindex, bool verbose); |
32e93fb7 DB |
122 | #else |
123 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
65fdae3d | 124 | const char *sec, __u32 ifindex, bool verbose) |
32e93fb7 DB |
125 | { |
126 | fprintf(stderr, "No ELF library support compiled in.\n"); | |
127 | errno = ENOSYS; | |
128 | return -1; | |
129 | } | |
130 | #endif | |
131 | ||
132 | static inline __u64 bpf_ptr_to_u64(const void *ptr) | |
133 | { | |
134 | return (__u64)(unsigned long)ptr; | |
135 | } | |
136 | ||
137 | static int bpf(int cmd, union bpf_attr *attr, unsigned int size) | |
138 | { | |
139 | #ifdef __NR_bpf | |
140 | return syscall(__NR_bpf, cmd, attr, size); | |
141 | #else | |
142 | fprintf(stderr, "No bpf syscall, kernel headers too old?\n"); | |
143 | errno = ENOSYS; | |
144 | return -1; | |
145 | #endif | |
146 | } | |
147 | ||
91d88eeb DB |
148 | static int bpf_map_update(int fd, const void *key, const void *value, |
149 | uint64_t flags) | |
32e93fb7 | 150 | { |
d17b136f | 151 | union bpf_attr attr = {}; |
67584e3a | 152 | |
67584e3a ND |
153 | attr.map_fd = fd; |
154 | attr.key = bpf_ptr_to_u64(key); | |
155 | attr.value = bpf_ptr_to_u64(value); | |
156 | attr.flags = flags; | |
32e93fb7 | 157 | |
91d88eeb | 158 | return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
32e93fb7 DB |
159 | } |
160 | ||
779525cd DB |
161 | static int bpf_prog_fd_by_id(uint32_t id) |
162 | { | |
163 | union bpf_attr attr = {}; | |
164 | ||
165 | attr.prog_id = id; | |
166 | ||
167 | return bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); | |
168 | } | |
169 | ||
170 | static int bpf_prog_info_by_fd(int fd, struct bpf_prog_info *info, | |
171 | uint32_t *info_len) | |
172 | { | |
173 | union bpf_attr attr = {}; | |
174 | int ret; | |
175 | ||
176 | attr.info.bpf_fd = fd; | |
177 | attr.info.info = bpf_ptr_to_u64(info); | |
178 | attr.info.info_len = *info_len; | |
179 | ||
180 | *info_len = 0; | |
181 | ret = bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); | |
182 | if (!ret) | |
183 | *info_len = attr.info.info_len; | |
184 | ||
185 | return ret; | |
186 | } | |
187 | ||
a0b5b7cf | 188 | int bpf_dump_prog_info(FILE *f, uint32_t id) |
779525cd DB |
189 | { |
190 | struct bpf_prog_info info = {}; | |
191 | uint32_t len = sizeof(info); | |
a0b5b7cf DB |
192 | int fd, ret, dump_ok = 0; |
193 | SPRINT_BUF(tmp); | |
779525cd | 194 | |
bc2d4d83 DB |
195 | open_json_object("prog"); |
196 | print_uint(PRINT_ANY, "id", "id %u ", id); | |
779525cd DB |
197 | |
198 | fd = bpf_prog_fd_by_id(id); | |
199 | if (fd < 0) | |
bc2d4d83 | 200 | goto out; |
779525cd DB |
201 | |
202 | ret = bpf_prog_info_by_fd(fd, &info, &len); | |
203 | if (!ret && len) { | |
bc2d4d83 DB |
204 | int jited = !!info.jited_prog_len; |
205 | ||
206 | print_string(PRINT_ANY, "tag", "tag %s ", | |
207 | hexstring_n2a(info.tag, sizeof(info.tag), | |
208 | tmp, sizeof(tmp))); | |
209 | print_uint(PRINT_JSON, "jited", NULL, jited); | |
210 | if (jited && !is_json_context()) | |
779525cd | 211 | fprintf(f, "jited "); |
a0b5b7cf | 212 | dump_ok = 1; |
779525cd DB |
213 | } |
214 | ||
215 | close(fd); | |
bc2d4d83 DB |
216 | out: |
217 | close_json_object(); | |
a0b5b7cf | 218 | return dump_ok; |
779525cd DB |
219 | } |
220 | ||
32e93fb7 DB |
221 | static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len, |
222 | char **bpf_string, bool *need_release, | |
223 | const char separator) | |
1d129d19 JP |
224 | { |
225 | char sp; | |
226 | ||
227 | if (from_file) { | |
228 | size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,"); | |
7c87c7fe | 229 | char *tmp_string, *pos, c_prev = ' '; |
1d129d19 | 230 | FILE *fp; |
7c87c7fe | 231 | int c; |
1d129d19 JP |
232 | |
233 | tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len; | |
3da3ebfc | 234 | tmp_string = pos = calloc(1, tmp_len); |
1d129d19 JP |
235 | if (tmp_string == NULL) |
236 | return -ENOMEM; | |
237 | ||
1d129d19 JP |
238 | fp = fopen(arg, "r"); |
239 | if (fp == NULL) { | |
240 | perror("Cannot fopen"); | |
241 | free(tmp_string); | |
242 | return -ENOENT; | |
243 | } | |
244 | ||
3da3ebfc PS |
245 | while ((c = fgetc(fp)) != EOF) { |
246 | switch (c) { | |
247 | case '\n': | |
248 | if (c_prev != ',') | |
249 | *(pos++) = ','; | |
7c87c7fe | 250 | c_prev = ','; |
3da3ebfc PS |
251 | break; |
252 | case ' ': | |
253 | case '\t': | |
254 | if (c_prev != ' ') | |
255 | *(pos++) = c; | |
7c87c7fe | 256 | c_prev = ' '; |
3da3ebfc PS |
257 | break; |
258 | default: | |
259 | *(pos++) = c; | |
7c87c7fe | 260 | c_prev = c; |
3da3ebfc PS |
261 | } |
262 | if (pos - tmp_string == tmp_len) | |
263 | break; | |
3da3ebfc PS |
264 | } |
265 | ||
266 | if (!feof(fp)) { | |
1d129d19 JP |
267 | free(tmp_string); |
268 | fclose(fp); | |
3da3ebfc | 269 | return -E2BIG; |
1d129d19 JP |
270 | } |
271 | ||
272 | fclose(fp); | |
3da3ebfc | 273 | *pos = 0; |
e4225669 | 274 | |
1d129d19 JP |
275 | *need_release = true; |
276 | *bpf_string = tmp_string; | |
277 | } else { | |
278 | *need_release = false; | |
279 | *bpf_string = arg; | |
280 | } | |
281 | ||
282 | if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 || | |
283 | sp != separator) { | |
284 | if (*need_release) | |
285 | free(*bpf_string); | |
286 | return -EINVAL; | |
287 | } | |
288 | ||
289 | return 0; | |
290 | } | |
291 | ||
32e93fb7 DB |
292 | static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops, |
293 | bool from_file) | |
1d129d19 JP |
294 | { |
295 | char *bpf_string, *token, separator = ','; | |
296 | int ret = 0, i = 0; | |
297 | bool need_release; | |
298 | __u16 bpf_len = 0; | |
299 | ||
300 | if (argc < 1) | |
301 | return -EINVAL; | |
302 | if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string, | |
303 | &need_release, separator)) | |
304 | return -EINVAL; | |
305 | if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) { | |
306 | ret = -EINVAL; | |
307 | goto out; | |
308 | } | |
309 | ||
310 | token = bpf_string; | |
311 | while ((token = strchr(token, separator)) && (++token)[0]) { | |
312 | if (i >= bpf_len) { | |
32a121cb | 313 | fprintf(stderr, "Real program length exceeds encoded length parameter!\n"); |
1d129d19 JP |
314 | ret = -EINVAL; |
315 | goto out; | |
316 | } | |
317 | ||
318 | if (sscanf(token, "%hu %hhu %hhu %u,", | |
319 | &bpf_ops[i].code, &bpf_ops[i].jt, | |
320 | &bpf_ops[i].jf, &bpf_ops[i].k) != 4) { | |
321 | fprintf(stderr, "Error at instruction %d!\n", i); | |
322 | ret = -EINVAL; | |
323 | goto out; | |
324 | } | |
325 | ||
326 | i++; | |
327 | } | |
328 | ||
329 | if (i != bpf_len) { | |
afc1a200 | 330 | fprintf(stderr, "Parsed program length is less than encoded length parameter!\n"); |
1d129d19 JP |
331 | ret = -EINVAL; |
332 | goto out; | |
333 | } | |
334 | ret = bpf_len; | |
1d129d19 JP |
335 | out: |
336 | if (need_release) | |
337 | free(bpf_string); | |
338 | ||
339 | return ret; | |
340 | } | |
341 | ||
52d57f6b | 342 | void bpf_print_ops(struct rtattr *bpf_ops, __u16 len) |
1d129d19 | 343 | { |
d896797c | 344 | struct sock_filter *ops = RTA_DATA(bpf_ops); |
1d129d19 JP |
345 | int i; |
346 | ||
347 | if (len == 0) | |
348 | return; | |
349 | ||
52d57f6b DC |
350 | open_json_object("bytecode"); |
351 | print_uint(PRINT_ANY, "length", "bytecode \'%u,", len); | |
352 | open_json_array(PRINT_JSON, "insns"); | |
353 | ||
354 | for (i = 0; i < len; i++) { | |
355 | open_json_object(NULL); | |
ca814443 DC |
356 | print_hu(PRINT_ANY, "code", "%hu ", ops[i].code); |
357 | print_hhu(PRINT_ANY, "jt", "%hhu ", ops[i].jt); | |
358 | print_hhu(PRINT_ANY, "jf", "%hhu ", ops[i].jf); | |
52d57f6b DC |
359 | if (i == len - 1) |
360 | print_uint(PRINT_ANY, "k", "%u\'", ops[i].k); | |
361 | else | |
362 | print_uint(PRINT_ANY, "k", "%u,", ops[i].k); | |
363 | close_json_object(); | |
364 | } | |
1d129d19 | 365 | |
52d57f6b DC |
366 | close_json_array(PRINT_JSON, NULL); |
367 | close_json_object(); | |
1d129d19 | 368 | } |
11c39b5e | 369 | |
afc1a200 DB |
370 | static void bpf_map_pin_report(const struct bpf_elf_map *pin, |
371 | const struct bpf_elf_map *obj) | |
372 | { | |
373 | fprintf(stderr, "Map specification differs from pinned file!\n"); | |
374 | ||
375 | if (obj->type != pin->type) | |
376 | fprintf(stderr, " - Type: %u (obj) != %u (pin)\n", | |
377 | obj->type, pin->type); | |
378 | if (obj->size_key != pin->size_key) | |
379 | fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n", | |
380 | obj->size_key, pin->size_key); | |
381 | if (obj->size_value != pin->size_value) | |
382 | fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n", | |
383 | obj->size_value, pin->size_value); | |
384 | if (obj->max_elem != pin->max_elem) | |
385 | fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n", | |
386 | obj->max_elem, pin->max_elem); | |
4dd3f50a DB |
387 | if (obj->flags != pin->flags) |
388 | fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n", | |
389 | obj->flags, pin->flags); | |
afc1a200 DB |
390 | |
391 | fprintf(stderr, "\n"); | |
392 | } | |
393 | ||
ecb05c0f DB |
394 | struct bpf_prog_data { |
395 | unsigned int type; | |
396 | unsigned int jited; | |
397 | }; | |
398 | ||
399 | struct bpf_map_ext { | |
400 | struct bpf_prog_data owner; | |
f823f360 DB |
401 | unsigned int btf_id_key; |
402 | unsigned int btf_id_val; | |
ecb05c0f DB |
403 | }; |
404 | ||
405 | static int bpf_derive_elf_map_from_fdinfo(int fd, struct bpf_elf_map *map, | |
406 | struct bpf_map_ext *ext) | |
9e607f2e | 407 | { |
ecb05c0f | 408 | unsigned int val, owner_type = 0, owner_jited = 0; |
358abfe0 | 409 | char file[PATH_MAX], buff[4096]; |
9e607f2e DB |
410 | FILE *fp; |
411 | ||
358abfe0 | 412 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); |
ecb05c0f | 413 | memset(map, 0, sizeof(*map)); |
9e607f2e DB |
414 | |
415 | fp = fopen(file, "r"); | |
416 | if (!fp) { | |
417 | fprintf(stderr, "No procfs support?!\n"); | |
418 | return -EIO; | |
419 | } | |
420 | ||
9e607f2e DB |
421 | while (fgets(buff, sizeof(buff), fp)) { |
422 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
ecb05c0f | 423 | map->type = val; |
9e607f2e | 424 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) |
ecb05c0f | 425 | map->size_key = val; |
9e607f2e | 426 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) |
ecb05c0f | 427 | map->size_value = val; |
9e607f2e | 428 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) |
ecb05c0f | 429 | map->max_elem = val; |
4dd3f50a | 430 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) |
ecb05c0f | 431 | map->flags = val; |
fb24802b DB |
432 | else if (sscanf(buff, "owner_prog_type:\t%i", &val) == 1) |
433 | owner_type = val; | |
ecb05c0f DB |
434 | else if (sscanf(buff, "owner_jited:\t%i", &val) == 1) |
435 | owner_jited = val; | |
9e607f2e DB |
436 | } |
437 | ||
438 | fclose(fp); | |
ecb05c0f DB |
439 | if (ext) { |
440 | memset(ext, 0, sizeof(*ext)); | |
441 | ext->owner.type = owner_type; | |
442 | ext->owner.jited = owner_jited; | |
443 | } | |
444 | ||
445 | return 0; | |
446 | } | |
447 | ||
448 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, | |
449 | struct bpf_map_ext *ext, int length, | |
450 | enum bpf_prog_type type) | |
451 | { | |
452 | struct bpf_elf_map tmp, zero = {}; | |
453 | int ret; | |
454 | ||
455 | ret = bpf_derive_elf_map_from_fdinfo(fd, &tmp, ext); | |
456 | if (ret < 0) | |
457 | return ret; | |
9e607f2e | 458 | |
fb24802b DB |
459 | /* The decision to reject this is on kernel side eventually, but |
460 | * at least give the user a chance to know what's wrong. | |
461 | */ | |
ecb05c0f | 462 | if (ext->owner.type && ext->owner.type != type) |
fb24802b | 463 | fprintf(stderr, "Program array map owner types differ: %u (obj) != %u (pin)\n", |
ecb05c0f | 464 | type, ext->owner.type); |
fb24802b | 465 | |
91d88eeb | 466 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
467 | return 0; |
468 | } else { | |
9e607f2e DB |
469 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, |
470 | * so just accept it. We know we do have an eBPF fd and in this | |
471 | * case, everything is 0. It is guaranteed that no such map exists | |
472 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
473 | */ | |
91d88eeb | 474 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
475 | return 0; |
476 | ||
afc1a200 | 477 | bpf_map_pin_report(&tmp, map); |
9e607f2e DB |
478 | return -EINVAL; |
479 | } | |
480 | } | |
481 | ||
91d88eeb DB |
482 | static int bpf_mnt_fs(const char *target) |
483 | { | |
484 | bool bind_done = false; | |
485 | ||
486 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
487 | if (errno != EINVAL || bind_done) { | |
488 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
489 | target, strerror(errno)); | |
490 | return -1; | |
491 | } | |
492 | ||
493 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
494 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
495 | target, target, strerror(errno)); | |
496 | return -1; | |
497 | } | |
498 | ||
499 | bind_done = true; | |
500 | } | |
501 | ||
e4225669 | 502 | if (mount("bpf", target, "bpf", 0, "mode=0700")) { |
91d88eeb DB |
503 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", |
504 | target, strerror(errno)); | |
505 | return -1; | |
506 | } | |
507 | ||
508 | return 0; | |
509 | } | |
510 | ||
95ae9a48 DB |
511 | static int bpf_mnt_check_target(const char *target) |
512 | { | |
513 | struct stat sb = {}; | |
514 | int ret; | |
515 | ||
516 | ret = stat(target, &sb); | |
517 | if (ret) { | |
518 | ret = mkdir(target, S_IRWXU); | |
519 | if (ret) { | |
520 | fprintf(stderr, "mkdir %s failed: %s\n", target, | |
521 | strerror(errno)); | |
522 | return ret; | |
523 | } | |
524 | } | |
525 | ||
526 | return 0; | |
527 | } | |
528 | ||
32e93fb7 DB |
529 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
530 | { | |
531 | struct statfs st_fs; | |
532 | ||
533 | if (statfs(mnt, &st_fs) < 0) | |
534 | return -ENOENT; | |
535 | if ((unsigned long)st_fs.f_type != magic) | |
536 | return -ENOENT; | |
537 | ||
538 | return 0; | |
539 | } | |
540 | ||
95ae9a48 DB |
541 | static const char *bpf_find_mntpt_single(unsigned long magic, char *mnt, |
542 | int len, const char *mntpt) | |
543 | { | |
544 | int ret; | |
545 | ||
546 | ret = bpf_valid_mntpt(mntpt, magic); | |
547 | if (!ret) { | |
18f156bf | 548 | strlcpy(mnt, mntpt, len); |
95ae9a48 DB |
549 | return mnt; |
550 | } | |
551 | ||
552 | return NULL; | |
553 | } | |
554 | ||
32e93fb7 DB |
555 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, |
556 | char *mnt, int len, | |
557 | const char * const *known_mnts) | |
558 | { | |
559 | const char * const *ptr; | |
560 | char type[100]; | |
561 | FILE *fp; | |
562 | ||
563 | if (known_mnts) { | |
564 | ptr = known_mnts; | |
565 | while (*ptr) { | |
95ae9a48 | 566 | if (bpf_find_mntpt_single(magic, mnt, len, *ptr)) |
32e93fb7 | 567 | return mnt; |
32e93fb7 DB |
568 | ptr++; |
569 | } | |
570 | } | |
571 | ||
c3724e4b PS |
572 | if (len != PATH_MAX) |
573 | return NULL; | |
574 | ||
32e93fb7 | 575 | fp = fopen("/proc/mounts", "r"); |
c3724e4b | 576 | if (fp == NULL) |
32e93fb7 DB |
577 | return NULL; |
578 | ||
579 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
580 | mnt, type) == 2) { | |
581 | if (strcmp(type, fstype) == 0) | |
582 | break; | |
583 | } | |
584 | ||
585 | fclose(fp); | |
586 | if (strcmp(type, fstype) != 0) | |
587 | return NULL; | |
588 | ||
589 | return mnt; | |
590 | } | |
591 | ||
592 | int bpf_trace_pipe(void) | |
593 | { | |
594 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
595 | static const char * const tracefs_known_mnts[] = { | |
596 | TRACE_DIR_MNT, | |
597 | "/sys/kernel/debug/tracing", | |
598 | "/tracing", | |
599 | "/trace", | |
600 | 0, | |
601 | }; | |
1b736dc4 | 602 | int fd_in, fd_out = STDERR_FILENO; |
358abfe0 | 603 | char tpipe[PATH_MAX]; |
32e93fb7 | 604 | const char *mnt; |
32e93fb7 DB |
605 | |
606 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
607 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
608 | if (!mnt) { | |
609 | fprintf(stderr, "tracefs not mounted?\n"); | |
610 | return -1; | |
611 | } | |
612 | ||
358abfe0 | 613 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); |
32e93fb7 | 614 | |
1b736dc4 DB |
615 | fd_in = open(tpipe, O_RDONLY); |
616 | if (fd_in < 0) | |
32e93fb7 DB |
617 | return -1; |
618 | ||
619 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
620 | while (1) { | |
621 | static char buff[4096]; | |
622 | ssize_t ret; | |
623 | ||
1b736dc4 DB |
624 | ret = read(fd_in, buff, sizeof(buff)); |
625 | if (ret > 0 && write(fd_out, buff, ret) == ret) | |
626 | continue; | |
627 | break; | |
32e93fb7 DB |
628 | } |
629 | ||
1b736dc4 DB |
630 | close(fd_in); |
631 | return -1; | |
32e93fb7 DB |
632 | } |
633 | ||
e4225669 | 634 | static int bpf_gen_global(const char *bpf_sub_dir) |
91d88eeb | 635 | { |
358abfe0 | 636 | char bpf_glo_dir[PATH_MAX]; |
e4225669 DB |
637 | int ret; |
638 | ||
358abfe0 AC |
639 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/", |
640 | bpf_sub_dir, BPF_DIR_GLOBALS); | |
e4225669 DB |
641 | |
642 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
643 | if (ret && errno != EEXIST) { | |
644 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
645 | strerror(errno)); | |
358abfe0 | 646 | return ret; |
e4225669 DB |
647 | } |
648 | ||
358abfe0 | 649 | return 0; |
e4225669 DB |
650 | } |
651 | ||
652 | static int bpf_gen_master(const char *base, const char *name) | |
653 | { | |
358abfe0 | 654 | char bpf_sub_dir[PATH_MAX + NAME_MAX + 1]; |
e4225669 DB |
655 | int ret; |
656 | ||
358abfe0 | 657 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name); |
e4225669 DB |
658 | |
659 | ret = mkdir(bpf_sub_dir, S_IRWXU); | |
660 | if (ret && errno != EEXIST) { | |
661 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir, | |
662 | strerror(errno)); | |
358abfe0 | 663 | return ret; |
e4225669 DB |
664 | } |
665 | ||
358abfe0 | 666 | return bpf_gen_global(bpf_sub_dir); |
e4225669 DB |
667 | } |
668 | ||
669 | static int bpf_slave_via_bind_mnt(const char *full_name, | |
670 | const char *full_link) | |
671 | { | |
672 | int ret; | |
673 | ||
674 | ret = mkdir(full_name, S_IRWXU); | |
675 | if (ret) { | |
676 | assert(errno != EEXIST); | |
677 | fprintf(stderr, "mkdir %s failed: %s\n", full_name, | |
678 | strerror(errno)); | |
679 | return ret; | |
680 | } | |
681 | ||
682 | ret = mount(full_link, full_name, "none", MS_BIND, NULL); | |
683 | if (ret) { | |
684 | rmdir(full_name); | |
685 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
686 | full_link, full_name, strerror(errno)); | |
687 | } | |
688 | ||
689 | return ret; | |
690 | } | |
691 | ||
692 | static int bpf_gen_slave(const char *base, const char *name, | |
693 | const char *link) | |
694 | { | |
358abfe0 AC |
695 | char bpf_lnk_dir[PATH_MAX + NAME_MAX + 1]; |
696 | char bpf_sub_dir[PATH_MAX + NAME_MAX]; | |
e4225669 DB |
697 | struct stat sb = {}; |
698 | int ret; | |
699 | ||
358abfe0 AC |
700 | snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link); |
701 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name); | |
e4225669 DB |
702 | |
703 | ret = symlink(bpf_lnk_dir, bpf_sub_dir); | |
704 | if (ret) { | |
705 | if (errno != EEXIST) { | |
706 | if (errno != EPERM) { | |
707 | fprintf(stderr, "symlink %s failed: %s\n", | |
708 | bpf_sub_dir, strerror(errno)); | |
358abfe0 | 709 | return ret; |
e4225669 DB |
710 | } |
711 | ||
358abfe0 AC |
712 | return bpf_slave_via_bind_mnt(bpf_sub_dir, |
713 | bpf_lnk_dir); | |
e4225669 DB |
714 | } |
715 | ||
716 | ret = lstat(bpf_sub_dir, &sb); | |
717 | if (ret) { | |
718 | fprintf(stderr, "lstat %s failed: %s\n", | |
719 | bpf_sub_dir, strerror(errno)); | |
358abfe0 | 720 | return ret; |
e4225669 DB |
721 | } |
722 | ||
358abfe0 AC |
723 | if ((sb.st_mode & S_IFMT) != S_IFLNK) |
724 | return bpf_gen_global(bpf_sub_dir); | |
e4225669 DB |
725 | } |
726 | ||
358abfe0 | 727 | return 0; |
e4225669 DB |
728 | } |
729 | ||
730 | static int bpf_gen_hierarchy(const char *base) | |
731 | { | |
732 | int ret, i; | |
733 | ||
734 | ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0])); | |
735 | for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++) | |
736 | ret = bpf_gen_slave(base, | |
737 | bpf_prog_to_subdir(__bpf_types[i]), | |
738 | bpf_prog_to_subdir(__bpf_types[0])); | |
739 | return ret; | |
740 | } | |
741 | ||
742 | static const char *bpf_get_work_dir(enum bpf_prog_type type) | |
743 | { | |
744 | static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT; | |
358abfe0 | 745 | static char bpf_wrk_dir[PATH_MAX]; |
91d88eeb | 746 | static const char *mnt; |
e4225669 | 747 | static bool bpf_mnt_cached; |
95ae9a48 | 748 | const char *mnt_env = getenv(BPF_ENV_MNT); |
91d88eeb DB |
749 | static const char * const bpf_known_mnts[] = { |
750 | BPF_DIR_MNT, | |
e4225669 | 751 | "/bpf", |
91d88eeb DB |
752 | 0, |
753 | }; | |
91d88eeb DB |
754 | int ret; |
755 | ||
e4225669 DB |
756 | if (bpf_mnt_cached) { |
757 | const char *out = mnt; | |
758 | ||
51361a9f | 759 | if (out && type) { |
e4225669 DB |
760 | snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/", |
761 | out, bpf_prog_to_subdir(type)); | |
762 | out = bpf_tmp; | |
763 | } | |
764 | return out; | |
765 | } | |
91d88eeb | 766 | |
95ae9a48 DB |
767 | if (mnt_env) |
768 | mnt = bpf_find_mntpt_single(BPF_FS_MAGIC, bpf_tmp, | |
769 | sizeof(bpf_tmp), mnt_env); | |
770 | else | |
771 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp, | |
772 | sizeof(bpf_tmp), bpf_known_mnts); | |
91d88eeb | 773 | if (!mnt) { |
95ae9a48 DB |
774 | mnt = mnt_env ? : BPF_DIR_MNT; |
775 | ret = bpf_mnt_check_target(mnt); | |
776 | if (!ret) | |
777 | ret = bpf_mnt_fs(mnt); | |
91d88eeb DB |
778 | if (ret) { |
779 | mnt = NULL; | |
780 | goto out; | |
781 | } | |
782 | } | |
783 | ||
358abfe0 | 784 | snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt); |
91d88eeb | 785 | |
e4225669 DB |
786 | ret = bpf_gen_hierarchy(bpf_wrk_dir); |
787 | if (ret) { | |
91d88eeb DB |
788 | mnt = NULL; |
789 | goto out; | |
790 | } | |
791 | ||
e4225669 | 792 | mnt = bpf_wrk_dir; |
91d88eeb DB |
793 | out: |
794 | bpf_mnt_cached = true; | |
91d88eeb DB |
795 | return mnt; |
796 | } | |
797 | ||
e4225669 | 798 | static int bpf_obj_get(const char *pathname, enum bpf_prog_type type) |
91d88eeb | 799 | { |
d17b136f | 800 | union bpf_attr attr = {}; |
91d88eeb DB |
801 | char tmp[PATH_MAX]; |
802 | ||
803 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
e4225669 | 804 | pathname[1] == ':' && bpf_get_work_dir(type)) { |
91d88eeb | 805 | snprintf(tmp, sizeof(tmp), "%s/%s", |
e4225669 | 806 | bpf_get_work_dir(type), pathname + 2); |
91d88eeb DB |
807 | pathname = tmp; |
808 | } | |
809 | ||
91d88eeb DB |
810 | attr.pathname = bpf_ptr_to_u64(pathname); |
811 | ||
812 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
813 | } | |
814 | ||
21856018 DB |
815 | static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type) |
816 | { | |
817 | int prog_fd = bpf_obj_get(pathname, type); | |
818 | ||
819 | if (prog_fd < 0) | |
820 | fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n", | |
821 | pathname, strerror(errno)); | |
822 | return prog_fd; | |
823 | } | |
824 | ||
3f0b9e62 | 825 | static int bpf_do_parse(struct bpf_cfg_in *cfg, const bool *opt_tbl) |
32e93fb7 | 826 | { |
32e93fb7 | 827 | const char *file, *section, *uds_name; |
32e93fb7 | 828 | bool verbose = false; |
e4225669 | 829 | int i, ret, argc; |
91d88eeb DB |
830 | char **argv; |
831 | ||
e4225669 DB |
832 | argv = cfg->argv; |
833 | argc = cfg->argc; | |
91d88eeb DB |
834 | |
835 | if (opt_tbl[CBPF_BYTECODE] && | |
836 | (matches(*argv, "bytecode") == 0 || | |
837 | strcmp(*argv, "bc") == 0)) { | |
f20ff2f1 | 838 | cfg->mode = CBPF_BYTECODE; |
91d88eeb DB |
839 | } else if (opt_tbl[CBPF_FILE] && |
840 | (matches(*argv, "bytecode-file") == 0 || | |
841 | strcmp(*argv, "bcf") == 0)) { | |
f20ff2f1 | 842 | cfg->mode = CBPF_FILE; |
91d88eeb DB |
843 | } else if (opt_tbl[EBPF_OBJECT] && |
844 | (matches(*argv, "object-file") == 0 || | |
845 | strcmp(*argv, "obj") == 0)) { | |
f20ff2f1 | 846 | cfg->mode = EBPF_OBJECT; |
91d88eeb DB |
847 | } else if (opt_tbl[EBPF_PINNED] && |
848 | (matches(*argv, "object-pinned") == 0 || | |
849 | matches(*argv, "pinned") == 0 || | |
850 | matches(*argv, "fd") == 0)) { | |
f20ff2f1 | 851 | cfg->mode = EBPF_PINNED; |
32e93fb7 DB |
852 | } else { |
853 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
854 | return -1; | |
855 | } | |
856 | ||
857 | NEXT_ARG(); | |
858 | file = section = uds_name = NULL; | |
f20ff2f1 | 859 | if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) { |
32e93fb7 DB |
860 | file = *argv; |
861 | NEXT_ARG_FWD(); | |
862 | ||
658cfebc | 863 | if (cfg->type == BPF_PROG_TYPE_UNSPEC) { |
91d88eeb DB |
864 | if (argc > 0 && matches(*argv, "type") == 0) { |
865 | NEXT_ARG(); | |
e4225669 DB |
866 | for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta); |
867 | i++) { | |
868 | if (!__bpf_prog_meta[i].type) | |
869 | continue; | |
870 | if (!matches(*argv, | |
871 | __bpf_prog_meta[i].type)) { | |
658cfebc | 872 | cfg->type = i; |
e4225669 DB |
873 | break; |
874 | } | |
875 | } | |
876 | ||
658cfebc | 877 | if (cfg->type == BPF_PROG_TYPE_UNSPEC) { |
91d88eeb DB |
878 | fprintf(stderr, "What type is \"%s\"?\n", |
879 | *argv); | |
880 | return -1; | |
881 | } | |
882 | NEXT_ARG_FWD(); | |
883 | } else { | |
658cfebc | 884 | cfg->type = BPF_PROG_TYPE_SCHED_CLS; |
91d88eeb DB |
885 | } |
886 | } | |
887 | ||
658cfebc | 888 | section = bpf_prog_to_default_section(cfg->type); |
32e93fb7 DB |
889 | if (argc > 0 && matches(*argv, "section") == 0) { |
890 | NEXT_ARG(); | |
891 | section = *argv; | |
892 | NEXT_ARG_FWD(); | |
893 | } | |
894 | ||
658cfebc | 895 | if (__bpf_prog_meta[cfg->type].may_uds_export) { |
e4225669 DB |
896 | uds_name = getenv(BPF_ENV_UDS); |
897 | if (argc > 0 && !uds_name && | |
898 | matches(*argv, "export") == 0) { | |
899 | NEXT_ARG(); | |
900 | uds_name = *argv; | |
901 | NEXT_ARG_FWD(); | |
902 | } | |
32e93fb7 DB |
903 | } |
904 | ||
905 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
906 | verbose = true; | |
907 | NEXT_ARG_FWD(); | |
908 | } | |
909 | ||
910 | PREV_ARG(); | |
911 | } | |
912 | ||
3f0b9e62 | 913 | if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) { |
51be7546 | 914 | ret = bpf_ops_parse(argc, argv, cfg->opcodes, |
f20ff2f1 | 915 | cfg->mode == CBPF_FILE); |
3f0b9e62 JK |
916 | cfg->n_opcodes = ret; |
917 | } else if (cfg->mode == EBPF_OBJECT) { | |
918 | ret = 0; /* program will be loaded by load stage */ | |
919 | } else if (cfg->mode == EBPF_PINNED) { | |
658cfebc | 920 | ret = bpf_obj_pinned(file, cfg->type); |
3f0b9e62 JK |
921 | cfg->prog_fd = ret; |
922 | } else { | |
32e93fb7 | 923 | return -1; |
3f0b9e62 | 924 | } |
32e93fb7 | 925 | |
e4225669 DB |
926 | cfg->object = file; |
927 | cfg->section = section; | |
928 | cfg->uds = uds_name; | |
929 | cfg->argc = argc; | |
930 | cfg->argv = argv; | |
3f0b9e62 | 931 | cfg->verbose = verbose; |
91d88eeb DB |
932 | |
933 | return ret; | |
934 | } | |
935 | ||
3f0b9e62 JK |
936 | static int bpf_do_load(struct bpf_cfg_in *cfg) |
937 | { | |
938 | if (cfg->mode == EBPF_OBJECT) { | |
939 | cfg->prog_fd = bpf_obj_open(cfg->object, cfg->type, | |
65fdae3d JK |
940 | cfg->section, cfg->ifindex, |
941 | cfg->verbose); | |
3f0b9e62 JK |
942 | return cfg->prog_fd; |
943 | } | |
944 | return 0; | |
945 | } | |
946 | ||
4a847fcb JK |
947 | int bpf_load_common(struct bpf_cfg_in *cfg, const struct bpf_cfg_ops *ops, |
948 | void *nl) | |
91d88eeb | 949 | { |
91d88eeb | 950 | char annotation[256]; |
91d88eeb DB |
951 | int ret; |
952 | ||
3f0b9e62 | 953 | ret = bpf_do_load(cfg); |
91d88eeb DB |
954 | if (ret < 0) |
955 | return ret; | |
956 | ||
f20ff2f1 | 957 | if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) |
3f0b9e62 | 958 | ops->cbpf_cb(nl, cfg->opcodes, cfg->n_opcodes); |
f20ff2f1 | 959 | if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) { |
32e93fb7 | 960 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
f20ff2f1 | 961 | basename(cfg->object), cfg->mode == EBPF_PINNED ? |
e4225669 | 962 | "*fsobj" : cfg->section); |
3f0b9e62 | 963 | ops->ebpf_cb(nl, cfg->prog_fd, annotation); |
32e93fb7 DB |
964 | } |
965 | ||
91d88eeb DB |
966 | return 0; |
967 | } | |
32e93fb7 | 968 | |
4a847fcb | 969 | int bpf_parse_common(struct bpf_cfg_in *cfg, const struct bpf_cfg_ops *ops) |
e4225669 DB |
970 | { |
971 | bool opt_tbl[BPF_MODE_MAX] = {}; | |
972 | ||
973 | if (ops->cbpf_cb) { | |
974 | opt_tbl[CBPF_BYTECODE] = true; | |
975 | opt_tbl[CBPF_FILE] = true; | |
976 | } | |
977 | ||
978 | if (ops->ebpf_cb) { | |
979 | opt_tbl[EBPF_OBJECT] = true; | |
980 | opt_tbl[EBPF_PINNED] = true; | |
981 | } | |
982 | ||
4a847fcb JK |
983 | return bpf_do_parse(cfg, opt_tbl); |
984 | } | |
985 | ||
986 | int bpf_parse_and_load_common(struct bpf_cfg_in *cfg, | |
987 | const struct bpf_cfg_ops *ops, void *nl) | |
988 | { | |
989 | int ret; | |
990 | ||
991 | ret = bpf_parse_common(cfg, ops); | |
992 | if (ret < 0) | |
993 | return ret; | |
994 | ||
995 | return bpf_load_common(cfg, ops, nl); | |
e4225669 DB |
996 | } |
997 | ||
91d88eeb DB |
998 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
999 | { | |
91d88eeb | 1000 | const bool opt_tbl[BPF_MODE_MAX] = { |
91d88eeb DB |
1001 | [EBPF_OBJECT] = true, |
1002 | [EBPF_PINNED] = true, | |
1003 | }; | |
1004 | const struct bpf_elf_map test = { | |
1005 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
1006 | .size_key = sizeof(int), | |
1007 | .size_value = sizeof(int), | |
1008 | }; | |
e4225669 | 1009 | struct bpf_cfg_in cfg = { |
658cfebc | 1010 | .type = BPF_PROG_TYPE_UNSPEC, |
e4225669 DB |
1011 | .argc = argc, |
1012 | .argv = argv, | |
1013 | }; | |
ecb05c0f | 1014 | struct bpf_map_ext ext = {}; |
91d88eeb | 1015 | int ret, prog_fd, map_fd; |
91d88eeb DB |
1016 | uint32_t map_key; |
1017 | ||
3f0b9e62 JK |
1018 | ret = bpf_do_parse(&cfg, opt_tbl); |
1019 | if (ret < 0) | |
1020 | return ret; | |
1021 | ||
1022 | ret = bpf_do_load(&cfg); | |
1023 | if (ret < 0) | |
1024 | return ret; | |
1025 | ||
1026 | prog_fd = cfg.prog_fd; | |
1027 | ||
91d88eeb DB |
1028 | if (key) { |
1029 | map_key = *key; | |
1030 | } else { | |
e4225669 | 1031 | ret = sscanf(cfg.section, "%*i/%i", &map_key); |
91d88eeb | 1032 | if (ret != 1) { |
32a121cb | 1033 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
1034 | ret = -EINVAL; |
1035 | goto out_prog; | |
1036 | } | |
1037 | } | |
32e93fb7 | 1038 | |
658cfebc | 1039 | map_fd = bpf_obj_get(map_path, cfg.type); |
91d88eeb DB |
1040 | if (map_fd < 0) { |
1041 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
1042 | map_path, strerror(errno)); | |
1043 | ret = map_fd; | |
1044 | goto out_prog; | |
1045 | } | |
1046 | ||
ecb05c0f | 1047 | ret = bpf_map_selfcheck_pinned(map_fd, &test, &ext, |
fb24802b | 1048 | offsetof(struct bpf_elf_map, max_elem), |
658cfebc | 1049 | cfg.type); |
91d88eeb DB |
1050 | if (ret < 0) { |
1051 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
1052 | goto out_map; | |
1053 | } | |
1054 | ||
1055 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
1056 | if (ret < 0) | |
1057 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
1058 | out_map: | |
1059 | close(map_fd); | |
1060 | out_prog: | |
1061 | close(prog_fd); | |
1062 | return ret; | |
32e93fb7 DB |
1063 | } |
1064 | ||
fc4ccce0 DA |
1065 | int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type) |
1066 | { | |
1067 | union bpf_attr attr = {}; | |
1068 | ||
1069 | attr.target_fd = target_fd; | |
1070 | attr.attach_bpf_fd = prog_fd; | |
1071 | attr.attach_type = type; | |
1072 | ||
1073 | return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
1074 | } | |
1075 | ||
1076 | int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type) | |
1077 | { | |
1078 | union bpf_attr attr = {}; | |
1079 | ||
1080 | attr.target_fd = target_fd; | |
1081 | attr.attach_type = type; | |
1082 | ||
1083 | return bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); | |
1084 | } | |
1085 | ||
65fdae3d JK |
1086 | static int bpf_prog_load_dev(enum bpf_prog_type type, |
1087 | const struct bpf_insn *insns, size_t size_insns, | |
1088 | const char *license, __u32 ifindex, | |
1089 | char *log, size_t size_log) | |
869d889e DA |
1090 | { |
1091 | union bpf_attr attr = {}; | |
1092 | ||
1093 | attr.prog_type = type; | |
1094 | attr.insns = bpf_ptr_to_u64(insns); | |
1095 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); | |
1096 | attr.license = bpf_ptr_to_u64(license); | |
65fdae3d | 1097 | attr.prog_ifindex = ifindex; |
869d889e DA |
1098 | |
1099 | if (size_log > 0) { | |
1100 | attr.log_buf = bpf_ptr_to_u64(log); | |
1101 | attr.log_size = size_log; | |
1102 | attr.log_level = 1; | |
1103 | } | |
1104 | ||
1105 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); | |
1106 | } | |
1107 | ||
65fdae3d JK |
1108 | int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, |
1109 | size_t size_insns, const char *license, char *log, | |
1110 | size_t size_log) | |
1111 | { | |
1112 | return bpf_prog_load_dev(type, insns, size_insns, license, 0, | |
1113 | log, size_log); | |
1114 | } | |
1115 | ||
6256f8c9 | 1116 | #ifdef HAVE_ELF |
32e93fb7 DB |
1117 | struct bpf_elf_prog { |
1118 | enum bpf_prog_type type; | |
b5cb33ae DB |
1119 | struct bpf_insn *insns; |
1120 | unsigned int insns_num; | |
32e93fb7 DB |
1121 | size_t size; |
1122 | const char *license; | |
1123 | }; | |
1124 | ||
f6793eec DB |
1125 | struct bpf_hash_entry { |
1126 | unsigned int pinning; | |
1127 | const char *subpath; | |
1128 | struct bpf_hash_entry *next; | |
1129 | }; | |
1130 | ||
ecb05c0f DB |
1131 | struct bpf_config { |
1132 | unsigned int jit_enabled; | |
1133 | }; | |
1134 | ||
f823f360 DB |
1135 | struct bpf_btf { |
1136 | const struct btf_header *hdr; | |
1137 | const void *raw; | |
1138 | const char *strings; | |
1139 | const struct btf_type **types; | |
1140 | int types_num; | |
1141 | }; | |
1142 | ||
32e93fb7 | 1143 | struct bpf_elf_ctx { |
ecb05c0f | 1144 | struct bpf_config cfg; |
32e93fb7 DB |
1145 | Elf *elf_fd; |
1146 | GElf_Ehdr elf_hdr; | |
1147 | Elf_Data *sym_tab; | |
1148 | Elf_Data *str_tab; | |
f823f360 | 1149 | Elf_Data *btf_data; |
6e5094db | 1150 | char obj_uid[64]; |
32e93fb7 | 1151 | int obj_fd; |
f823f360 | 1152 | int btf_fd; |
32e93fb7 DB |
1153 | int map_fds[ELF_MAX_MAPS]; |
1154 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
ecb05c0f | 1155 | struct bpf_map_ext maps_ext[ELF_MAX_MAPS]; |
b5cb33ae | 1156 | struct bpf_elf_prog prog_text; |
f823f360 | 1157 | struct bpf_btf btf; |
32e93fb7 DB |
1158 | int sym_num; |
1159 | int map_num; | |
e4225669 | 1160 | int map_len; |
32e93fb7 DB |
1161 | bool *sec_done; |
1162 | int sec_maps; | |
b5cb33ae | 1163 | int sec_text; |
f823f360 | 1164 | int sec_btf; |
32e93fb7 DB |
1165 | char license[ELF_MAX_LICENSE_LEN]; |
1166 | enum bpf_prog_type type; | |
65fdae3d | 1167 | __u32 ifindex; |
32e93fb7 | 1168 | bool verbose; |
6e5094db | 1169 | bool noafalg; |
32e93fb7 | 1170 | struct bpf_elf_st stat; |
f6793eec | 1171 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
1172 | char *log; |
1173 | size_t log_size; | |
32e93fb7 DB |
1174 | }; |
1175 | ||
6256f8c9 | 1176 | struct bpf_elf_sec_data { |
32e93fb7 DB |
1177 | GElf_Shdr sec_hdr; |
1178 | Elf_Data *sec_data; | |
1179 | const char *sec_name; | |
6256f8c9 DB |
1180 | }; |
1181 | ||
1182 | struct bpf_map_data { | |
32e93fb7 DB |
1183 | int *fds; |
1184 | const char *obj; | |
1185 | struct bpf_elf_st *st; | |
1186 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
1187 | }; |
1188 | ||
f823f360 DB |
1189 | static bool bpf_log_has_data(struct bpf_elf_ctx *ctx) |
1190 | { | |
1191 | return ctx->log && ctx->log[0]; | |
1192 | } | |
1193 | ||
f31645d1 DB |
1194 | static __check_format_string(2, 3) void |
1195 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
1196 | { |
1197 | va_list vl; | |
1198 | ||
1199 | va_start(vl, format); | |
1200 | vfprintf(stderr, format, vl); | |
1201 | va_end(vl); | |
1202 | ||
f823f360 | 1203 | if (bpf_log_has_data(ctx)) { |
afc1a200 DB |
1204 | if (ctx->verbose) { |
1205 | fprintf(stderr, "%s\n", ctx->log); | |
1206 | } else { | |
1207 | unsigned int off = 0, len = strlen(ctx->log); | |
1208 | ||
1209 | if (len > BPF_MAX_LOG) { | |
1210 | off = len - BPF_MAX_LOG; | |
1211 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
1212 | off); | |
1213 | } | |
1214 | fprintf(stderr, "%s\n", ctx->log + off); | |
1215 | } | |
1216 | ||
f31645d1 DB |
1217 | memset(ctx->log, 0, ctx->log_size); |
1218 | } | |
1219 | } | |
1220 | ||
1221 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
1222 | { | |
0f74d0f3 | 1223 | const size_t log_max = UINT_MAX >> 8; |
f31645d1 | 1224 | size_t log_size = ctx->log_size; |
f6a54d72 | 1225 | char *ptr; |
f31645d1 DB |
1226 | |
1227 | if (!ctx->log) { | |
1228 | log_size = 65536; | |
0f74d0f3 | 1229 | } else if (log_size < log_max) { |
f31645d1 | 1230 | log_size <<= 1; |
0f74d0f3 TG |
1231 | if (log_size > log_max) |
1232 | log_size = log_max; | |
1233 | } else { | |
1234 | return -EINVAL; | |
d937a74b | 1235 | } |
f31645d1 DB |
1236 | |
1237 | ptr = realloc(ctx->log, log_size); | |
1238 | if (!ptr) | |
1239 | return -ENOMEM; | |
1240 | ||
f6a54d72 | 1241 | ptr[0] = 0; |
f31645d1 DB |
1242 | ctx->log = ptr; |
1243 | ctx->log_size = log_size; | |
1244 | ||
1245 | return 0; | |
11c39b5e DB |
1246 | } |
1247 | ||
4dd3f50a DB |
1248 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
1249 | uint32_t size_value, uint32_t max_elem, | |
f823f360 DB |
1250 | uint32_t flags, int inner_fd, int btf_fd, |
1251 | uint32_t ifindex, uint32_t btf_id_key, | |
1252 | uint32_t btf_id_val) | |
11c39b5e | 1253 | { |
d17b136f | 1254 | union bpf_attr attr = {}; |
67584e3a | 1255 | |
67584e3a ND |
1256 | attr.map_type = type; |
1257 | attr.key_size = size_key; | |
612ff099 | 1258 | attr.value_size = inner_fd ? sizeof(int) : size_value; |
67584e3a | 1259 | attr.max_entries = max_elem; |
4dd3f50a | 1260 | attr.map_flags = flags; |
612ff099 | 1261 | attr.inner_map_fd = inner_fd; |
5691e6bc | 1262 | attr.map_ifindex = ifindex; |
f823f360 DB |
1263 | attr.btf_fd = btf_fd; |
1264 | attr.btf_key_type_id = btf_id_key; | |
1265 | attr.btf_value_type_id = btf_id_val; | |
11c39b5e DB |
1266 | |
1267 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
1268 | } | |
11c39b5e | 1269 | |
f823f360 DB |
1270 | static int bpf_btf_load(void *btf, size_t size_btf, |
1271 | char *log, size_t size_log) | |
1272 | { | |
1273 | union bpf_attr attr = {}; | |
1274 | ||
1275 | attr.btf = bpf_ptr_to_u64(btf); | |
1276 | attr.btf_size = size_btf; | |
1277 | ||
1278 | if (size_log > 0) { | |
1279 | attr.btf_log_buf = bpf_ptr_to_u64(log); | |
1280 | attr.btf_log_size = size_log; | |
1281 | attr.btf_log_level = 1; | |
1282 | } | |
1283 | ||
1284 | return bpf(BPF_BTF_LOAD, &attr, sizeof(attr)); | |
1285 | } | |
1286 | ||
32e93fb7 | 1287 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 1288 | { |
d17b136f | 1289 | union bpf_attr attr = {}; |
67584e3a | 1290 | |
67584e3a ND |
1291 | attr.pathname = bpf_ptr_to_u64(pathname); |
1292 | attr.bpf_fd = fd; | |
32e93fb7 DB |
1293 | |
1294 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
1295 | } | |
11c39b5e | 1296 | |
32e93fb7 DB |
1297 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
1298 | { | |
1299 | struct sockaddr_alg alg = { | |
1300 | .salg_family = AF_ALG, | |
1301 | .salg_type = "hash", | |
1302 | .salg_name = "sha1", | |
1303 | }; | |
1304 | int ret, cfd, ofd, ffd; | |
1305 | struct stat stbuff; | |
1306 | ssize_t size; | |
1307 | ||
1308 | if (!object || len != 20) | |
1309 | return -EINVAL; | |
1310 | ||
1311 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
6e5094db | 1312 | if (cfd < 0) |
32e93fb7 | 1313 | return cfd; |
32e93fb7 DB |
1314 | |
1315 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
6e5094db | 1316 | if (ret < 0) |
32e93fb7 | 1317 | goto out_cfd; |
32e93fb7 DB |
1318 | |
1319 | ofd = accept(cfd, NULL, 0); | |
1320 | if (ofd < 0) { | |
32e93fb7 DB |
1321 | ret = ofd; |
1322 | goto out_cfd; | |
1323 | } | |
1324 | ||
1325 | ffd = open(object, O_RDONLY); | |
1326 | if (ffd < 0) { | |
1327 | fprintf(stderr, "Error opening object %s: %s\n", | |
1328 | object, strerror(errno)); | |
1329 | ret = ffd; | |
1330 | goto out_ofd; | |
1331 | } | |
1332 | ||
32a121cb | 1333 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
1334 | if (ret < 0) { |
1335 | fprintf(stderr, "Error doing fstat: %s\n", | |
1336 | strerror(errno)); | |
1337 | goto out_ffd; | |
d937a74b | 1338 | } |
11c39b5e | 1339 | |
32e93fb7 DB |
1340 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
1341 | if (size != stbuff.st_size) { | |
1342 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
1343 | size, stbuff.st_size, strerror(errno)); | |
1344 | ret = -1; | |
1345 | goto out_ffd; | |
1346 | } | |
1347 | ||
1348 | size = read(ofd, out, len); | |
1349 | if (size != len) { | |
1350 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
1351 | size, len, strerror(errno)); | |
1352 | ret = -1; | |
1353 | } else { | |
1354 | ret = 0; | |
1355 | } | |
1356 | out_ffd: | |
1357 | close(ffd); | |
1358 | out_ofd: | |
1359 | close(ofd); | |
1360 | out_cfd: | |
1361 | close(cfd); | |
1362 | return ret; | |
11c39b5e DB |
1363 | } |
1364 | ||
6e5094db | 1365 | static void bpf_init_env(void) |
32e93fb7 DB |
1366 | { |
1367 | struct rlimit limit = { | |
1368 | .rlim_cur = RLIM_INFINITY, | |
1369 | .rlim_max = RLIM_INFINITY, | |
1370 | }; | |
1371 | ||
1372 | /* Don't bother in case we fail! */ | |
1373 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
1374 | ||
6e5094db | 1375 | if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) |
32a121cb | 1376 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
6256f8c9 DB |
1377 | } |
1378 | ||
f6793eec DB |
1379 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
1380 | uint32_t pinning) | |
1381 | { | |
1382 | struct bpf_hash_entry *entry; | |
1383 | ||
1384 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1385 | while (entry && entry->pinning != pinning) | |
1386 | entry = entry->next; | |
1387 | ||
1388 | return entry ? entry->subpath : NULL; | |
1389 | } | |
1390 | ||
1391 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
1392 | uint32_t pinning) | |
11c39b5e | 1393 | { |
32e93fb7 DB |
1394 | switch (pinning) { |
1395 | case PIN_OBJECT_NS: | |
1396 | case PIN_GLOBAL_NS: | |
1397 | return false; | |
1398 | case PIN_NONE: | |
32e93fb7 | 1399 | return true; |
f6793eec DB |
1400 | default: |
1401 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
1402 | } |
1403 | } | |
1404 | ||
1405 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 1406 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
1407 | { |
1408 | switch (pinning) { | |
1409 | case PIN_OBJECT_NS: | |
e4225669 DB |
1410 | snprintf(pathname, len, "%s/%s/%s", |
1411 | bpf_get_work_dir(ctx->type), | |
6e5094db | 1412 | ctx->obj_uid, name); |
32e93fb7 DB |
1413 | break; |
1414 | case PIN_GLOBAL_NS: | |
e4225669 DB |
1415 | snprintf(pathname, len, "%s/%s/%s", |
1416 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1417 | BPF_DIR_GLOBALS, name); |
1418 | break; | |
f6793eec | 1419 | default: |
e4225669 DB |
1420 | snprintf(pathname, len, "%s/../%s/%s", |
1421 | bpf_get_work_dir(ctx->type), | |
f6793eec DB |
1422 | bpf_custom_pinning(ctx, pinning), name); |
1423 | break; | |
32e93fb7 DB |
1424 | } |
1425 | } | |
1426 | ||
f6793eec DB |
1427 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1428 | uint32_t pinning) | |
32e93fb7 DB |
1429 | { |
1430 | char pathname[PATH_MAX]; | |
1431 | ||
e4225669 | 1432 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
32e93fb7 DB |
1433 | return 0; |
1434 | ||
f6793eec | 1435 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
e4225669 | 1436 | return bpf_obj_get(pathname, ctx->type); |
32e93fb7 DB |
1437 | } |
1438 | ||
e4225669 | 1439 | static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx) |
32e93fb7 | 1440 | { |
358abfe0 | 1441 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1442 | int ret; |
1443 | ||
358abfe0 AC |
1444 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type), |
1445 | ctx->obj_uid); | |
f6793eec DB |
1446 | |
1447 | ret = mkdir(tmp, S_IRWXU); | |
1448 | if (ret && errno != EEXIST) { | |
1449 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
358abfe0 | 1450 | return ret; |
f6793eec DB |
1451 | } |
1452 | ||
358abfe0 | 1453 | return 0; |
f6793eec DB |
1454 | } |
1455 | ||
e4225669 DB |
1456 | static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx, |
1457 | const char *todo) | |
f6793eec | 1458 | { |
358abfe0 | 1459 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; |
f6793eec DB |
1460 | int ret; |
1461 | ||
358abfe0 AC |
1462 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type)); |
1463 | snprintf(rem, sizeof(rem), "%s/", todo); | |
c0325b06 | 1464 | sub = strtok(rem, "/"); |
358abfe0 | 1465 | |
f6793eec DB |
1466 | while (sub) { |
1467 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1468 | return -EINVAL; | |
1469 | ||
1470 | strcat(tmp, sub); | |
1471 | strcat(tmp, "/"); | |
32e93fb7 | 1472 | |
f6793eec | 1473 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1474 | if (ret && errno != EEXIST) { |
f6793eec | 1475 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 | 1476 | strerror(errno)); |
358abfe0 | 1477 | return ret; |
32e93fb7 | 1478 | } |
f6793eec DB |
1479 | |
1480 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1481 | } |
1482 | ||
358abfe0 | 1483 | return 0; |
f6793eec DB |
1484 | } |
1485 | ||
1486 | static int bpf_place_pinned(int fd, const char *name, | |
1487 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1488 | { | |
1489 | char pathname[PATH_MAX]; | |
1490 | const char *tmp; | |
1491 | int ret = 0; | |
1492 | ||
e4225669 | 1493 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
f6793eec DB |
1494 | return 0; |
1495 | ||
1496 | if (pinning == PIN_OBJECT_NS) | |
e4225669 | 1497 | ret = bpf_make_obj_path(ctx); |
f6793eec | 1498 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) |
e4225669 | 1499 | ret = bpf_make_custom_path(ctx, tmp); |
f6793eec DB |
1500 | if (ret < 0) |
1501 | return ret; | |
1502 | ||
1503 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1504 | return bpf_obj_pin(fd, pathname); |
1505 | } | |
1506 | ||
f31645d1 DB |
1507 | static void bpf_prog_report(int fd, const char *section, |
1508 | const struct bpf_elf_prog *prog, | |
1509 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1510 | { |
afc1a200 DB |
1511 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1512 | ||
1513 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1514 | fd < 0 ? "rejected: " : "loaded", |
1515 | fd < 0 ? strerror(errno) : "", | |
1516 | fd < 0 ? errno : fd); | |
1517 | ||
1518 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1519 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1520 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1521 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1522 | ||
1523 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1524 | } | |
32e93fb7 | 1525 | |
f31645d1 DB |
1526 | static int bpf_prog_attach(const char *section, |
1527 | const struct bpf_elf_prog *prog, | |
1528 | struct bpf_elf_ctx *ctx) | |
1529 | { | |
1530 | int tries = 0, fd; | |
1531 | retry: | |
32e93fb7 | 1532 | errno = 0; |
65fdae3d JK |
1533 | fd = bpf_prog_load_dev(prog->type, prog->insns, prog->size, |
1534 | prog->license, ctx->ifindex, | |
1535 | ctx->log, ctx->log_size); | |
f31645d1 DB |
1536 | if (fd < 0 || ctx->verbose) { |
1537 | /* The verifier log is pretty chatty, sometimes so chatty | |
1538 | * on larger programs, that we could fail to dump everything | |
1539 | * into our buffer. Still, try to give a debuggable error | |
1540 | * log for the user, so enlarge it and re-fail. | |
1541 | */ | |
1542 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
0f74d0f3 | 1543 | if (tries++ < 10 && !bpf_log_realloc(ctx)) |
f31645d1 DB |
1544 | goto retry; |
1545 | ||
32a121cb | 1546 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1547 | ctx->log_size, tries); |
1548 | return fd; | |
1549 | } | |
1550 | ||
1551 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1552 | } |
1553 | ||
1554 | return fd; | |
1555 | } | |
1556 | ||
f31645d1 DB |
1557 | static void bpf_map_report(int fd, const char *name, |
1558 | const struct bpf_elf_map *map, | |
612ff099 | 1559 | struct bpf_elf_ctx *ctx, int inner_fd) |
f31645d1 DB |
1560 | { |
1561 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1562 | fd < 0 ? "rejected: " : "loaded", | |
1563 | fd < 0 ? strerror(errno) : "", | |
1564 | fd < 0 ? errno : fd); | |
1565 | ||
1566 | fprintf(stderr, " - Type: %u\n", map->type); | |
1567 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1568 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1569 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
612ff099 DB |
1570 | fprintf(stderr, " - Size value: %u\n", |
1571 | inner_fd ? (int)sizeof(int) : map->size_value); | |
4dd3f50a DB |
1572 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1573 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1574 | } |
1575 | ||
612ff099 DB |
1576 | static int bpf_find_map_id(const struct bpf_elf_ctx *ctx, uint32_t id) |
1577 | { | |
1578 | int i; | |
1579 | ||
1580 | for (i = 0; i < ctx->map_num; i++) { | |
1581 | if (ctx->maps[i].id != id) | |
1582 | continue; | |
1583 | if (ctx->map_fds[i] < 0) | |
1584 | return -EINVAL; | |
1585 | ||
1586 | return ctx->map_fds[i]; | |
1587 | } | |
1588 | ||
1589 | return -ENOENT; | |
1590 | } | |
1591 | ||
0b5eadc5 | 1592 | static void bpf_report_map_in_map(int outer_fd, uint32_t idx) |
612ff099 DB |
1593 | { |
1594 | struct bpf_elf_map outer_map; | |
1595 | int ret; | |
1596 | ||
1597 | fprintf(stderr, "Cannot insert map into map! "); | |
1598 | ||
ecb05c0f | 1599 | ret = bpf_derive_elf_map_from_fdinfo(outer_fd, &outer_map, NULL); |
612ff099 DB |
1600 | if (!ret) { |
1601 | if (idx >= outer_map.max_elem && | |
1602 | outer_map.type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
1603 | fprintf(stderr, "Outer map has %u elements, index %u is invalid!\n", | |
1604 | outer_map.max_elem, idx); | |
1605 | return; | |
1606 | } | |
1607 | } | |
1608 | ||
1609 | fprintf(stderr, "Different map specs used for outer and inner map?\n"); | |
1610 | } | |
1611 | ||
1612 | static bool bpf_is_map_in_map_type(const struct bpf_elf_map *map) | |
1613 | { | |
1614 | return map->type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1615 | map->type == BPF_MAP_TYPE_HASH_OF_MAPS; | |
1616 | } | |
1617 | ||
33fde2b6 SH |
1618 | static bool bpf_map_offload_neutral(enum bpf_map_type type) |
1619 | { | |
1620 | return type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; | |
1621 | } | |
1622 | ||
ecb05c0f DB |
1623 | static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx, |
1624 | const struct bpf_elf_map *map, struct bpf_map_ext *ext, | |
1625 | int *have_map_in_map) | |
32e93fb7 | 1626 | { |
0c0394ff | 1627 | int fd, ifindex, ret, map_inner_fd = 0; |
e4c4685f | 1628 | bool retried = false; |
32e93fb7 | 1629 | |
e4c4685f | 1630 | probe: |
f6793eec | 1631 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1632 | if (fd > 0) { |
ecb05c0f | 1633 | ret = bpf_map_selfcheck_pinned(fd, map, ext, |
91d88eeb | 1634 | offsetof(struct bpf_elf_map, |
fb24802b | 1635 | id), ctx->type); |
9e607f2e DB |
1636 | if (ret < 0) { |
1637 | close(fd); | |
1638 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1639 | name); | |
1640 | return ret; | |
1641 | } | |
f31645d1 | 1642 | if (ctx->verbose) |
32e93fb7 DB |
1643 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1644 | name); | |
1645 | return fd; | |
1646 | } | |
1647 | ||
612ff099 DB |
1648 | if (have_map_in_map && bpf_is_map_in_map_type(map)) { |
1649 | (*have_map_in_map)++; | |
1650 | if (map->inner_id) | |
1651 | return 0; | |
1652 | fprintf(stderr, "Map \'%s\' cannot be created since no inner map ID defined!\n", | |
1653 | name); | |
1654 | return -EINVAL; | |
1655 | } | |
1656 | ||
1657 | if (!have_map_in_map && bpf_is_map_in_map_type(map)) { | |
1658 | map_inner_fd = bpf_find_map_id(ctx, map->inner_id); | |
1659 | if (map_inner_fd < 0) { | |
1660 | fprintf(stderr, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n", | |
1661 | name, map->inner_id); | |
1662 | return -EINVAL; | |
1663 | } | |
1664 | } | |
1665 | ||
0c0394ff | 1666 | ifindex = bpf_map_offload_neutral(map->type) ? 0 : ctx->ifindex; |
32e93fb7 DB |
1667 | errno = 0; |
1668 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
f823f360 DB |
1669 | map->max_elem, map->flags, map_inner_fd, ctx->btf_fd, |
1670 | ifindex, ext->btf_id_key, ext->btf_id_val); | |
5691e6bc | 1671 | |
f31645d1 | 1672 | if (fd < 0 || ctx->verbose) { |
612ff099 | 1673 | bpf_map_report(fd, name, map, ctx, map_inner_fd); |
32e93fb7 DB |
1674 | if (fd < 0) |
1675 | return fd; | |
1676 | } | |
1677 | ||
f6793eec | 1678 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
e4c4685f JS |
1679 | if (ret < 0) { |
1680 | close(fd); | |
1681 | if (!retried && errno == EEXIST) { | |
1682 | retried = true; | |
1683 | goto probe; | |
1684 | } | |
32e93fb7 DB |
1685 | fprintf(stderr, "Could not pin %s map: %s\n", name, |
1686 | strerror(errno)); | |
32e93fb7 DB |
1687 | return ret; |
1688 | } | |
1689 | ||
1690 | return fd; | |
1691 | } | |
1692 | ||
32e93fb7 DB |
1693 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1694 | const GElf_Sym *sym) | |
1695 | { | |
1696 | return ctx->str_tab->d_buf + sym->st_name; | |
1697 | } | |
1698 | ||
f823f360 DB |
1699 | static int bpf_btf_find(struct bpf_elf_ctx *ctx, const char *name) |
1700 | { | |
1701 | const struct btf_type *type; | |
1702 | const char *res; | |
1703 | int id; | |
1704 | ||
1705 | for (id = 1; id < ctx->btf.types_num; id++) { | |
1706 | type = ctx->btf.types[id]; | |
1707 | if (type->name_off >= ctx->btf.hdr->str_len) | |
1708 | continue; | |
1709 | res = &ctx->btf.strings[type->name_off]; | |
1710 | if (!strcmp(res, name)) | |
1711 | return id; | |
1712 | } | |
1713 | ||
1714 | return -ENOENT; | |
1715 | } | |
1716 | ||
1717 | static int bpf_btf_find_kv(struct bpf_elf_ctx *ctx, const struct bpf_elf_map *map, | |
1718 | const char *name, uint32_t *id_key, uint32_t *id_val) | |
1719 | { | |
1720 | const struct btf_member *key, *val; | |
1721 | const struct btf_type *type; | |
1722 | char btf_name[512]; | |
1723 | const char *res; | |
1724 | int id; | |
1725 | ||
1726 | snprintf(btf_name, sizeof(btf_name), "____btf_map_%s", name); | |
1727 | id = bpf_btf_find(ctx, btf_name); | |
1728 | if (id < 0) | |
1729 | return id; | |
1730 | ||
1731 | type = ctx->btf.types[id]; | |
1732 | if (BTF_INFO_KIND(type->info) != BTF_KIND_STRUCT) | |
1733 | return -EINVAL; | |
1734 | if (BTF_INFO_VLEN(type->info) != 2) | |
1735 | return -EINVAL; | |
1736 | ||
1737 | key = ((void *) type) + sizeof(*type); | |
1738 | val = key + 1; | |
1739 | if (!key->type || key->type >= ctx->btf.types_num || | |
1740 | !val->type || val->type >= ctx->btf.types_num) | |
1741 | return -EINVAL; | |
1742 | ||
1743 | if (key->name_off >= ctx->btf.hdr->str_len || | |
1744 | val->name_off >= ctx->btf.hdr->str_len) | |
1745 | return -EINVAL; | |
1746 | ||
1747 | res = &ctx->btf.strings[key->name_off]; | |
1748 | if (strcmp(res, "key")) | |
1749 | return -EINVAL; | |
1750 | ||
1751 | res = &ctx->btf.strings[val->name_off]; | |
1752 | if (strcmp(res, "value")) | |
1753 | return -EINVAL; | |
1754 | ||
1755 | *id_key = key->type; | |
1756 | *id_val = val->type; | |
1757 | return 0; | |
1758 | } | |
1759 | ||
1760 | static void bpf_btf_annotate(struct bpf_elf_ctx *ctx, int which, const char *name) | |
1761 | { | |
1762 | uint32_t id_key = 0, id_val = 0; | |
1763 | ||
1764 | if (!bpf_btf_find_kv(ctx, &ctx->maps[which], name, &id_key, &id_val)) { | |
1765 | ctx->maps_ext[which].btf_id_key = id_key; | |
1766 | ctx->maps_ext[which].btf_id_val = id_val; | |
1767 | } | |
1768 | } | |
1769 | ||
32e93fb7 DB |
1770 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) |
1771 | { | |
f823f360 | 1772 | const char *name; |
32e93fb7 | 1773 | GElf_Sym sym; |
11c39b5e DB |
1774 | int i; |
1775 | ||
32e93fb7 | 1776 | for (i = 0; i < ctx->sym_num; i++) { |
1a7d3ad8 | 1777 | int type; |
7a04dd84 | 1778 | |
32e93fb7 DB |
1779 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) |
1780 | continue; | |
1781 | ||
1a7d3ad8 | 1782 | type = GELF_ST_TYPE(sym.st_info); |
5230a2ed | 1783 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
7a04dd84 | 1784 | (type != STT_NOTYPE && type != STT_OBJECT) || |
32e93fb7 | 1785 | sym.st_shndx != ctx->sec_maps || |
e4225669 | 1786 | sym.st_value / ctx->map_len != which) |
32e93fb7 DB |
1787 | continue; |
1788 | ||
f823f360 DB |
1789 | name = bpf_str_tab_name(ctx, &sym); |
1790 | bpf_btf_annotate(ctx, which, name); | |
1791 | return name; | |
11c39b5e | 1792 | } |
32e93fb7 DB |
1793 | |
1794 | return NULL; | |
11c39b5e DB |
1795 | } |
1796 | ||
32e93fb7 | 1797 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1798 | { |
612ff099 | 1799 | int i, j, ret, fd, inner_fd, inner_idx, have_map_in_map = 0; |
32e93fb7 | 1800 | const char *map_name; |
11c39b5e | 1801 | |
32e93fb7 | 1802 | for (i = 0; i < ctx->map_num; i++) { |
6e5094db DB |
1803 | if (ctx->maps[i].pinning == PIN_OBJECT_NS && |
1804 | ctx->noafalg) { | |
1805 | fprintf(stderr, "Missing kernel AF_ALG support for PIN_OBJECT_NS!\n"); | |
1806 | return -ENOTSUP; | |
1807 | } | |
1808 | ||
32e93fb7 DB |
1809 | map_name = bpf_map_fetch_name(ctx, i); |
1810 | if (!map_name) | |
1811 | return -EIO; | |
11c39b5e | 1812 | |
ecb05c0f DB |
1813 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1814 | &ctx->maps_ext[i], &have_map_in_map); | |
612ff099 DB |
1815 | if (fd < 0) |
1816 | return fd; | |
1817 | ||
1818 | ctx->map_fds[i] = !fd ? -1 : fd; | |
1819 | } | |
1820 | ||
1821 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { | |
1822 | if (ctx->map_fds[i] >= 0) | |
1823 | continue; | |
1824 | ||
1825 | map_name = bpf_map_fetch_name(ctx, i); | |
1826 | if (!map_name) | |
1827 | return -EIO; | |
1828 | ||
ecb05c0f DB |
1829 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1830 | &ctx->maps_ext[i], NULL); | |
32e93fb7 DB |
1831 | if (fd < 0) |
1832 | return fd; | |
11c39b5e | 1833 | |
32e93fb7 | 1834 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1835 | } |
1836 | ||
612ff099 DB |
1837 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { |
1838 | if (!ctx->maps[i].id || | |
1839 | ctx->maps[i].inner_id || | |
1840 | ctx->maps[i].inner_idx == -1) | |
1841 | continue; | |
1842 | ||
1843 | inner_fd = ctx->map_fds[i]; | |
1844 | inner_idx = ctx->maps[i].inner_idx; | |
1845 | ||
1846 | for (j = 0; j < ctx->map_num; j++) { | |
1847 | if (!bpf_is_map_in_map_type(&ctx->maps[j])) | |
1848 | continue; | |
1849 | if (ctx->maps[j].inner_id != ctx->maps[i].id) | |
1850 | continue; | |
1851 | ||
1852 | ret = bpf_map_update(ctx->map_fds[j], &inner_idx, | |
1853 | &inner_fd, BPF_ANY); | |
1854 | if (ret < 0) { | |
1855 | bpf_report_map_in_map(ctx->map_fds[j], | |
0b5eadc5 | 1856 | inner_idx); |
612ff099 DB |
1857 | return ret; |
1858 | } | |
1859 | } | |
1860 | } | |
1861 | ||
11c39b5e | 1862 | return 0; |
11c39b5e DB |
1863 | } |
1864 | ||
e4225669 DB |
1865 | static int bpf_map_num_sym(struct bpf_elf_ctx *ctx) |
1866 | { | |
1867 | int i, num = 0; | |
1868 | GElf_Sym sym; | |
1869 | ||
1870 | for (i = 0; i < ctx->sym_num; i++) { | |
1a7d3ad8 | 1871 | int type; |
7a04dd84 | 1872 | |
e4225669 DB |
1873 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) |
1874 | continue; | |
1875 | ||
1a7d3ad8 | 1876 | type = GELF_ST_TYPE(sym.st_info); |
e4225669 | 1877 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
7a04dd84 | 1878 | (type != STT_NOTYPE && type != STT_OBJECT) || |
e4225669 DB |
1879 | sym.st_shndx != ctx->sec_maps) |
1880 | continue; | |
1881 | num++; | |
1882 | } | |
1883 | ||
1884 | return num; | |
1885 | } | |
1886 | ||
32e93fb7 DB |
1887 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1888 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1889 | { |
32e93fb7 | 1890 | Elf_Data *sec_edata; |
11c39b5e DB |
1891 | GElf_Shdr sec_hdr; |
1892 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1893 | char *sec_name; |
1894 | ||
32e93fb7 | 1895 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1896 | |
32e93fb7 | 1897 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1898 | if (!sec_fd) |
1899 | return -EINVAL; | |
11c39b5e DB |
1900 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1901 | return -EIO; | |
1902 | ||
32e93fb7 | 1903 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1904 | sec_hdr.sh_name); |
1905 | if (!sec_name || !sec_hdr.sh_size) | |
1906 | return -ENOENT; | |
1907 | ||
1908 | sec_edata = elf_getdata(sec_fd, NULL); | |
1909 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1910 | return -EIO; | |
1911 | ||
32e93fb7 | 1912 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1913 | |
32e93fb7 DB |
1914 | data->sec_name = sec_name; |
1915 | data->sec_data = sec_edata; | |
11c39b5e DB |
1916 | return 0; |
1917 | } | |
1918 | ||
e4225669 DB |
1919 | struct bpf_elf_map_min { |
1920 | __u32 type; | |
1921 | __u32 size_key; | |
1922 | __u32 size_value; | |
1923 | __u32 max_elem; | |
1924 | }; | |
11c39b5e | 1925 | |
e4225669 DB |
1926 | static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section, |
1927 | struct bpf_elf_sec_data *data) | |
1928 | { | |
1929 | ctx->map_num = data->sec_data->d_size; | |
32e93fb7 DB |
1930 | ctx->sec_maps = section; |
1931 | ctx->sec_done[section] = true; | |
11c39b5e | 1932 | |
e4225669 | 1933 | if (ctx->map_num > sizeof(ctx->maps)) { |
32e93fb7 DB |
1934 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); |
1935 | return -ENOMEM; | |
1936 | } | |
11c39b5e | 1937 | |
e4225669 DB |
1938 | memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num); |
1939 | return 0; | |
1940 | } | |
1941 | ||
1942 | static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end) | |
1943 | { | |
1944 | GElf_Sym sym; | |
1945 | int off, i; | |
1946 | ||
1947 | for (off = 0; off < end; off += ctx->map_len) { | |
1948 | /* Order doesn't need to be linear here, hence we walk | |
1949 | * the table again. | |
1950 | */ | |
1951 | for (i = 0; i < ctx->sym_num; i++) { | |
1a7d3ad8 | 1952 | int type; |
7a04dd84 | 1953 | |
e4225669 DB |
1954 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) |
1955 | continue; | |
1a7d3ad8 QM |
1956 | |
1957 | type = GELF_ST_TYPE(sym.st_info); | |
e4225669 | 1958 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
7a04dd84 | 1959 | (type != STT_NOTYPE && type != STT_OBJECT) || |
e4225669 DB |
1960 | sym.st_shndx != ctx->sec_maps) |
1961 | continue; | |
1962 | if (sym.st_value == off) | |
1963 | break; | |
1964 | if (i == ctx->sym_num - 1) | |
1965 | return -1; | |
1966 | } | |
1967 | } | |
1968 | ||
1969 | return off == end ? 0 : -1; | |
1970 | } | |
1971 | ||
1972 | static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx) | |
1973 | { | |
1974 | struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {}; | |
1975 | int i, sym_num = bpf_map_num_sym(ctx); | |
1976 | __u8 *buff; | |
1977 | ||
1978 | if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) { | |
1979 | fprintf(stderr, "%u maps not supported in current map section!\n", | |
1980 | sym_num); | |
1981 | return -EINVAL; | |
1982 | } | |
1983 | ||
1984 | if (ctx->map_num % sym_num != 0 || | |
1985 | ctx->map_num % sizeof(__u32) != 0) { | |
1986 | fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n"); | |
1987 | return -EINVAL; | |
1988 | } | |
1989 | ||
1990 | ctx->map_len = ctx->map_num / sym_num; | |
1991 | if (bpf_map_verify_all_offs(ctx, ctx->map_num)) { | |
1992 | fprintf(stderr, "Different struct bpf_elf_map in use!\n"); | |
1993 | return -EINVAL; | |
1994 | } | |
1995 | ||
1996 | if (ctx->map_len == sizeof(struct bpf_elf_map)) { | |
1997 | ctx->map_num = sym_num; | |
1998 | return 0; | |
1999 | } else if (ctx->map_len > sizeof(struct bpf_elf_map)) { | |
2000 | fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n"); | |
2001 | return -EINVAL; | |
2002 | } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) { | |
2003 | fprintf(stderr, "struct bpf_elf_map too small, not supported!\n"); | |
2004 | return -EINVAL; | |
2005 | } | |
2006 | ||
2007 | ctx->map_num = sym_num; | |
2008 | for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num; | |
2009 | i++, buff += ctx->map_len) { | |
2010 | /* The fixup leaves the rest of the members as zero, which | |
2011 | * is fine currently, but option exist to set some other | |
2012 | * default value as well when needed in future. | |
2013 | */ | |
2014 | memcpy(&fixup[i], buff, ctx->map_len); | |
2015 | } | |
2016 | ||
2017 | memcpy(ctx->maps, fixup, sizeof(fixup)); | |
282a1fe1 DB |
2018 | if (ctx->verbose) |
2019 | printf("%zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n", | |
2020 | sizeof(struct bpf_elf_map) - ctx->map_len); | |
32e93fb7 DB |
2021 | return 0; |
2022 | } | |
11c39b5e | 2023 | |
32e93fb7 DB |
2024 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
2025 | struct bpf_elf_sec_data *data) | |
2026 | { | |
2027 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
2028 | return -ENOMEM; | |
11c39b5e | 2029 | |
32e93fb7 DB |
2030 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
2031 | ctx->sec_done[section] = true; | |
2032 | return 0; | |
2033 | } | |
11c39b5e | 2034 | |
32e93fb7 DB |
2035 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
2036 | struct bpf_elf_sec_data *data) | |
2037 | { | |
2038 | ctx->sym_tab = data->sec_data; | |
2039 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
2040 | ctx->sec_done[section] = true; | |
11c39b5e DB |
2041 | return 0; |
2042 | } | |
2043 | ||
32e93fb7 DB |
2044 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
2045 | struct bpf_elf_sec_data *data) | |
11c39b5e | 2046 | { |
32e93fb7 DB |
2047 | ctx->str_tab = data->sec_data; |
2048 | ctx->sec_done[section] = true; | |
2049 | return 0; | |
2050 | } | |
11c39b5e | 2051 | |
b5cb33ae DB |
2052 | static int bpf_fetch_text(struct bpf_elf_ctx *ctx, int section, |
2053 | struct bpf_elf_sec_data *data) | |
2054 | { | |
2055 | ctx->sec_text = section; | |
2056 | ctx->sec_done[section] = true; | |
2057 | return 0; | |
2058 | } | |
2059 | ||
f823f360 DB |
2060 | static void bpf_btf_report(int fd, struct bpf_elf_ctx *ctx) |
2061 | { | |
2062 | fprintf(stderr, "\nBTF debug data section \'.BTF\' %s%s (%d)!\n", | |
2063 | fd < 0 ? "rejected: " : "loaded", | |
2064 | fd < 0 ? strerror(errno) : "", | |
2065 | fd < 0 ? errno : fd); | |
2066 | ||
2067 | fprintf(stderr, " - Length: %zu\n", ctx->btf_data->d_size); | |
2068 | ||
2069 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
2070 | } | |
2071 | ||
2072 | static int bpf_btf_attach(struct bpf_elf_ctx *ctx) | |
2073 | { | |
2074 | int tries = 0, fd; | |
2075 | retry: | |
2076 | errno = 0; | |
2077 | fd = bpf_btf_load(ctx->btf_data->d_buf, ctx->btf_data->d_size, | |
2078 | ctx->log, ctx->log_size); | |
2079 | if (fd < 0 || ctx->verbose) { | |
2080 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
2081 | if (tries++ < 10 && !bpf_log_realloc(ctx)) | |
2082 | goto retry; | |
2083 | ||
2084 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", | |
2085 | ctx->log_size, tries); | |
2086 | return fd; | |
2087 | } | |
2088 | ||
2089 | if (bpf_log_has_data(ctx)) | |
2090 | bpf_btf_report(fd, ctx); | |
2091 | } | |
2092 | ||
2093 | return fd; | |
2094 | } | |
2095 | ||
2096 | static int bpf_fetch_btf_begin(struct bpf_elf_ctx *ctx, int section, | |
2097 | struct bpf_elf_sec_data *data) | |
2098 | { | |
2099 | ctx->btf_data = data->sec_data; | |
2100 | ctx->sec_btf = section; | |
2101 | ctx->sec_done[section] = true; | |
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | static int bpf_btf_check_header(struct bpf_elf_ctx *ctx) | |
2106 | { | |
2107 | const struct btf_header *hdr = ctx->btf_data->d_buf; | |
2108 | const char *str_start, *str_end; | |
2109 | unsigned int data_len; | |
2110 | ||
2111 | if (hdr->magic != BTF_MAGIC) { | |
2112 | fprintf(stderr, "Object has wrong BTF magic: %x, expected: %x!\n", | |
2113 | hdr->magic, BTF_MAGIC); | |
2114 | return -EINVAL; | |
2115 | } | |
2116 | ||
2117 | if (hdr->version != BTF_VERSION) { | |
2118 | fprintf(stderr, "Object has wrong BTF version: %u, expected: %u!\n", | |
2119 | hdr->version, BTF_VERSION); | |
2120 | return -EINVAL; | |
2121 | } | |
2122 | ||
2123 | if (hdr->flags) { | |
2124 | fprintf(stderr, "Object has unsupported BTF flags %x!\n", | |
2125 | hdr->flags); | |
2126 | return -EINVAL; | |
2127 | } | |
2128 | ||
2129 | data_len = ctx->btf_data->d_size - sizeof(*hdr); | |
2130 | if (data_len < hdr->type_off || | |
2131 | data_len < hdr->str_off || | |
2132 | data_len < hdr->type_len + hdr->str_len || | |
2133 | hdr->type_off >= hdr->str_off || | |
2134 | hdr->type_off + hdr->type_len != hdr->str_off || | |
2135 | hdr->str_off + hdr->str_len != data_len || | |
2136 | (hdr->type_off & (sizeof(uint32_t) - 1))) { | |
2137 | fprintf(stderr, "Object has malformed BTF data!\n"); | |
2138 | return -EINVAL; | |
2139 | } | |
2140 | ||
2141 | ctx->btf.hdr = hdr; | |
2142 | ctx->btf.raw = hdr + 1; | |
2143 | ||
2144 | str_start = ctx->btf.raw + hdr->str_off; | |
2145 | str_end = str_start + hdr->str_len; | |
2146 | if (!hdr->str_len || | |
2147 | hdr->str_len - 1 > BTF_MAX_NAME_OFFSET || | |
2148 | str_start[0] || str_end[-1]) { | |
2149 | fprintf(stderr, "Object has malformed BTF string data!\n"); | |
2150 | return -EINVAL; | |
2151 | } | |
2152 | ||
2153 | ctx->btf.strings = str_start; | |
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | static int bpf_btf_register_type(struct bpf_elf_ctx *ctx, | |
2158 | const struct btf_type *type) | |
2159 | { | |
2160 | int cur = ctx->btf.types_num, num = cur + 1; | |
2161 | const struct btf_type **types; | |
2162 | ||
2163 | types = realloc(ctx->btf.types, num * sizeof(type)); | |
2164 | if (!types) { | |
2165 | free(ctx->btf.types); | |
2166 | ctx->btf.types = NULL; | |
2167 | ctx->btf.types_num = 0; | |
2168 | return -ENOMEM; | |
2169 | } | |
2170 | ||
2171 | ctx->btf.types = types; | |
2172 | ctx->btf.types[cur] = type; | |
2173 | ctx->btf.types_num = num; | |
2174 | return 0; | |
2175 | } | |
2176 | ||
2177 | static struct btf_type btf_type_void; | |
2178 | ||
2179 | static int bpf_btf_prep_type_data(struct bpf_elf_ctx *ctx) | |
2180 | { | |
2181 | const void *type_cur = ctx->btf.raw + ctx->btf.hdr->type_off; | |
2182 | const void *type_end = ctx->btf.raw + ctx->btf.hdr->str_off; | |
2183 | const struct btf_type *type; | |
2184 | uint16_t var_len; | |
2185 | int ret, kind; | |
2186 | ||
2187 | ret = bpf_btf_register_type(ctx, &btf_type_void); | |
2188 | if (ret < 0) | |
2189 | return ret; | |
2190 | ||
2191 | while (type_cur < type_end) { | |
2192 | type = type_cur; | |
2193 | type_cur += sizeof(*type); | |
2194 | ||
2195 | var_len = BTF_INFO_VLEN(type->info); | |
2196 | kind = BTF_INFO_KIND(type->info); | |
2197 | ||
2198 | switch (kind) { | |
2199 | case BTF_KIND_INT: | |
2200 | type_cur += sizeof(int); | |
2201 | break; | |
2202 | case BTF_KIND_ARRAY: | |
2203 | type_cur += sizeof(struct btf_array); | |
2204 | break; | |
2205 | case BTF_KIND_STRUCT: | |
2206 | case BTF_KIND_UNION: | |
2207 | type_cur += var_len * sizeof(struct btf_member); | |
2208 | break; | |
2209 | case BTF_KIND_ENUM: | |
2210 | type_cur += var_len * sizeof(struct btf_enum); | |
2211 | break; | |
3da6d055 YS |
2212 | case BTF_KIND_FUNC_PROTO: |
2213 | type_cur += var_len * sizeof(struct btf_param); | |
2214 | break; | |
f823f360 DB |
2215 | case BTF_KIND_TYPEDEF: |
2216 | case BTF_KIND_PTR: | |
2217 | case BTF_KIND_FWD: | |
2218 | case BTF_KIND_VOLATILE: | |
2219 | case BTF_KIND_CONST: | |
2220 | case BTF_KIND_RESTRICT: | |
3da6d055 | 2221 | case BTF_KIND_FUNC: |
f823f360 DB |
2222 | break; |
2223 | default: | |
2224 | fprintf(stderr, "Object has unknown BTF type: %u!\n", kind); | |
2225 | return -EINVAL; | |
2226 | } | |
2227 | ||
2228 | ret = bpf_btf_register_type(ctx, type); | |
2229 | if (ret < 0) | |
2230 | return ret; | |
2231 | } | |
2232 | ||
2233 | return 0; | |
2234 | } | |
2235 | ||
2236 | static int bpf_btf_prep_data(struct bpf_elf_ctx *ctx) | |
2237 | { | |
2238 | int ret = bpf_btf_check_header(ctx); | |
2239 | ||
2240 | if (!ret) | |
2241 | return bpf_btf_prep_type_data(ctx); | |
2242 | return ret; | |
2243 | } | |
2244 | ||
2245 | static void bpf_fetch_btf_end(struct bpf_elf_ctx *ctx) | |
2246 | { | |
2247 | int fd = bpf_btf_attach(ctx); | |
2248 | ||
2249 | if (fd < 0) | |
2250 | return; | |
2251 | ctx->btf_fd = fd; | |
2252 | if (bpf_btf_prep_data(ctx) < 0) { | |
2253 | close(ctx->btf_fd); | |
2254 | ctx->btf_fd = 0; | |
2255 | } | |
2256 | } | |
2257 | ||
afc1a200 DB |
2258 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
2259 | { | |
2260 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
2261 | } | |
2262 | ||
f823f360 DB |
2263 | static bool bpf_has_btf_data(const struct bpf_elf_ctx *ctx) |
2264 | { | |
2265 | return ctx->sec_btf; | |
2266 | } | |
2267 | ||
b5cb33ae DB |
2268 | static bool bpf_has_call_data(const struct bpf_elf_ctx *ctx) |
2269 | { | |
2270 | return ctx->sec_text; | |
2271 | } | |
2272 | ||
2273 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx, bool check_text_sec) | |
32e93fb7 DB |
2274 | { |
2275 | struct bpf_elf_sec_data data; | |
2276 | int i, ret = -1; | |
11c39b5e | 2277 | |
32e93fb7 DB |
2278 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2279 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
2280 | if (ret < 0) |
2281 | continue; | |
2282 | ||
cce3d466 DB |
2283 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2284 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
e4225669 | 2285 | ret = bpf_fetch_maps_begin(ctx, i, &data); |
cce3d466 DB |
2286 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2287 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 2288 | ret = bpf_fetch_license(ctx, i, &data); |
b5cb33ae DB |
2289 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2290 | (data.sec_hdr.sh_flags & SHF_EXECINSTR) && | |
2291 | !strcmp(data.sec_name, ".text") && | |
2292 | check_text_sec) | |
2293 | ret = bpf_fetch_text(ctx, i, &data); | |
cce3d466 DB |
2294 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
2295 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
2296 | ret = bpf_fetch_symtab(ctx, i, &data); |
2297 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 2298 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 | 2299 | ret = bpf_fetch_strtab(ctx, i, &data); |
f823f360 DB |
2300 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2301 | !strcmp(data.sec_name, ".BTF")) | |
2302 | ret = bpf_fetch_btf_begin(ctx, i, &data); | |
32e93fb7 | 2303 | if (ret < 0) { |
afc1a200 | 2304 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 2305 | i); |
e4225669 | 2306 | return ret; |
11c39b5e | 2307 | } |
32e93fb7 DB |
2308 | } |
2309 | ||
f823f360 DB |
2310 | if (bpf_has_btf_data(ctx)) |
2311 | bpf_fetch_btf_end(ctx); | |
afc1a200 | 2312 | if (bpf_has_map_data(ctx)) { |
e4225669 DB |
2313 | ret = bpf_fetch_maps_end(ctx); |
2314 | if (ret < 0) { | |
2315 | fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n"); | |
2316 | return ret; | |
2317 | } | |
2318 | ||
32e93fb7 DB |
2319 | ret = bpf_maps_attach_all(ctx); |
2320 | if (ret < 0) { | |
2321 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
2322 | return ret; | |
11c39b5e DB |
2323 | } |
2324 | } | |
2325 | ||
2326 | return ret; | |
2327 | } | |
2328 | ||
e4225669 DB |
2329 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section, |
2330 | bool *sseen) | |
11c39b5e | 2331 | { |
32e93fb7 DB |
2332 | struct bpf_elf_sec_data data; |
2333 | struct bpf_elf_prog prog; | |
2334 | int ret, i, fd = -1; | |
11c39b5e | 2335 | |
32e93fb7 DB |
2336 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2337 | if (ctx->sec_done[i]) | |
11c39b5e DB |
2338 | continue; |
2339 | ||
32e93fb7 | 2340 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
2341 | if (ret < 0 || |
2342 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
b5cb33ae | 2343 | (data.sec_hdr.sh_flags & SHF_EXECINSTR) && |
cce3d466 | 2344 | !strcmp(data.sec_name, section))) |
11c39b5e DB |
2345 | continue; |
2346 | ||
e4225669 DB |
2347 | *sseen = true; |
2348 | ||
32e93fb7 | 2349 | memset(&prog, 0, sizeof(prog)); |
b5cb33ae DB |
2350 | prog.type = ctx->type; |
2351 | prog.license = ctx->license; | |
2352 | prog.size = data.sec_data->d_size; | |
2353 | prog.insns_num = prog.size / sizeof(struct bpf_insn); | |
2354 | prog.insns = data.sec_data->d_buf; | |
11c39b5e | 2355 | |
f31645d1 | 2356 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 2357 | if (fd < 0) |
e4225669 | 2358 | return fd; |
11c39b5e | 2359 | |
32e93fb7 | 2360 | ctx->sec_done[i] = true; |
11c39b5e DB |
2361 | break; |
2362 | } | |
2363 | ||
32e93fb7 | 2364 | return fd; |
11c39b5e DB |
2365 | } |
2366 | ||
b5cb33ae DB |
2367 | struct bpf_relo_props { |
2368 | struct bpf_tail_call { | |
2369 | unsigned int total; | |
2370 | unsigned int jited; | |
2371 | } tc; | |
2372 | int main_num; | |
ecb05c0f DB |
2373 | }; |
2374 | ||
b5cb33ae DB |
2375 | static int bpf_apply_relo_map(struct bpf_elf_ctx *ctx, struct bpf_elf_prog *prog, |
2376 | GElf_Rel *relo, GElf_Sym *sym, | |
2377 | struct bpf_relo_props *props) | |
2378 | { | |
2379 | unsigned int insn_off = relo->r_offset / sizeof(struct bpf_insn); | |
2380 | unsigned int map_idx = sym->st_value / ctx->map_len; | |
2381 | ||
2382 | if (insn_off >= prog->insns_num) | |
2383 | return -EINVAL; | |
2384 | if (prog->insns[insn_off].code != (BPF_LD | BPF_IMM | BPF_DW)) { | |
2385 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", | |
2386 | insn_off); | |
2387 | return -EINVAL; | |
2388 | } | |
2389 | ||
2390 | if (map_idx >= ARRAY_SIZE(ctx->map_fds)) | |
2391 | return -EINVAL; | |
2392 | if (!ctx->map_fds[map_idx]) | |
2393 | return -EINVAL; | |
2394 | if (ctx->maps[map_idx].type == BPF_MAP_TYPE_PROG_ARRAY) { | |
2395 | props->tc.total++; | |
2396 | if (ctx->maps_ext[map_idx].owner.jited || | |
2397 | (ctx->maps_ext[map_idx].owner.type == 0 && | |
2398 | ctx->cfg.jit_enabled)) | |
2399 | props->tc.jited++; | |
2400 | } | |
2401 | ||
2402 | prog->insns[insn_off].src_reg = BPF_PSEUDO_MAP_FD; | |
2403 | prog->insns[insn_off].imm = ctx->map_fds[map_idx]; | |
2404 | return 0; | |
2405 | } | |
2406 | ||
2407 | static int bpf_apply_relo_call(struct bpf_elf_ctx *ctx, struct bpf_elf_prog *prog, | |
2408 | GElf_Rel *relo, GElf_Sym *sym, | |
2409 | struct bpf_relo_props *props) | |
2410 | { | |
2411 | unsigned int insn_off = relo->r_offset / sizeof(struct bpf_insn); | |
2412 | struct bpf_elf_prog *prog_text = &ctx->prog_text; | |
2413 | ||
2414 | if (insn_off >= prog->insns_num) | |
2415 | return -EINVAL; | |
2416 | if (prog->insns[insn_off].code != (BPF_JMP | BPF_CALL) && | |
2417 | prog->insns[insn_off].src_reg != BPF_PSEUDO_CALL) { | |
2418 | fprintf(stderr, "ELF contains relo data for non call instruction at offset %u! Compiler bug?!\n", | |
2419 | insn_off); | |
2420 | return -EINVAL; | |
2421 | } | |
2422 | ||
2423 | if (!props->main_num) { | |
2424 | struct bpf_insn *insns = realloc(prog->insns, | |
2425 | prog->size + prog_text->size); | |
2426 | if (!insns) | |
2427 | return -ENOMEM; | |
2428 | ||
2429 | memcpy(insns + prog->insns_num, prog_text->insns, | |
2430 | prog_text->size); | |
2431 | props->main_num = prog->insns_num; | |
2432 | prog->insns = insns; | |
2433 | prog->insns_num += prog_text->insns_num; | |
2434 | prog->size += prog_text->size; | |
2435 | } | |
2436 | ||
2437 | prog->insns[insn_off].imm += props->main_num - insn_off; | |
2438 | return 0; | |
2439 | } | |
2440 | ||
32e93fb7 DB |
2441 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
2442 | struct bpf_elf_sec_data *data_relo, | |
b5cb33ae DB |
2443 | struct bpf_elf_prog *prog, |
2444 | struct bpf_relo_props *props) | |
11c39b5e | 2445 | { |
32e93fb7 DB |
2446 | GElf_Shdr *rhdr = &data_relo->sec_hdr; |
2447 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
11c39b5e | 2448 | |
32e93fb7 | 2449 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
32e93fb7 DB |
2450 | GElf_Rel relo; |
2451 | GElf_Sym sym; | |
b5cb33ae | 2452 | int ret = -EIO; |
32e93fb7 DB |
2453 | |
2454 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
2455 | return -EIO; | |
32e93fb7 DB |
2456 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) |
2457 | return -EIO; | |
2458 | ||
b5cb33ae DB |
2459 | if (sym.st_shndx == ctx->sec_maps) |
2460 | ret = bpf_apply_relo_map(ctx, prog, &relo, &sym, props); | |
2461 | else if (sym.st_shndx == ctx->sec_text) | |
2462 | ret = bpf_apply_relo_call(ctx, prog, &relo, &sym, props); | |
2463 | else | |
2464 | fprintf(stderr, "ELF contains non-{map,call} related relo data in entry %u pointing to section %u! Compiler bug?!\n", | |
2465 | relo_ent, sym.st_shndx); | |
2466 | if (ret < 0) | |
2467 | return ret; | |
32e93fb7 DB |
2468 | } |
2469 | ||
2470 | return 0; | |
2471 | } | |
2472 | ||
afc1a200 | 2473 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
b5cb33ae | 2474 | bool *lderr, bool *sseen, struct bpf_elf_prog *prog) |
32e93fb7 DB |
2475 | { |
2476 | struct bpf_elf_sec_data data_relo, data_insn; | |
32e93fb7 DB |
2477 | int ret, idx, i, fd = -1; |
2478 | ||
2479 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
b5cb33ae | 2480 | struct bpf_relo_props props = {}; |
ecb05c0f | 2481 | |
32e93fb7 DB |
2482 | ret = bpf_fill_section_data(ctx, i, &data_relo); |
2483 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
2484 | continue; |
2485 | ||
32e93fb7 | 2486 | idx = data_relo.sec_hdr.sh_info; |
e4225669 | 2487 | |
32e93fb7 | 2488 | ret = bpf_fill_section_data(ctx, idx, &data_insn); |
cce3d466 DB |
2489 | if (ret < 0 || |
2490 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
b5cb33ae | 2491 | (data_insn.sec_hdr.sh_flags & SHF_EXECINSTR) && |
cce3d466 | 2492 | !strcmp(data_insn.sec_name, section))) |
11c39b5e | 2493 | continue; |
b5cb33ae DB |
2494 | if (sseen) |
2495 | *sseen = true; | |
2496 | ||
2497 | memset(prog, 0, sizeof(*prog)); | |
2498 | prog->type = ctx->type; | |
2499 | prog->license = ctx->license; | |
2500 | prog->size = data_insn.sec_data->d_size; | |
2501 | prog->insns_num = prog->size / sizeof(struct bpf_insn); | |
2502 | prog->insns = malloc(prog->size); | |
2503 | if (!prog->insns) { | |
2504 | *lderr = true; | |
2505 | return -ENOMEM; | |
2506 | } | |
32e93fb7 | 2507 | |
b5cb33ae | 2508 | memcpy(prog->insns, data_insn.sec_data->d_buf, prog->size); |
e4225669 | 2509 | |
b5cb33ae | 2510 | ret = bpf_apply_relo_data(ctx, &data_relo, prog, &props); |
c9c3720d DB |
2511 | if (ret < 0) { |
2512 | *lderr = true; | |
b5cb33ae DB |
2513 | if (ctx->sec_text != idx) |
2514 | free(prog->insns); | |
e4225669 | 2515 | return ret; |
c9c3720d | 2516 | } |
b5cb33ae DB |
2517 | if (ctx->sec_text == idx) { |
2518 | fd = 0; | |
2519 | goto out; | |
2520 | } | |
11c39b5e | 2521 | |
b5cb33ae DB |
2522 | fd = bpf_prog_attach(section, prog, ctx); |
2523 | free(prog->insns); | |
afc1a200 DB |
2524 | if (fd < 0) { |
2525 | *lderr = true; | |
b5cb33ae | 2526 | if (props.tc.total) { |
ecb05c0f | 2527 | if (ctx->cfg.jit_enabled && |
b5cb33ae | 2528 | props.tc.total != props.tc.jited) |
ecb05c0f | 2529 | fprintf(stderr, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n", |
b5cb33ae | 2530 | props.tc.jited, props.tc.total); |
ecb05c0f | 2531 | if (!ctx->cfg.jit_enabled && |
b5cb33ae | 2532 | props.tc.jited) |
ecb05c0f | 2533 | fprintf(stderr, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n", |
b5cb33ae | 2534 | props.tc.jited, props.tc.total); |
ecb05c0f | 2535 | } |
e4225669 | 2536 | return fd; |
afc1a200 | 2537 | } |
b5cb33ae | 2538 | out: |
32e93fb7 DB |
2539 | ctx->sec_done[i] = true; |
2540 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
2541 | break; |
2542 | } | |
2543 | ||
32e93fb7 | 2544 | return fd; |
11c39b5e DB |
2545 | } |
2546 | ||
32e93fb7 | 2547 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 2548 | { |
e4225669 | 2549 | bool lderr = false, sseen = false; |
b5cb33ae | 2550 | struct bpf_elf_prog prog; |
473d7840 DB |
2551 | int ret = -1; |
2552 | ||
b5cb33ae DB |
2553 | if (bpf_has_call_data(ctx)) { |
2554 | ret = bpf_fetch_prog_relo(ctx, ".text", &lderr, NULL, | |
2555 | &ctx->prog_text); | |
2556 | if (ret < 0) | |
2557 | return ret; | |
2558 | } | |
2559 | ||
2560 | if (bpf_has_map_data(ctx) || bpf_has_call_data(ctx)) | |
2561 | ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen, &prog); | |
afc1a200 | 2562 | if (ret < 0 && !lderr) |
e4225669 DB |
2563 | ret = bpf_fetch_prog(ctx, section, &sseen); |
2564 | if (ret < 0 && !sseen) | |
2565 | fprintf(stderr, "Program section \'%s\' not found in ELF file!\n", | |
2566 | section); | |
473d7840 DB |
2567 | return ret; |
2568 | } | |
2569 | ||
910b543d DB |
2570 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
2571 | { | |
2572 | int i; | |
2573 | ||
2574 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
2575 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
2576 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
2577 | return i; | |
2578 | return -1; | |
2579 | } | |
2580 | ||
ecb05c0f DB |
2581 | struct bpf_jited_aux { |
2582 | int prog_fd; | |
2583 | int map_fd; | |
2584 | struct bpf_prog_data prog; | |
2585 | struct bpf_map_ext map; | |
2586 | }; | |
2587 | ||
2588 | static int bpf_derive_prog_from_fdinfo(int fd, struct bpf_prog_data *prog) | |
2589 | { | |
358abfe0 | 2590 | char file[PATH_MAX], buff[4096]; |
ecb05c0f DB |
2591 | unsigned int val; |
2592 | FILE *fp; | |
2593 | ||
358abfe0 | 2594 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); |
ecb05c0f DB |
2595 | memset(prog, 0, sizeof(*prog)); |
2596 | ||
2597 | fp = fopen(file, "r"); | |
2598 | if (!fp) { | |
2599 | fprintf(stderr, "No procfs support?!\n"); | |
2600 | return -EIO; | |
2601 | } | |
2602 | ||
2603 | while (fgets(buff, sizeof(buff), fp)) { | |
2604 | if (sscanf(buff, "prog_type:\t%u", &val) == 1) | |
2605 | prog->type = val; | |
2606 | else if (sscanf(buff, "prog_jited:\t%u", &val) == 1) | |
2607 | prog->jited = val; | |
2608 | } | |
2609 | ||
2610 | fclose(fp); | |
2611 | return 0; | |
2612 | } | |
2613 | ||
2614 | static int bpf_tail_call_get_aux(struct bpf_jited_aux *aux) | |
2615 | { | |
2616 | struct bpf_elf_map tmp; | |
2617 | int ret; | |
2618 | ||
2619 | ret = bpf_derive_elf_map_from_fdinfo(aux->map_fd, &tmp, &aux->map); | |
2620 | if (!ret) | |
2621 | ret = bpf_derive_prog_from_fdinfo(aux->prog_fd, &aux->prog); | |
2622 | ||
2623 | return ret; | |
2624 | } | |
2625 | ||
32e93fb7 | 2626 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 2627 | { |
32e93fb7 DB |
2628 | struct bpf_elf_sec_data data; |
2629 | uint32_t map_id, key_id; | |
910b543d | 2630 | int fd, i, ret, idx; |
473d7840 | 2631 | |
32e93fb7 DB |
2632 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2633 | if (ctx->sec_done[i]) | |
473d7840 DB |
2634 | continue; |
2635 | ||
32e93fb7 | 2636 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
2637 | if (ret < 0) |
2638 | continue; | |
2639 | ||
910b543d DB |
2640 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
2641 | if (ret != 2) | |
32e93fb7 | 2642 | continue; |
910b543d DB |
2643 | |
2644 | idx = bpf_find_map_by_id(ctx, map_id); | |
2645 | if (idx < 0) | |
473d7840 DB |
2646 | continue; |
2647 | ||
32e93fb7 DB |
2648 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
2649 | if (fd < 0) | |
473d7840 DB |
2650 | return -EIO; |
2651 | ||
910b543d DB |
2652 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
2653 | &fd, BPF_ANY); | |
afc1a200 | 2654 | if (ret < 0) { |
ecb05c0f DB |
2655 | struct bpf_jited_aux aux = {}; |
2656 | ||
2657 | ret = -errno; | |
2658 | if (errno == E2BIG) { | |
afc1a200 DB |
2659 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", |
2660 | key_id, map_id); | |
ecb05c0f DB |
2661 | return ret; |
2662 | } | |
2663 | ||
2664 | aux.map_fd = ctx->map_fds[idx]; | |
2665 | aux.prog_fd = fd; | |
2666 | ||
2667 | if (bpf_tail_call_get_aux(&aux)) | |
2668 | return ret; | |
2669 | if (!aux.map.owner.type) | |
2670 | return ret; | |
2671 | ||
2672 | if (aux.prog.type != aux.map.owner.type) | |
2673 | fprintf(stderr, "Tail call map owned by prog type %u, but prog type is %u!\n", | |
2674 | aux.map.owner.type, aux.prog.type); | |
2675 | if (aux.prog.jited != aux.map.owner.jited) | |
2676 | fprintf(stderr, "Tail call map %s jited, but prog %s!\n", | |
2677 | aux.map.owner.jited ? "is" : "not", | |
2678 | aux.prog.jited ? "is" : "not"); | |
2679 | return ret; | |
afc1a200 | 2680 | } |
473d7840 | 2681 | |
32e93fb7 | 2682 | ctx->sec_done[i] = true; |
473d7840 DB |
2683 | } |
2684 | ||
2685 | return 0; | |
2686 | } | |
2687 | ||
32e93fb7 | 2688 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 2689 | { |
32e93fb7 DB |
2690 | struct stat st; |
2691 | int ret; | |
11c39b5e | 2692 | |
32e93fb7 | 2693 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 2694 | |
32e93fb7 DB |
2695 | ret = fstat(ctx->obj_fd, &st); |
2696 | if (ret < 0) { | |
2697 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
2698 | strerror(errno)); | |
2699 | return; | |
2700 | } | |
11c39b5e | 2701 | |
32e93fb7 DB |
2702 | ctx->stat.st_dev = st.st_dev; |
2703 | ctx->stat.st_ino = st.st_ino; | |
2704 | } | |
2705 | ||
f6793eec DB |
2706 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
2707 | { | |
2708 | char buff[PATH_MAX]; | |
2709 | ||
2710 | while (fgets(buff, sizeof(buff), fp)) { | |
2711 | char *ptr = buff; | |
2712 | ||
2713 | while (*ptr == ' ' || *ptr == '\t') | |
2714 | ptr++; | |
2715 | ||
2716 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
2717 | continue; | |
2718 | ||
2719 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
2720 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
2721 | strcpy(path, ptr); | |
2722 | return -1; | |
2723 | } | |
2724 | ||
2725 | return 1; | |
2726 | } | |
2727 | ||
2728 | return 0; | |
2729 | } | |
2730 | ||
2731 | static bool bpf_pinning_reserved(uint32_t pinning) | |
2732 | { | |
2733 | switch (pinning) { | |
2734 | case PIN_NONE: | |
2735 | case PIN_OBJECT_NS: | |
2736 | case PIN_GLOBAL_NS: | |
2737 | return true; | |
2738 | default: | |
2739 | return false; | |
2740 | } | |
2741 | } | |
2742 | ||
2743 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
2744 | { | |
2745 | struct bpf_hash_entry *entry; | |
d17b136f | 2746 | char subpath[PATH_MAX] = {}; |
f6793eec DB |
2747 | uint32_t pinning; |
2748 | FILE *fp; | |
2749 | int ret; | |
2750 | ||
2751 | fp = fopen(db_file, "r"); | |
2752 | if (!fp) | |
2753 | return; | |
2754 | ||
f6793eec DB |
2755 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { |
2756 | if (ret == -1) { | |
2757 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
2758 | db_file, subpath); | |
2759 | fclose(fp); | |
2760 | return; | |
2761 | } | |
2762 | ||
2763 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
2764 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
2765 | db_file, pinning); | |
f6793eec DB |
2766 | continue; |
2767 | } | |
2768 | ||
2769 | entry = malloc(sizeof(*entry)); | |
2770 | if (!entry) { | |
2771 | fprintf(stderr, "No memory left for db entry!\n"); | |
2772 | continue; | |
2773 | } | |
2774 | ||
2775 | entry->pinning = pinning; | |
2776 | entry->subpath = strdup(subpath); | |
2777 | if (!entry->subpath) { | |
2778 | fprintf(stderr, "No memory left for db entry!\n"); | |
2779 | free(entry); | |
2780 | continue; | |
2781 | } | |
2782 | ||
2783 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
2784 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
2785 | } | |
2786 | ||
2787 | fclose(fp); | |
2788 | } | |
2789 | ||
2790 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
2791 | { | |
2792 | struct bpf_hash_entry *entry; | |
2793 | int i; | |
2794 | ||
2795 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
2796 | while ((entry = ctx->ht[i]) != NULL) { | |
2797 | ctx->ht[i] = entry->next; | |
2798 | free((char *)entry->subpath); | |
2799 | free(entry); | |
2800 | } | |
2801 | } | |
2802 | } | |
2803 | ||
8187b012 DB |
2804 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
2805 | { | |
2806 | if (ctx->elf_hdr.e_type != ET_REL || | |
e77fa41d DB |
2807 | (ctx->elf_hdr.e_machine != EM_NONE && |
2808 | ctx->elf_hdr.e_machine != EM_BPF) || | |
8187b012 DB |
2809 | ctx->elf_hdr.e_version != EV_CURRENT) { |
2810 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
2811 | return -EINVAL; | |
2812 | } | |
2813 | ||
2814 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
2815 | default: | |
2816 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
2817 | return -EINVAL; | |
2818 | case ELFDATA2LSB: | |
2819 | if (htons(1) == 1) { | |
2820 | fprintf(stderr, | |
2821 | "We are big endian, eBPF object is little endian!\n"); | |
2822 | return -EIO; | |
2823 | } | |
2824 | break; | |
2825 | case ELFDATA2MSB: | |
2826 | if (htons(1) != 1) { | |
2827 | fprintf(stderr, | |
2828 | "We are little endian, eBPF object is big endian!\n"); | |
2829 | return -EIO; | |
2830 | } | |
2831 | break; | |
2832 | } | |
2833 | ||
2834 | return 0; | |
2835 | } | |
2836 | ||
ecb05c0f DB |
2837 | static void bpf_get_cfg(struct bpf_elf_ctx *ctx) |
2838 | { | |
2839 | static const char *path_jit = "/proc/sys/net/core/bpf_jit_enable"; | |
2840 | int fd; | |
2841 | ||
2842 | fd = open(path_jit, O_RDONLY); | |
2843 | if (fd > 0) { | |
2844 | char tmp[16] = {}; | |
2845 | ||
2846 | if (read(fd, tmp, sizeof(tmp)) > 0) | |
2847 | ctx->cfg.jit_enabled = atoi(tmp); | |
2848 | close(fd); | |
2849 | } | |
2850 | } | |
2851 | ||
32e93fb7 | 2852 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
65fdae3d JK |
2853 | enum bpf_prog_type type, __u32 ifindex, |
2854 | bool verbose) | |
32e93fb7 | 2855 | { |
6e5094db DB |
2856 | uint8_t tmp[20]; |
2857 | int ret; | |
32e93fb7 | 2858 | |
6e5094db DB |
2859 | if (elf_version(EV_CURRENT) == EV_NONE) |
2860 | return -EINVAL; | |
2861 | ||
2862 | bpf_init_env(); | |
32e93fb7 DB |
2863 | |
2864 | memset(ctx, 0, sizeof(*ctx)); | |
ecb05c0f | 2865 | bpf_get_cfg(ctx); |
6e5094db DB |
2866 | |
2867 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); | |
2868 | if (ret) | |
2869 | ctx->noafalg = true; | |
2870 | else | |
2871 | hexstring_n2a(tmp, sizeof(tmp), ctx->obj_uid, | |
2872 | sizeof(ctx->obj_uid)); | |
2873 | ||
32e93fb7 DB |
2874 | ctx->verbose = verbose; |
2875 | ctx->type = type; | |
65fdae3d | 2876 | ctx->ifindex = ifindex; |
32e93fb7 DB |
2877 | |
2878 | ctx->obj_fd = open(pathname, O_RDONLY); | |
2879 | if (ctx->obj_fd < 0) | |
2880 | return ctx->obj_fd; | |
2881 | ||
2882 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
2883 | if (!ctx->elf_fd) { | |
11c39b5e | 2884 | ret = -EINVAL; |
32e93fb7 | 2885 | goto out_fd; |
11c39b5e DB |
2886 | } |
2887 | ||
8187b012 DB |
2888 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
2889 | ret = -EINVAL; | |
2890 | goto out_fd; | |
2891 | } | |
2892 | ||
32e93fb7 DB |
2893 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
2894 | &ctx->elf_hdr) { | |
11c39b5e DB |
2895 | ret = -EIO; |
2896 | goto out_elf; | |
2897 | } | |
2898 | ||
8187b012 DB |
2899 | ret = bpf_elf_check_ehdr(ctx); |
2900 | if (ret < 0) | |
2901 | goto out_elf; | |
2902 | ||
32e93fb7 DB |
2903 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
2904 | sizeof(*(ctx->sec_done))); | |
2905 | if (!ctx->sec_done) { | |
11c39b5e DB |
2906 | ret = -ENOMEM; |
2907 | goto out_elf; | |
2908 | } | |
2909 | ||
f31645d1 DB |
2910 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
2911 | ret = -ENOMEM; | |
2912 | goto out_free; | |
2913 | } | |
2914 | ||
32e93fb7 | 2915 | bpf_save_finfo(ctx); |
f6793eec DB |
2916 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
2917 | ||
32e93fb7 | 2918 | return 0; |
f31645d1 DB |
2919 | out_free: |
2920 | free(ctx->sec_done); | |
32e93fb7 DB |
2921 | out_elf: |
2922 | elf_end(ctx->elf_fd); | |
2923 | out_fd: | |
2924 | close(ctx->obj_fd); | |
2925 | return ret; | |
2926 | } | |
d937a74b | 2927 | |
32e93fb7 DB |
2928 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
2929 | { | |
2930 | int i, count = 0; | |
11c39b5e | 2931 | |
32e93fb7 DB |
2932 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
2933 | if (!ctx->map_fds[i]) | |
2934 | break; | |
2935 | count++; | |
2936 | } | |
473d7840 | 2937 | |
32e93fb7 DB |
2938 | return count; |
2939 | } | |
6256f8c9 | 2940 | |
32e93fb7 DB |
2941 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
2942 | { | |
2943 | int i; | |
2944 | ||
2945 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
2946 | if (ctx->map_fds[i]) | |
2947 | close(ctx->map_fds[i]); | |
473d7840 | 2948 | } |
f823f360 DB |
2949 | |
2950 | if (ctx->btf_fd) | |
2951 | close(ctx->btf_fd); | |
2952 | free(ctx->btf.types); | |
32e93fb7 DB |
2953 | } |
2954 | ||
2955 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
2956 | { | |
2957 | if (failure) | |
2958 | bpf_maps_teardown(ctx); | |
473d7840 | 2959 | |
f6793eec | 2960 | bpf_hash_destroy(ctx); |
f31645d1 | 2961 | |
b5cb33ae | 2962 | free(ctx->prog_text.insns); |
32e93fb7 | 2963 | free(ctx->sec_done); |
f31645d1 DB |
2964 | free(ctx->log); |
2965 | ||
32e93fb7 DB |
2966 | elf_end(ctx->elf_fd); |
2967 | close(ctx->obj_fd); | |
2968 | } | |
6256f8c9 | 2969 | |
32e93fb7 | 2970 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 2971 | |
32e93fb7 | 2972 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
65fdae3d | 2973 | const char *section, __u32 ifindex, bool verbose) |
32e93fb7 DB |
2974 | { |
2975 | struct bpf_elf_ctx *ctx = &__ctx; | |
2976 | int fd = 0, ret; | |
6256f8c9 | 2977 | |
65fdae3d | 2978 | ret = bpf_elf_ctx_init(ctx, pathname, type, ifindex, verbose); |
32e93fb7 DB |
2979 | if (ret < 0) { |
2980 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
2981 | return ret; | |
2982 | } | |
6256f8c9 | 2983 | |
b5cb33ae | 2984 | ret = bpf_fetch_ancillary(ctx, strcmp(section, ".text")); |
32e93fb7 DB |
2985 | if (ret < 0) { |
2986 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
2987 | goto out; | |
2988 | } | |
2989 | ||
2990 | fd = bpf_fetch_prog_sec(ctx, section); | |
2991 | if (fd < 0) { | |
2992 | fprintf(stderr, "Error fetching program/map!\n"); | |
2993 | ret = fd; | |
2994 | goto out; | |
2995 | } | |
2996 | ||
2997 | ret = bpf_fill_prog_arrays(ctx); | |
2998 | if (ret < 0) | |
2999 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 3000 | out: |
32e93fb7 DB |
3001 | bpf_elf_ctx_destroy(ctx, ret < 0); |
3002 | if (ret < 0) { | |
3003 | if (fd) | |
3004 | close(fd); | |
3005 | return ret; | |
3006 | } | |
3007 | ||
3008 | return fd; | |
6256f8c9 | 3009 | } |
11c39b5e | 3010 | |
6256f8c9 | 3011 | static int |
4bd62446 DB |
3012 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
3013 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 | 3014 | { |
d17b136f PS |
3015 | struct bpf_map_set_msg msg = { |
3016 | .aux.uds_ver = BPF_SCM_AUX_VER, | |
3017 | .aux.num_ent = entries, | |
3018 | }; | |
6256f8c9 DB |
3019 | int *cmsg_buf, min_fd; |
3020 | char *amsg_buf; | |
3021 | int i; | |
3022 | ||
08a93b32 | 3023 | strlcpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); |
6256f8c9 DB |
3024 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); |
3025 | ||
3026 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
3027 | amsg_buf = (char *)msg.aux.ent; | |
3028 | ||
4bd62446 | 3029 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
3030 | int ret; |
3031 | ||
4bd62446 | 3032 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
3033 | bpf_map_set_init_single(&msg, min_fd); |
3034 | ||
3035 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
3036 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
3037 | ||
3038 | ret = sendmsg(fd, &msg.hdr, 0); | |
3039 | if (ret <= 0) | |
3040 | return ret ? : -1; | |
3041 | } | |
3042 | ||
3043 | return 0; | |
11c39b5e DB |
3044 | } |
3045 | ||
4bd62446 DB |
3046 | static int |
3047 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
3048 | unsigned int entries) | |
3049 | { | |
3050 | struct bpf_map_set_msg msg; | |
3051 | int *cmsg_buf, min_fd; | |
3052 | char *amsg_buf, *mmsg_buf; | |
3053 | unsigned int needed = 1; | |
3054 | int i; | |
3055 | ||
3056 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
3057 | amsg_buf = (char *)msg.aux.ent; | |
3058 | mmsg_buf = (char *)&msg.aux; | |
3059 | ||
3060 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
3061 | struct cmsghdr *cmsg; | |
3062 | int ret; | |
3063 | ||
3064 | min_fd = min(entries, entries - i); | |
3065 | bpf_map_set_init_single(&msg, min_fd); | |
3066 | ||
3067 | ret = recvmsg(fd, &msg.hdr, 0); | |
3068 | if (ret <= 0) | |
3069 | return ret ? : -1; | |
3070 | ||
3071 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
3072 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
3073 | return -EINVAL; | |
3074 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
3075 | return -EIO; | |
3076 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
3077 | return -ENOSYS; | |
3078 | ||
3079 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
3080 | if (min_fd > entries || min_fd <= 0) | |
3081 | return -EINVAL; | |
3082 | ||
3083 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
3084 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
3085 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
3086 | ||
3087 | needed = aux->num_ent; | |
3088 | } | |
3089 | ||
3090 | return 0; | |
3091 | } | |
3092 | ||
3093 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 3094 | { |
32e93fb7 | 3095 | struct bpf_elf_ctx *ctx = &__ctx; |
d17b136f PS |
3096 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
3097 | struct bpf_map_data bpf_aux = { | |
3098 | .fds = ctx->map_fds, | |
3099 | .ent = ctx->maps, | |
3100 | .st = &ctx->stat, | |
3101 | .obj = obj, | |
3102 | }; | |
6256f8c9 DB |
3103 | int fd, ret; |
3104 | ||
3105 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
3106 | if (fd < 0) { | |
3107 | fprintf(stderr, "Cannot open socket: %s\n", | |
3108 | strerror(errno)); | |
3109 | return -1; | |
3110 | } | |
3111 | ||
08a93b32 | 3112 | strlcpy(addr.sun_path, path, sizeof(addr.sun_path)); |
6256f8c9 DB |
3113 | |
3114 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
3115 | if (ret < 0) { | |
3116 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
3117 | path, strerror(errno)); | |
3118 | return -1; | |
3119 | } | |
3120 | ||
4bd62446 | 3121 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 3122 | bpf_maps_count(ctx)); |
6256f8c9 | 3123 | if (ret < 0) |
4bd62446 DB |
3124 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
3125 | path, strerror(errno)); | |
3126 | ||
32e93fb7 | 3127 | bpf_maps_teardown(ctx); |
4bd62446 DB |
3128 | close(fd); |
3129 | return ret; | |
3130 | } | |
3131 | ||
3132 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
3133 | unsigned int entries) | |
3134 | { | |
d17b136f | 3135 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
4bd62446 DB |
3136 | int fd, ret; |
3137 | ||
3138 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
3139 | if (fd < 0) { | |
3140 | fprintf(stderr, "Cannot open socket: %s\n", | |
3141 | strerror(errno)); | |
3142 | return -1; | |
3143 | } | |
3144 | ||
08a93b32 | 3145 | strlcpy(addr.sun_path, path, sizeof(addr.sun_path)); |
4bd62446 DB |
3146 | |
3147 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
3148 | if (ret < 0) { | |
3149 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
3150 | strerror(errno)); | |
3151 | return -1; | |
3152 | } | |
3153 | ||
3154 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
3155 | if (ret < 0) | |
3156 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
3157 | path, strerror(errno)); |
3158 | ||
4bd62446 | 3159 | unlink(addr.sun_path); |
6256f8c9 DB |
3160 | close(fd); |
3161 | return ret; | |
3162 | } | |
11c39b5e | 3163 | #endif /* HAVE_ELF */ |