]>
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 | ||
342 | void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len) | |
343 | { | |
d896797c | 344 | struct sock_filter *ops = RTA_DATA(bpf_ops); |
1d129d19 JP |
345 | int i; |
346 | ||
347 | if (len == 0) | |
348 | return; | |
349 | ||
350 | fprintf(f, "bytecode \'%u,", len); | |
351 | ||
352 | for (i = 0; i < len - 1; i++) | |
353 | fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt, | |
354 | ops[i].jf, ops[i].k); | |
355 | ||
6256f8c9 | 356 | fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt, |
1d129d19 JP |
357 | ops[i].jf, ops[i].k); |
358 | } | |
11c39b5e | 359 | |
afc1a200 DB |
360 | static void bpf_map_pin_report(const struct bpf_elf_map *pin, |
361 | const struct bpf_elf_map *obj) | |
362 | { | |
363 | fprintf(stderr, "Map specification differs from pinned file!\n"); | |
364 | ||
365 | if (obj->type != pin->type) | |
366 | fprintf(stderr, " - Type: %u (obj) != %u (pin)\n", | |
367 | obj->type, pin->type); | |
368 | if (obj->size_key != pin->size_key) | |
369 | fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n", | |
370 | obj->size_key, pin->size_key); | |
371 | if (obj->size_value != pin->size_value) | |
372 | fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n", | |
373 | obj->size_value, pin->size_value); | |
374 | if (obj->max_elem != pin->max_elem) | |
375 | fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n", | |
376 | obj->max_elem, pin->max_elem); | |
4dd3f50a DB |
377 | if (obj->flags != pin->flags) |
378 | fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n", | |
379 | obj->flags, pin->flags); | |
afc1a200 DB |
380 | |
381 | fprintf(stderr, "\n"); | |
382 | } | |
383 | ||
ecb05c0f DB |
384 | struct bpf_prog_data { |
385 | unsigned int type; | |
386 | unsigned int jited; | |
387 | }; | |
388 | ||
389 | struct bpf_map_ext { | |
390 | struct bpf_prog_data owner; | |
f823f360 DB |
391 | unsigned int btf_id_key; |
392 | unsigned int btf_id_val; | |
ecb05c0f DB |
393 | }; |
394 | ||
395 | static int bpf_derive_elf_map_from_fdinfo(int fd, struct bpf_elf_map *map, | |
396 | struct bpf_map_ext *ext) | |
9e607f2e | 397 | { |
ecb05c0f | 398 | unsigned int val, owner_type = 0, owner_jited = 0; |
9e607f2e | 399 | char file[PATH_MAX], buff[4096]; |
9e607f2e DB |
400 | FILE *fp; |
401 | ||
402 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
ecb05c0f | 403 | memset(map, 0, sizeof(*map)); |
9e607f2e DB |
404 | |
405 | fp = fopen(file, "r"); | |
406 | if (!fp) { | |
407 | fprintf(stderr, "No procfs support?!\n"); | |
408 | return -EIO; | |
409 | } | |
410 | ||
9e607f2e DB |
411 | while (fgets(buff, sizeof(buff), fp)) { |
412 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
ecb05c0f | 413 | map->type = val; |
9e607f2e | 414 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) |
ecb05c0f | 415 | map->size_key = val; |
9e607f2e | 416 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) |
ecb05c0f | 417 | map->size_value = val; |
9e607f2e | 418 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) |
ecb05c0f | 419 | map->max_elem = val; |
4dd3f50a | 420 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) |
ecb05c0f | 421 | map->flags = val; |
fb24802b DB |
422 | else if (sscanf(buff, "owner_prog_type:\t%i", &val) == 1) |
423 | owner_type = val; | |
ecb05c0f DB |
424 | else if (sscanf(buff, "owner_jited:\t%i", &val) == 1) |
425 | owner_jited = val; | |
9e607f2e DB |
426 | } |
427 | ||
428 | fclose(fp); | |
ecb05c0f DB |
429 | if (ext) { |
430 | memset(ext, 0, sizeof(*ext)); | |
431 | ext->owner.type = owner_type; | |
432 | ext->owner.jited = owner_jited; | |
433 | } | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
438 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, | |
439 | struct bpf_map_ext *ext, int length, | |
440 | enum bpf_prog_type type) | |
441 | { | |
442 | struct bpf_elf_map tmp, zero = {}; | |
443 | int ret; | |
444 | ||
445 | ret = bpf_derive_elf_map_from_fdinfo(fd, &tmp, ext); | |
446 | if (ret < 0) | |
447 | return ret; | |
9e607f2e | 448 | |
fb24802b DB |
449 | /* The decision to reject this is on kernel side eventually, but |
450 | * at least give the user a chance to know what's wrong. | |
451 | */ | |
ecb05c0f | 452 | if (ext->owner.type && ext->owner.type != type) |
fb24802b | 453 | fprintf(stderr, "Program array map owner types differ: %u (obj) != %u (pin)\n", |
ecb05c0f | 454 | type, ext->owner.type); |
fb24802b | 455 | |
91d88eeb | 456 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
457 | return 0; |
458 | } else { | |
9e607f2e DB |
459 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, |
460 | * so just accept it. We know we do have an eBPF fd and in this | |
461 | * case, everything is 0. It is guaranteed that no such map exists | |
462 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
463 | */ | |
91d88eeb | 464 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
465 | return 0; |
466 | ||
afc1a200 | 467 | bpf_map_pin_report(&tmp, map); |
9e607f2e DB |
468 | return -EINVAL; |
469 | } | |
470 | } | |
471 | ||
91d88eeb DB |
472 | static int bpf_mnt_fs(const char *target) |
473 | { | |
474 | bool bind_done = false; | |
475 | ||
476 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
477 | if (errno != EINVAL || bind_done) { | |
478 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
479 | target, strerror(errno)); | |
480 | return -1; | |
481 | } | |
482 | ||
483 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
484 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
485 | target, target, strerror(errno)); | |
486 | return -1; | |
487 | } | |
488 | ||
489 | bind_done = true; | |
490 | } | |
491 | ||
e4225669 | 492 | if (mount("bpf", target, "bpf", 0, "mode=0700")) { |
91d88eeb DB |
493 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", |
494 | target, strerror(errno)); | |
495 | return -1; | |
496 | } | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
95ae9a48 DB |
501 | static int bpf_mnt_check_target(const char *target) |
502 | { | |
503 | struct stat sb = {}; | |
504 | int ret; | |
505 | ||
506 | ret = stat(target, &sb); | |
507 | if (ret) { | |
508 | ret = mkdir(target, S_IRWXU); | |
509 | if (ret) { | |
510 | fprintf(stderr, "mkdir %s failed: %s\n", target, | |
511 | strerror(errno)); | |
512 | return ret; | |
513 | } | |
514 | } | |
515 | ||
516 | return 0; | |
517 | } | |
518 | ||
32e93fb7 DB |
519 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
520 | { | |
521 | struct statfs st_fs; | |
522 | ||
523 | if (statfs(mnt, &st_fs) < 0) | |
524 | return -ENOENT; | |
525 | if ((unsigned long)st_fs.f_type != magic) | |
526 | return -ENOENT; | |
527 | ||
528 | return 0; | |
529 | } | |
530 | ||
95ae9a48 DB |
531 | static const char *bpf_find_mntpt_single(unsigned long magic, char *mnt, |
532 | int len, const char *mntpt) | |
533 | { | |
534 | int ret; | |
535 | ||
536 | ret = bpf_valid_mntpt(mntpt, magic); | |
537 | if (!ret) { | |
18f156bf | 538 | strlcpy(mnt, mntpt, len); |
95ae9a48 DB |
539 | return mnt; |
540 | } | |
541 | ||
542 | return NULL; | |
543 | } | |
544 | ||
32e93fb7 DB |
545 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, |
546 | char *mnt, int len, | |
547 | const char * const *known_mnts) | |
548 | { | |
549 | const char * const *ptr; | |
550 | char type[100]; | |
551 | FILE *fp; | |
552 | ||
553 | if (known_mnts) { | |
554 | ptr = known_mnts; | |
555 | while (*ptr) { | |
95ae9a48 | 556 | if (bpf_find_mntpt_single(magic, mnt, len, *ptr)) |
32e93fb7 | 557 | return mnt; |
32e93fb7 DB |
558 | ptr++; |
559 | } | |
560 | } | |
561 | ||
c3724e4b PS |
562 | if (len != PATH_MAX) |
563 | return NULL; | |
564 | ||
32e93fb7 | 565 | fp = fopen("/proc/mounts", "r"); |
c3724e4b | 566 | if (fp == NULL) |
32e93fb7 DB |
567 | return NULL; |
568 | ||
569 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
570 | mnt, type) == 2) { | |
571 | if (strcmp(type, fstype) == 0) | |
572 | break; | |
573 | } | |
574 | ||
575 | fclose(fp); | |
576 | if (strcmp(type, fstype) != 0) | |
577 | return NULL; | |
578 | ||
579 | return mnt; | |
580 | } | |
581 | ||
582 | int bpf_trace_pipe(void) | |
583 | { | |
584 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
585 | static const char * const tracefs_known_mnts[] = { | |
586 | TRACE_DIR_MNT, | |
587 | "/sys/kernel/debug/tracing", | |
588 | "/tracing", | |
589 | "/trace", | |
590 | 0, | |
591 | }; | |
1b736dc4 | 592 | int fd_in, fd_out = STDERR_FILENO; |
32e93fb7 DB |
593 | char tpipe[PATH_MAX]; |
594 | const char *mnt; | |
32e93fb7 DB |
595 | |
596 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
597 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
598 | if (!mnt) { | |
599 | fprintf(stderr, "tracefs not mounted?\n"); | |
600 | return -1; | |
601 | } | |
602 | ||
603 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
604 | ||
1b736dc4 DB |
605 | fd_in = open(tpipe, O_RDONLY); |
606 | if (fd_in < 0) | |
32e93fb7 DB |
607 | return -1; |
608 | ||
609 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
610 | while (1) { | |
611 | static char buff[4096]; | |
612 | ssize_t ret; | |
613 | ||
1b736dc4 DB |
614 | ret = read(fd_in, buff, sizeof(buff)); |
615 | if (ret > 0 && write(fd_out, buff, ret) == ret) | |
616 | continue; | |
617 | break; | |
32e93fb7 DB |
618 | } |
619 | ||
1b736dc4 DB |
620 | close(fd_in); |
621 | return -1; | |
32e93fb7 DB |
622 | } |
623 | ||
e4225669 | 624 | static int bpf_gen_global(const char *bpf_sub_dir) |
91d88eeb | 625 | { |
e4225669 DB |
626 | char bpf_glo_dir[PATH_MAX]; |
627 | int ret; | |
628 | ||
629 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/", | |
630 | bpf_sub_dir, BPF_DIR_GLOBALS); | |
631 | ||
632 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
633 | if (ret && errno != EEXIST) { | |
634 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
635 | strerror(errno)); | |
636 | return ret; | |
637 | } | |
638 | ||
639 | return 0; | |
640 | } | |
641 | ||
642 | static int bpf_gen_master(const char *base, const char *name) | |
643 | { | |
260a92af | 644 | char bpf_sub_dir[PATH_MAX + NAME_MAX + 1]; |
e4225669 DB |
645 | int ret; |
646 | ||
647 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name); | |
648 | ||
649 | ret = mkdir(bpf_sub_dir, S_IRWXU); | |
650 | if (ret && errno != EEXIST) { | |
651 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir, | |
652 | strerror(errno)); | |
653 | return ret; | |
654 | } | |
655 | ||
656 | return bpf_gen_global(bpf_sub_dir); | |
657 | } | |
658 | ||
659 | static int bpf_slave_via_bind_mnt(const char *full_name, | |
660 | const char *full_link) | |
661 | { | |
662 | int ret; | |
663 | ||
664 | ret = mkdir(full_name, S_IRWXU); | |
665 | if (ret) { | |
666 | assert(errno != EEXIST); | |
667 | fprintf(stderr, "mkdir %s failed: %s\n", full_name, | |
668 | strerror(errno)); | |
669 | return ret; | |
670 | } | |
671 | ||
672 | ret = mount(full_link, full_name, "none", MS_BIND, NULL); | |
673 | if (ret) { | |
674 | rmdir(full_name); | |
675 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
676 | full_link, full_name, strerror(errno)); | |
677 | } | |
678 | ||
679 | return ret; | |
680 | } | |
681 | ||
682 | static int bpf_gen_slave(const char *base, const char *name, | |
683 | const char *link) | |
684 | { | |
260a92af SH |
685 | char bpf_lnk_dir[PATH_MAX + NAME_MAX + 1]; |
686 | char bpf_sub_dir[PATH_MAX + NAME_MAX]; | |
e4225669 DB |
687 | struct stat sb = {}; |
688 | int ret; | |
689 | ||
690 | snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link); | |
691 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name); | |
692 | ||
693 | ret = symlink(bpf_lnk_dir, bpf_sub_dir); | |
694 | if (ret) { | |
695 | if (errno != EEXIST) { | |
696 | if (errno != EPERM) { | |
697 | fprintf(stderr, "symlink %s failed: %s\n", | |
698 | bpf_sub_dir, strerror(errno)); | |
699 | return ret; | |
700 | } | |
701 | ||
702 | return bpf_slave_via_bind_mnt(bpf_sub_dir, | |
703 | bpf_lnk_dir); | |
704 | } | |
705 | ||
706 | ret = lstat(bpf_sub_dir, &sb); | |
707 | if (ret) { | |
708 | fprintf(stderr, "lstat %s failed: %s\n", | |
709 | bpf_sub_dir, strerror(errno)); | |
710 | return ret; | |
711 | } | |
712 | ||
713 | if ((sb.st_mode & S_IFMT) != S_IFLNK) | |
714 | return bpf_gen_global(bpf_sub_dir); | |
715 | } | |
716 | ||
717 | return 0; | |
718 | } | |
719 | ||
720 | static int bpf_gen_hierarchy(const char *base) | |
721 | { | |
722 | int ret, i; | |
723 | ||
724 | ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0])); | |
725 | for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++) | |
726 | ret = bpf_gen_slave(base, | |
727 | bpf_prog_to_subdir(__bpf_types[i]), | |
728 | bpf_prog_to_subdir(__bpf_types[0])); | |
729 | return ret; | |
730 | } | |
731 | ||
732 | static const char *bpf_get_work_dir(enum bpf_prog_type type) | |
733 | { | |
734 | static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT; | |
735 | static char bpf_wrk_dir[PATH_MAX]; | |
91d88eeb | 736 | static const char *mnt; |
e4225669 | 737 | static bool bpf_mnt_cached; |
95ae9a48 | 738 | const char *mnt_env = getenv(BPF_ENV_MNT); |
91d88eeb DB |
739 | static const char * const bpf_known_mnts[] = { |
740 | BPF_DIR_MNT, | |
e4225669 | 741 | "/bpf", |
91d88eeb DB |
742 | 0, |
743 | }; | |
91d88eeb DB |
744 | int ret; |
745 | ||
e4225669 DB |
746 | if (bpf_mnt_cached) { |
747 | const char *out = mnt; | |
748 | ||
51361a9f | 749 | if (out && type) { |
e4225669 DB |
750 | snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/", |
751 | out, bpf_prog_to_subdir(type)); | |
752 | out = bpf_tmp; | |
753 | } | |
754 | return out; | |
755 | } | |
91d88eeb | 756 | |
95ae9a48 DB |
757 | if (mnt_env) |
758 | mnt = bpf_find_mntpt_single(BPF_FS_MAGIC, bpf_tmp, | |
759 | sizeof(bpf_tmp), mnt_env); | |
760 | else | |
761 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp, | |
762 | sizeof(bpf_tmp), bpf_known_mnts); | |
91d88eeb | 763 | if (!mnt) { |
95ae9a48 DB |
764 | mnt = mnt_env ? : BPF_DIR_MNT; |
765 | ret = bpf_mnt_check_target(mnt); | |
766 | if (!ret) | |
767 | ret = bpf_mnt_fs(mnt); | |
91d88eeb DB |
768 | if (ret) { |
769 | mnt = NULL; | |
770 | goto out; | |
771 | } | |
772 | } | |
773 | ||
e4225669 | 774 | snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt); |
91d88eeb | 775 | |
e4225669 DB |
776 | ret = bpf_gen_hierarchy(bpf_wrk_dir); |
777 | if (ret) { | |
91d88eeb DB |
778 | mnt = NULL; |
779 | goto out; | |
780 | } | |
781 | ||
e4225669 | 782 | mnt = bpf_wrk_dir; |
91d88eeb DB |
783 | out: |
784 | bpf_mnt_cached = true; | |
91d88eeb DB |
785 | return mnt; |
786 | } | |
787 | ||
e4225669 | 788 | static int bpf_obj_get(const char *pathname, enum bpf_prog_type type) |
91d88eeb | 789 | { |
d17b136f | 790 | union bpf_attr attr = {}; |
91d88eeb DB |
791 | char tmp[PATH_MAX]; |
792 | ||
793 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
e4225669 | 794 | pathname[1] == ':' && bpf_get_work_dir(type)) { |
91d88eeb | 795 | snprintf(tmp, sizeof(tmp), "%s/%s", |
e4225669 | 796 | bpf_get_work_dir(type), pathname + 2); |
91d88eeb DB |
797 | pathname = tmp; |
798 | } | |
799 | ||
91d88eeb DB |
800 | attr.pathname = bpf_ptr_to_u64(pathname); |
801 | ||
802 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
803 | } | |
804 | ||
21856018 DB |
805 | static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type) |
806 | { | |
807 | int prog_fd = bpf_obj_get(pathname, type); | |
808 | ||
809 | if (prog_fd < 0) | |
810 | fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n", | |
811 | pathname, strerror(errno)); | |
812 | return prog_fd; | |
813 | } | |
814 | ||
3f0b9e62 | 815 | static int bpf_do_parse(struct bpf_cfg_in *cfg, const bool *opt_tbl) |
32e93fb7 | 816 | { |
32e93fb7 | 817 | const char *file, *section, *uds_name; |
32e93fb7 | 818 | bool verbose = false; |
e4225669 | 819 | int i, ret, argc; |
91d88eeb DB |
820 | char **argv; |
821 | ||
e4225669 DB |
822 | argv = cfg->argv; |
823 | argc = cfg->argc; | |
91d88eeb DB |
824 | |
825 | if (opt_tbl[CBPF_BYTECODE] && | |
826 | (matches(*argv, "bytecode") == 0 || | |
827 | strcmp(*argv, "bc") == 0)) { | |
f20ff2f1 | 828 | cfg->mode = CBPF_BYTECODE; |
91d88eeb DB |
829 | } else if (opt_tbl[CBPF_FILE] && |
830 | (matches(*argv, "bytecode-file") == 0 || | |
831 | strcmp(*argv, "bcf") == 0)) { | |
f20ff2f1 | 832 | cfg->mode = CBPF_FILE; |
91d88eeb DB |
833 | } else if (opt_tbl[EBPF_OBJECT] && |
834 | (matches(*argv, "object-file") == 0 || | |
835 | strcmp(*argv, "obj") == 0)) { | |
f20ff2f1 | 836 | cfg->mode = EBPF_OBJECT; |
91d88eeb DB |
837 | } else if (opt_tbl[EBPF_PINNED] && |
838 | (matches(*argv, "object-pinned") == 0 || | |
839 | matches(*argv, "pinned") == 0 || | |
840 | matches(*argv, "fd") == 0)) { | |
f20ff2f1 | 841 | cfg->mode = EBPF_PINNED; |
32e93fb7 DB |
842 | } else { |
843 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
844 | return -1; | |
845 | } | |
846 | ||
847 | NEXT_ARG(); | |
848 | file = section = uds_name = NULL; | |
f20ff2f1 | 849 | if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) { |
32e93fb7 DB |
850 | file = *argv; |
851 | NEXT_ARG_FWD(); | |
852 | ||
658cfebc | 853 | if (cfg->type == BPF_PROG_TYPE_UNSPEC) { |
91d88eeb DB |
854 | if (argc > 0 && matches(*argv, "type") == 0) { |
855 | NEXT_ARG(); | |
e4225669 DB |
856 | for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta); |
857 | i++) { | |
858 | if (!__bpf_prog_meta[i].type) | |
859 | continue; | |
860 | if (!matches(*argv, | |
861 | __bpf_prog_meta[i].type)) { | |
658cfebc | 862 | cfg->type = i; |
e4225669 DB |
863 | break; |
864 | } | |
865 | } | |
866 | ||
658cfebc | 867 | if (cfg->type == BPF_PROG_TYPE_UNSPEC) { |
91d88eeb DB |
868 | fprintf(stderr, "What type is \"%s\"?\n", |
869 | *argv); | |
870 | return -1; | |
871 | } | |
872 | NEXT_ARG_FWD(); | |
873 | } else { | |
658cfebc | 874 | cfg->type = BPF_PROG_TYPE_SCHED_CLS; |
91d88eeb DB |
875 | } |
876 | } | |
877 | ||
658cfebc | 878 | section = bpf_prog_to_default_section(cfg->type); |
32e93fb7 DB |
879 | if (argc > 0 && matches(*argv, "section") == 0) { |
880 | NEXT_ARG(); | |
881 | section = *argv; | |
882 | NEXT_ARG_FWD(); | |
883 | } | |
884 | ||
658cfebc | 885 | if (__bpf_prog_meta[cfg->type].may_uds_export) { |
e4225669 DB |
886 | uds_name = getenv(BPF_ENV_UDS); |
887 | if (argc > 0 && !uds_name && | |
888 | matches(*argv, "export") == 0) { | |
889 | NEXT_ARG(); | |
890 | uds_name = *argv; | |
891 | NEXT_ARG_FWD(); | |
892 | } | |
32e93fb7 DB |
893 | } |
894 | ||
895 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
896 | verbose = true; | |
897 | NEXT_ARG_FWD(); | |
898 | } | |
899 | ||
900 | PREV_ARG(); | |
901 | } | |
902 | ||
3f0b9e62 | 903 | if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) { |
51be7546 | 904 | ret = bpf_ops_parse(argc, argv, cfg->opcodes, |
f20ff2f1 | 905 | cfg->mode == CBPF_FILE); |
3f0b9e62 JK |
906 | cfg->n_opcodes = ret; |
907 | } else if (cfg->mode == EBPF_OBJECT) { | |
908 | ret = 0; /* program will be loaded by load stage */ | |
909 | } else if (cfg->mode == EBPF_PINNED) { | |
658cfebc | 910 | ret = bpf_obj_pinned(file, cfg->type); |
3f0b9e62 JK |
911 | cfg->prog_fd = ret; |
912 | } else { | |
32e93fb7 | 913 | return -1; |
3f0b9e62 | 914 | } |
32e93fb7 | 915 | |
e4225669 DB |
916 | cfg->object = file; |
917 | cfg->section = section; | |
918 | cfg->uds = uds_name; | |
919 | cfg->argc = argc; | |
920 | cfg->argv = argv; | |
3f0b9e62 | 921 | cfg->verbose = verbose; |
91d88eeb DB |
922 | |
923 | return ret; | |
924 | } | |
925 | ||
3f0b9e62 JK |
926 | static int bpf_do_load(struct bpf_cfg_in *cfg) |
927 | { | |
928 | if (cfg->mode == EBPF_OBJECT) { | |
929 | cfg->prog_fd = bpf_obj_open(cfg->object, cfg->type, | |
65fdae3d JK |
930 | cfg->section, cfg->ifindex, |
931 | cfg->verbose); | |
3f0b9e62 JK |
932 | return cfg->prog_fd; |
933 | } | |
934 | return 0; | |
935 | } | |
936 | ||
4a847fcb JK |
937 | int bpf_load_common(struct bpf_cfg_in *cfg, const struct bpf_cfg_ops *ops, |
938 | void *nl) | |
91d88eeb | 939 | { |
91d88eeb | 940 | char annotation[256]; |
91d88eeb DB |
941 | int ret; |
942 | ||
3f0b9e62 | 943 | ret = bpf_do_load(cfg); |
91d88eeb DB |
944 | if (ret < 0) |
945 | return ret; | |
946 | ||
f20ff2f1 | 947 | if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) |
3f0b9e62 | 948 | ops->cbpf_cb(nl, cfg->opcodes, cfg->n_opcodes); |
f20ff2f1 | 949 | if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) { |
32e93fb7 | 950 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
f20ff2f1 | 951 | basename(cfg->object), cfg->mode == EBPF_PINNED ? |
e4225669 | 952 | "*fsobj" : cfg->section); |
3f0b9e62 | 953 | ops->ebpf_cb(nl, cfg->prog_fd, annotation); |
32e93fb7 DB |
954 | } |
955 | ||
91d88eeb DB |
956 | return 0; |
957 | } | |
32e93fb7 | 958 | |
4a847fcb | 959 | int bpf_parse_common(struct bpf_cfg_in *cfg, const struct bpf_cfg_ops *ops) |
e4225669 DB |
960 | { |
961 | bool opt_tbl[BPF_MODE_MAX] = {}; | |
962 | ||
963 | if (ops->cbpf_cb) { | |
964 | opt_tbl[CBPF_BYTECODE] = true; | |
965 | opt_tbl[CBPF_FILE] = true; | |
966 | } | |
967 | ||
968 | if (ops->ebpf_cb) { | |
969 | opt_tbl[EBPF_OBJECT] = true; | |
970 | opt_tbl[EBPF_PINNED] = true; | |
971 | } | |
972 | ||
4a847fcb JK |
973 | return bpf_do_parse(cfg, opt_tbl); |
974 | } | |
975 | ||
976 | int bpf_parse_and_load_common(struct bpf_cfg_in *cfg, | |
977 | const struct bpf_cfg_ops *ops, void *nl) | |
978 | { | |
979 | int ret; | |
980 | ||
981 | ret = bpf_parse_common(cfg, ops); | |
982 | if (ret < 0) | |
983 | return ret; | |
984 | ||
985 | return bpf_load_common(cfg, ops, nl); | |
e4225669 DB |
986 | } |
987 | ||
91d88eeb DB |
988 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
989 | { | |
91d88eeb | 990 | const bool opt_tbl[BPF_MODE_MAX] = { |
91d88eeb DB |
991 | [EBPF_OBJECT] = true, |
992 | [EBPF_PINNED] = true, | |
993 | }; | |
994 | const struct bpf_elf_map test = { | |
995 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
996 | .size_key = sizeof(int), | |
997 | .size_value = sizeof(int), | |
998 | }; | |
e4225669 | 999 | struct bpf_cfg_in cfg = { |
658cfebc | 1000 | .type = BPF_PROG_TYPE_UNSPEC, |
e4225669 DB |
1001 | .argc = argc, |
1002 | .argv = argv, | |
1003 | }; | |
ecb05c0f | 1004 | struct bpf_map_ext ext = {}; |
91d88eeb | 1005 | int ret, prog_fd, map_fd; |
91d88eeb DB |
1006 | uint32_t map_key; |
1007 | ||
3f0b9e62 JK |
1008 | ret = bpf_do_parse(&cfg, opt_tbl); |
1009 | if (ret < 0) | |
1010 | return ret; | |
1011 | ||
1012 | ret = bpf_do_load(&cfg); | |
1013 | if (ret < 0) | |
1014 | return ret; | |
1015 | ||
1016 | prog_fd = cfg.prog_fd; | |
1017 | ||
91d88eeb DB |
1018 | if (key) { |
1019 | map_key = *key; | |
1020 | } else { | |
e4225669 | 1021 | ret = sscanf(cfg.section, "%*i/%i", &map_key); |
91d88eeb | 1022 | if (ret != 1) { |
32a121cb | 1023 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
1024 | ret = -EINVAL; |
1025 | goto out_prog; | |
1026 | } | |
1027 | } | |
32e93fb7 | 1028 | |
658cfebc | 1029 | map_fd = bpf_obj_get(map_path, cfg.type); |
91d88eeb DB |
1030 | if (map_fd < 0) { |
1031 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
1032 | map_path, strerror(errno)); | |
1033 | ret = map_fd; | |
1034 | goto out_prog; | |
1035 | } | |
1036 | ||
ecb05c0f | 1037 | ret = bpf_map_selfcheck_pinned(map_fd, &test, &ext, |
fb24802b | 1038 | offsetof(struct bpf_elf_map, max_elem), |
658cfebc | 1039 | cfg.type); |
91d88eeb DB |
1040 | if (ret < 0) { |
1041 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
1042 | goto out_map; | |
1043 | } | |
1044 | ||
1045 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
1046 | if (ret < 0) | |
1047 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
1048 | out_map: | |
1049 | close(map_fd); | |
1050 | out_prog: | |
1051 | close(prog_fd); | |
1052 | return ret; | |
32e93fb7 DB |
1053 | } |
1054 | ||
fc4ccce0 DA |
1055 | int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type) |
1056 | { | |
1057 | union bpf_attr attr = {}; | |
1058 | ||
1059 | attr.target_fd = target_fd; | |
1060 | attr.attach_bpf_fd = prog_fd; | |
1061 | attr.attach_type = type; | |
1062 | ||
1063 | return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
1064 | } | |
1065 | ||
1066 | int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type) | |
1067 | { | |
1068 | union bpf_attr attr = {}; | |
1069 | ||
1070 | attr.target_fd = target_fd; | |
1071 | attr.attach_type = type; | |
1072 | ||
1073 | return bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); | |
1074 | } | |
1075 | ||
65fdae3d JK |
1076 | static int bpf_prog_load_dev(enum bpf_prog_type type, |
1077 | const struct bpf_insn *insns, size_t size_insns, | |
1078 | const char *license, __u32 ifindex, | |
1079 | char *log, size_t size_log) | |
869d889e DA |
1080 | { |
1081 | union bpf_attr attr = {}; | |
1082 | ||
1083 | attr.prog_type = type; | |
1084 | attr.insns = bpf_ptr_to_u64(insns); | |
1085 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); | |
1086 | attr.license = bpf_ptr_to_u64(license); | |
65fdae3d | 1087 | attr.prog_ifindex = ifindex; |
869d889e DA |
1088 | |
1089 | if (size_log > 0) { | |
1090 | attr.log_buf = bpf_ptr_to_u64(log); | |
1091 | attr.log_size = size_log; | |
1092 | attr.log_level = 1; | |
1093 | } | |
1094 | ||
1095 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); | |
1096 | } | |
1097 | ||
65fdae3d JK |
1098 | int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, |
1099 | size_t size_insns, const char *license, char *log, | |
1100 | size_t size_log) | |
1101 | { | |
1102 | return bpf_prog_load_dev(type, insns, size_insns, license, 0, | |
1103 | log, size_log); | |
1104 | } | |
1105 | ||
6256f8c9 | 1106 | #ifdef HAVE_ELF |
32e93fb7 DB |
1107 | struct bpf_elf_prog { |
1108 | enum bpf_prog_type type; | |
b5cb33ae DB |
1109 | struct bpf_insn *insns; |
1110 | unsigned int insns_num; | |
32e93fb7 DB |
1111 | size_t size; |
1112 | const char *license; | |
1113 | }; | |
1114 | ||
f6793eec DB |
1115 | struct bpf_hash_entry { |
1116 | unsigned int pinning; | |
1117 | const char *subpath; | |
1118 | struct bpf_hash_entry *next; | |
1119 | }; | |
1120 | ||
ecb05c0f DB |
1121 | struct bpf_config { |
1122 | unsigned int jit_enabled; | |
1123 | }; | |
1124 | ||
f823f360 DB |
1125 | struct bpf_btf { |
1126 | const struct btf_header *hdr; | |
1127 | const void *raw; | |
1128 | const char *strings; | |
1129 | const struct btf_type **types; | |
1130 | int types_num; | |
1131 | }; | |
1132 | ||
32e93fb7 | 1133 | struct bpf_elf_ctx { |
ecb05c0f | 1134 | struct bpf_config cfg; |
32e93fb7 DB |
1135 | Elf *elf_fd; |
1136 | GElf_Ehdr elf_hdr; | |
1137 | Elf_Data *sym_tab; | |
1138 | Elf_Data *str_tab; | |
f823f360 | 1139 | Elf_Data *btf_data; |
6e5094db | 1140 | char obj_uid[64]; |
32e93fb7 | 1141 | int obj_fd; |
f823f360 | 1142 | int btf_fd; |
32e93fb7 DB |
1143 | int map_fds[ELF_MAX_MAPS]; |
1144 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
ecb05c0f | 1145 | struct bpf_map_ext maps_ext[ELF_MAX_MAPS]; |
b5cb33ae | 1146 | struct bpf_elf_prog prog_text; |
f823f360 | 1147 | struct bpf_btf btf; |
32e93fb7 DB |
1148 | int sym_num; |
1149 | int map_num; | |
e4225669 | 1150 | int map_len; |
32e93fb7 DB |
1151 | bool *sec_done; |
1152 | int sec_maps; | |
b5cb33ae | 1153 | int sec_text; |
f823f360 | 1154 | int sec_btf; |
32e93fb7 DB |
1155 | char license[ELF_MAX_LICENSE_LEN]; |
1156 | enum bpf_prog_type type; | |
65fdae3d | 1157 | __u32 ifindex; |
32e93fb7 | 1158 | bool verbose; |
6e5094db | 1159 | bool noafalg; |
32e93fb7 | 1160 | struct bpf_elf_st stat; |
f6793eec | 1161 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
1162 | char *log; |
1163 | size_t log_size; | |
32e93fb7 DB |
1164 | }; |
1165 | ||
6256f8c9 | 1166 | struct bpf_elf_sec_data { |
32e93fb7 DB |
1167 | GElf_Shdr sec_hdr; |
1168 | Elf_Data *sec_data; | |
1169 | const char *sec_name; | |
6256f8c9 DB |
1170 | }; |
1171 | ||
1172 | struct bpf_map_data { | |
32e93fb7 DB |
1173 | int *fds; |
1174 | const char *obj; | |
1175 | struct bpf_elf_st *st; | |
1176 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
1177 | }; |
1178 | ||
f823f360 DB |
1179 | static bool bpf_log_has_data(struct bpf_elf_ctx *ctx) |
1180 | { | |
1181 | return ctx->log && ctx->log[0]; | |
1182 | } | |
1183 | ||
f31645d1 DB |
1184 | static __check_format_string(2, 3) void |
1185 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
1186 | { |
1187 | va_list vl; | |
1188 | ||
1189 | va_start(vl, format); | |
1190 | vfprintf(stderr, format, vl); | |
1191 | va_end(vl); | |
1192 | ||
f823f360 | 1193 | if (bpf_log_has_data(ctx)) { |
afc1a200 DB |
1194 | if (ctx->verbose) { |
1195 | fprintf(stderr, "%s\n", ctx->log); | |
1196 | } else { | |
1197 | unsigned int off = 0, len = strlen(ctx->log); | |
1198 | ||
1199 | if (len > BPF_MAX_LOG) { | |
1200 | off = len - BPF_MAX_LOG; | |
1201 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
1202 | off); | |
1203 | } | |
1204 | fprintf(stderr, "%s\n", ctx->log + off); | |
1205 | } | |
1206 | ||
f31645d1 DB |
1207 | memset(ctx->log, 0, ctx->log_size); |
1208 | } | |
1209 | } | |
1210 | ||
1211 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
1212 | { | |
0f74d0f3 | 1213 | const size_t log_max = UINT_MAX >> 8; |
f31645d1 | 1214 | size_t log_size = ctx->log_size; |
f6a54d72 | 1215 | char *ptr; |
f31645d1 DB |
1216 | |
1217 | if (!ctx->log) { | |
1218 | log_size = 65536; | |
0f74d0f3 | 1219 | } else if (log_size < log_max) { |
f31645d1 | 1220 | log_size <<= 1; |
0f74d0f3 TG |
1221 | if (log_size > log_max) |
1222 | log_size = log_max; | |
1223 | } else { | |
1224 | return -EINVAL; | |
d937a74b | 1225 | } |
f31645d1 DB |
1226 | |
1227 | ptr = realloc(ctx->log, log_size); | |
1228 | if (!ptr) | |
1229 | return -ENOMEM; | |
1230 | ||
f6a54d72 | 1231 | ptr[0] = 0; |
f31645d1 DB |
1232 | ctx->log = ptr; |
1233 | ctx->log_size = log_size; | |
1234 | ||
1235 | return 0; | |
11c39b5e DB |
1236 | } |
1237 | ||
4dd3f50a DB |
1238 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
1239 | uint32_t size_value, uint32_t max_elem, | |
f823f360 DB |
1240 | uint32_t flags, int inner_fd, int btf_fd, |
1241 | uint32_t ifindex, uint32_t btf_id_key, | |
1242 | uint32_t btf_id_val) | |
11c39b5e | 1243 | { |
d17b136f | 1244 | union bpf_attr attr = {}; |
67584e3a | 1245 | |
67584e3a ND |
1246 | attr.map_type = type; |
1247 | attr.key_size = size_key; | |
612ff099 | 1248 | attr.value_size = inner_fd ? sizeof(int) : size_value; |
67584e3a | 1249 | attr.max_entries = max_elem; |
4dd3f50a | 1250 | attr.map_flags = flags; |
612ff099 | 1251 | attr.inner_map_fd = inner_fd; |
5691e6bc | 1252 | attr.map_ifindex = ifindex; |
f823f360 DB |
1253 | attr.btf_fd = btf_fd; |
1254 | attr.btf_key_type_id = btf_id_key; | |
1255 | attr.btf_value_type_id = btf_id_val; | |
11c39b5e DB |
1256 | |
1257 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
1258 | } | |
11c39b5e | 1259 | |
f823f360 DB |
1260 | static int bpf_btf_load(void *btf, size_t size_btf, |
1261 | char *log, size_t size_log) | |
1262 | { | |
1263 | union bpf_attr attr = {}; | |
1264 | ||
1265 | attr.btf = bpf_ptr_to_u64(btf); | |
1266 | attr.btf_size = size_btf; | |
1267 | ||
1268 | if (size_log > 0) { | |
1269 | attr.btf_log_buf = bpf_ptr_to_u64(log); | |
1270 | attr.btf_log_size = size_log; | |
1271 | attr.btf_log_level = 1; | |
1272 | } | |
1273 | ||
1274 | return bpf(BPF_BTF_LOAD, &attr, sizeof(attr)); | |
1275 | } | |
1276 | ||
32e93fb7 | 1277 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 1278 | { |
d17b136f | 1279 | union bpf_attr attr = {}; |
67584e3a | 1280 | |
67584e3a ND |
1281 | attr.pathname = bpf_ptr_to_u64(pathname); |
1282 | attr.bpf_fd = fd; | |
32e93fb7 DB |
1283 | |
1284 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
1285 | } | |
11c39b5e | 1286 | |
32e93fb7 DB |
1287 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
1288 | { | |
1289 | struct sockaddr_alg alg = { | |
1290 | .salg_family = AF_ALG, | |
1291 | .salg_type = "hash", | |
1292 | .salg_name = "sha1", | |
1293 | }; | |
1294 | int ret, cfd, ofd, ffd; | |
1295 | struct stat stbuff; | |
1296 | ssize_t size; | |
1297 | ||
1298 | if (!object || len != 20) | |
1299 | return -EINVAL; | |
1300 | ||
1301 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
6e5094db | 1302 | if (cfd < 0) |
32e93fb7 | 1303 | return cfd; |
32e93fb7 DB |
1304 | |
1305 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
6e5094db | 1306 | if (ret < 0) |
32e93fb7 | 1307 | goto out_cfd; |
32e93fb7 DB |
1308 | |
1309 | ofd = accept(cfd, NULL, 0); | |
1310 | if (ofd < 0) { | |
32e93fb7 DB |
1311 | ret = ofd; |
1312 | goto out_cfd; | |
1313 | } | |
1314 | ||
1315 | ffd = open(object, O_RDONLY); | |
1316 | if (ffd < 0) { | |
1317 | fprintf(stderr, "Error opening object %s: %s\n", | |
1318 | object, strerror(errno)); | |
1319 | ret = ffd; | |
1320 | goto out_ofd; | |
1321 | } | |
1322 | ||
32a121cb | 1323 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
1324 | if (ret < 0) { |
1325 | fprintf(stderr, "Error doing fstat: %s\n", | |
1326 | strerror(errno)); | |
1327 | goto out_ffd; | |
d937a74b | 1328 | } |
11c39b5e | 1329 | |
32e93fb7 DB |
1330 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
1331 | if (size != stbuff.st_size) { | |
1332 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
1333 | size, stbuff.st_size, strerror(errno)); | |
1334 | ret = -1; | |
1335 | goto out_ffd; | |
1336 | } | |
1337 | ||
1338 | size = read(ofd, out, len); | |
1339 | if (size != len) { | |
1340 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
1341 | size, len, strerror(errno)); | |
1342 | ret = -1; | |
1343 | } else { | |
1344 | ret = 0; | |
1345 | } | |
1346 | out_ffd: | |
1347 | close(ffd); | |
1348 | out_ofd: | |
1349 | close(ofd); | |
1350 | out_cfd: | |
1351 | close(cfd); | |
1352 | return ret; | |
11c39b5e DB |
1353 | } |
1354 | ||
6e5094db | 1355 | static void bpf_init_env(void) |
32e93fb7 DB |
1356 | { |
1357 | struct rlimit limit = { | |
1358 | .rlim_cur = RLIM_INFINITY, | |
1359 | .rlim_max = RLIM_INFINITY, | |
1360 | }; | |
1361 | ||
1362 | /* Don't bother in case we fail! */ | |
1363 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
1364 | ||
6e5094db | 1365 | if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) |
32a121cb | 1366 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
6256f8c9 DB |
1367 | } |
1368 | ||
f6793eec DB |
1369 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
1370 | uint32_t pinning) | |
1371 | { | |
1372 | struct bpf_hash_entry *entry; | |
1373 | ||
1374 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1375 | while (entry && entry->pinning != pinning) | |
1376 | entry = entry->next; | |
1377 | ||
1378 | return entry ? entry->subpath : NULL; | |
1379 | } | |
1380 | ||
1381 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
1382 | uint32_t pinning) | |
11c39b5e | 1383 | { |
32e93fb7 DB |
1384 | switch (pinning) { |
1385 | case PIN_OBJECT_NS: | |
1386 | case PIN_GLOBAL_NS: | |
1387 | return false; | |
1388 | case PIN_NONE: | |
32e93fb7 | 1389 | return true; |
f6793eec DB |
1390 | default: |
1391 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
1392 | } |
1393 | } | |
1394 | ||
1395 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 1396 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
1397 | { |
1398 | switch (pinning) { | |
1399 | case PIN_OBJECT_NS: | |
e4225669 DB |
1400 | snprintf(pathname, len, "%s/%s/%s", |
1401 | bpf_get_work_dir(ctx->type), | |
6e5094db | 1402 | ctx->obj_uid, name); |
32e93fb7 DB |
1403 | break; |
1404 | case PIN_GLOBAL_NS: | |
e4225669 DB |
1405 | snprintf(pathname, len, "%s/%s/%s", |
1406 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1407 | BPF_DIR_GLOBALS, name); |
1408 | break; | |
f6793eec | 1409 | default: |
e4225669 DB |
1410 | snprintf(pathname, len, "%s/../%s/%s", |
1411 | bpf_get_work_dir(ctx->type), | |
f6793eec DB |
1412 | bpf_custom_pinning(ctx, pinning), name); |
1413 | break; | |
32e93fb7 DB |
1414 | } |
1415 | } | |
1416 | ||
f6793eec DB |
1417 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1418 | uint32_t pinning) | |
32e93fb7 DB |
1419 | { |
1420 | char pathname[PATH_MAX]; | |
1421 | ||
e4225669 | 1422 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
32e93fb7 DB |
1423 | return 0; |
1424 | ||
f6793eec | 1425 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
e4225669 | 1426 | return bpf_obj_get(pathname, ctx->type); |
32e93fb7 DB |
1427 | } |
1428 | ||
e4225669 | 1429 | static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx) |
32e93fb7 | 1430 | { |
f6793eec | 1431 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1432 | int ret; |
1433 | ||
e4225669 | 1434 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type), |
6e5094db | 1435 | ctx->obj_uid); |
f6793eec DB |
1436 | |
1437 | ret = mkdir(tmp, S_IRWXU); | |
1438 | if (ret && errno != EEXIST) { | |
1439 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1440 | return ret; | |
1441 | } | |
1442 | ||
1443 | return 0; | |
1444 | } | |
1445 | ||
e4225669 DB |
1446 | static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx, |
1447 | const char *todo) | |
f6793eec DB |
1448 | { |
1449 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1450 | int ret; | |
1451 | ||
e4225669 | 1452 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type)); |
f6793eec DB |
1453 | snprintf(rem, sizeof(rem), "%s/", todo); |
1454 | sub = strtok(rem, "/"); | |
32e93fb7 | 1455 | |
f6793eec DB |
1456 | while (sub) { |
1457 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1458 | return -EINVAL; | |
1459 | ||
1460 | strcat(tmp, sub); | |
1461 | strcat(tmp, "/"); | |
32e93fb7 | 1462 | |
f6793eec | 1463 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1464 | if (ret && errno != EEXIST) { |
f6793eec | 1465 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1466 | strerror(errno)); |
1467 | return ret; | |
1468 | } | |
f6793eec DB |
1469 | |
1470 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1471 | } |
1472 | ||
f6793eec DB |
1473 | return 0; |
1474 | } | |
1475 | ||
1476 | static int bpf_place_pinned(int fd, const char *name, | |
1477 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1478 | { | |
1479 | char pathname[PATH_MAX]; | |
1480 | const char *tmp; | |
1481 | int ret = 0; | |
1482 | ||
e4225669 | 1483 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
f6793eec DB |
1484 | return 0; |
1485 | ||
1486 | if (pinning == PIN_OBJECT_NS) | |
e4225669 | 1487 | ret = bpf_make_obj_path(ctx); |
f6793eec | 1488 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) |
e4225669 | 1489 | ret = bpf_make_custom_path(ctx, tmp); |
f6793eec DB |
1490 | if (ret < 0) |
1491 | return ret; | |
1492 | ||
1493 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1494 | return bpf_obj_pin(fd, pathname); |
1495 | } | |
1496 | ||
f31645d1 DB |
1497 | static void bpf_prog_report(int fd, const char *section, |
1498 | const struct bpf_elf_prog *prog, | |
1499 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1500 | { |
afc1a200 DB |
1501 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1502 | ||
1503 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1504 | fd < 0 ? "rejected: " : "loaded", |
1505 | fd < 0 ? strerror(errno) : "", | |
1506 | fd < 0 ? errno : fd); | |
1507 | ||
1508 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1509 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1510 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1511 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1512 | ||
1513 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1514 | } | |
32e93fb7 | 1515 | |
f31645d1 DB |
1516 | static int bpf_prog_attach(const char *section, |
1517 | const struct bpf_elf_prog *prog, | |
1518 | struct bpf_elf_ctx *ctx) | |
1519 | { | |
1520 | int tries = 0, fd; | |
1521 | retry: | |
32e93fb7 | 1522 | errno = 0; |
65fdae3d JK |
1523 | fd = bpf_prog_load_dev(prog->type, prog->insns, prog->size, |
1524 | prog->license, ctx->ifindex, | |
1525 | ctx->log, ctx->log_size); | |
f31645d1 DB |
1526 | if (fd < 0 || ctx->verbose) { |
1527 | /* The verifier log is pretty chatty, sometimes so chatty | |
1528 | * on larger programs, that we could fail to dump everything | |
1529 | * into our buffer. Still, try to give a debuggable error | |
1530 | * log for the user, so enlarge it and re-fail. | |
1531 | */ | |
1532 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
0f74d0f3 | 1533 | if (tries++ < 10 && !bpf_log_realloc(ctx)) |
f31645d1 DB |
1534 | goto retry; |
1535 | ||
32a121cb | 1536 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1537 | ctx->log_size, tries); |
1538 | return fd; | |
1539 | } | |
1540 | ||
1541 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1542 | } |
1543 | ||
1544 | return fd; | |
1545 | } | |
1546 | ||
f31645d1 DB |
1547 | static void bpf_map_report(int fd, const char *name, |
1548 | const struct bpf_elf_map *map, | |
612ff099 | 1549 | struct bpf_elf_ctx *ctx, int inner_fd) |
f31645d1 DB |
1550 | { |
1551 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1552 | fd < 0 ? "rejected: " : "loaded", | |
1553 | fd < 0 ? strerror(errno) : "", | |
1554 | fd < 0 ? errno : fd); | |
1555 | ||
1556 | fprintf(stderr, " - Type: %u\n", map->type); | |
1557 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1558 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1559 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
612ff099 DB |
1560 | fprintf(stderr, " - Size value: %u\n", |
1561 | inner_fd ? (int)sizeof(int) : map->size_value); | |
4dd3f50a DB |
1562 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1563 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1564 | } |
1565 | ||
612ff099 DB |
1566 | static int bpf_find_map_id(const struct bpf_elf_ctx *ctx, uint32_t id) |
1567 | { | |
1568 | int i; | |
1569 | ||
1570 | for (i = 0; i < ctx->map_num; i++) { | |
1571 | if (ctx->maps[i].id != id) | |
1572 | continue; | |
1573 | if (ctx->map_fds[i] < 0) | |
1574 | return -EINVAL; | |
1575 | ||
1576 | return ctx->map_fds[i]; | |
1577 | } | |
1578 | ||
1579 | return -ENOENT; | |
1580 | } | |
1581 | ||
0b5eadc5 | 1582 | static void bpf_report_map_in_map(int outer_fd, uint32_t idx) |
612ff099 DB |
1583 | { |
1584 | struct bpf_elf_map outer_map; | |
1585 | int ret; | |
1586 | ||
1587 | fprintf(stderr, "Cannot insert map into map! "); | |
1588 | ||
ecb05c0f | 1589 | ret = bpf_derive_elf_map_from_fdinfo(outer_fd, &outer_map, NULL); |
612ff099 DB |
1590 | if (!ret) { |
1591 | if (idx >= outer_map.max_elem && | |
1592 | outer_map.type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
1593 | fprintf(stderr, "Outer map has %u elements, index %u is invalid!\n", | |
1594 | outer_map.max_elem, idx); | |
1595 | return; | |
1596 | } | |
1597 | } | |
1598 | ||
1599 | fprintf(stderr, "Different map specs used for outer and inner map?\n"); | |
1600 | } | |
1601 | ||
1602 | static bool bpf_is_map_in_map_type(const struct bpf_elf_map *map) | |
1603 | { | |
1604 | return map->type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1605 | map->type == BPF_MAP_TYPE_HASH_OF_MAPS; | |
1606 | } | |
1607 | ||
33fde2b6 SH |
1608 | static bool bpf_map_offload_neutral(enum bpf_map_type type) |
1609 | { | |
1610 | return type == BPF_MAP_TYPE_PERF_EVENT_ARRAY; | |
1611 | } | |
1612 | ||
ecb05c0f DB |
1613 | static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx, |
1614 | const struct bpf_elf_map *map, struct bpf_map_ext *ext, | |
1615 | int *have_map_in_map) | |
32e93fb7 | 1616 | { |
0c0394ff | 1617 | int fd, ifindex, ret, map_inner_fd = 0; |
32e93fb7 | 1618 | |
f6793eec | 1619 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1620 | if (fd > 0) { |
ecb05c0f | 1621 | ret = bpf_map_selfcheck_pinned(fd, map, ext, |
91d88eeb | 1622 | offsetof(struct bpf_elf_map, |
fb24802b | 1623 | id), ctx->type); |
9e607f2e DB |
1624 | if (ret < 0) { |
1625 | close(fd); | |
1626 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1627 | name); | |
1628 | return ret; | |
1629 | } | |
f31645d1 | 1630 | if (ctx->verbose) |
32e93fb7 DB |
1631 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1632 | name); | |
1633 | return fd; | |
1634 | } | |
1635 | ||
612ff099 DB |
1636 | if (have_map_in_map && bpf_is_map_in_map_type(map)) { |
1637 | (*have_map_in_map)++; | |
1638 | if (map->inner_id) | |
1639 | return 0; | |
1640 | fprintf(stderr, "Map \'%s\' cannot be created since no inner map ID defined!\n", | |
1641 | name); | |
1642 | return -EINVAL; | |
1643 | } | |
1644 | ||
1645 | if (!have_map_in_map && bpf_is_map_in_map_type(map)) { | |
1646 | map_inner_fd = bpf_find_map_id(ctx, map->inner_id); | |
1647 | if (map_inner_fd < 0) { | |
1648 | fprintf(stderr, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n", | |
1649 | name, map->inner_id); | |
1650 | return -EINVAL; | |
1651 | } | |
1652 | } | |
1653 | ||
0c0394ff | 1654 | ifindex = bpf_map_offload_neutral(map->type) ? 0 : ctx->ifindex; |
32e93fb7 DB |
1655 | errno = 0; |
1656 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
f823f360 DB |
1657 | map->max_elem, map->flags, map_inner_fd, ctx->btf_fd, |
1658 | ifindex, ext->btf_id_key, ext->btf_id_val); | |
5691e6bc | 1659 | |
f31645d1 | 1660 | if (fd < 0 || ctx->verbose) { |
612ff099 | 1661 | bpf_map_report(fd, name, map, ctx, map_inner_fd); |
32e93fb7 DB |
1662 | if (fd < 0) |
1663 | return fd; | |
1664 | } | |
1665 | ||
f6793eec | 1666 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1667 | if (ret < 0 && errno != EEXIST) { |
1668 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1669 | strerror(errno)); | |
1670 | close(fd); | |
1671 | return ret; | |
1672 | } | |
1673 | ||
1674 | return fd; | |
1675 | } | |
1676 | ||
32e93fb7 DB |
1677 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1678 | const GElf_Sym *sym) | |
1679 | { | |
1680 | return ctx->str_tab->d_buf + sym->st_name; | |
1681 | } | |
1682 | ||
f823f360 DB |
1683 | static int bpf_btf_find(struct bpf_elf_ctx *ctx, const char *name) |
1684 | { | |
1685 | const struct btf_type *type; | |
1686 | const char *res; | |
1687 | int id; | |
1688 | ||
1689 | for (id = 1; id < ctx->btf.types_num; id++) { | |
1690 | type = ctx->btf.types[id]; | |
1691 | if (type->name_off >= ctx->btf.hdr->str_len) | |
1692 | continue; | |
1693 | res = &ctx->btf.strings[type->name_off]; | |
1694 | if (!strcmp(res, name)) | |
1695 | return id; | |
1696 | } | |
1697 | ||
1698 | return -ENOENT; | |
1699 | } | |
1700 | ||
1701 | static int bpf_btf_find_kv(struct bpf_elf_ctx *ctx, const struct bpf_elf_map *map, | |
1702 | const char *name, uint32_t *id_key, uint32_t *id_val) | |
1703 | { | |
1704 | const struct btf_member *key, *val; | |
1705 | const struct btf_type *type; | |
1706 | char btf_name[512]; | |
1707 | const char *res; | |
1708 | int id; | |
1709 | ||
1710 | snprintf(btf_name, sizeof(btf_name), "____btf_map_%s", name); | |
1711 | id = bpf_btf_find(ctx, btf_name); | |
1712 | if (id < 0) | |
1713 | return id; | |
1714 | ||
1715 | type = ctx->btf.types[id]; | |
1716 | if (BTF_INFO_KIND(type->info) != BTF_KIND_STRUCT) | |
1717 | return -EINVAL; | |
1718 | if (BTF_INFO_VLEN(type->info) != 2) | |
1719 | return -EINVAL; | |
1720 | ||
1721 | key = ((void *) type) + sizeof(*type); | |
1722 | val = key + 1; | |
1723 | if (!key->type || key->type >= ctx->btf.types_num || | |
1724 | !val->type || val->type >= ctx->btf.types_num) | |
1725 | return -EINVAL; | |
1726 | ||
1727 | if (key->name_off >= ctx->btf.hdr->str_len || | |
1728 | val->name_off >= ctx->btf.hdr->str_len) | |
1729 | return -EINVAL; | |
1730 | ||
1731 | res = &ctx->btf.strings[key->name_off]; | |
1732 | if (strcmp(res, "key")) | |
1733 | return -EINVAL; | |
1734 | ||
1735 | res = &ctx->btf.strings[val->name_off]; | |
1736 | if (strcmp(res, "value")) | |
1737 | return -EINVAL; | |
1738 | ||
1739 | *id_key = key->type; | |
1740 | *id_val = val->type; | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | static void bpf_btf_annotate(struct bpf_elf_ctx *ctx, int which, const char *name) | |
1745 | { | |
1746 | uint32_t id_key = 0, id_val = 0; | |
1747 | ||
1748 | if (!bpf_btf_find_kv(ctx, &ctx->maps[which], name, &id_key, &id_val)) { | |
1749 | ctx->maps_ext[which].btf_id_key = id_key; | |
1750 | ctx->maps_ext[which].btf_id_val = id_val; | |
1751 | } | |
1752 | } | |
1753 | ||
32e93fb7 DB |
1754 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) |
1755 | { | |
f823f360 | 1756 | const char *name; |
32e93fb7 | 1757 | GElf_Sym sym; |
11c39b5e DB |
1758 | int i; |
1759 | ||
32e93fb7 | 1760 | for (i = 0; i < ctx->sym_num; i++) { |
1a7d3ad8 | 1761 | int type; |
7a04dd84 | 1762 | |
32e93fb7 DB |
1763 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) |
1764 | continue; | |
1765 | ||
1a7d3ad8 | 1766 | type = GELF_ST_TYPE(sym.st_info); |
5230a2ed | 1767 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
7a04dd84 | 1768 | (type != STT_NOTYPE && type != STT_OBJECT) || |
32e93fb7 | 1769 | sym.st_shndx != ctx->sec_maps || |
e4225669 | 1770 | sym.st_value / ctx->map_len != which) |
32e93fb7 DB |
1771 | continue; |
1772 | ||
f823f360 DB |
1773 | name = bpf_str_tab_name(ctx, &sym); |
1774 | bpf_btf_annotate(ctx, which, name); | |
1775 | return name; | |
11c39b5e | 1776 | } |
32e93fb7 DB |
1777 | |
1778 | return NULL; | |
11c39b5e DB |
1779 | } |
1780 | ||
32e93fb7 | 1781 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1782 | { |
612ff099 | 1783 | int i, j, ret, fd, inner_fd, inner_idx, have_map_in_map = 0; |
32e93fb7 | 1784 | const char *map_name; |
11c39b5e | 1785 | |
32e93fb7 | 1786 | for (i = 0; i < ctx->map_num; i++) { |
6e5094db DB |
1787 | if (ctx->maps[i].pinning == PIN_OBJECT_NS && |
1788 | ctx->noafalg) { | |
1789 | fprintf(stderr, "Missing kernel AF_ALG support for PIN_OBJECT_NS!\n"); | |
1790 | return -ENOTSUP; | |
1791 | } | |
1792 | ||
32e93fb7 DB |
1793 | map_name = bpf_map_fetch_name(ctx, i); |
1794 | if (!map_name) | |
1795 | return -EIO; | |
11c39b5e | 1796 | |
ecb05c0f DB |
1797 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1798 | &ctx->maps_ext[i], &have_map_in_map); | |
612ff099 DB |
1799 | if (fd < 0) |
1800 | return fd; | |
1801 | ||
1802 | ctx->map_fds[i] = !fd ? -1 : fd; | |
1803 | } | |
1804 | ||
1805 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { | |
1806 | if (ctx->map_fds[i] >= 0) | |
1807 | continue; | |
1808 | ||
1809 | map_name = bpf_map_fetch_name(ctx, i); | |
1810 | if (!map_name) | |
1811 | return -EIO; | |
1812 | ||
ecb05c0f DB |
1813 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1814 | &ctx->maps_ext[i], NULL); | |
32e93fb7 DB |
1815 | if (fd < 0) |
1816 | return fd; | |
11c39b5e | 1817 | |
32e93fb7 | 1818 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1819 | } |
1820 | ||
612ff099 DB |
1821 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { |
1822 | if (!ctx->maps[i].id || | |
1823 | ctx->maps[i].inner_id || | |
1824 | ctx->maps[i].inner_idx == -1) | |
1825 | continue; | |
1826 | ||
1827 | inner_fd = ctx->map_fds[i]; | |
1828 | inner_idx = ctx->maps[i].inner_idx; | |
1829 | ||
1830 | for (j = 0; j < ctx->map_num; j++) { | |
1831 | if (!bpf_is_map_in_map_type(&ctx->maps[j])) | |
1832 | continue; | |
1833 | if (ctx->maps[j].inner_id != ctx->maps[i].id) | |
1834 | continue; | |
1835 | ||
1836 | ret = bpf_map_update(ctx->map_fds[j], &inner_idx, | |
1837 | &inner_fd, BPF_ANY); | |
1838 | if (ret < 0) { | |
1839 | bpf_report_map_in_map(ctx->map_fds[j], | |
0b5eadc5 | 1840 | inner_idx); |
612ff099 DB |
1841 | return ret; |
1842 | } | |
1843 | } | |
1844 | } | |
1845 | ||
11c39b5e | 1846 | return 0; |
11c39b5e DB |
1847 | } |
1848 | ||
e4225669 DB |
1849 | static int bpf_map_num_sym(struct bpf_elf_ctx *ctx) |
1850 | { | |
1851 | int i, num = 0; | |
1852 | GElf_Sym sym; | |
1853 | ||
1854 | for (i = 0; i < ctx->sym_num; i++) { | |
1a7d3ad8 | 1855 | int type; |
7a04dd84 | 1856 | |
e4225669 DB |
1857 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) |
1858 | continue; | |
1859 | ||
1a7d3ad8 | 1860 | type = GELF_ST_TYPE(sym.st_info); |
e4225669 | 1861 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
7a04dd84 | 1862 | (type != STT_NOTYPE && type != STT_OBJECT) || |
e4225669 DB |
1863 | sym.st_shndx != ctx->sec_maps) |
1864 | continue; | |
1865 | num++; | |
1866 | } | |
1867 | ||
1868 | return num; | |
1869 | } | |
1870 | ||
32e93fb7 DB |
1871 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1872 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1873 | { |
32e93fb7 | 1874 | Elf_Data *sec_edata; |
11c39b5e DB |
1875 | GElf_Shdr sec_hdr; |
1876 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1877 | char *sec_name; |
1878 | ||
32e93fb7 | 1879 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1880 | |
32e93fb7 | 1881 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1882 | if (!sec_fd) |
1883 | return -EINVAL; | |
11c39b5e DB |
1884 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1885 | return -EIO; | |
1886 | ||
32e93fb7 | 1887 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1888 | sec_hdr.sh_name); |
1889 | if (!sec_name || !sec_hdr.sh_size) | |
1890 | return -ENOENT; | |
1891 | ||
1892 | sec_edata = elf_getdata(sec_fd, NULL); | |
1893 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1894 | return -EIO; | |
1895 | ||
32e93fb7 | 1896 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1897 | |
32e93fb7 DB |
1898 | data->sec_name = sec_name; |
1899 | data->sec_data = sec_edata; | |
11c39b5e DB |
1900 | return 0; |
1901 | } | |
1902 | ||
e4225669 DB |
1903 | struct bpf_elf_map_min { |
1904 | __u32 type; | |
1905 | __u32 size_key; | |
1906 | __u32 size_value; | |
1907 | __u32 max_elem; | |
1908 | }; | |
11c39b5e | 1909 | |
e4225669 DB |
1910 | static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section, |
1911 | struct bpf_elf_sec_data *data) | |
1912 | { | |
1913 | ctx->map_num = data->sec_data->d_size; | |
32e93fb7 DB |
1914 | ctx->sec_maps = section; |
1915 | ctx->sec_done[section] = true; | |
11c39b5e | 1916 | |
e4225669 | 1917 | if (ctx->map_num > sizeof(ctx->maps)) { |
32e93fb7 DB |
1918 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); |
1919 | return -ENOMEM; | |
1920 | } | |
11c39b5e | 1921 | |
e4225669 DB |
1922 | memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num); |
1923 | return 0; | |
1924 | } | |
1925 | ||
1926 | static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end) | |
1927 | { | |
1928 | GElf_Sym sym; | |
1929 | int off, i; | |
1930 | ||
1931 | for (off = 0; off < end; off += ctx->map_len) { | |
1932 | /* Order doesn't need to be linear here, hence we walk | |
1933 | * the table again. | |
1934 | */ | |
1935 | for (i = 0; i < ctx->sym_num; i++) { | |
1a7d3ad8 | 1936 | int type; |
7a04dd84 | 1937 | |
e4225669 DB |
1938 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) |
1939 | continue; | |
1a7d3ad8 QM |
1940 | |
1941 | type = GELF_ST_TYPE(sym.st_info); | |
e4225669 | 1942 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
7a04dd84 | 1943 | (type != STT_NOTYPE && type != STT_OBJECT) || |
e4225669 DB |
1944 | sym.st_shndx != ctx->sec_maps) |
1945 | continue; | |
1946 | if (sym.st_value == off) | |
1947 | break; | |
1948 | if (i == ctx->sym_num - 1) | |
1949 | return -1; | |
1950 | } | |
1951 | } | |
1952 | ||
1953 | return off == end ? 0 : -1; | |
1954 | } | |
1955 | ||
1956 | static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx) | |
1957 | { | |
1958 | struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {}; | |
1959 | int i, sym_num = bpf_map_num_sym(ctx); | |
1960 | __u8 *buff; | |
1961 | ||
1962 | if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) { | |
1963 | fprintf(stderr, "%u maps not supported in current map section!\n", | |
1964 | sym_num); | |
1965 | return -EINVAL; | |
1966 | } | |
1967 | ||
1968 | if (ctx->map_num % sym_num != 0 || | |
1969 | ctx->map_num % sizeof(__u32) != 0) { | |
1970 | fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n"); | |
1971 | return -EINVAL; | |
1972 | } | |
1973 | ||
1974 | ctx->map_len = ctx->map_num / sym_num; | |
1975 | if (bpf_map_verify_all_offs(ctx, ctx->map_num)) { | |
1976 | fprintf(stderr, "Different struct bpf_elf_map in use!\n"); | |
1977 | return -EINVAL; | |
1978 | } | |
1979 | ||
1980 | if (ctx->map_len == sizeof(struct bpf_elf_map)) { | |
1981 | ctx->map_num = sym_num; | |
1982 | return 0; | |
1983 | } else if (ctx->map_len > sizeof(struct bpf_elf_map)) { | |
1984 | fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n"); | |
1985 | return -EINVAL; | |
1986 | } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) { | |
1987 | fprintf(stderr, "struct bpf_elf_map too small, not supported!\n"); | |
1988 | return -EINVAL; | |
1989 | } | |
1990 | ||
1991 | ctx->map_num = sym_num; | |
1992 | for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num; | |
1993 | i++, buff += ctx->map_len) { | |
1994 | /* The fixup leaves the rest of the members as zero, which | |
1995 | * is fine currently, but option exist to set some other | |
1996 | * default value as well when needed in future. | |
1997 | */ | |
1998 | memcpy(&fixup[i], buff, ctx->map_len); | |
1999 | } | |
2000 | ||
2001 | memcpy(ctx->maps, fixup, sizeof(fixup)); | |
282a1fe1 DB |
2002 | if (ctx->verbose) |
2003 | printf("%zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n", | |
2004 | sizeof(struct bpf_elf_map) - ctx->map_len); | |
32e93fb7 DB |
2005 | return 0; |
2006 | } | |
11c39b5e | 2007 | |
32e93fb7 DB |
2008 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
2009 | struct bpf_elf_sec_data *data) | |
2010 | { | |
2011 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
2012 | return -ENOMEM; | |
11c39b5e | 2013 | |
32e93fb7 DB |
2014 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
2015 | ctx->sec_done[section] = true; | |
2016 | return 0; | |
2017 | } | |
11c39b5e | 2018 | |
32e93fb7 DB |
2019 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
2020 | struct bpf_elf_sec_data *data) | |
2021 | { | |
2022 | ctx->sym_tab = data->sec_data; | |
2023 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
2024 | ctx->sec_done[section] = true; | |
11c39b5e DB |
2025 | return 0; |
2026 | } | |
2027 | ||
32e93fb7 DB |
2028 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
2029 | struct bpf_elf_sec_data *data) | |
11c39b5e | 2030 | { |
32e93fb7 DB |
2031 | ctx->str_tab = data->sec_data; |
2032 | ctx->sec_done[section] = true; | |
2033 | return 0; | |
2034 | } | |
11c39b5e | 2035 | |
b5cb33ae DB |
2036 | static int bpf_fetch_text(struct bpf_elf_ctx *ctx, int section, |
2037 | struct bpf_elf_sec_data *data) | |
2038 | { | |
2039 | ctx->sec_text = section; | |
2040 | ctx->sec_done[section] = true; | |
2041 | return 0; | |
2042 | } | |
2043 | ||
f823f360 DB |
2044 | static void bpf_btf_report(int fd, struct bpf_elf_ctx *ctx) |
2045 | { | |
2046 | fprintf(stderr, "\nBTF debug data section \'.BTF\' %s%s (%d)!\n", | |
2047 | fd < 0 ? "rejected: " : "loaded", | |
2048 | fd < 0 ? strerror(errno) : "", | |
2049 | fd < 0 ? errno : fd); | |
2050 | ||
2051 | fprintf(stderr, " - Length: %zu\n", ctx->btf_data->d_size); | |
2052 | ||
2053 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
2054 | } | |
2055 | ||
2056 | static int bpf_btf_attach(struct bpf_elf_ctx *ctx) | |
2057 | { | |
2058 | int tries = 0, fd; | |
2059 | retry: | |
2060 | errno = 0; | |
2061 | fd = bpf_btf_load(ctx->btf_data->d_buf, ctx->btf_data->d_size, | |
2062 | ctx->log, ctx->log_size); | |
2063 | if (fd < 0 || ctx->verbose) { | |
2064 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
2065 | if (tries++ < 10 && !bpf_log_realloc(ctx)) | |
2066 | goto retry; | |
2067 | ||
2068 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", | |
2069 | ctx->log_size, tries); | |
2070 | return fd; | |
2071 | } | |
2072 | ||
2073 | if (bpf_log_has_data(ctx)) | |
2074 | bpf_btf_report(fd, ctx); | |
2075 | } | |
2076 | ||
2077 | return fd; | |
2078 | } | |
2079 | ||
2080 | static int bpf_fetch_btf_begin(struct bpf_elf_ctx *ctx, int section, | |
2081 | struct bpf_elf_sec_data *data) | |
2082 | { | |
2083 | ctx->btf_data = data->sec_data; | |
2084 | ctx->sec_btf = section; | |
2085 | ctx->sec_done[section] = true; | |
2086 | return 0; | |
2087 | } | |
2088 | ||
2089 | static int bpf_btf_check_header(struct bpf_elf_ctx *ctx) | |
2090 | { | |
2091 | const struct btf_header *hdr = ctx->btf_data->d_buf; | |
2092 | const char *str_start, *str_end; | |
2093 | unsigned int data_len; | |
2094 | ||
2095 | if (hdr->magic != BTF_MAGIC) { | |
2096 | fprintf(stderr, "Object has wrong BTF magic: %x, expected: %x!\n", | |
2097 | hdr->magic, BTF_MAGIC); | |
2098 | return -EINVAL; | |
2099 | } | |
2100 | ||
2101 | if (hdr->version != BTF_VERSION) { | |
2102 | fprintf(stderr, "Object has wrong BTF version: %u, expected: %u!\n", | |
2103 | hdr->version, BTF_VERSION); | |
2104 | return -EINVAL; | |
2105 | } | |
2106 | ||
2107 | if (hdr->flags) { | |
2108 | fprintf(stderr, "Object has unsupported BTF flags %x!\n", | |
2109 | hdr->flags); | |
2110 | return -EINVAL; | |
2111 | } | |
2112 | ||
2113 | data_len = ctx->btf_data->d_size - sizeof(*hdr); | |
2114 | if (data_len < hdr->type_off || | |
2115 | data_len < hdr->str_off || | |
2116 | data_len < hdr->type_len + hdr->str_len || | |
2117 | hdr->type_off >= hdr->str_off || | |
2118 | hdr->type_off + hdr->type_len != hdr->str_off || | |
2119 | hdr->str_off + hdr->str_len != data_len || | |
2120 | (hdr->type_off & (sizeof(uint32_t) - 1))) { | |
2121 | fprintf(stderr, "Object has malformed BTF data!\n"); | |
2122 | return -EINVAL; | |
2123 | } | |
2124 | ||
2125 | ctx->btf.hdr = hdr; | |
2126 | ctx->btf.raw = hdr + 1; | |
2127 | ||
2128 | str_start = ctx->btf.raw + hdr->str_off; | |
2129 | str_end = str_start + hdr->str_len; | |
2130 | if (!hdr->str_len || | |
2131 | hdr->str_len - 1 > BTF_MAX_NAME_OFFSET || | |
2132 | str_start[0] || str_end[-1]) { | |
2133 | fprintf(stderr, "Object has malformed BTF string data!\n"); | |
2134 | return -EINVAL; | |
2135 | } | |
2136 | ||
2137 | ctx->btf.strings = str_start; | |
2138 | return 0; | |
2139 | } | |
2140 | ||
2141 | static int bpf_btf_register_type(struct bpf_elf_ctx *ctx, | |
2142 | const struct btf_type *type) | |
2143 | { | |
2144 | int cur = ctx->btf.types_num, num = cur + 1; | |
2145 | const struct btf_type **types; | |
2146 | ||
2147 | types = realloc(ctx->btf.types, num * sizeof(type)); | |
2148 | if (!types) { | |
2149 | free(ctx->btf.types); | |
2150 | ctx->btf.types = NULL; | |
2151 | ctx->btf.types_num = 0; | |
2152 | return -ENOMEM; | |
2153 | } | |
2154 | ||
2155 | ctx->btf.types = types; | |
2156 | ctx->btf.types[cur] = type; | |
2157 | ctx->btf.types_num = num; | |
2158 | return 0; | |
2159 | } | |
2160 | ||
2161 | static struct btf_type btf_type_void; | |
2162 | ||
2163 | static int bpf_btf_prep_type_data(struct bpf_elf_ctx *ctx) | |
2164 | { | |
2165 | const void *type_cur = ctx->btf.raw + ctx->btf.hdr->type_off; | |
2166 | const void *type_end = ctx->btf.raw + ctx->btf.hdr->str_off; | |
2167 | const struct btf_type *type; | |
2168 | uint16_t var_len; | |
2169 | int ret, kind; | |
2170 | ||
2171 | ret = bpf_btf_register_type(ctx, &btf_type_void); | |
2172 | if (ret < 0) | |
2173 | return ret; | |
2174 | ||
2175 | while (type_cur < type_end) { | |
2176 | type = type_cur; | |
2177 | type_cur += sizeof(*type); | |
2178 | ||
2179 | var_len = BTF_INFO_VLEN(type->info); | |
2180 | kind = BTF_INFO_KIND(type->info); | |
2181 | ||
2182 | switch (kind) { | |
2183 | case BTF_KIND_INT: | |
2184 | type_cur += sizeof(int); | |
2185 | break; | |
2186 | case BTF_KIND_ARRAY: | |
2187 | type_cur += sizeof(struct btf_array); | |
2188 | break; | |
2189 | case BTF_KIND_STRUCT: | |
2190 | case BTF_KIND_UNION: | |
2191 | type_cur += var_len * sizeof(struct btf_member); | |
2192 | break; | |
2193 | case BTF_KIND_ENUM: | |
2194 | type_cur += var_len * sizeof(struct btf_enum); | |
2195 | break; | |
2196 | case BTF_KIND_TYPEDEF: | |
2197 | case BTF_KIND_PTR: | |
2198 | case BTF_KIND_FWD: | |
2199 | case BTF_KIND_VOLATILE: | |
2200 | case BTF_KIND_CONST: | |
2201 | case BTF_KIND_RESTRICT: | |
2202 | break; | |
2203 | default: | |
2204 | fprintf(stderr, "Object has unknown BTF type: %u!\n", kind); | |
2205 | return -EINVAL; | |
2206 | } | |
2207 | ||
2208 | ret = bpf_btf_register_type(ctx, type); | |
2209 | if (ret < 0) | |
2210 | return ret; | |
2211 | } | |
2212 | ||
2213 | return 0; | |
2214 | } | |
2215 | ||
2216 | static int bpf_btf_prep_data(struct bpf_elf_ctx *ctx) | |
2217 | { | |
2218 | int ret = bpf_btf_check_header(ctx); | |
2219 | ||
2220 | if (!ret) | |
2221 | return bpf_btf_prep_type_data(ctx); | |
2222 | return ret; | |
2223 | } | |
2224 | ||
2225 | static void bpf_fetch_btf_end(struct bpf_elf_ctx *ctx) | |
2226 | { | |
2227 | int fd = bpf_btf_attach(ctx); | |
2228 | ||
2229 | if (fd < 0) | |
2230 | return; | |
2231 | ctx->btf_fd = fd; | |
2232 | if (bpf_btf_prep_data(ctx) < 0) { | |
2233 | close(ctx->btf_fd); | |
2234 | ctx->btf_fd = 0; | |
2235 | } | |
2236 | } | |
2237 | ||
afc1a200 DB |
2238 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
2239 | { | |
2240 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
2241 | } | |
2242 | ||
f823f360 DB |
2243 | static bool bpf_has_btf_data(const struct bpf_elf_ctx *ctx) |
2244 | { | |
2245 | return ctx->sec_btf; | |
2246 | } | |
2247 | ||
b5cb33ae DB |
2248 | static bool bpf_has_call_data(const struct bpf_elf_ctx *ctx) |
2249 | { | |
2250 | return ctx->sec_text; | |
2251 | } | |
2252 | ||
2253 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx, bool check_text_sec) | |
32e93fb7 DB |
2254 | { |
2255 | struct bpf_elf_sec_data data; | |
2256 | int i, ret = -1; | |
11c39b5e | 2257 | |
32e93fb7 DB |
2258 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2259 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
2260 | if (ret < 0) |
2261 | continue; | |
2262 | ||
cce3d466 DB |
2263 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2264 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
e4225669 | 2265 | ret = bpf_fetch_maps_begin(ctx, i, &data); |
cce3d466 DB |
2266 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2267 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 2268 | ret = bpf_fetch_license(ctx, i, &data); |
b5cb33ae DB |
2269 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2270 | (data.sec_hdr.sh_flags & SHF_EXECINSTR) && | |
2271 | !strcmp(data.sec_name, ".text") && | |
2272 | check_text_sec) | |
2273 | ret = bpf_fetch_text(ctx, i, &data); | |
cce3d466 DB |
2274 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
2275 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
2276 | ret = bpf_fetch_symtab(ctx, i, &data); |
2277 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 2278 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 | 2279 | ret = bpf_fetch_strtab(ctx, i, &data); |
f823f360 DB |
2280 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
2281 | !strcmp(data.sec_name, ".BTF")) | |
2282 | ret = bpf_fetch_btf_begin(ctx, i, &data); | |
32e93fb7 | 2283 | if (ret < 0) { |
afc1a200 | 2284 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 2285 | i); |
e4225669 | 2286 | return ret; |
11c39b5e | 2287 | } |
32e93fb7 DB |
2288 | } |
2289 | ||
f823f360 DB |
2290 | if (bpf_has_btf_data(ctx)) |
2291 | bpf_fetch_btf_end(ctx); | |
afc1a200 | 2292 | if (bpf_has_map_data(ctx)) { |
e4225669 DB |
2293 | ret = bpf_fetch_maps_end(ctx); |
2294 | if (ret < 0) { | |
2295 | fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n"); | |
2296 | return ret; | |
2297 | } | |
2298 | ||
32e93fb7 DB |
2299 | ret = bpf_maps_attach_all(ctx); |
2300 | if (ret < 0) { | |
2301 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
2302 | return ret; | |
11c39b5e DB |
2303 | } |
2304 | } | |
2305 | ||
2306 | return ret; | |
2307 | } | |
2308 | ||
e4225669 DB |
2309 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section, |
2310 | bool *sseen) | |
11c39b5e | 2311 | { |
32e93fb7 DB |
2312 | struct bpf_elf_sec_data data; |
2313 | struct bpf_elf_prog prog; | |
2314 | int ret, i, fd = -1; | |
11c39b5e | 2315 | |
32e93fb7 DB |
2316 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2317 | if (ctx->sec_done[i]) | |
11c39b5e DB |
2318 | continue; |
2319 | ||
32e93fb7 | 2320 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
2321 | if (ret < 0 || |
2322 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
b5cb33ae | 2323 | (data.sec_hdr.sh_flags & SHF_EXECINSTR) && |
cce3d466 | 2324 | !strcmp(data.sec_name, section))) |
11c39b5e DB |
2325 | continue; |
2326 | ||
e4225669 DB |
2327 | *sseen = true; |
2328 | ||
32e93fb7 | 2329 | memset(&prog, 0, sizeof(prog)); |
b5cb33ae DB |
2330 | prog.type = ctx->type; |
2331 | prog.license = ctx->license; | |
2332 | prog.size = data.sec_data->d_size; | |
2333 | prog.insns_num = prog.size / sizeof(struct bpf_insn); | |
2334 | prog.insns = data.sec_data->d_buf; | |
11c39b5e | 2335 | |
f31645d1 | 2336 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 2337 | if (fd < 0) |
e4225669 | 2338 | return fd; |
11c39b5e | 2339 | |
32e93fb7 | 2340 | ctx->sec_done[i] = true; |
11c39b5e DB |
2341 | break; |
2342 | } | |
2343 | ||
32e93fb7 | 2344 | return fd; |
11c39b5e DB |
2345 | } |
2346 | ||
b5cb33ae DB |
2347 | struct bpf_relo_props { |
2348 | struct bpf_tail_call { | |
2349 | unsigned int total; | |
2350 | unsigned int jited; | |
2351 | } tc; | |
2352 | int main_num; | |
ecb05c0f DB |
2353 | }; |
2354 | ||
b5cb33ae DB |
2355 | static int bpf_apply_relo_map(struct bpf_elf_ctx *ctx, struct bpf_elf_prog *prog, |
2356 | GElf_Rel *relo, GElf_Sym *sym, | |
2357 | struct bpf_relo_props *props) | |
2358 | { | |
2359 | unsigned int insn_off = relo->r_offset / sizeof(struct bpf_insn); | |
2360 | unsigned int map_idx = sym->st_value / ctx->map_len; | |
2361 | ||
2362 | if (insn_off >= prog->insns_num) | |
2363 | return -EINVAL; | |
2364 | if (prog->insns[insn_off].code != (BPF_LD | BPF_IMM | BPF_DW)) { | |
2365 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", | |
2366 | insn_off); | |
2367 | return -EINVAL; | |
2368 | } | |
2369 | ||
2370 | if (map_idx >= ARRAY_SIZE(ctx->map_fds)) | |
2371 | return -EINVAL; | |
2372 | if (!ctx->map_fds[map_idx]) | |
2373 | return -EINVAL; | |
2374 | if (ctx->maps[map_idx].type == BPF_MAP_TYPE_PROG_ARRAY) { | |
2375 | props->tc.total++; | |
2376 | if (ctx->maps_ext[map_idx].owner.jited || | |
2377 | (ctx->maps_ext[map_idx].owner.type == 0 && | |
2378 | ctx->cfg.jit_enabled)) | |
2379 | props->tc.jited++; | |
2380 | } | |
2381 | ||
2382 | prog->insns[insn_off].src_reg = BPF_PSEUDO_MAP_FD; | |
2383 | prog->insns[insn_off].imm = ctx->map_fds[map_idx]; | |
2384 | return 0; | |
2385 | } | |
2386 | ||
2387 | static int bpf_apply_relo_call(struct bpf_elf_ctx *ctx, struct bpf_elf_prog *prog, | |
2388 | GElf_Rel *relo, GElf_Sym *sym, | |
2389 | struct bpf_relo_props *props) | |
2390 | { | |
2391 | unsigned int insn_off = relo->r_offset / sizeof(struct bpf_insn); | |
2392 | struct bpf_elf_prog *prog_text = &ctx->prog_text; | |
2393 | ||
2394 | if (insn_off >= prog->insns_num) | |
2395 | return -EINVAL; | |
2396 | if (prog->insns[insn_off].code != (BPF_JMP | BPF_CALL) && | |
2397 | prog->insns[insn_off].src_reg != BPF_PSEUDO_CALL) { | |
2398 | fprintf(stderr, "ELF contains relo data for non call instruction at offset %u! Compiler bug?!\n", | |
2399 | insn_off); | |
2400 | return -EINVAL; | |
2401 | } | |
2402 | ||
2403 | if (!props->main_num) { | |
2404 | struct bpf_insn *insns = realloc(prog->insns, | |
2405 | prog->size + prog_text->size); | |
2406 | if (!insns) | |
2407 | return -ENOMEM; | |
2408 | ||
2409 | memcpy(insns + prog->insns_num, prog_text->insns, | |
2410 | prog_text->size); | |
2411 | props->main_num = prog->insns_num; | |
2412 | prog->insns = insns; | |
2413 | prog->insns_num += prog_text->insns_num; | |
2414 | prog->size += prog_text->size; | |
2415 | } | |
2416 | ||
2417 | prog->insns[insn_off].imm += props->main_num - insn_off; | |
2418 | return 0; | |
2419 | } | |
2420 | ||
32e93fb7 DB |
2421 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
2422 | struct bpf_elf_sec_data *data_relo, | |
b5cb33ae DB |
2423 | struct bpf_elf_prog *prog, |
2424 | struct bpf_relo_props *props) | |
11c39b5e | 2425 | { |
32e93fb7 DB |
2426 | GElf_Shdr *rhdr = &data_relo->sec_hdr; |
2427 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
11c39b5e | 2428 | |
32e93fb7 | 2429 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
32e93fb7 DB |
2430 | GElf_Rel relo; |
2431 | GElf_Sym sym; | |
b5cb33ae | 2432 | int ret = -EIO; |
32e93fb7 DB |
2433 | |
2434 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
2435 | return -EIO; | |
32e93fb7 DB |
2436 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) |
2437 | return -EIO; | |
2438 | ||
b5cb33ae DB |
2439 | if (sym.st_shndx == ctx->sec_maps) |
2440 | ret = bpf_apply_relo_map(ctx, prog, &relo, &sym, props); | |
2441 | else if (sym.st_shndx == ctx->sec_text) | |
2442 | ret = bpf_apply_relo_call(ctx, prog, &relo, &sym, props); | |
2443 | else | |
2444 | fprintf(stderr, "ELF contains non-{map,call} related relo data in entry %u pointing to section %u! Compiler bug?!\n", | |
2445 | relo_ent, sym.st_shndx); | |
2446 | if (ret < 0) | |
2447 | return ret; | |
32e93fb7 DB |
2448 | } |
2449 | ||
2450 | return 0; | |
2451 | } | |
2452 | ||
afc1a200 | 2453 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
b5cb33ae | 2454 | bool *lderr, bool *sseen, struct bpf_elf_prog *prog) |
32e93fb7 DB |
2455 | { |
2456 | struct bpf_elf_sec_data data_relo, data_insn; | |
32e93fb7 DB |
2457 | int ret, idx, i, fd = -1; |
2458 | ||
2459 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
b5cb33ae | 2460 | struct bpf_relo_props props = {}; |
ecb05c0f | 2461 | |
32e93fb7 DB |
2462 | ret = bpf_fill_section_data(ctx, i, &data_relo); |
2463 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
2464 | continue; |
2465 | ||
32e93fb7 | 2466 | idx = data_relo.sec_hdr.sh_info; |
e4225669 | 2467 | |
32e93fb7 | 2468 | ret = bpf_fill_section_data(ctx, idx, &data_insn); |
cce3d466 DB |
2469 | if (ret < 0 || |
2470 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
b5cb33ae | 2471 | (data_insn.sec_hdr.sh_flags & SHF_EXECINSTR) && |
cce3d466 | 2472 | !strcmp(data_insn.sec_name, section))) |
11c39b5e | 2473 | continue; |
b5cb33ae DB |
2474 | if (sseen) |
2475 | *sseen = true; | |
2476 | ||
2477 | memset(prog, 0, sizeof(*prog)); | |
2478 | prog->type = ctx->type; | |
2479 | prog->license = ctx->license; | |
2480 | prog->size = data_insn.sec_data->d_size; | |
2481 | prog->insns_num = prog->size / sizeof(struct bpf_insn); | |
2482 | prog->insns = malloc(prog->size); | |
2483 | if (!prog->insns) { | |
2484 | *lderr = true; | |
2485 | return -ENOMEM; | |
2486 | } | |
32e93fb7 | 2487 | |
b5cb33ae | 2488 | memcpy(prog->insns, data_insn.sec_data->d_buf, prog->size); |
e4225669 | 2489 | |
b5cb33ae | 2490 | ret = bpf_apply_relo_data(ctx, &data_relo, prog, &props); |
c9c3720d DB |
2491 | if (ret < 0) { |
2492 | *lderr = true; | |
b5cb33ae DB |
2493 | if (ctx->sec_text != idx) |
2494 | free(prog->insns); | |
e4225669 | 2495 | return ret; |
c9c3720d | 2496 | } |
b5cb33ae DB |
2497 | if (ctx->sec_text == idx) { |
2498 | fd = 0; | |
2499 | goto out; | |
2500 | } | |
11c39b5e | 2501 | |
b5cb33ae DB |
2502 | fd = bpf_prog_attach(section, prog, ctx); |
2503 | free(prog->insns); | |
afc1a200 DB |
2504 | if (fd < 0) { |
2505 | *lderr = true; | |
b5cb33ae | 2506 | if (props.tc.total) { |
ecb05c0f | 2507 | if (ctx->cfg.jit_enabled && |
b5cb33ae | 2508 | props.tc.total != props.tc.jited) |
ecb05c0f | 2509 | fprintf(stderr, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n", |
b5cb33ae | 2510 | props.tc.jited, props.tc.total); |
ecb05c0f | 2511 | if (!ctx->cfg.jit_enabled && |
b5cb33ae | 2512 | props.tc.jited) |
ecb05c0f | 2513 | fprintf(stderr, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n", |
b5cb33ae | 2514 | props.tc.jited, props.tc.total); |
ecb05c0f | 2515 | } |
e4225669 | 2516 | return fd; |
afc1a200 | 2517 | } |
b5cb33ae | 2518 | out: |
32e93fb7 DB |
2519 | ctx->sec_done[i] = true; |
2520 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
2521 | break; |
2522 | } | |
2523 | ||
32e93fb7 | 2524 | return fd; |
11c39b5e DB |
2525 | } |
2526 | ||
32e93fb7 | 2527 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 2528 | { |
e4225669 | 2529 | bool lderr = false, sseen = false; |
b5cb33ae | 2530 | struct bpf_elf_prog prog; |
473d7840 DB |
2531 | int ret = -1; |
2532 | ||
b5cb33ae DB |
2533 | if (bpf_has_call_data(ctx)) { |
2534 | ret = bpf_fetch_prog_relo(ctx, ".text", &lderr, NULL, | |
2535 | &ctx->prog_text); | |
2536 | if (ret < 0) | |
2537 | return ret; | |
2538 | } | |
2539 | ||
2540 | if (bpf_has_map_data(ctx) || bpf_has_call_data(ctx)) | |
2541 | ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen, &prog); | |
afc1a200 | 2542 | if (ret < 0 && !lderr) |
e4225669 DB |
2543 | ret = bpf_fetch_prog(ctx, section, &sseen); |
2544 | if (ret < 0 && !sseen) | |
2545 | fprintf(stderr, "Program section \'%s\' not found in ELF file!\n", | |
2546 | section); | |
473d7840 DB |
2547 | return ret; |
2548 | } | |
2549 | ||
910b543d DB |
2550 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
2551 | { | |
2552 | int i; | |
2553 | ||
2554 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
2555 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
2556 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
2557 | return i; | |
2558 | return -1; | |
2559 | } | |
2560 | ||
ecb05c0f DB |
2561 | struct bpf_jited_aux { |
2562 | int prog_fd; | |
2563 | int map_fd; | |
2564 | struct bpf_prog_data prog; | |
2565 | struct bpf_map_ext map; | |
2566 | }; | |
2567 | ||
2568 | static int bpf_derive_prog_from_fdinfo(int fd, struct bpf_prog_data *prog) | |
2569 | { | |
2570 | char file[PATH_MAX], buff[4096]; | |
2571 | unsigned int val; | |
2572 | FILE *fp; | |
2573 | ||
2574 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
2575 | memset(prog, 0, sizeof(*prog)); | |
2576 | ||
2577 | fp = fopen(file, "r"); | |
2578 | if (!fp) { | |
2579 | fprintf(stderr, "No procfs support?!\n"); | |
2580 | return -EIO; | |
2581 | } | |
2582 | ||
2583 | while (fgets(buff, sizeof(buff), fp)) { | |
2584 | if (sscanf(buff, "prog_type:\t%u", &val) == 1) | |
2585 | prog->type = val; | |
2586 | else if (sscanf(buff, "prog_jited:\t%u", &val) == 1) | |
2587 | prog->jited = val; | |
2588 | } | |
2589 | ||
2590 | fclose(fp); | |
2591 | return 0; | |
2592 | } | |
2593 | ||
2594 | static int bpf_tail_call_get_aux(struct bpf_jited_aux *aux) | |
2595 | { | |
2596 | struct bpf_elf_map tmp; | |
2597 | int ret; | |
2598 | ||
2599 | ret = bpf_derive_elf_map_from_fdinfo(aux->map_fd, &tmp, &aux->map); | |
2600 | if (!ret) | |
2601 | ret = bpf_derive_prog_from_fdinfo(aux->prog_fd, &aux->prog); | |
2602 | ||
2603 | return ret; | |
2604 | } | |
2605 | ||
32e93fb7 | 2606 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 2607 | { |
32e93fb7 DB |
2608 | struct bpf_elf_sec_data data; |
2609 | uint32_t map_id, key_id; | |
910b543d | 2610 | int fd, i, ret, idx; |
473d7840 | 2611 | |
32e93fb7 DB |
2612 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2613 | if (ctx->sec_done[i]) | |
473d7840 DB |
2614 | continue; |
2615 | ||
32e93fb7 | 2616 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
2617 | if (ret < 0) |
2618 | continue; | |
2619 | ||
910b543d DB |
2620 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
2621 | if (ret != 2) | |
32e93fb7 | 2622 | continue; |
910b543d DB |
2623 | |
2624 | idx = bpf_find_map_by_id(ctx, map_id); | |
2625 | if (idx < 0) | |
473d7840 DB |
2626 | continue; |
2627 | ||
32e93fb7 DB |
2628 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
2629 | if (fd < 0) | |
473d7840 DB |
2630 | return -EIO; |
2631 | ||
910b543d DB |
2632 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
2633 | &fd, BPF_ANY); | |
afc1a200 | 2634 | if (ret < 0) { |
ecb05c0f DB |
2635 | struct bpf_jited_aux aux = {}; |
2636 | ||
2637 | ret = -errno; | |
2638 | if (errno == E2BIG) { | |
afc1a200 DB |
2639 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", |
2640 | key_id, map_id); | |
ecb05c0f DB |
2641 | return ret; |
2642 | } | |
2643 | ||
2644 | aux.map_fd = ctx->map_fds[idx]; | |
2645 | aux.prog_fd = fd; | |
2646 | ||
2647 | if (bpf_tail_call_get_aux(&aux)) | |
2648 | return ret; | |
2649 | if (!aux.map.owner.type) | |
2650 | return ret; | |
2651 | ||
2652 | if (aux.prog.type != aux.map.owner.type) | |
2653 | fprintf(stderr, "Tail call map owned by prog type %u, but prog type is %u!\n", | |
2654 | aux.map.owner.type, aux.prog.type); | |
2655 | if (aux.prog.jited != aux.map.owner.jited) | |
2656 | fprintf(stderr, "Tail call map %s jited, but prog %s!\n", | |
2657 | aux.map.owner.jited ? "is" : "not", | |
2658 | aux.prog.jited ? "is" : "not"); | |
2659 | return ret; | |
afc1a200 | 2660 | } |
473d7840 | 2661 | |
32e93fb7 | 2662 | ctx->sec_done[i] = true; |
473d7840 DB |
2663 | } |
2664 | ||
2665 | return 0; | |
2666 | } | |
2667 | ||
32e93fb7 | 2668 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 2669 | { |
32e93fb7 DB |
2670 | struct stat st; |
2671 | int ret; | |
11c39b5e | 2672 | |
32e93fb7 | 2673 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 2674 | |
32e93fb7 DB |
2675 | ret = fstat(ctx->obj_fd, &st); |
2676 | if (ret < 0) { | |
2677 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
2678 | strerror(errno)); | |
2679 | return; | |
2680 | } | |
11c39b5e | 2681 | |
32e93fb7 DB |
2682 | ctx->stat.st_dev = st.st_dev; |
2683 | ctx->stat.st_ino = st.st_ino; | |
2684 | } | |
2685 | ||
f6793eec DB |
2686 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
2687 | { | |
2688 | char buff[PATH_MAX]; | |
2689 | ||
2690 | while (fgets(buff, sizeof(buff), fp)) { | |
2691 | char *ptr = buff; | |
2692 | ||
2693 | while (*ptr == ' ' || *ptr == '\t') | |
2694 | ptr++; | |
2695 | ||
2696 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
2697 | continue; | |
2698 | ||
2699 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
2700 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
2701 | strcpy(path, ptr); | |
2702 | return -1; | |
2703 | } | |
2704 | ||
2705 | return 1; | |
2706 | } | |
2707 | ||
2708 | return 0; | |
2709 | } | |
2710 | ||
2711 | static bool bpf_pinning_reserved(uint32_t pinning) | |
2712 | { | |
2713 | switch (pinning) { | |
2714 | case PIN_NONE: | |
2715 | case PIN_OBJECT_NS: | |
2716 | case PIN_GLOBAL_NS: | |
2717 | return true; | |
2718 | default: | |
2719 | return false; | |
2720 | } | |
2721 | } | |
2722 | ||
2723 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
2724 | { | |
2725 | struct bpf_hash_entry *entry; | |
d17b136f | 2726 | char subpath[PATH_MAX] = {}; |
f6793eec DB |
2727 | uint32_t pinning; |
2728 | FILE *fp; | |
2729 | int ret; | |
2730 | ||
2731 | fp = fopen(db_file, "r"); | |
2732 | if (!fp) | |
2733 | return; | |
2734 | ||
f6793eec DB |
2735 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { |
2736 | if (ret == -1) { | |
2737 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
2738 | db_file, subpath); | |
2739 | fclose(fp); | |
2740 | return; | |
2741 | } | |
2742 | ||
2743 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
2744 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
2745 | db_file, pinning); | |
f6793eec DB |
2746 | continue; |
2747 | } | |
2748 | ||
2749 | entry = malloc(sizeof(*entry)); | |
2750 | if (!entry) { | |
2751 | fprintf(stderr, "No memory left for db entry!\n"); | |
2752 | continue; | |
2753 | } | |
2754 | ||
2755 | entry->pinning = pinning; | |
2756 | entry->subpath = strdup(subpath); | |
2757 | if (!entry->subpath) { | |
2758 | fprintf(stderr, "No memory left for db entry!\n"); | |
2759 | free(entry); | |
2760 | continue; | |
2761 | } | |
2762 | ||
2763 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
2764 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
2765 | } | |
2766 | ||
2767 | fclose(fp); | |
2768 | } | |
2769 | ||
2770 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
2771 | { | |
2772 | struct bpf_hash_entry *entry; | |
2773 | int i; | |
2774 | ||
2775 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
2776 | while ((entry = ctx->ht[i]) != NULL) { | |
2777 | ctx->ht[i] = entry->next; | |
2778 | free((char *)entry->subpath); | |
2779 | free(entry); | |
2780 | } | |
2781 | } | |
2782 | } | |
2783 | ||
8187b012 DB |
2784 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
2785 | { | |
2786 | if (ctx->elf_hdr.e_type != ET_REL || | |
e77fa41d DB |
2787 | (ctx->elf_hdr.e_machine != EM_NONE && |
2788 | ctx->elf_hdr.e_machine != EM_BPF) || | |
8187b012 DB |
2789 | ctx->elf_hdr.e_version != EV_CURRENT) { |
2790 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
2791 | return -EINVAL; | |
2792 | } | |
2793 | ||
2794 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
2795 | default: | |
2796 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
2797 | return -EINVAL; | |
2798 | case ELFDATA2LSB: | |
2799 | if (htons(1) == 1) { | |
2800 | fprintf(stderr, | |
2801 | "We are big endian, eBPF object is little endian!\n"); | |
2802 | return -EIO; | |
2803 | } | |
2804 | break; | |
2805 | case ELFDATA2MSB: | |
2806 | if (htons(1) != 1) { | |
2807 | fprintf(stderr, | |
2808 | "We are little endian, eBPF object is big endian!\n"); | |
2809 | return -EIO; | |
2810 | } | |
2811 | break; | |
2812 | } | |
2813 | ||
2814 | return 0; | |
2815 | } | |
2816 | ||
ecb05c0f DB |
2817 | static void bpf_get_cfg(struct bpf_elf_ctx *ctx) |
2818 | { | |
2819 | static const char *path_jit = "/proc/sys/net/core/bpf_jit_enable"; | |
2820 | int fd; | |
2821 | ||
2822 | fd = open(path_jit, O_RDONLY); | |
2823 | if (fd > 0) { | |
2824 | char tmp[16] = {}; | |
2825 | ||
2826 | if (read(fd, tmp, sizeof(tmp)) > 0) | |
2827 | ctx->cfg.jit_enabled = atoi(tmp); | |
2828 | close(fd); | |
2829 | } | |
2830 | } | |
2831 | ||
32e93fb7 | 2832 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
65fdae3d JK |
2833 | enum bpf_prog_type type, __u32 ifindex, |
2834 | bool verbose) | |
32e93fb7 | 2835 | { |
6e5094db DB |
2836 | uint8_t tmp[20]; |
2837 | int ret; | |
32e93fb7 | 2838 | |
6e5094db DB |
2839 | if (elf_version(EV_CURRENT) == EV_NONE) |
2840 | return -EINVAL; | |
2841 | ||
2842 | bpf_init_env(); | |
32e93fb7 DB |
2843 | |
2844 | memset(ctx, 0, sizeof(*ctx)); | |
ecb05c0f | 2845 | bpf_get_cfg(ctx); |
6e5094db DB |
2846 | |
2847 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); | |
2848 | if (ret) | |
2849 | ctx->noafalg = true; | |
2850 | else | |
2851 | hexstring_n2a(tmp, sizeof(tmp), ctx->obj_uid, | |
2852 | sizeof(ctx->obj_uid)); | |
2853 | ||
32e93fb7 DB |
2854 | ctx->verbose = verbose; |
2855 | ctx->type = type; | |
65fdae3d | 2856 | ctx->ifindex = ifindex; |
32e93fb7 DB |
2857 | |
2858 | ctx->obj_fd = open(pathname, O_RDONLY); | |
2859 | if (ctx->obj_fd < 0) | |
2860 | return ctx->obj_fd; | |
2861 | ||
2862 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
2863 | if (!ctx->elf_fd) { | |
11c39b5e | 2864 | ret = -EINVAL; |
32e93fb7 | 2865 | goto out_fd; |
11c39b5e DB |
2866 | } |
2867 | ||
8187b012 DB |
2868 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
2869 | ret = -EINVAL; | |
2870 | goto out_fd; | |
2871 | } | |
2872 | ||
32e93fb7 DB |
2873 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
2874 | &ctx->elf_hdr) { | |
11c39b5e DB |
2875 | ret = -EIO; |
2876 | goto out_elf; | |
2877 | } | |
2878 | ||
8187b012 DB |
2879 | ret = bpf_elf_check_ehdr(ctx); |
2880 | if (ret < 0) | |
2881 | goto out_elf; | |
2882 | ||
32e93fb7 DB |
2883 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
2884 | sizeof(*(ctx->sec_done))); | |
2885 | if (!ctx->sec_done) { | |
11c39b5e DB |
2886 | ret = -ENOMEM; |
2887 | goto out_elf; | |
2888 | } | |
2889 | ||
f31645d1 DB |
2890 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
2891 | ret = -ENOMEM; | |
2892 | goto out_free; | |
2893 | } | |
2894 | ||
32e93fb7 | 2895 | bpf_save_finfo(ctx); |
f6793eec DB |
2896 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
2897 | ||
32e93fb7 | 2898 | return 0; |
f31645d1 DB |
2899 | out_free: |
2900 | free(ctx->sec_done); | |
32e93fb7 DB |
2901 | out_elf: |
2902 | elf_end(ctx->elf_fd); | |
2903 | out_fd: | |
2904 | close(ctx->obj_fd); | |
2905 | return ret; | |
2906 | } | |
d937a74b | 2907 | |
32e93fb7 DB |
2908 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
2909 | { | |
2910 | int i, count = 0; | |
11c39b5e | 2911 | |
32e93fb7 DB |
2912 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
2913 | if (!ctx->map_fds[i]) | |
2914 | break; | |
2915 | count++; | |
2916 | } | |
473d7840 | 2917 | |
32e93fb7 DB |
2918 | return count; |
2919 | } | |
6256f8c9 | 2920 | |
32e93fb7 DB |
2921 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
2922 | { | |
2923 | int i; | |
2924 | ||
2925 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
2926 | if (ctx->map_fds[i]) | |
2927 | close(ctx->map_fds[i]); | |
473d7840 | 2928 | } |
f823f360 DB |
2929 | |
2930 | if (ctx->btf_fd) | |
2931 | close(ctx->btf_fd); | |
2932 | free(ctx->btf.types); | |
32e93fb7 DB |
2933 | } |
2934 | ||
2935 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
2936 | { | |
2937 | if (failure) | |
2938 | bpf_maps_teardown(ctx); | |
473d7840 | 2939 | |
f6793eec | 2940 | bpf_hash_destroy(ctx); |
f31645d1 | 2941 | |
b5cb33ae | 2942 | free(ctx->prog_text.insns); |
32e93fb7 | 2943 | free(ctx->sec_done); |
f31645d1 DB |
2944 | free(ctx->log); |
2945 | ||
32e93fb7 DB |
2946 | elf_end(ctx->elf_fd); |
2947 | close(ctx->obj_fd); | |
2948 | } | |
6256f8c9 | 2949 | |
32e93fb7 | 2950 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 2951 | |
32e93fb7 | 2952 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
65fdae3d | 2953 | const char *section, __u32 ifindex, bool verbose) |
32e93fb7 DB |
2954 | { |
2955 | struct bpf_elf_ctx *ctx = &__ctx; | |
2956 | int fd = 0, ret; | |
6256f8c9 | 2957 | |
65fdae3d | 2958 | ret = bpf_elf_ctx_init(ctx, pathname, type, ifindex, verbose); |
32e93fb7 DB |
2959 | if (ret < 0) { |
2960 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
2961 | return ret; | |
2962 | } | |
6256f8c9 | 2963 | |
b5cb33ae | 2964 | ret = bpf_fetch_ancillary(ctx, strcmp(section, ".text")); |
32e93fb7 DB |
2965 | if (ret < 0) { |
2966 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
2967 | goto out; | |
2968 | } | |
2969 | ||
2970 | fd = bpf_fetch_prog_sec(ctx, section); | |
2971 | if (fd < 0) { | |
2972 | fprintf(stderr, "Error fetching program/map!\n"); | |
2973 | ret = fd; | |
2974 | goto out; | |
2975 | } | |
2976 | ||
2977 | ret = bpf_fill_prog_arrays(ctx); | |
2978 | if (ret < 0) | |
2979 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 2980 | out: |
32e93fb7 DB |
2981 | bpf_elf_ctx_destroy(ctx, ret < 0); |
2982 | if (ret < 0) { | |
2983 | if (fd) | |
2984 | close(fd); | |
2985 | return ret; | |
2986 | } | |
2987 | ||
2988 | return fd; | |
6256f8c9 | 2989 | } |
11c39b5e | 2990 | |
6256f8c9 | 2991 | static int |
4bd62446 DB |
2992 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
2993 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 | 2994 | { |
d17b136f PS |
2995 | struct bpf_map_set_msg msg = { |
2996 | .aux.uds_ver = BPF_SCM_AUX_VER, | |
2997 | .aux.num_ent = entries, | |
2998 | }; | |
6256f8c9 DB |
2999 | int *cmsg_buf, min_fd; |
3000 | char *amsg_buf; | |
3001 | int i; | |
3002 | ||
08a93b32 | 3003 | strlcpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); |
6256f8c9 DB |
3004 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); |
3005 | ||
3006 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
3007 | amsg_buf = (char *)msg.aux.ent; | |
3008 | ||
4bd62446 | 3009 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
3010 | int ret; |
3011 | ||
4bd62446 | 3012 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
3013 | bpf_map_set_init_single(&msg, min_fd); |
3014 | ||
3015 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
3016 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
3017 | ||
3018 | ret = sendmsg(fd, &msg.hdr, 0); | |
3019 | if (ret <= 0) | |
3020 | return ret ? : -1; | |
3021 | } | |
3022 | ||
3023 | return 0; | |
11c39b5e DB |
3024 | } |
3025 | ||
4bd62446 DB |
3026 | static int |
3027 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
3028 | unsigned int entries) | |
3029 | { | |
3030 | struct bpf_map_set_msg msg; | |
3031 | int *cmsg_buf, min_fd; | |
3032 | char *amsg_buf, *mmsg_buf; | |
3033 | unsigned int needed = 1; | |
3034 | int i; | |
3035 | ||
3036 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
3037 | amsg_buf = (char *)msg.aux.ent; | |
3038 | mmsg_buf = (char *)&msg.aux; | |
3039 | ||
3040 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
3041 | struct cmsghdr *cmsg; | |
3042 | int ret; | |
3043 | ||
3044 | min_fd = min(entries, entries - i); | |
3045 | bpf_map_set_init_single(&msg, min_fd); | |
3046 | ||
3047 | ret = recvmsg(fd, &msg.hdr, 0); | |
3048 | if (ret <= 0) | |
3049 | return ret ? : -1; | |
3050 | ||
3051 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
3052 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
3053 | return -EINVAL; | |
3054 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
3055 | return -EIO; | |
3056 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
3057 | return -ENOSYS; | |
3058 | ||
3059 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
3060 | if (min_fd > entries || min_fd <= 0) | |
3061 | return -EINVAL; | |
3062 | ||
3063 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
3064 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
3065 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
3066 | ||
3067 | needed = aux->num_ent; | |
3068 | } | |
3069 | ||
3070 | return 0; | |
3071 | } | |
3072 | ||
3073 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 3074 | { |
32e93fb7 | 3075 | struct bpf_elf_ctx *ctx = &__ctx; |
d17b136f PS |
3076 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
3077 | struct bpf_map_data bpf_aux = { | |
3078 | .fds = ctx->map_fds, | |
3079 | .ent = ctx->maps, | |
3080 | .st = &ctx->stat, | |
3081 | .obj = obj, | |
3082 | }; | |
6256f8c9 DB |
3083 | int fd, ret; |
3084 | ||
3085 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
3086 | if (fd < 0) { | |
3087 | fprintf(stderr, "Cannot open socket: %s\n", | |
3088 | strerror(errno)); | |
3089 | return -1; | |
3090 | } | |
3091 | ||
08a93b32 | 3092 | strlcpy(addr.sun_path, path, sizeof(addr.sun_path)); |
6256f8c9 DB |
3093 | |
3094 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
3095 | if (ret < 0) { | |
3096 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
3097 | path, strerror(errno)); | |
3098 | return -1; | |
3099 | } | |
3100 | ||
4bd62446 | 3101 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 3102 | bpf_maps_count(ctx)); |
6256f8c9 | 3103 | if (ret < 0) |
4bd62446 DB |
3104 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
3105 | path, strerror(errno)); | |
3106 | ||
32e93fb7 | 3107 | bpf_maps_teardown(ctx); |
4bd62446 DB |
3108 | close(fd); |
3109 | return ret; | |
3110 | } | |
3111 | ||
3112 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
3113 | unsigned int entries) | |
3114 | { | |
d17b136f | 3115 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
4bd62446 DB |
3116 | int fd, ret; |
3117 | ||
3118 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
3119 | if (fd < 0) { | |
3120 | fprintf(stderr, "Cannot open socket: %s\n", | |
3121 | strerror(errno)); | |
3122 | return -1; | |
3123 | } | |
3124 | ||
08a93b32 | 3125 | strlcpy(addr.sun_path, path, sizeof(addr.sun_path)); |
4bd62446 DB |
3126 | |
3127 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
3128 | if (ret < 0) { | |
3129 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
3130 | strerror(errno)); | |
3131 | return -1; | |
3132 | } | |
3133 | ||
3134 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
3135 | if (ret < 0) | |
3136 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
3137 | path, strerror(errno)); |
3138 | ||
4bd62446 | 3139 | unlink(addr.sun_path); |
6256f8c9 DB |
3140 | close(fd); |
3141 | return ret; | |
3142 | } | |
11c39b5e | 3143 | #endif /* HAVE_ELF */ |