]>
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 | }, | |
e4225669 DB |
98 | }; |
99 | ||
100 | static const char *bpf_prog_to_subdir(enum bpf_prog_type type) | |
101 | { | |
102 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
103 | __bpf_prog_meta[type].subdir); | |
104 | return __bpf_prog_meta[type].subdir; | |
105 | } | |
106 | ||
107 | const char *bpf_prog_to_default_section(enum bpf_prog_type type) | |
108 | { | |
109 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
110 | __bpf_prog_meta[type].section); | |
111 | return __bpf_prog_meta[type].section; | |
112 | } | |
e77fa41d | 113 | |
32e93fb7 DB |
114 | #ifdef HAVE_ELF |
115 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
116 | const char *sec, bool verbose); | |
117 | #else | |
118 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
119 | const char *sec, bool verbose) | |
120 | { | |
121 | fprintf(stderr, "No ELF library support compiled in.\n"); | |
122 | errno = ENOSYS; | |
123 | return -1; | |
124 | } | |
125 | #endif | |
126 | ||
127 | static inline __u64 bpf_ptr_to_u64(const void *ptr) | |
128 | { | |
129 | return (__u64)(unsigned long)ptr; | |
130 | } | |
131 | ||
132 | static int bpf(int cmd, union bpf_attr *attr, unsigned int size) | |
133 | { | |
134 | #ifdef __NR_bpf | |
135 | return syscall(__NR_bpf, cmd, attr, size); | |
136 | #else | |
137 | fprintf(stderr, "No bpf syscall, kernel headers too old?\n"); | |
138 | errno = ENOSYS; | |
139 | return -1; | |
140 | #endif | |
141 | } | |
142 | ||
91d88eeb DB |
143 | static int bpf_map_update(int fd, const void *key, const void *value, |
144 | uint64_t flags) | |
32e93fb7 | 145 | { |
d17b136f | 146 | union bpf_attr attr = {}; |
67584e3a | 147 | |
67584e3a ND |
148 | attr.map_fd = fd; |
149 | attr.key = bpf_ptr_to_u64(key); | |
150 | attr.value = bpf_ptr_to_u64(value); | |
151 | attr.flags = flags; | |
32e93fb7 | 152 | |
91d88eeb | 153 | return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
32e93fb7 DB |
154 | } |
155 | ||
779525cd DB |
156 | static int bpf_prog_fd_by_id(uint32_t id) |
157 | { | |
158 | union bpf_attr attr = {}; | |
159 | ||
160 | attr.prog_id = id; | |
161 | ||
162 | return bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr)); | |
163 | } | |
164 | ||
165 | static int bpf_prog_info_by_fd(int fd, struct bpf_prog_info *info, | |
166 | uint32_t *info_len) | |
167 | { | |
168 | union bpf_attr attr = {}; | |
169 | int ret; | |
170 | ||
171 | attr.info.bpf_fd = fd; | |
172 | attr.info.info = bpf_ptr_to_u64(info); | |
173 | attr.info.info_len = *info_len; | |
174 | ||
175 | *info_len = 0; | |
176 | ret = bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr)); | |
177 | if (!ret) | |
178 | *info_len = attr.info.info_len; | |
179 | ||
180 | return ret; | |
181 | } | |
182 | ||
a0b5b7cf | 183 | int bpf_dump_prog_info(FILE *f, uint32_t id) |
779525cd DB |
184 | { |
185 | struct bpf_prog_info info = {}; | |
186 | uint32_t len = sizeof(info); | |
a0b5b7cf DB |
187 | int fd, ret, dump_ok = 0; |
188 | SPRINT_BUF(tmp); | |
779525cd | 189 | |
bc2d4d83 DB |
190 | open_json_object("prog"); |
191 | print_uint(PRINT_ANY, "id", "id %u ", id); | |
779525cd DB |
192 | |
193 | fd = bpf_prog_fd_by_id(id); | |
194 | if (fd < 0) | |
bc2d4d83 | 195 | goto out; |
779525cd DB |
196 | |
197 | ret = bpf_prog_info_by_fd(fd, &info, &len); | |
198 | if (!ret && len) { | |
bc2d4d83 DB |
199 | int jited = !!info.jited_prog_len; |
200 | ||
201 | print_string(PRINT_ANY, "tag", "tag %s ", | |
202 | hexstring_n2a(info.tag, sizeof(info.tag), | |
203 | tmp, sizeof(tmp))); | |
204 | print_uint(PRINT_JSON, "jited", NULL, jited); | |
205 | if (jited && !is_json_context()) | |
779525cd | 206 | fprintf(f, "jited "); |
a0b5b7cf | 207 | dump_ok = 1; |
779525cd DB |
208 | } |
209 | ||
210 | close(fd); | |
bc2d4d83 DB |
211 | out: |
212 | close_json_object(); | |
a0b5b7cf | 213 | return dump_ok; |
779525cd DB |
214 | } |
215 | ||
32e93fb7 DB |
216 | static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len, |
217 | char **bpf_string, bool *need_release, | |
218 | const char separator) | |
1d129d19 JP |
219 | { |
220 | char sp; | |
221 | ||
222 | if (from_file) { | |
223 | size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,"); | |
7c87c7fe | 224 | char *tmp_string, *pos, c_prev = ' '; |
1d129d19 | 225 | FILE *fp; |
7c87c7fe | 226 | int c; |
1d129d19 JP |
227 | |
228 | tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len; | |
3da3ebfc | 229 | tmp_string = pos = calloc(1, tmp_len); |
1d129d19 JP |
230 | if (tmp_string == NULL) |
231 | return -ENOMEM; | |
232 | ||
1d129d19 JP |
233 | fp = fopen(arg, "r"); |
234 | if (fp == NULL) { | |
235 | perror("Cannot fopen"); | |
236 | free(tmp_string); | |
237 | return -ENOENT; | |
238 | } | |
239 | ||
3da3ebfc PS |
240 | while ((c = fgetc(fp)) != EOF) { |
241 | switch (c) { | |
242 | case '\n': | |
243 | if (c_prev != ',') | |
244 | *(pos++) = ','; | |
7c87c7fe | 245 | c_prev = ','; |
3da3ebfc PS |
246 | break; |
247 | case ' ': | |
248 | case '\t': | |
249 | if (c_prev != ' ') | |
250 | *(pos++) = c; | |
7c87c7fe | 251 | c_prev = ' '; |
3da3ebfc PS |
252 | break; |
253 | default: | |
254 | *(pos++) = c; | |
7c87c7fe | 255 | c_prev = c; |
3da3ebfc PS |
256 | } |
257 | if (pos - tmp_string == tmp_len) | |
258 | break; | |
3da3ebfc PS |
259 | } |
260 | ||
261 | if (!feof(fp)) { | |
1d129d19 JP |
262 | free(tmp_string); |
263 | fclose(fp); | |
3da3ebfc | 264 | return -E2BIG; |
1d129d19 JP |
265 | } |
266 | ||
267 | fclose(fp); | |
3da3ebfc | 268 | *pos = 0; |
e4225669 | 269 | |
1d129d19 JP |
270 | *need_release = true; |
271 | *bpf_string = tmp_string; | |
272 | } else { | |
273 | *need_release = false; | |
274 | *bpf_string = arg; | |
275 | } | |
276 | ||
277 | if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 || | |
278 | sp != separator) { | |
279 | if (*need_release) | |
280 | free(*bpf_string); | |
281 | return -EINVAL; | |
282 | } | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
32e93fb7 DB |
287 | static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops, |
288 | bool from_file) | |
1d129d19 JP |
289 | { |
290 | char *bpf_string, *token, separator = ','; | |
291 | int ret = 0, i = 0; | |
292 | bool need_release; | |
293 | __u16 bpf_len = 0; | |
294 | ||
295 | if (argc < 1) | |
296 | return -EINVAL; | |
297 | if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string, | |
298 | &need_release, separator)) | |
299 | return -EINVAL; | |
300 | if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) { | |
301 | ret = -EINVAL; | |
302 | goto out; | |
303 | } | |
304 | ||
305 | token = bpf_string; | |
306 | while ((token = strchr(token, separator)) && (++token)[0]) { | |
307 | if (i >= bpf_len) { | |
32a121cb | 308 | fprintf(stderr, "Real program length exceeds encoded length parameter!\n"); |
1d129d19 JP |
309 | ret = -EINVAL; |
310 | goto out; | |
311 | } | |
312 | ||
313 | if (sscanf(token, "%hu %hhu %hhu %u,", | |
314 | &bpf_ops[i].code, &bpf_ops[i].jt, | |
315 | &bpf_ops[i].jf, &bpf_ops[i].k) != 4) { | |
316 | fprintf(stderr, "Error at instruction %d!\n", i); | |
317 | ret = -EINVAL; | |
318 | goto out; | |
319 | } | |
320 | ||
321 | i++; | |
322 | } | |
323 | ||
324 | if (i != bpf_len) { | |
afc1a200 | 325 | fprintf(stderr, "Parsed program length is less than encoded length parameter!\n"); |
1d129d19 JP |
326 | ret = -EINVAL; |
327 | goto out; | |
328 | } | |
329 | ret = bpf_len; | |
1d129d19 JP |
330 | out: |
331 | if (need_release) | |
332 | free(bpf_string); | |
333 | ||
334 | return ret; | |
335 | } | |
336 | ||
337 | void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len) | |
338 | { | |
d896797c | 339 | struct sock_filter *ops = RTA_DATA(bpf_ops); |
1d129d19 JP |
340 | int i; |
341 | ||
342 | if (len == 0) | |
343 | return; | |
344 | ||
345 | fprintf(f, "bytecode \'%u,", len); | |
346 | ||
347 | for (i = 0; i < len - 1; i++) | |
348 | fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt, | |
349 | ops[i].jf, ops[i].k); | |
350 | ||
6256f8c9 | 351 | fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt, |
1d129d19 JP |
352 | ops[i].jf, ops[i].k); |
353 | } | |
11c39b5e | 354 | |
afc1a200 DB |
355 | static void bpf_map_pin_report(const struct bpf_elf_map *pin, |
356 | const struct bpf_elf_map *obj) | |
357 | { | |
358 | fprintf(stderr, "Map specification differs from pinned file!\n"); | |
359 | ||
360 | if (obj->type != pin->type) | |
361 | fprintf(stderr, " - Type: %u (obj) != %u (pin)\n", | |
362 | obj->type, pin->type); | |
363 | if (obj->size_key != pin->size_key) | |
364 | fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n", | |
365 | obj->size_key, pin->size_key); | |
366 | if (obj->size_value != pin->size_value) | |
367 | fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n", | |
368 | obj->size_value, pin->size_value); | |
369 | if (obj->max_elem != pin->max_elem) | |
370 | fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n", | |
371 | obj->max_elem, pin->max_elem); | |
4dd3f50a DB |
372 | if (obj->flags != pin->flags) |
373 | fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n", | |
374 | obj->flags, pin->flags); | |
afc1a200 DB |
375 | |
376 | fprintf(stderr, "\n"); | |
377 | } | |
378 | ||
ecb05c0f DB |
379 | struct bpf_prog_data { |
380 | unsigned int type; | |
381 | unsigned int jited; | |
382 | }; | |
383 | ||
384 | struct bpf_map_ext { | |
385 | struct bpf_prog_data owner; | |
386 | }; | |
387 | ||
388 | static int bpf_derive_elf_map_from_fdinfo(int fd, struct bpf_elf_map *map, | |
389 | struct bpf_map_ext *ext) | |
9e607f2e | 390 | { |
ecb05c0f | 391 | unsigned int val, owner_type = 0, owner_jited = 0; |
9e607f2e | 392 | char file[PATH_MAX], buff[4096]; |
9e607f2e DB |
393 | FILE *fp; |
394 | ||
395 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
ecb05c0f | 396 | memset(map, 0, sizeof(*map)); |
9e607f2e DB |
397 | |
398 | fp = fopen(file, "r"); | |
399 | if (!fp) { | |
400 | fprintf(stderr, "No procfs support?!\n"); | |
401 | return -EIO; | |
402 | } | |
403 | ||
9e607f2e DB |
404 | while (fgets(buff, sizeof(buff), fp)) { |
405 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
ecb05c0f | 406 | map->type = val; |
9e607f2e | 407 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) |
ecb05c0f | 408 | map->size_key = val; |
9e607f2e | 409 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) |
ecb05c0f | 410 | map->size_value = val; |
9e607f2e | 411 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) |
ecb05c0f | 412 | map->max_elem = val; |
4dd3f50a | 413 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) |
ecb05c0f | 414 | map->flags = val; |
fb24802b DB |
415 | else if (sscanf(buff, "owner_prog_type:\t%i", &val) == 1) |
416 | owner_type = val; | |
ecb05c0f DB |
417 | else if (sscanf(buff, "owner_jited:\t%i", &val) == 1) |
418 | owner_jited = val; | |
9e607f2e DB |
419 | } |
420 | ||
421 | fclose(fp); | |
ecb05c0f DB |
422 | if (ext) { |
423 | memset(ext, 0, sizeof(*ext)); | |
424 | ext->owner.type = owner_type; | |
425 | ext->owner.jited = owner_jited; | |
426 | } | |
427 | ||
428 | return 0; | |
429 | } | |
430 | ||
431 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, | |
432 | struct bpf_map_ext *ext, int length, | |
433 | enum bpf_prog_type type) | |
434 | { | |
435 | struct bpf_elf_map tmp, zero = {}; | |
436 | int ret; | |
437 | ||
438 | ret = bpf_derive_elf_map_from_fdinfo(fd, &tmp, ext); | |
439 | if (ret < 0) | |
440 | return ret; | |
9e607f2e | 441 | |
fb24802b DB |
442 | /* The decision to reject this is on kernel side eventually, but |
443 | * at least give the user a chance to know what's wrong. | |
444 | */ | |
ecb05c0f | 445 | if (ext->owner.type && ext->owner.type != type) |
fb24802b | 446 | fprintf(stderr, "Program array map owner types differ: %u (obj) != %u (pin)\n", |
ecb05c0f | 447 | type, ext->owner.type); |
fb24802b | 448 | |
91d88eeb | 449 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
450 | return 0; |
451 | } else { | |
9e607f2e DB |
452 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, |
453 | * so just accept it. We know we do have an eBPF fd and in this | |
454 | * case, everything is 0. It is guaranteed that no such map exists | |
455 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
456 | */ | |
91d88eeb | 457 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
458 | return 0; |
459 | ||
afc1a200 | 460 | bpf_map_pin_report(&tmp, map); |
9e607f2e DB |
461 | return -EINVAL; |
462 | } | |
463 | } | |
464 | ||
91d88eeb DB |
465 | static int bpf_mnt_fs(const char *target) |
466 | { | |
467 | bool bind_done = false; | |
468 | ||
469 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
470 | if (errno != EINVAL || bind_done) { | |
471 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
472 | target, strerror(errno)); | |
473 | return -1; | |
474 | } | |
475 | ||
476 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
477 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
478 | target, target, strerror(errno)); | |
479 | return -1; | |
480 | } | |
481 | ||
482 | bind_done = true; | |
483 | } | |
484 | ||
e4225669 | 485 | if (mount("bpf", target, "bpf", 0, "mode=0700")) { |
91d88eeb DB |
486 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", |
487 | target, strerror(errno)); | |
488 | return -1; | |
489 | } | |
490 | ||
491 | return 0; | |
492 | } | |
493 | ||
95ae9a48 DB |
494 | static int bpf_mnt_check_target(const char *target) |
495 | { | |
496 | struct stat sb = {}; | |
497 | int ret; | |
498 | ||
499 | ret = stat(target, &sb); | |
500 | if (ret) { | |
501 | ret = mkdir(target, S_IRWXU); | |
502 | if (ret) { | |
503 | fprintf(stderr, "mkdir %s failed: %s\n", target, | |
504 | strerror(errno)); | |
505 | return ret; | |
506 | } | |
507 | } | |
508 | ||
509 | return 0; | |
510 | } | |
511 | ||
32e93fb7 DB |
512 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
513 | { | |
514 | struct statfs st_fs; | |
515 | ||
516 | if (statfs(mnt, &st_fs) < 0) | |
517 | return -ENOENT; | |
518 | if ((unsigned long)st_fs.f_type != magic) | |
519 | return -ENOENT; | |
520 | ||
521 | return 0; | |
522 | } | |
523 | ||
95ae9a48 DB |
524 | static const char *bpf_find_mntpt_single(unsigned long magic, char *mnt, |
525 | int len, const char *mntpt) | |
526 | { | |
527 | int ret; | |
528 | ||
529 | ret = bpf_valid_mntpt(mntpt, magic); | |
530 | if (!ret) { | |
18f156bf | 531 | strlcpy(mnt, mntpt, len); |
95ae9a48 DB |
532 | return mnt; |
533 | } | |
534 | ||
535 | return NULL; | |
536 | } | |
537 | ||
32e93fb7 DB |
538 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, |
539 | char *mnt, int len, | |
540 | const char * const *known_mnts) | |
541 | { | |
542 | const char * const *ptr; | |
543 | char type[100]; | |
544 | FILE *fp; | |
545 | ||
546 | if (known_mnts) { | |
547 | ptr = known_mnts; | |
548 | while (*ptr) { | |
95ae9a48 | 549 | if (bpf_find_mntpt_single(magic, mnt, len, *ptr)) |
32e93fb7 | 550 | return mnt; |
32e93fb7 DB |
551 | ptr++; |
552 | } | |
553 | } | |
554 | ||
c3724e4b PS |
555 | if (len != PATH_MAX) |
556 | return NULL; | |
557 | ||
32e93fb7 | 558 | fp = fopen("/proc/mounts", "r"); |
c3724e4b | 559 | if (fp == NULL) |
32e93fb7 DB |
560 | return NULL; |
561 | ||
562 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
563 | mnt, type) == 2) { | |
564 | if (strcmp(type, fstype) == 0) | |
565 | break; | |
566 | } | |
567 | ||
568 | fclose(fp); | |
569 | if (strcmp(type, fstype) != 0) | |
570 | return NULL; | |
571 | ||
572 | return mnt; | |
573 | } | |
574 | ||
575 | int bpf_trace_pipe(void) | |
576 | { | |
577 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
578 | static const char * const tracefs_known_mnts[] = { | |
579 | TRACE_DIR_MNT, | |
580 | "/sys/kernel/debug/tracing", | |
581 | "/tracing", | |
582 | "/trace", | |
583 | 0, | |
584 | }; | |
1b736dc4 | 585 | int fd_in, fd_out = STDERR_FILENO; |
32e93fb7 DB |
586 | char tpipe[PATH_MAX]; |
587 | const char *mnt; | |
32e93fb7 DB |
588 | |
589 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
590 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
591 | if (!mnt) { | |
592 | fprintf(stderr, "tracefs not mounted?\n"); | |
593 | return -1; | |
594 | } | |
595 | ||
596 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
597 | ||
1b736dc4 DB |
598 | fd_in = open(tpipe, O_RDONLY); |
599 | if (fd_in < 0) | |
32e93fb7 DB |
600 | return -1; |
601 | ||
602 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
603 | while (1) { | |
604 | static char buff[4096]; | |
605 | ssize_t ret; | |
606 | ||
1b736dc4 DB |
607 | ret = read(fd_in, buff, sizeof(buff)); |
608 | if (ret > 0 && write(fd_out, buff, ret) == ret) | |
609 | continue; | |
610 | break; | |
32e93fb7 DB |
611 | } |
612 | ||
1b736dc4 DB |
613 | close(fd_in); |
614 | return -1; | |
32e93fb7 DB |
615 | } |
616 | ||
e4225669 | 617 | static int bpf_gen_global(const char *bpf_sub_dir) |
91d88eeb | 618 | { |
e4225669 DB |
619 | char bpf_glo_dir[PATH_MAX]; |
620 | int ret; | |
621 | ||
622 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/", | |
623 | bpf_sub_dir, BPF_DIR_GLOBALS); | |
624 | ||
625 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
626 | if (ret && errno != EEXIST) { | |
627 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
628 | strerror(errno)); | |
629 | return ret; | |
630 | } | |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
635 | static int bpf_gen_master(const char *base, const char *name) | |
636 | { | |
637 | char bpf_sub_dir[PATH_MAX]; | |
638 | int ret; | |
639 | ||
640 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name); | |
641 | ||
642 | ret = mkdir(bpf_sub_dir, S_IRWXU); | |
643 | if (ret && errno != EEXIST) { | |
644 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir, | |
645 | strerror(errno)); | |
646 | return ret; | |
647 | } | |
648 | ||
649 | return bpf_gen_global(bpf_sub_dir); | |
650 | } | |
651 | ||
652 | static int bpf_slave_via_bind_mnt(const char *full_name, | |
653 | const char *full_link) | |
654 | { | |
655 | int ret; | |
656 | ||
657 | ret = mkdir(full_name, S_IRWXU); | |
658 | if (ret) { | |
659 | assert(errno != EEXIST); | |
660 | fprintf(stderr, "mkdir %s failed: %s\n", full_name, | |
661 | strerror(errno)); | |
662 | return ret; | |
663 | } | |
664 | ||
665 | ret = mount(full_link, full_name, "none", MS_BIND, NULL); | |
666 | if (ret) { | |
667 | rmdir(full_name); | |
668 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
669 | full_link, full_name, strerror(errno)); | |
670 | } | |
671 | ||
672 | return ret; | |
673 | } | |
674 | ||
675 | static int bpf_gen_slave(const char *base, const char *name, | |
676 | const char *link) | |
677 | { | |
678 | char bpf_lnk_dir[PATH_MAX]; | |
679 | char bpf_sub_dir[PATH_MAX]; | |
680 | struct stat sb = {}; | |
681 | int ret; | |
682 | ||
683 | snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link); | |
684 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name); | |
685 | ||
686 | ret = symlink(bpf_lnk_dir, bpf_sub_dir); | |
687 | if (ret) { | |
688 | if (errno != EEXIST) { | |
689 | if (errno != EPERM) { | |
690 | fprintf(stderr, "symlink %s failed: %s\n", | |
691 | bpf_sub_dir, strerror(errno)); | |
692 | return ret; | |
693 | } | |
694 | ||
695 | return bpf_slave_via_bind_mnt(bpf_sub_dir, | |
696 | bpf_lnk_dir); | |
697 | } | |
698 | ||
699 | ret = lstat(bpf_sub_dir, &sb); | |
700 | if (ret) { | |
701 | fprintf(stderr, "lstat %s failed: %s\n", | |
702 | bpf_sub_dir, strerror(errno)); | |
703 | return ret; | |
704 | } | |
705 | ||
706 | if ((sb.st_mode & S_IFMT) != S_IFLNK) | |
707 | return bpf_gen_global(bpf_sub_dir); | |
708 | } | |
709 | ||
710 | return 0; | |
711 | } | |
712 | ||
713 | static int bpf_gen_hierarchy(const char *base) | |
714 | { | |
715 | int ret, i; | |
716 | ||
717 | ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0])); | |
718 | for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++) | |
719 | ret = bpf_gen_slave(base, | |
720 | bpf_prog_to_subdir(__bpf_types[i]), | |
721 | bpf_prog_to_subdir(__bpf_types[0])); | |
722 | return ret; | |
723 | } | |
724 | ||
725 | static const char *bpf_get_work_dir(enum bpf_prog_type type) | |
726 | { | |
727 | static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT; | |
728 | static char bpf_wrk_dir[PATH_MAX]; | |
91d88eeb | 729 | static const char *mnt; |
e4225669 | 730 | static bool bpf_mnt_cached; |
95ae9a48 | 731 | const char *mnt_env = getenv(BPF_ENV_MNT); |
91d88eeb DB |
732 | static const char * const bpf_known_mnts[] = { |
733 | BPF_DIR_MNT, | |
e4225669 | 734 | "/bpf", |
91d88eeb DB |
735 | 0, |
736 | }; | |
91d88eeb DB |
737 | int ret; |
738 | ||
e4225669 DB |
739 | if (bpf_mnt_cached) { |
740 | const char *out = mnt; | |
741 | ||
51361a9f | 742 | if (out && type) { |
e4225669 DB |
743 | snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/", |
744 | out, bpf_prog_to_subdir(type)); | |
745 | out = bpf_tmp; | |
746 | } | |
747 | return out; | |
748 | } | |
91d88eeb | 749 | |
95ae9a48 DB |
750 | if (mnt_env) |
751 | mnt = bpf_find_mntpt_single(BPF_FS_MAGIC, bpf_tmp, | |
752 | sizeof(bpf_tmp), mnt_env); | |
753 | else | |
754 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp, | |
755 | sizeof(bpf_tmp), bpf_known_mnts); | |
91d88eeb | 756 | if (!mnt) { |
95ae9a48 DB |
757 | mnt = mnt_env ? : BPF_DIR_MNT; |
758 | ret = bpf_mnt_check_target(mnt); | |
759 | if (!ret) | |
760 | ret = bpf_mnt_fs(mnt); | |
91d88eeb DB |
761 | if (ret) { |
762 | mnt = NULL; | |
763 | goto out; | |
764 | } | |
765 | } | |
766 | ||
e4225669 | 767 | snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt); |
91d88eeb | 768 | |
e4225669 DB |
769 | ret = bpf_gen_hierarchy(bpf_wrk_dir); |
770 | if (ret) { | |
91d88eeb DB |
771 | mnt = NULL; |
772 | goto out; | |
773 | } | |
774 | ||
e4225669 | 775 | mnt = bpf_wrk_dir; |
91d88eeb DB |
776 | out: |
777 | bpf_mnt_cached = true; | |
91d88eeb DB |
778 | return mnt; |
779 | } | |
780 | ||
e4225669 | 781 | static int bpf_obj_get(const char *pathname, enum bpf_prog_type type) |
91d88eeb | 782 | { |
d17b136f | 783 | union bpf_attr attr = {}; |
91d88eeb DB |
784 | char tmp[PATH_MAX]; |
785 | ||
786 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
e4225669 | 787 | pathname[1] == ':' && bpf_get_work_dir(type)) { |
91d88eeb | 788 | snprintf(tmp, sizeof(tmp), "%s/%s", |
e4225669 | 789 | bpf_get_work_dir(type), pathname + 2); |
91d88eeb DB |
790 | pathname = tmp; |
791 | } | |
792 | ||
91d88eeb DB |
793 | attr.pathname = bpf_ptr_to_u64(pathname); |
794 | ||
795 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
796 | } | |
797 | ||
21856018 DB |
798 | static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type) |
799 | { | |
800 | int prog_fd = bpf_obj_get(pathname, type); | |
801 | ||
802 | if (prog_fd < 0) | |
803 | fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n", | |
804 | pathname, strerror(errno)); | |
805 | return prog_fd; | |
806 | } | |
807 | ||
3f0b9e62 | 808 | static int bpf_do_parse(struct bpf_cfg_in *cfg, const bool *opt_tbl) |
32e93fb7 | 809 | { |
32e93fb7 | 810 | const char *file, *section, *uds_name; |
32e93fb7 | 811 | bool verbose = false; |
e4225669 | 812 | int i, ret, argc; |
91d88eeb DB |
813 | char **argv; |
814 | ||
e4225669 DB |
815 | argv = cfg->argv; |
816 | argc = cfg->argc; | |
91d88eeb DB |
817 | |
818 | if (opt_tbl[CBPF_BYTECODE] && | |
819 | (matches(*argv, "bytecode") == 0 || | |
820 | strcmp(*argv, "bc") == 0)) { | |
f20ff2f1 | 821 | cfg->mode = CBPF_BYTECODE; |
91d88eeb DB |
822 | } else if (opt_tbl[CBPF_FILE] && |
823 | (matches(*argv, "bytecode-file") == 0 || | |
824 | strcmp(*argv, "bcf") == 0)) { | |
f20ff2f1 | 825 | cfg->mode = CBPF_FILE; |
91d88eeb DB |
826 | } else if (opt_tbl[EBPF_OBJECT] && |
827 | (matches(*argv, "object-file") == 0 || | |
828 | strcmp(*argv, "obj") == 0)) { | |
f20ff2f1 | 829 | cfg->mode = EBPF_OBJECT; |
91d88eeb DB |
830 | } else if (opt_tbl[EBPF_PINNED] && |
831 | (matches(*argv, "object-pinned") == 0 || | |
832 | matches(*argv, "pinned") == 0 || | |
833 | matches(*argv, "fd") == 0)) { | |
f20ff2f1 | 834 | cfg->mode = EBPF_PINNED; |
32e93fb7 DB |
835 | } else { |
836 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
837 | return -1; | |
838 | } | |
839 | ||
840 | NEXT_ARG(); | |
841 | file = section = uds_name = NULL; | |
f20ff2f1 | 842 | if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) { |
32e93fb7 DB |
843 | file = *argv; |
844 | NEXT_ARG_FWD(); | |
845 | ||
658cfebc | 846 | if (cfg->type == BPF_PROG_TYPE_UNSPEC) { |
91d88eeb DB |
847 | if (argc > 0 && matches(*argv, "type") == 0) { |
848 | NEXT_ARG(); | |
e4225669 DB |
849 | for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta); |
850 | i++) { | |
851 | if (!__bpf_prog_meta[i].type) | |
852 | continue; | |
853 | if (!matches(*argv, | |
854 | __bpf_prog_meta[i].type)) { | |
658cfebc | 855 | cfg->type = i; |
e4225669 DB |
856 | break; |
857 | } | |
858 | } | |
859 | ||
658cfebc | 860 | if (cfg->type == BPF_PROG_TYPE_UNSPEC) { |
91d88eeb DB |
861 | fprintf(stderr, "What type is \"%s\"?\n", |
862 | *argv); | |
863 | return -1; | |
864 | } | |
865 | NEXT_ARG_FWD(); | |
866 | } else { | |
658cfebc | 867 | cfg->type = BPF_PROG_TYPE_SCHED_CLS; |
91d88eeb DB |
868 | } |
869 | } | |
870 | ||
658cfebc | 871 | section = bpf_prog_to_default_section(cfg->type); |
32e93fb7 DB |
872 | if (argc > 0 && matches(*argv, "section") == 0) { |
873 | NEXT_ARG(); | |
874 | section = *argv; | |
875 | NEXT_ARG_FWD(); | |
876 | } | |
877 | ||
658cfebc | 878 | if (__bpf_prog_meta[cfg->type].may_uds_export) { |
e4225669 DB |
879 | uds_name = getenv(BPF_ENV_UDS); |
880 | if (argc > 0 && !uds_name && | |
881 | matches(*argv, "export") == 0) { | |
882 | NEXT_ARG(); | |
883 | uds_name = *argv; | |
884 | NEXT_ARG_FWD(); | |
885 | } | |
32e93fb7 DB |
886 | } |
887 | ||
888 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
889 | verbose = true; | |
890 | NEXT_ARG_FWD(); | |
891 | } | |
892 | ||
893 | PREV_ARG(); | |
894 | } | |
895 | ||
3f0b9e62 | 896 | if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) { |
51be7546 | 897 | ret = bpf_ops_parse(argc, argv, cfg->opcodes, |
f20ff2f1 | 898 | cfg->mode == CBPF_FILE); |
3f0b9e62 JK |
899 | cfg->n_opcodes = ret; |
900 | } else if (cfg->mode == EBPF_OBJECT) { | |
901 | ret = 0; /* program will be loaded by load stage */ | |
902 | } else if (cfg->mode == EBPF_PINNED) { | |
658cfebc | 903 | ret = bpf_obj_pinned(file, cfg->type); |
3f0b9e62 JK |
904 | cfg->prog_fd = ret; |
905 | } else { | |
32e93fb7 | 906 | return -1; |
3f0b9e62 | 907 | } |
32e93fb7 | 908 | |
e4225669 DB |
909 | cfg->object = file; |
910 | cfg->section = section; | |
911 | cfg->uds = uds_name; | |
912 | cfg->argc = argc; | |
913 | cfg->argv = argv; | |
3f0b9e62 | 914 | cfg->verbose = verbose; |
91d88eeb DB |
915 | |
916 | return ret; | |
917 | } | |
918 | ||
3f0b9e62 JK |
919 | static int bpf_do_load(struct bpf_cfg_in *cfg) |
920 | { | |
921 | if (cfg->mode == EBPF_OBJECT) { | |
922 | cfg->prog_fd = bpf_obj_open(cfg->object, cfg->type, | |
923 | cfg->section, cfg->verbose); | |
924 | return cfg->prog_fd; | |
925 | } | |
926 | return 0; | |
927 | } | |
928 | ||
658cfebc | 929 | static int bpf_parse_opt_tbl(struct bpf_cfg_in *cfg, |
e4225669 DB |
930 | const struct bpf_cfg_ops *ops, void *nl, |
931 | const bool *opt_tbl) | |
91d88eeb | 932 | { |
91d88eeb | 933 | char annotation[256]; |
91d88eeb DB |
934 | int ret; |
935 | ||
3f0b9e62 JK |
936 | ret = bpf_do_parse(cfg, opt_tbl); |
937 | if (ret < 0) | |
938 | return ret; | |
939 | ||
940 | ret = bpf_do_load(cfg); | |
91d88eeb DB |
941 | if (ret < 0) |
942 | return ret; | |
943 | ||
f20ff2f1 | 944 | if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) |
3f0b9e62 | 945 | ops->cbpf_cb(nl, cfg->opcodes, cfg->n_opcodes); |
f20ff2f1 | 946 | if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) { |
32e93fb7 | 947 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
f20ff2f1 | 948 | basename(cfg->object), cfg->mode == EBPF_PINNED ? |
e4225669 | 949 | "*fsobj" : cfg->section); |
3f0b9e62 | 950 | ops->ebpf_cb(nl, cfg->prog_fd, annotation); |
32e93fb7 DB |
951 | } |
952 | ||
91d88eeb DB |
953 | return 0; |
954 | } | |
32e93fb7 | 955 | |
399db839 JK |
956 | int bpf_parse_and_load_common(struct bpf_cfg_in *cfg, |
957 | const struct bpf_cfg_ops *ops, void *nl) | |
e4225669 DB |
958 | { |
959 | bool opt_tbl[BPF_MODE_MAX] = {}; | |
960 | ||
961 | if (ops->cbpf_cb) { | |
962 | opt_tbl[CBPF_BYTECODE] = true; | |
963 | opt_tbl[CBPF_FILE] = true; | |
964 | } | |
965 | ||
966 | if (ops->ebpf_cb) { | |
967 | opt_tbl[EBPF_OBJECT] = true; | |
968 | opt_tbl[EBPF_PINNED] = true; | |
969 | } | |
970 | ||
658cfebc | 971 | return bpf_parse_opt_tbl(cfg, ops, nl, opt_tbl); |
e4225669 DB |
972 | } |
973 | ||
91d88eeb DB |
974 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
975 | { | |
91d88eeb | 976 | const bool opt_tbl[BPF_MODE_MAX] = { |
91d88eeb DB |
977 | [EBPF_OBJECT] = true, |
978 | [EBPF_PINNED] = true, | |
979 | }; | |
980 | const struct bpf_elf_map test = { | |
981 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
982 | .size_key = sizeof(int), | |
983 | .size_value = sizeof(int), | |
984 | }; | |
e4225669 | 985 | struct bpf_cfg_in cfg = { |
658cfebc | 986 | .type = BPF_PROG_TYPE_UNSPEC, |
e4225669 DB |
987 | .argc = argc, |
988 | .argv = argv, | |
989 | }; | |
ecb05c0f | 990 | struct bpf_map_ext ext = {}; |
91d88eeb | 991 | int ret, prog_fd, map_fd; |
91d88eeb DB |
992 | uint32_t map_key; |
993 | ||
3f0b9e62 JK |
994 | ret = bpf_do_parse(&cfg, opt_tbl); |
995 | if (ret < 0) | |
996 | return ret; | |
997 | ||
998 | ret = bpf_do_load(&cfg); | |
999 | if (ret < 0) | |
1000 | return ret; | |
1001 | ||
1002 | prog_fd = cfg.prog_fd; | |
1003 | ||
91d88eeb DB |
1004 | if (key) { |
1005 | map_key = *key; | |
1006 | } else { | |
e4225669 | 1007 | ret = sscanf(cfg.section, "%*i/%i", &map_key); |
91d88eeb | 1008 | if (ret != 1) { |
32a121cb | 1009 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
1010 | ret = -EINVAL; |
1011 | goto out_prog; | |
1012 | } | |
1013 | } | |
32e93fb7 | 1014 | |
658cfebc | 1015 | map_fd = bpf_obj_get(map_path, cfg.type); |
91d88eeb DB |
1016 | if (map_fd < 0) { |
1017 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
1018 | map_path, strerror(errno)); | |
1019 | ret = map_fd; | |
1020 | goto out_prog; | |
1021 | } | |
1022 | ||
ecb05c0f | 1023 | ret = bpf_map_selfcheck_pinned(map_fd, &test, &ext, |
fb24802b | 1024 | offsetof(struct bpf_elf_map, max_elem), |
658cfebc | 1025 | cfg.type); |
91d88eeb DB |
1026 | if (ret < 0) { |
1027 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
1028 | goto out_map; | |
1029 | } | |
1030 | ||
1031 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
1032 | if (ret < 0) | |
1033 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
1034 | out_map: | |
1035 | close(map_fd); | |
1036 | out_prog: | |
1037 | close(prog_fd); | |
1038 | return ret; | |
32e93fb7 DB |
1039 | } |
1040 | ||
fc4ccce0 DA |
1041 | int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type) |
1042 | { | |
1043 | union bpf_attr attr = {}; | |
1044 | ||
1045 | attr.target_fd = target_fd; | |
1046 | attr.attach_bpf_fd = prog_fd; | |
1047 | attr.attach_type = type; | |
1048 | ||
1049 | return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
1050 | } | |
1051 | ||
1052 | int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type) | |
1053 | { | |
1054 | union bpf_attr attr = {}; | |
1055 | ||
1056 | attr.target_fd = target_fd; | |
1057 | attr.attach_type = type; | |
1058 | ||
1059 | return bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); | |
1060 | } | |
1061 | ||
869d889e DA |
1062 | int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, |
1063 | size_t size_insns, const char *license, char *log, | |
1064 | size_t size_log) | |
1065 | { | |
1066 | union bpf_attr attr = {}; | |
1067 | ||
1068 | attr.prog_type = type; | |
1069 | attr.insns = bpf_ptr_to_u64(insns); | |
1070 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); | |
1071 | attr.license = bpf_ptr_to_u64(license); | |
1072 | ||
1073 | if (size_log > 0) { | |
1074 | attr.log_buf = bpf_ptr_to_u64(log); | |
1075 | attr.log_size = size_log; | |
1076 | attr.log_level = 1; | |
1077 | } | |
1078 | ||
1079 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); | |
1080 | } | |
1081 | ||
6256f8c9 | 1082 | #ifdef HAVE_ELF |
32e93fb7 DB |
1083 | struct bpf_elf_prog { |
1084 | enum bpf_prog_type type; | |
1085 | const struct bpf_insn *insns; | |
1086 | size_t size; | |
1087 | const char *license; | |
1088 | }; | |
1089 | ||
f6793eec DB |
1090 | struct bpf_hash_entry { |
1091 | unsigned int pinning; | |
1092 | const char *subpath; | |
1093 | struct bpf_hash_entry *next; | |
1094 | }; | |
1095 | ||
ecb05c0f DB |
1096 | struct bpf_config { |
1097 | unsigned int jit_enabled; | |
1098 | }; | |
1099 | ||
32e93fb7 | 1100 | struct bpf_elf_ctx { |
ecb05c0f | 1101 | struct bpf_config cfg; |
32e93fb7 DB |
1102 | Elf *elf_fd; |
1103 | GElf_Ehdr elf_hdr; | |
1104 | Elf_Data *sym_tab; | |
1105 | Elf_Data *str_tab; | |
1106 | int obj_fd; | |
1107 | int map_fds[ELF_MAX_MAPS]; | |
1108 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
ecb05c0f | 1109 | struct bpf_map_ext maps_ext[ELF_MAX_MAPS]; |
32e93fb7 DB |
1110 | int sym_num; |
1111 | int map_num; | |
e4225669 | 1112 | int map_len; |
32e93fb7 DB |
1113 | bool *sec_done; |
1114 | int sec_maps; | |
1115 | char license[ELF_MAX_LICENSE_LEN]; | |
1116 | enum bpf_prog_type type; | |
1117 | bool verbose; | |
1118 | struct bpf_elf_st stat; | |
f6793eec | 1119 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
1120 | char *log; |
1121 | size_t log_size; | |
32e93fb7 DB |
1122 | }; |
1123 | ||
6256f8c9 | 1124 | struct bpf_elf_sec_data { |
32e93fb7 DB |
1125 | GElf_Shdr sec_hdr; |
1126 | Elf_Data *sec_data; | |
1127 | const char *sec_name; | |
6256f8c9 DB |
1128 | }; |
1129 | ||
1130 | struct bpf_map_data { | |
32e93fb7 DB |
1131 | int *fds; |
1132 | const char *obj; | |
1133 | struct bpf_elf_st *st; | |
1134 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
1135 | }; |
1136 | ||
f31645d1 DB |
1137 | static __check_format_string(2, 3) void |
1138 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
1139 | { |
1140 | va_list vl; | |
1141 | ||
1142 | va_start(vl, format); | |
1143 | vfprintf(stderr, format, vl); | |
1144 | va_end(vl); | |
1145 | ||
f31645d1 | 1146 | if (ctx->log && ctx->log[0]) { |
afc1a200 DB |
1147 | if (ctx->verbose) { |
1148 | fprintf(stderr, "%s\n", ctx->log); | |
1149 | } else { | |
1150 | unsigned int off = 0, len = strlen(ctx->log); | |
1151 | ||
1152 | if (len > BPF_MAX_LOG) { | |
1153 | off = len - BPF_MAX_LOG; | |
1154 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
1155 | off); | |
1156 | } | |
1157 | fprintf(stderr, "%s\n", ctx->log + off); | |
1158 | } | |
1159 | ||
f31645d1 DB |
1160 | memset(ctx->log, 0, ctx->log_size); |
1161 | } | |
1162 | } | |
1163 | ||
1164 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
1165 | { | |
0f74d0f3 | 1166 | const size_t log_max = UINT_MAX >> 8; |
f31645d1 | 1167 | size_t log_size = ctx->log_size; |
f6a54d72 | 1168 | char *ptr; |
f31645d1 DB |
1169 | |
1170 | if (!ctx->log) { | |
1171 | log_size = 65536; | |
0f74d0f3 | 1172 | } else if (log_size < log_max) { |
f31645d1 | 1173 | log_size <<= 1; |
0f74d0f3 TG |
1174 | if (log_size > log_max) |
1175 | log_size = log_max; | |
1176 | } else { | |
1177 | return -EINVAL; | |
d937a74b | 1178 | } |
f31645d1 DB |
1179 | |
1180 | ptr = realloc(ctx->log, log_size); | |
1181 | if (!ptr) | |
1182 | return -ENOMEM; | |
1183 | ||
f6a54d72 | 1184 | ptr[0] = 0; |
f31645d1 DB |
1185 | ctx->log = ptr; |
1186 | ctx->log_size = log_size; | |
1187 | ||
1188 | return 0; | |
11c39b5e DB |
1189 | } |
1190 | ||
4dd3f50a DB |
1191 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
1192 | uint32_t size_value, uint32_t max_elem, | |
612ff099 | 1193 | uint32_t flags, int inner_fd) |
11c39b5e | 1194 | { |
d17b136f | 1195 | union bpf_attr attr = {}; |
67584e3a | 1196 | |
67584e3a ND |
1197 | attr.map_type = type; |
1198 | attr.key_size = size_key; | |
612ff099 | 1199 | attr.value_size = inner_fd ? sizeof(int) : size_value; |
67584e3a | 1200 | attr.max_entries = max_elem; |
4dd3f50a | 1201 | attr.map_flags = flags; |
612ff099 | 1202 | attr.inner_map_fd = inner_fd; |
11c39b5e DB |
1203 | |
1204 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
1205 | } | |
11c39b5e | 1206 | |
32e93fb7 | 1207 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 1208 | { |
d17b136f | 1209 | union bpf_attr attr = {}; |
67584e3a | 1210 | |
67584e3a ND |
1211 | attr.pathname = bpf_ptr_to_u64(pathname); |
1212 | attr.bpf_fd = fd; | |
32e93fb7 DB |
1213 | |
1214 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
1215 | } | |
11c39b5e | 1216 | |
32e93fb7 DB |
1217 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
1218 | { | |
1219 | struct sockaddr_alg alg = { | |
1220 | .salg_family = AF_ALG, | |
1221 | .salg_type = "hash", | |
1222 | .salg_name = "sha1", | |
1223 | }; | |
1224 | int ret, cfd, ofd, ffd; | |
1225 | struct stat stbuff; | |
1226 | ssize_t size; | |
1227 | ||
1228 | if (!object || len != 20) | |
1229 | return -EINVAL; | |
1230 | ||
1231 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
1232 | if (cfd < 0) { | |
1233 | fprintf(stderr, "Cannot get AF_ALG socket: %s\n", | |
1234 | strerror(errno)); | |
1235 | return cfd; | |
1236 | } | |
1237 | ||
1238 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
1239 | if (ret < 0) { | |
1240 | fprintf(stderr, "Error binding socket: %s\n", strerror(errno)); | |
1241 | goto out_cfd; | |
1242 | } | |
1243 | ||
1244 | ofd = accept(cfd, NULL, 0); | |
1245 | if (ofd < 0) { | |
1246 | fprintf(stderr, "Error accepting socket: %s\n", | |
1247 | strerror(errno)); | |
1248 | ret = ofd; | |
1249 | goto out_cfd; | |
1250 | } | |
1251 | ||
1252 | ffd = open(object, O_RDONLY); | |
1253 | if (ffd < 0) { | |
1254 | fprintf(stderr, "Error opening object %s: %s\n", | |
1255 | object, strerror(errno)); | |
1256 | ret = ffd; | |
1257 | goto out_ofd; | |
1258 | } | |
1259 | ||
32a121cb | 1260 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
1261 | if (ret < 0) { |
1262 | fprintf(stderr, "Error doing fstat: %s\n", | |
1263 | strerror(errno)); | |
1264 | goto out_ffd; | |
d937a74b | 1265 | } |
11c39b5e | 1266 | |
32e93fb7 DB |
1267 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
1268 | if (size != stbuff.st_size) { | |
1269 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
1270 | size, stbuff.st_size, strerror(errno)); | |
1271 | ret = -1; | |
1272 | goto out_ffd; | |
1273 | } | |
1274 | ||
1275 | size = read(ofd, out, len); | |
1276 | if (size != len) { | |
1277 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
1278 | size, len, strerror(errno)); | |
1279 | ret = -1; | |
1280 | } else { | |
1281 | ret = 0; | |
1282 | } | |
1283 | out_ffd: | |
1284 | close(ffd); | |
1285 | out_ofd: | |
1286 | close(ofd); | |
1287 | out_cfd: | |
1288 | close(cfd); | |
1289 | return ret; | |
11c39b5e DB |
1290 | } |
1291 | ||
32e93fb7 | 1292 | static const char *bpf_get_obj_uid(const char *pathname) |
11c39b5e | 1293 | { |
32a121cb | 1294 | static bool bpf_uid_cached; |
32e93fb7 DB |
1295 | static char bpf_uid[64]; |
1296 | uint8_t tmp[20]; | |
1297 | int ret; | |
11c39b5e | 1298 | |
32e93fb7 DB |
1299 | if (bpf_uid_cached) |
1300 | goto done; | |
11c39b5e | 1301 | |
32e93fb7 DB |
1302 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); |
1303 | if (ret) { | |
1304 | fprintf(stderr, "Object hashing failed!\n"); | |
1305 | return NULL; | |
1306 | } | |
1307 | ||
1308 | hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid)); | |
1309 | bpf_uid_cached = true; | |
1310 | done: | |
1311 | return bpf_uid; | |
11c39b5e DB |
1312 | } |
1313 | ||
32e93fb7 DB |
1314 | static int bpf_init_env(const char *pathname) |
1315 | { | |
1316 | struct rlimit limit = { | |
1317 | .rlim_cur = RLIM_INFINITY, | |
1318 | .rlim_max = RLIM_INFINITY, | |
1319 | }; | |
1320 | ||
1321 | /* Don't bother in case we fail! */ | |
1322 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
1323 | ||
e4225669 | 1324 | if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) { |
32a121cb | 1325 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
32e93fb7 DB |
1326 | return 0; |
1327 | } | |
1328 | ||
1329 | if (!bpf_get_obj_uid(pathname)) | |
1330 | return -1; | |
1331 | ||
1332 | return 0; | |
6256f8c9 DB |
1333 | } |
1334 | ||
f6793eec DB |
1335 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
1336 | uint32_t pinning) | |
1337 | { | |
1338 | struct bpf_hash_entry *entry; | |
1339 | ||
1340 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1341 | while (entry && entry->pinning != pinning) | |
1342 | entry = entry->next; | |
1343 | ||
1344 | return entry ? entry->subpath : NULL; | |
1345 | } | |
1346 | ||
1347 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
1348 | uint32_t pinning) | |
11c39b5e | 1349 | { |
32e93fb7 DB |
1350 | switch (pinning) { |
1351 | case PIN_OBJECT_NS: | |
1352 | case PIN_GLOBAL_NS: | |
1353 | return false; | |
1354 | case PIN_NONE: | |
32e93fb7 | 1355 | return true; |
f6793eec DB |
1356 | default: |
1357 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
1358 | } |
1359 | } | |
1360 | ||
1361 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 1362 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
1363 | { |
1364 | switch (pinning) { | |
1365 | case PIN_OBJECT_NS: | |
e4225669 DB |
1366 | snprintf(pathname, len, "%s/%s/%s", |
1367 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1368 | bpf_get_obj_uid(NULL), name); |
1369 | break; | |
1370 | case PIN_GLOBAL_NS: | |
e4225669 DB |
1371 | snprintf(pathname, len, "%s/%s/%s", |
1372 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1373 | BPF_DIR_GLOBALS, name); |
1374 | break; | |
f6793eec | 1375 | default: |
e4225669 DB |
1376 | snprintf(pathname, len, "%s/../%s/%s", |
1377 | bpf_get_work_dir(ctx->type), | |
f6793eec DB |
1378 | bpf_custom_pinning(ctx, pinning), name); |
1379 | break; | |
32e93fb7 DB |
1380 | } |
1381 | } | |
1382 | ||
f6793eec DB |
1383 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1384 | uint32_t pinning) | |
32e93fb7 DB |
1385 | { |
1386 | char pathname[PATH_MAX]; | |
1387 | ||
e4225669 | 1388 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
32e93fb7 DB |
1389 | return 0; |
1390 | ||
f6793eec | 1391 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
e4225669 | 1392 | return bpf_obj_get(pathname, ctx->type); |
32e93fb7 DB |
1393 | } |
1394 | ||
e4225669 | 1395 | static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx) |
32e93fb7 | 1396 | { |
f6793eec | 1397 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1398 | int ret; |
1399 | ||
e4225669 | 1400 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type), |
f6793eec DB |
1401 | bpf_get_obj_uid(NULL)); |
1402 | ||
1403 | ret = mkdir(tmp, S_IRWXU); | |
1404 | if (ret && errno != EEXIST) { | |
1405 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1406 | return ret; | |
1407 | } | |
1408 | ||
1409 | return 0; | |
1410 | } | |
1411 | ||
e4225669 DB |
1412 | static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx, |
1413 | const char *todo) | |
f6793eec DB |
1414 | { |
1415 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1416 | int ret; | |
1417 | ||
e4225669 | 1418 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type)); |
f6793eec DB |
1419 | snprintf(rem, sizeof(rem), "%s/", todo); |
1420 | sub = strtok(rem, "/"); | |
32e93fb7 | 1421 | |
f6793eec DB |
1422 | while (sub) { |
1423 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1424 | return -EINVAL; | |
1425 | ||
1426 | strcat(tmp, sub); | |
1427 | strcat(tmp, "/"); | |
32e93fb7 | 1428 | |
f6793eec | 1429 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1430 | if (ret && errno != EEXIST) { |
f6793eec | 1431 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1432 | strerror(errno)); |
1433 | return ret; | |
1434 | } | |
f6793eec DB |
1435 | |
1436 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1437 | } |
1438 | ||
f6793eec DB |
1439 | return 0; |
1440 | } | |
1441 | ||
1442 | static int bpf_place_pinned(int fd, const char *name, | |
1443 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1444 | { | |
1445 | char pathname[PATH_MAX]; | |
1446 | const char *tmp; | |
1447 | int ret = 0; | |
1448 | ||
e4225669 | 1449 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
f6793eec DB |
1450 | return 0; |
1451 | ||
1452 | if (pinning == PIN_OBJECT_NS) | |
e4225669 | 1453 | ret = bpf_make_obj_path(ctx); |
f6793eec | 1454 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) |
e4225669 | 1455 | ret = bpf_make_custom_path(ctx, tmp); |
f6793eec DB |
1456 | if (ret < 0) |
1457 | return ret; | |
1458 | ||
1459 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1460 | return bpf_obj_pin(fd, pathname); |
1461 | } | |
1462 | ||
f31645d1 DB |
1463 | static void bpf_prog_report(int fd, const char *section, |
1464 | const struct bpf_elf_prog *prog, | |
1465 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1466 | { |
afc1a200 DB |
1467 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1468 | ||
1469 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1470 | fd < 0 ? "rejected: " : "loaded", |
1471 | fd < 0 ? strerror(errno) : "", | |
1472 | fd < 0 ? errno : fd); | |
1473 | ||
1474 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1475 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1476 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1477 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1478 | ||
1479 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1480 | } | |
32e93fb7 | 1481 | |
f31645d1 DB |
1482 | static int bpf_prog_attach(const char *section, |
1483 | const struct bpf_elf_prog *prog, | |
1484 | struct bpf_elf_ctx *ctx) | |
1485 | { | |
1486 | int tries = 0, fd; | |
1487 | retry: | |
32e93fb7 DB |
1488 | errno = 0; |
1489 | fd = bpf_prog_load(prog->type, prog->insns, prog->size, | |
f31645d1 DB |
1490 | prog->license, ctx->log, ctx->log_size); |
1491 | if (fd < 0 || ctx->verbose) { | |
1492 | /* The verifier log is pretty chatty, sometimes so chatty | |
1493 | * on larger programs, that we could fail to dump everything | |
1494 | * into our buffer. Still, try to give a debuggable error | |
1495 | * log for the user, so enlarge it and re-fail. | |
1496 | */ | |
1497 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
0f74d0f3 | 1498 | if (tries++ < 10 && !bpf_log_realloc(ctx)) |
f31645d1 DB |
1499 | goto retry; |
1500 | ||
32a121cb | 1501 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1502 | ctx->log_size, tries); |
1503 | return fd; | |
1504 | } | |
1505 | ||
1506 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1507 | } |
1508 | ||
1509 | return fd; | |
1510 | } | |
1511 | ||
f31645d1 DB |
1512 | static void bpf_map_report(int fd, const char *name, |
1513 | const struct bpf_elf_map *map, | |
612ff099 | 1514 | struct bpf_elf_ctx *ctx, int inner_fd) |
f31645d1 DB |
1515 | { |
1516 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1517 | fd < 0 ? "rejected: " : "loaded", | |
1518 | fd < 0 ? strerror(errno) : "", | |
1519 | fd < 0 ? errno : fd); | |
1520 | ||
1521 | fprintf(stderr, " - Type: %u\n", map->type); | |
1522 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1523 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1524 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
612ff099 DB |
1525 | fprintf(stderr, " - Size value: %u\n", |
1526 | inner_fd ? (int)sizeof(int) : map->size_value); | |
4dd3f50a DB |
1527 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1528 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1529 | } |
1530 | ||
612ff099 DB |
1531 | static int bpf_find_map_id(const struct bpf_elf_ctx *ctx, uint32_t id) |
1532 | { | |
1533 | int i; | |
1534 | ||
1535 | for (i = 0; i < ctx->map_num; i++) { | |
1536 | if (ctx->maps[i].id != id) | |
1537 | continue; | |
1538 | if (ctx->map_fds[i] < 0) | |
1539 | return -EINVAL; | |
1540 | ||
1541 | return ctx->map_fds[i]; | |
1542 | } | |
1543 | ||
1544 | return -ENOENT; | |
1545 | } | |
1546 | ||
0b5eadc5 | 1547 | static void bpf_report_map_in_map(int outer_fd, uint32_t idx) |
612ff099 DB |
1548 | { |
1549 | struct bpf_elf_map outer_map; | |
1550 | int ret; | |
1551 | ||
1552 | fprintf(stderr, "Cannot insert map into map! "); | |
1553 | ||
ecb05c0f | 1554 | ret = bpf_derive_elf_map_from_fdinfo(outer_fd, &outer_map, NULL); |
612ff099 DB |
1555 | if (!ret) { |
1556 | if (idx >= outer_map.max_elem && | |
1557 | outer_map.type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
1558 | fprintf(stderr, "Outer map has %u elements, index %u is invalid!\n", | |
1559 | outer_map.max_elem, idx); | |
1560 | return; | |
1561 | } | |
1562 | } | |
1563 | ||
1564 | fprintf(stderr, "Different map specs used for outer and inner map?\n"); | |
1565 | } | |
1566 | ||
1567 | static bool bpf_is_map_in_map_type(const struct bpf_elf_map *map) | |
1568 | { | |
1569 | return map->type == BPF_MAP_TYPE_ARRAY_OF_MAPS || | |
1570 | map->type == BPF_MAP_TYPE_HASH_OF_MAPS; | |
1571 | } | |
1572 | ||
ecb05c0f DB |
1573 | static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx, |
1574 | const struct bpf_elf_map *map, struct bpf_map_ext *ext, | |
1575 | int *have_map_in_map) | |
32e93fb7 | 1576 | { |
612ff099 | 1577 | int fd, ret, map_inner_fd = 0; |
32e93fb7 | 1578 | |
f6793eec | 1579 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1580 | if (fd > 0) { |
ecb05c0f | 1581 | ret = bpf_map_selfcheck_pinned(fd, map, ext, |
91d88eeb | 1582 | offsetof(struct bpf_elf_map, |
fb24802b | 1583 | id), ctx->type); |
9e607f2e DB |
1584 | if (ret < 0) { |
1585 | close(fd); | |
1586 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1587 | name); | |
1588 | return ret; | |
1589 | } | |
f31645d1 | 1590 | if (ctx->verbose) |
32e93fb7 DB |
1591 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1592 | name); | |
1593 | return fd; | |
1594 | } | |
1595 | ||
612ff099 DB |
1596 | if (have_map_in_map && bpf_is_map_in_map_type(map)) { |
1597 | (*have_map_in_map)++; | |
1598 | if (map->inner_id) | |
1599 | return 0; | |
1600 | fprintf(stderr, "Map \'%s\' cannot be created since no inner map ID defined!\n", | |
1601 | name); | |
1602 | return -EINVAL; | |
1603 | } | |
1604 | ||
1605 | if (!have_map_in_map && bpf_is_map_in_map_type(map)) { | |
1606 | map_inner_fd = bpf_find_map_id(ctx, map->inner_id); | |
1607 | if (map_inner_fd < 0) { | |
1608 | fprintf(stderr, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n", | |
1609 | name, map->inner_id); | |
1610 | return -EINVAL; | |
1611 | } | |
1612 | } | |
1613 | ||
32e93fb7 DB |
1614 | errno = 0; |
1615 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
612ff099 | 1616 | map->max_elem, map->flags, map_inner_fd); |
f31645d1 | 1617 | if (fd < 0 || ctx->verbose) { |
612ff099 | 1618 | bpf_map_report(fd, name, map, ctx, map_inner_fd); |
32e93fb7 DB |
1619 | if (fd < 0) |
1620 | return fd; | |
1621 | } | |
1622 | ||
f6793eec | 1623 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1624 | if (ret < 0 && errno != EEXIST) { |
1625 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1626 | strerror(errno)); | |
1627 | close(fd); | |
1628 | return ret; | |
1629 | } | |
1630 | ||
1631 | return fd; | |
1632 | } | |
1633 | ||
32e93fb7 DB |
1634 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1635 | const GElf_Sym *sym) | |
1636 | { | |
1637 | return ctx->str_tab->d_buf + sym->st_name; | |
1638 | } | |
1639 | ||
1640 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) | |
1641 | { | |
1642 | GElf_Sym sym; | |
11c39b5e DB |
1643 | int i; |
1644 | ||
32e93fb7 DB |
1645 | for (i = 0; i < ctx->sym_num; i++) { |
1646 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1647 | continue; | |
1648 | ||
5230a2ed DB |
1649 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
1650 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
32e93fb7 | 1651 | sym.st_shndx != ctx->sec_maps || |
e4225669 | 1652 | sym.st_value / ctx->map_len != which) |
32e93fb7 DB |
1653 | continue; |
1654 | ||
1655 | return bpf_str_tab_name(ctx, &sym); | |
11c39b5e | 1656 | } |
32e93fb7 DB |
1657 | |
1658 | return NULL; | |
11c39b5e DB |
1659 | } |
1660 | ||
32e93fb7 | 1661 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1662 | { |
612ff099 | 1663 | int i, j, ret, fd, inner_fd, inner_idx, have_map_in_map = 0; |
32e93fb7 | 1664 | const char *map_name; |
11c39b5e | 1665 | |
32e93fb7 DB |
1666 | for (i = 0; i < ctx->map_num; i++) { |
1667 | map_name = bpf_map_fetch_name(ctx, i); | |
1668 | if (!map_name) | |
1669 | return -EIO; | |
11c39b5e | 1670 | |
ecb05c0f DB |
1671 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1672 | &ctx->maps_ext[i], &have_map_in_map); | |
612ff099 DB |
1673 | if (fd < 0) |
1674 | return fd; | |
1675 | ||
1676 | ctx->map_fds[i] = !fd ? -1 : fd; | |
1677 | } | |
1678 | ||
1679 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { | |
1680 | if (ctx->map_fds[i] >= 0) | |
1681 | continue; | |
1682 | ||
1683 | map_name = bpf_map_fetch_name(ctx, i); | |
1684 | if (!map_name) | |
1685 | return -EIO; | |
1686 | ||
ecb05c0f DB |
1687 | fd = bpf_map_attach(map_name, ctx, &ctx->maps[i], |
1688 | &ctx->maps_ext[i], NULL); | |
32e93fb7 DB |
1689 | if (fd < 0) |
1690 | return fd; | |
11c39b5e | 1691 | |
32e93fb7 | 1692 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1693 | } |
1694 | ||
612ff099 DB |
1695 | for (i = 0; have_map_in_map && i < ctx->map_num; i++) { |
1696 | if (!ctx->maps[i].id || | |
1697 | ctx->maps[i].inner_id || | |
1698 | ctx->maps[i].inner_idx == -1) | |
1699 | continue; | |
1700 | ||
1701 | inner_fd = ctx->map_fds[i]; | |
1702 | inner_idx = ctx->maps[i].inner_idx; | |
1703 | ||
1704 | for (j = 0; j < ctx->map_num; j++) { | |
1705 | if (!bpf_is_map_in_map_type(&ctx->maps[j])) | |
1706 | continue; | |
1707 | if (ctx->maps[j].inner_id != ctx->maps[i].id) | |
1708 | continue; | |
1709 | ||
1710 | ret = bpf_map_update(ctx->map_fds[j], &inner_idx, | |
1711 | &inner_fd, BPF_ANY); | |
1712 | if (ret < 0) { | |
1713 | bpf_report_map_in_map(ctx->map_fds[j], | |
0b5eadc5 | 1714 | inner_idx); |
612ff099 DB |
1715 | return ret; |
1716 | } | |
1717 | } | |
1718 | } | |
1719 | ||
11c39b5e | 1720 | return 0; |
11c39b5e DB |
1721 | } |
1722 | ||
e4225669 DB |
1723 | static int bpf_map_num_sym(struct bpf_elf_ctx *ctx) |
1724 | { | |
1725 | int i, num = 0; | |
1726 | GElf_Sym sym; | |
1727 | ||
1728 | for (i = 0; i < ctx->sym_num; i++) { | |
1729 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1730 | continue; | |
1731 | ||
1732 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1733 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1734 | sym.st_shndx != ctx->sec_maps) | |
1735 | continue; | |
1736 | num++; | |
1737 | } | |
1738 | ||
1739 | return num; | |
1740 | } | |
1741 | ||
32e93fb7 DB |
1742 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1743 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1744 | { |
32e93fb7 | 1745 | Elf_Data *sec_edata; |
11c39b5e DB |
1746 | GElf_Shdr sec_hdr; |
1747 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1748 | char *sec_name; |
1749 | ||
32e93fb7 | 1750 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1751 | |
32e93fb7 | 1752 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1753 | if (!sec_fd) |
1754 | return -EINVAL; | |
11c39b5e DB |
1755 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1756 | return -EIO; | |
1757 | ||
32e93fb7 | 1758 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1759 | sec_hdr.sh_name); |
1760 | if (!sec_name || !sec_hdr.sh_size) | |
1761 | return -ENOENT; | |
1762 | ||
1763 | sec_edata = elf_getdata(sec_fd, NULL); | |
1764 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1765 | return -EIO; | |
1766 | ||
32e93fb7 | 1767 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1768 | |
32e93fb7 DB |
1769 | data->sec_name = sec_name; |
1770 | data->sec_data = sec_edata; | |
11c39b5e DB |
1771 | return 0; |
1772 | } | |
1773 | ||
e4225669 DB |
1774 | struct bpf_elf_map_min { |
1775 | __u32 type; | |
1776 | __u32 size_key; | |
1777 | __u32 size_value; | |
1778 | __u32 max_elem; | |
1779 | }; | |
11c39b5e | 1780 | |
e4225669 DB |
1781 | static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section, |
1782 | struct bpf_elf_sec_data *data) | |
1783 | { | |
1784 | ctx->map_num = data->sec_data->d_size; | |
32e93fb7 DB |
1785 | ctx->sec_maps = section; |
1786 | ctx->sec_done[section] = true; | |
11c39b5e | 1787 | |
e4225669 | 1788 | if (ctx->map_num > sizeof(ctx->maps)) { |
32e93fb7 DB |
1789 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); |
1790 | return -ENOMEM; | |
1791 | } | |
11c39b5e | 1792 | |
e4225669 DB |
1793 | memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num); |
1794 | return 0; | |
1795 | } | |
1796 | ||
1797 | static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end) | |
1798 | { | |
1799 | GElf_Sym sym; | |
1800 | int off, i; | |
1801 | ||
1802 | for (off = 0; off < end; off += ctx->map_len) { | |
1803 | /* Order doesn't need to be linear here, hence we walk | |
1804 | * the table again. | |
1805 | */ | |
1806 | for (i = 0; i < ctx->sym_num; i++) { | |
1807 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1808 | continue; | |
1809 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1810 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1811 | sym.st_shndx != ctx->sec_maps) | |
1812 | continue; | |
1813 | if (sym.st_value == off) | |
1814 | break; | |
1815 | if (i == ctx->sym_num - 1) | |
1816 | return -1; | |
1817 | } | |
1818 | } | |
1819 | ||
1820 | return off == end ? 0 : -1; | |
1821 | } | |
1822 | ||
1823 | static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx) | |
1824 | { | |
1825 | struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {}; | |
1826 | int i, sym_num = bpf_map_num_sym(ctx); | |
1827 | __u8 *buff; | |
1828 | ||
1829 | if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) { | |
1830 | fprintf(stderr, "%u maps not supported in current map section!\n", | |
1831 | sym_num); | |
1832 | return -EINVAL; | |
1833 | } | |
1834 | ||
1835 | if (ctx->map_num % sym_num != 0 || | |
1836 | ctx->map_num % sizeof(__u32) != 0) { | |
1837 | fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n"); | |
1838 | return -EINVAL; | |
1839 | } | |
1840 | ||
1841 | ctx->map_len = ctx->map_num / sym_num; | |
1842 | if (bpf_map_verify_all_offs(ctx, ctx->map_num)) { | |
1843 | fprintf(stderr, "Different struct bpf_elf_map in use!\n"); | |
1844 | return -EINVAL; | |
1845 | } | |
1846 | ||
1847 | if (ctx->map_len == sizeof(struct bpf_elf_map)) { | |
1848 | ctx->map_num = sym_num; | |
1849 | return 0; | |
1850 | } else if (ctx->map_len > sizeof(struct bpf_elf_map)) { | |
1851 | fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n"); | |
1852 | return -EINVAL; | |
1853 | } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) { | |
1854 | fprintf(stderr, "struct bpf_elf_map too small, not supported!\n"); | |
1855 | return -EINVAL; | |
1856 | } | |
1857 | ||
1858 | ctx->map_num = sym_num; | |
1859 | for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num; | |
1860 | i++, buff += ctx->map_len) { | |
1861 | /* The fixup leaves the rest of the members as zero, which | |
1862 | * is fine currently, but option exist to set some other | |
1863 | * default value as well when needed in future. | |
1864 | */ | |
1865 | memcpy(&fixup[i], buff, ctx->map_len); | |
1866 | } | |
1867 | ||
1868 | memcpy(ctx->maps, fixup, sizeof(fixup)); | |
1869 | ||
1870 | printf("Note: %zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n", | |
1871 | sizeof(struct bpf_elf_map) - ctx->map_len); | |
32e93fb7 DB |
1872 | return 0; |
1873 | } | |
11c39b5e | 1874 | |
32e93fb7 DB |
1875 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
1876 | struct bpf_elf_sec_data *data) | |
1877 | { | |
1878 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
1879 | return -ENOMEM; | |
11c39b5e | 1880 | |
32e93fb7 DB |
1881 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
1882 | ctx->sec_done[section] = true; | |
1883 | return 0; | |
1884 | } | |
11c39b5e | 1885 | |
32e93fb7 DB |
1886 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
1887 | struct bpf_elf_sec_data *data) | |
1888 | { | |
1889 | ctx->sym_tab = data->sec_data; | |
1890 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
1891 | ctx->sec_done[section] = true; | |
11c39b5e DB |
1892 | return 0; |
1893 | } | |
1894 | ||
32e93fb7 DB |
1895 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
1896 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1897 | { |
32e93fb7 DB |
1898 | ctx->str_tab = data->sec_data; |
1899 | ctx->sec_done[section] = true; | |
1900 | return 0; | |
1901 | } | |
11c39b5e | 1902 | |
afc1a200 DB |
1903 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
1904 | { | |
1905 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
1906 | } | |
1907 | ||
32e93fb7 DB |
1908 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx) |
1909 | { | |
1910 | struct bpf_elf_sec_data data; | |
1911 | int i, ret = -1; | |
11c39b5e | 1912 | |
32e93fb7 DB |
1913 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1914 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
1915 | if (ret < 0) |
1916 | continue; | |
1917 | ||
cce3d466 DB |
1918 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1919 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
e4225669 | 1920 | ret = bpf_fetch_maps_begin(ctx, i, &data); |
cce3d466 DB |
1921 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1922 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 1923 | ret = bpf_fetch_license(ctx, i, &data); |
cce3d466 DB |
1924 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
1925 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
1926 | ret = bpf_fetch_symtab(ctx, i, &data); |
1927 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 1928 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 DB |
1929 | ret = bpf_fetch_strtab(ctx, i, &data); |
1930 | if (ret < 0) { | |
afc1a200 | 1931 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 1932 | i); |
e4225669 | 1933 | return ret; |
11c39b5e | 1934 | } |
32e93fb7 DB |
1935 | } |
1936 | ||
afc1a200 | 1937 | if (bpf_has_map_data(ctx)) { |
e4225669 DB |
1938 | ret = bpf_fetch_maps_end(ctx); |
1939 | if (ret < 0) { | |
1940 | fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n"); | |
1941 | return ret; | |
1942 | } | |
1943 | ||
32e93fb7 DB |
1944 | ret = bpf_maps_attach_all(ctx); |
1945 | if (ret < 0) { | |
1946 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
1947 | return ret; | |
11c39b5e DB |
1948 | } |
1949 | } | |
1950 | ||
1951 | return ret; | |
1952 | } | |
1953 | ||
e4225669 DB |
1954 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section, |
1955 | bool *sseen) | |
11c39b5e | 1956 | { |
32e93fb7 DB |
1957 | struct bpf_elf_sec_data data; |
1958 | struct bpf_elf_prog prog; | |
1959 | int ret, i, fd = -1; | |
11c39b5e | 1960 | |
32e93fb7 DB |
1961 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1962 | if (ctx->sec_done[i]) | |
11c39b5e DB |
1963 | continue; |
1964 | ||
32e93fb7 | 1965 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
1966 | if (ret < 0 || |
1967 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
1968 | data.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1969 | !strcmp(data.sec_name, section))) | |
11c39b5e DB |
1970 | continue; |
1971 | ||
e4225669 DB |
1972 | *sseen = true; |
1973 | ||
32e93fb7 DB |
1974 | memset(&prog, 0, sizeof(prog)); |
1975 | prog.type = ctx->type; | |
1976 | prog.insns = data.sec_data->d_buf; | |
1977 | prog.size = data.sec_data->d_size; | |
1978 | prog.license = ctx->license; | |
11c39b5e | 1979 | |
f31645d1 | 1980 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1981 | if (fd < 0) |
e4225669 | 1982 | return fd; |
11c39b5e | 1983 | |
32e93fb7 | 1984 | ctx->sec_done[i] = true; |
11c39b5e DB |
1985 | break; |
1986 | } | |
1987 | ||
32e93fb7 | 1988 | return fd; |
11c39b5e DB |
1989 | } |
1990 | ||
ecb05c0f DB |
1991 | struct bpf_tail_call_props { |
1992 | unsigned int total; | |
1993 | unsigned int jited; | |
1994 | }; | |
1995 | ||
32e93fb7 DB |
1996 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
1997 | struct bpf_elf_sec_data *data_relo, | |
ecb05c0f DB |
1998 | struct bpf_elf_sec_data *data_insn, |
1999 | struct bpf_tail_call_props *props) | |
11c39b5e | 2000 | { |
32e93fb7 DB |
2001 | Elf_Data *idata = data_insn->sec_data; |
2002 | GElf_Shdr *rhdr = &data_relo->sec_hdr; | |
2003 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
2004 | struct bpf_insn *insns = idata->d_buf; | |
2005 | unsigned int num_insns = idata->d_size / sizeof(*insns); | |
11c39b5e | 2006 | |
32e93fb7 DB |
2007 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
2008 | unsigned int ioff, rmap; | |
2009 | GElf_Rel relo; | |
2010 | GElf_Sym sym; | |
2011 | ||
2012 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
2013 | return -EIO; | |
2014 | ||
2015 | ioff = relo.r_offset / sizeof(struct bpf_insn); | |
2016 | if (ioff >= num_insns || | |
a576c6b9 | 2017 | insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) { |
32a121cb | 2018 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", |
a576c6b9 DB |
2019 | ioff); |
2020 | if (ioff < num_insns && | |
2021 | insns[ioff].code == (BPF_JMP | BPF_CALL)) | |
32a121cb | 2022 | fprintf(stderr, " - Try to annotate functions with always_inline attribute!\n"); |
32e93fb7 | 2023 | return -EINVAL; |
a576c6b9 | 2024 | } |
32e93fb7 DB |
2025 | |
2026 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) | |
2027 | return -EIO; | |
2486337a | 2028 | if (sym.st_shndx != ctx->sec_maps) { |
32a121cb | 2029 | fprintf(stderr, "ELF contains non-map related relo data in entry %u pointing to section %u! Compiler bug?!\n", |
2486337a DB |
2030 | relo_ent, sym.st_shndx); |
2031 | return -EIO; | |
2032 | } | |
32e93fb7 | 2033 | |
e4225669 | 2034 | rmap = sym.st_value / ctx->map_len; |
32e93fb7 DB |
2035 | if (rmap >= ARRAY_SIZE(ctx->map_fds)) |
2036 | return -EINVAL; | |
2037 | if (!ctx->map_fds[rmap]) | |
2038 | return -EINVAL; | |
ecb05c0f DB |
2039 | if (ctx->maps[rmap].type == BPF_MAP_TYPE_PROG_ARRAY) { |
2040 | props->total++; | |
2041 | if (ctx->maps_ext[rmap].owner.jited || | |
2042 | (ctx->maps_ext[rmap].owner.type == 0 && | |
2043 | ctx->cfg.jit_enabled)) | |
2044 | props->jited++; | |
2045 | } | |
32e93fb7 DB |
2046 | |
2047 | if (ctx->verbose) | |
32a121cb | 2048 | fprintf(stderr, "Map \'%s\' (%d) injected into prog section \'%s\' at offset %u!\n", |
32e93fb7 DB |
2049 | bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap], |
2050 | data_insn->sec_name, ioff); | |
11c39b5e | 2051 | |
32e93fb7 DB |
2052 | insns[ioff].src_reg = BPF_PSEUDO_MAP_FD; |
2053 | insns[ioff].imm = ctx->map_fds[rmap]; | |
2054 | } | |
2055 | ||
2056 | return 0; | |
2057 | } | |
2058 | ||
afc1a200 | 2059 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
e4225669 | 2060 | bool *lderr, bool *sseen) |
32e93fb7 DB |
2061 | { |
2062 | struct bpf_elf_sec_data data_relo, data_insn; | |
2063 | struct bpf_elf_prog prog; | |
2064 | int ret, idx, i, fd = -1; | |
2065 | ||
2066 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
ecb05c0f DB |
2067 | struct bpf_tail_call_props props = {}; |
2068 | ||
32e93fb7 DB |
2069 | ret = bpf_fill_section_data(ctx, i, &data_relo); |
2070 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
2071 | continue; |
2072 | ||
32e93fb7 | 2073 | idx = data_relo.sec_hdr.sh_info; |
e4225669 | 2074 | |
32e93fb7 | 2075 | ret = bpf_fill_section_data(ctx, idx, &data_insn); |
cce3d466 DB |
2076 | if (ret < 0 || |
2077 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
2078 | data_insn.sec_hdr.sh_flags & SHF_EXECINSTR && | |
2079 | !strcmp(data_insn.sec_name, section))) | |
11c39b5e | 2080 | continue; |
32e93fb7 | 2081 | |
e4225669 DB |
2082 | *sseen = true; |
2083 | ||
ecb05c0f | 2084 | ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn, &props); |
c9c3720d DB |
2085 | if (ret < 0) { |
2086 | *lderr = true; | |
e4225669 | 2087 | return ret; |
c9c3720d | 2088 | } |
11c39b5e | 2089 | |
32e93fb7 DB |
2090 | memset(&prog, 0, sizeof(prog)); |
2091 | prog.type = ctx->type; | |
2092 | prog.insns = data_insn.sec_data->d_buf; | |
2093 | prog.size = data_insn.sec_data->d_size; | |
2094 | prog.license = ctx->license; | |
2095 | ||
f31645d1 | 2096 | fd = bpf_prog_attach(section, &prog, ctx); |
afc1a200 DB |
2097 | if (fd < 0) { |
2098 | *lderr = true; | |
ecb05c0f DB |
2099 | if (props.total) { |
2100 | if (ctx->cfg.jit_enabled && | |
2101 | props.total != props.jited) | |
2102 | fprintf(stderr, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n", | |
2103 | props.jited, props.total); | |
2104 | if (!ctx->cfg.jit_enabled && | |
2105 | props.jited) | |
2106 | fprintf(stderr, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n", | |
2107 | props.jited, props.total); | |
2108 | } | |
e4225669 | 2109 | return fd; |
afc1a200 | 2110 | } |
11c39b5e | 2111 | |
32e93fb7 DB |
2112 | ctx->sec_done[i] = true; |
2113 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
2114 | break; |
2115 | } | |
2116 | ||
32e93fb7 | 2117 | return fd; |
11c39b5e DB |
2118 | } |
2119 | ||
32e93fb7 | 2120 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 2121 | { |
e4225669 | 2122 | bool lderr = false, sseen = false; |
473d7840 DB |
2123 | int ret = -1; |
2124 | ||
afc1a200 | 2125 | if (bpf_has_map_data(ctx)) |
e4225669 | 2126 | ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen); |
afc1a200 | 2127 | if (ret < 0 && !lderr) |
e4225669 DB |
2128 | ret = bpf_fetch_prog(ctx, section, &sseen); |
2129 | if (ret < 0 && !sseen) | |
2130 | fprintf(stderr, "Program section \'%s\' not found in ELF file!\n", | |
2131 | section); | |
473d7840 DB |
2132 | return ret; |
2133 | } | |
2134 | ||
910b543d DB |
2135 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
2136 | { | |
2137 | int i; | |
2138 | ||
2139 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
2140 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
2141 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
2142 | return i; | |
2143 | return -1; | |
2144 | } | |
2145 | ||
ecb05c0f DB |
2146 | struct bpf_jited_aux { |
2147 | int prog_fd; | |
2148 | int map_fd; | |
2149 | struct bpf_prog_data prog; | |
2150 | struct bpf_map_ext map; | |
2151 | }; | |
2152 | ||
2153 | static int bpf_derive_prog_from_fdinfo(int fd, struct bpf_prog_data *prog) | |
2154 | { | |
2155 | char file[PATH_MAX], buff[4096]; | |
2156 | unsigned int val; | |
2157 | FILE *fp; | |
2158 | ||
2159 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
2160 | memset(prog, 0, sizeof(*prog)); | |
2161 | ||
2162 | fp = fopen(file, "r"); | |
2163 | if (!fp) { | |
2164 | fprintf(stderr, "No procfs support?!\n"); | |
2165 | return -EIO; | |
2166 | } | |
2167 | ||
2168 | while (fgets(buff, sizeof(buff), fp)) { | |
2169 | if (sscanf(buff, "prog_type:\t%u", &val) == 1) | |
2170 | prog->type = val; | |
2171 | else if (sscanf(buff, "prog_jited:\t%u", &val) == 1) | |
2172 | prog->jited = val; | |
2173 | } | |
2174 | ||
2175 | fclose(fp); | |
2176 | return 0; | |
2177 | } | |
2178 | ||
2179 | static int bpf_tail_call_get_aux(struct bpf_jited_aux *aux) | |
2180 | { | |
2181 | struct bpf_elf_map tmp; | |
2182 | int ret; | |
2183 | ||
2184 | ret = bpf_derive_elf_map_from_fdinfo(aux->map_fd, &tmp, &aux->map); | |
2185 | if (!ret) | |
2186 | ret = bpf_derive_prog_from_fdinfo(aux->prog_fd, &aux->prog); | |
2187 | ||
2188 | return ret; | |
2189 | } | |
2190 | ||
32e93fb7 | 2191 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 2192 | { |
32e93fb7 DB |
2193 | struct bpf_elf_sec_data data; |
2194 | uint32_t map_id, key_id; | |
910b543d | 2195 | int fd, i, ret, idx; |
473d7840 | 2196 | |
32e93fb7 DB |
2197 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
2198 | if (ctx->sec_done[i]) | |
473d7840 DB |
2199 | continue; |
2200 | ||
32e93fb7 | 2201 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
2202 | if (ret < 0) |
2203 | continue; | |
2204 | ||
910b543d DB |
2205 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
2206 | if (ret != 2) | |
32e93fb7 | 2207 | continue; |
910b543d DB |
2208 | |
2209 | idx = bpf_find_map_by_id(ctx, map_id); | |
2210 | if (idx < 0) | |
473d7840 DB |
2211 | continue; |
2212 | ||
32e93fb7 DB |
2213 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
2214 | if (fd < 0) | |
473d7840 DB |
2215 | return -EIO; |
2216 | ||
910b543d DB |
2217 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
2218 | &fd, BPF_ANY); | |
afc1a200 | 2219 | if (ret < 0) { |
ecb05c0f DB |
2220 | struct bpf_jited_aux aux = {}; |
2221 | ||
2222 | ret = -errno; | |
2223 | if (errno == E2BIG) { | |
afc1a200 DB |
2224 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", |
2225 | key_id, map_id); | |
ecb05c0f DB |
2226 | return ret; |
2227 | } | |
2228 | ||
2229 | aux.map_fd = ctx->map_fds[idx]; | |
2230 | aux.prog_fd = fd; | |
2231 | ||
2232 | if (bpf_tail_call_get_aux(&aux)) | |
2233 | return ret; | |
2234 | if (!aux.map.owner.type) | |
2235 | return ret; | |
2236 | ||
2237 | if (aux.prog.type != aux.map.owner.type) | |
2238 | fprintf(stderr, "Tail call map owned by prog type %u, but prog type is %u!\n", | |
2239 | aux.map.owner.type, aux.prog.type); | |
2240 | if (aux.prog.jited != aux.map.owner.jited) | |
2241 | fprintf(stderr, "Tail call map %s jited, but prog %s!\n", | |
2242 | aux.map.owner.jited ? "is" : "not", | |
2243 | aux.prog.jited ? "is" : "not"); | |
2244 | return ret; | |
afc1a200 | 2245 | } |
473d7840 | 2246 | |
32e93fb7 | 2247 | ctx->sec_done[i] = true; |
473d7840 DB |
2248 | } |
2249 | ||
2250 | return 0; | |
2251 | } | |
2252 | ||
32e93fb7 | 2253 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 2254 | { |
32e93fb7 DB |
2255 | struct stat st; |
2256 | int ret; | |
11c39b5e | 2257 | |
32e93fb7 | 2258 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 2259 | |
32e93fb7 DB |
2260 | ret = fstat(ctx->obj_fd, &st); |
2261 | if (ret < 0) { | |
2262 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
2263 | strerror(errno)); | |
2264 | return; | |
2265 | } | |
11c39b5e | 2266 | |
32e93fb7 DB |
2267 | ctx->stat.st_dev = st.st_dev; |
2268 | ctx->stat.st_ino = st.st_ino; | |
2269 | } | |
2270 | ||
f6793eec DB |
2271 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
2272 | { | |
2273 | char buff[PATH_MAX]; | |
2274 | ||
2275 | while (fgets(buff, sizeof(buff), fp)) { | |
2276 | char *ptr = buff; | |
2277 | ||
2278 | while (*ptr == ' ' || *ptr == '\t') | |
2279 | ptr++; | |
2280 | ||
2281 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
2282 | continue; | |
2283 | ||
2284 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
2285 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
2286 | strcpy(path, ptr); | |
2287 | return -1; | |
2288 | } | |
2289 | ||
2290 | return 1; | |
2291 | } | |
2292 | ||
2293 | return 0; | |
2294 | } | |
2295 | ||
2296 | static bool bpf_pinning_reserved(uint32_t pinning) | |
2297 | { | |
2298 | switch (pinning) { | |
2299 | case PIN_NONE: | |
2300 | case PIN_OBJECT_NS: | |
2301 | case PIN_GLOBAL_NS: | |
2302 | return true; | |
2303 | default: | |
2304 | return false; | |
2305 | } | |
2306 | } | |
2307 | ||
2308 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
2309 | { | |
2310 | struct bpf_hash_entry *entry; | |
d17b136f | 2311 | char subpath[PATH_MAX] = {}; |
f6793eec DB |
2312 | uint32_t pinning; |
2313 | FILE *fp; | |
2314 | int ret; | |
2315 | ||
2316 | fp = fopen(db_file, "r"); | |
2317 | if (!fp) | |
2318 | return; | |
2319 | ||
f6793eec DB |
2320 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { |
2321 | if (ret == -1) { | |
2322 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
2323 | db_file, subpath); | |
2324 | fclose(fp); | |
2325 | return; | |
2326 | } | |
2327 | ||
2328 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
2329 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
2330 | db_file, pinning); | |
f6793eec DB |
2331 | continue; |
2332 | } | |
2333 | ||
2334 | entry = malloc(sizeof(*entry)); | |
2335 | if (!entry) { | |
2336 | fprintf(stderr, "No memory left for db entry!\n"); | |
2337 | continue; | |
2338 | } | |
2339 | ||
2340 | entry->pinning = pinning; | |
2341 | entry->subpath = strdup(subpath); | |
2342 | if (!entry->subpath) { | |
2343 | fprintf(stderr, "No memory left for db entry!\n"); | |
2344 | free(entry); | |
2345 | continue; | |
2346 | } | |
2347 | ||
2348 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
2349 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
2350 | } | |
2351 | ||
2352 | fclose(fp); | |
2353 | } | |
2354 | ||
2355 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
2356 | { | |
2357 | struct bpf_hash_entry *entry; | |
2358 | int i; | |
2359 | ||
2360 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
2361 | while ((entry = ctx->ht[i]) != NULL) { | |
2362 | ctx->ht[i] = entry->next; | |
2363 | free((char *)entry->subpath); | |
2364 | free(entry); | |
2365 | } | |
2366 | } | |
2367 | } | |
2368 | ||
8187b012 DB |
2369 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
2370 | { | |
2371 | if (ctx->elf_hdr.e_type != ET_REL || | |
e77fa41d DB |
2372 | (ctx->elf_hdr.e_machine != EM_NONE && |
2373 | ctx->elf_hdr.e_machine != EM_BPF) || | |
8187b012 DB |
2374 | ctx->elf_hdr.e_version != EV_CURRENT) { |
2375 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
2376 | return -EINVAL; | |
2377 | } | |
2378 | ||
2379 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
2380 | default: | |
2381 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
2382 | return -EINVAL; | |
2383 | case ELFDATA2LSB: | |
2384 | if (htons(1) == 1) { | |
2385 | fprintf(stderr, | |
2386 | "We are big endian, eBPF object is little endian!\n"); | |
2387 | return -EIO; | |
2388 | } | |
2389 | break; | |
2390 | case ELFDATA2MSB: | |
2391 | if (htons(1) != 1) { | |
2392 | fprintf(stderr, | |
2393 | "We are little endian, eBPF object is big endian!\n"); | |
2394 | return -EIO; | |
2395 | } | |
2396 | break; | |
2397 | } | |
2398 | ||
2399 | return 0; | |
2400 | } | |
2401 | ||
ecb05c0f DB |
2402 | static void bpf_get_cfg(struct bpf_elf_ctx *ctx) |
2403 | { | |
2404 | static const char *path_jit = "/proc/sys/net/core/bpf_jit_enable"; | |
2405 | int fd; | |
2406 | ||
2407 | fd = open(path_jit, O_RDONLY); | |
2408 | if (fd > 0) { | |
2409 | char tmp[16] = {}; | |
2410 | ||
2411 | if (read(fd, tmp, sizeof(tmp)) > 0) | |
2412 | ctx->cfg.jit_enabled = atoi(tmp); | |
2413 | close(fd); | |
2414 | } | |
2415 | } | |
2416 | ||
32e93fb7 DB |
2417 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
2418 | enum bpf_prog_type type, bool verbose) | |
2419 | { | |
2420 | int ret = -EINVAL; | |
2421 | ||
2422 | if (elf_version(EV_CURRENT) == EV_NONE || | |
2423 | bpf_init_env(pathname)) | |
2424 | return ret; | |
2425 | ||
2426 | memset(ctx, 0, sizeof(*ctx)); | |
ecb05c0f | 2427 | bpf_get_cfg(ctx); |
32e93fb7 DB |
2428 | ctx->verbose = verbose; |
2429 | ctx->type = type; | |
2430 | ||
2431 | ctx->obj_fd = open(pathname, O_RDONLY); | |
2432 | if (ctx->obj_fd < 0) | |
2433 | return ctx->obj_fd; | |
2434 | ||
2435 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
2436 | if (!ctx->elf_fd) { | |
11c39b5e | 2437 | ret = -EINVAL; |
32e93fb7 | 2438 | goto out_fd; |
11c39b5e DB |
2439 | } |
2440 | ||
8187b012 DB |
2441 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
2442 | ret = -EINVAL; | |
2443 | goto out_fd; | |
2444 | } | |
2445 | ||
32e93fb7 DB |
2446 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
2447 | &ctx->elf_hdr) { | |
11c39b5e DB |
2448 | ret = -EIO; |
2449 | goto out_elf; | |
2450 | } | |
2451 | ||
8187b012 DB |
2452 | ret = bpf_elf_check_ehdr(ctx); |
2453 | if (ret < 0) | |
2454 | goto out_elf; | |
2455 | ||
32e93fb7 DB |
2456 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
2457 | sizeof(*(ctx->sec_done))); | |
2458 | if (!ctx->sec_done) { | |
11c39b5e DB |
2459 | ret = -ENOMEM; |
2460 | goto out_elf; | |
2461 | } | |
2462 | ||
f31645d1 DB |
2463 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
2464 | ret = -ENOMEM; | |
2465 | goto out_free; | |
2466 | } | |
2467 | ||
32e93fb7 | 2468 | bpf_save_finfo(ctx); |
f6793eec DB |
2469 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
2470 | ||
32e93fb7 | 2471 | return 0; |
f31645d1 DB |
2472 | out_free: |
2473 | free(ctx->sec_done); | |
32e93fb7 DB |
2474 | out_elf: |
2475 | elf_end(ctx->elf_fd); | |
2476 | out_fd: | |
2477 | close(ctx->obj_fd); | |
2478 | return ret; | |
2479 | } | |
d937a74b | 2480 | |
32e93fb7 DB |
2481 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
2482 | { | |
2483 | int i, count = 0; | |
11c39b5e | 2484 | |
32e93fb7 DB |
2485 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
2486 | if (!ctx->map_fds[i]) | |
2487 | break; | |
2488 | count++; | |
2489 | } | |
473d7840 | 2490 | |
32e93fb7 DB |
2491 | return count; |
2492 | } | |
6256f8c9 | 2493 | |
32e93fb7 DB |
2494 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
2495 | { | |
2496 | int i; | |
2497 | ||
2498 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
2499 | if (ctx->map_fds[i]) | |
2500 | close(ctx->map_fds[i]); | |
473d7840 | 2501 | } |
32e93fb7 DB |
2502 | } |
2503 | ||
2504 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
2505 | { | |
2506 | if (failure) | |
2507 | bpf_maps_teardown(ctx); | |
473d7840 | 2508 | |
f6793eec | 2509 | bpf_hash_destroy(ctx); |
f31645d1 | 2510 | |
32e93fb7 | 2511 | free(ctx->sec_done); |
f31645d1 DB |
2512 | free(ctx->log); |
2513 | ||
32e93fb7 DB |
2514 | elf_end(ctx->elf_fd); |
2515 | close(ctx->obj_fd); | |
2516 | } | |
6256f8c9 | 2517 | |
32e93fb7 | 2518 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 2519 | |
32e93fb7 DB |
2520 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
2521 | const char *section, bool verbose) | |
2522 | { | |
2523 | struct bpf_elf_ctx *ctx = &__ctx; | |
2524 | int fd = 0, ret; | |
6256f8c9 | 2525 | |
32e93fb7 DB |
2526 | ret = bpf_elf_ctx_init(ctx, pathname, type, verbose); |
2527 | if (ret < 0) { | |
2528 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
2529 | return ret; | |
2530 | } | |
6256f8c9 | 2531 | |
32e93fb7 DB |
2532 | ret = bpf_fetch_ancillary(ctx); |
2533 | if (ret < 0) { | |
2534 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
2535 | goto out; | |
2536 | } | |
2537 | ||
2538 | fd = bpf_fetch_prog_sec(ctx, section); | |
2539 | if (fd < 0) { | |
2540 | fprintf(stderr, "Error fetching program/map!\n"); | |
2541 | ret = fd; | |
2542 | goto out; | |
2543 | } | |
2544 | ||
2545 | ret = bpf_fill_prog_arrays(ctx); | |
2546 | if (ret < 0) | |
2547 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 2548 | out: |
32e93fb7 DB |
2549 | bpf_elf_ctx_destroy(ctx, ret < 0); |
2550 | if (ret < 0) { | |
2551 | if (fd) | |
2552 | close(fd); | |
2553 | return ret; | |
2554 | } | |
2555 | ||
2556 | return fd; | |
6256f8c9 | 2557 | } |
11c39b5e | 2558 | |
6256f8c9 | 2559 | static int |
4bd62446 DB |
2560 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
2561 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 | 2562 | { |
d17b136f PS |
2563 | struct bpf_map_set_msg msg = { |
2564 | .aux.uds_ver = BPF_SCM_AUX_VER, | |
2565 | .aux.num_ent = entries, | |
2566 | }; | |
6256f8c9 DB |
2567 | int *cmsg_buf, min_fd; |
2568 | char *amsg_buf; | |
2569 | int i; | |
2570 | ||
6256f8c9 DB |
2571 | strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); |
2572 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); | |
2573 | ||
2574 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
2575 | amsg_buf = (char *)msg.aux.ent; | |
2576 | ||
4bd62446 | 2577 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
2578 | int ret; |
2579 | ||
4bd62446 | 2580 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
2581 | bpf_map_set_init_single(&msg, min_fd); |
2582 | ||
2583 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
2584 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
2585 | ||
2586 | ret = sendmsg(fd, &msg.hdr, 0); | |
2587 | if (ret <= 0) | |
2588 | return ret ? : -1; | |
2589 | } | |
2590 | ||
2591 | return 0; | |
11c39b5e DB |
2592 | } |
2593 | ||
4bd62446 DB |
2594 | static int |
2595 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
2596 | unsigned int entries) | |
2597 | { | |
2598 | struct bpf_map_set_msg msg; | |
2599 | int *cmsg_buf, min_fd; | |
2600 | char *amsg_buf, *mmsg_buf; | |
2601 | unsigned int needed = 1; | |
2602 | int i; | |
2603 | ||
2604 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
2605 | amsg_buf = (char *)msg.aux.ent; | |
2606 | mmsg_buf = (char *)&msg.aux; | |
2607 | ||
2608 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
2609 | struct cmsghdr *cmsg; | |
2610 | int ret; | |
2611 | ||
2612 | min_fd = min(entries, entries - i); | |
2613 | bpf_map_set_init_single(&msg, min_fd); | |
2614 | ||
2615 | ret = recvmsg(fd, &msg.hdr, 0); | |
2616 | if (ret <= 0) | |
2617 | return ret ? : -1; | |
2618 | ||
2619 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
2620 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
2621 | return -EINVAL; | |
2622 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
2623 | return -EIO; | |
2624 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
2625 | return -ENOSYS; | |
2626 | ||
2627 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
2628 | if (min_fd > entries || min_fd <= 0) | |
2629 | return -EINVAL; | |
2630 | ||
2631 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
2632 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
2633 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
2634 | ||
2635 | needed = aux->num_ent; | |
2636 | } | |
2637 | ||
2638 | return 0; | |
2639 | } | |
2640 | ||
2641 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 2642 | { |
32e93fb7 | 2643 | struct bpf_elf_ctx *ctx = &__ctx; |
d17b136f PS |
2644 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
2645 | struct bpf_map_data bpf_aux = { | |
2646 | .fds = ctx->map_fds, | |
2647 | .ent = ctx->maps, | |
2648 | .st = &ctx->stat, | |
2649 | .obj = obj, | |
2650 | }; | |
6256f8c9 DB |
2651 | int fd, ret; |
2652 | ||
2653 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2654 | if (fd < 0) { | |
2655 | fprintf(stderr, "Cannot open socket: %s\n", | |
2656 | strerror(errno)); | |
2657 | return -1; | |
2658 | } | |
2659 | ||
6256f8c9 DB |
2660 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2661 | ||
2662 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2663 | if (ret < 0) { | |
2664 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
2665 | path, strerror(errno)); | |
2666 | return -1; | |
2667 | } | |
2668 | ||
4bd62446 | 2669 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 2670 | bpf_maps_count(ctx)); |
6256f8c9 | 2671 | if (ret < 0) |
4bd62446 DB |
2672 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
2673 | path, strerror(errno)); | |
2674 | ||
32e93fb7 | 2675 | bpf_maps_teardown(ctx); |
4bd62446 DB |
2676 | close(fd); |
2677 | return ret; | |
2678 | } | |
2679 | ||
2680 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
2681 | unsigned int entries) | |
2682 | { | |
d17b136f | 2683 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
4bd62446 DB |
2684 | int fd, ret; |
2685 | ||
2686 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2687 | if (fd < 0) { | |
2688 | fprintf(stderr, "Cannot open socket: %s\n", | |
2689 | strerror(errno)); | |
2690 | return -1; | |
2691 | } | |
2692 | ||
4bd62446 DB |
2693 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2694 | ||
2695 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2696 | if (ret < 0) { | |
2697 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
2698 | strerror(errno)); | |
2699 | return -1; | |
2700 | } | |
2701 | ||
2702 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
2703 | if (ret < 0) | |
2704 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
2705 | path, strerror(errno)); |
2706 | ||
4bd62446 | 2707 | unlink(addr.sun_path); |
6256f8c9 DB |
2708 | close(fd); |
2709 | return ret; | |
2710 | } | |
11c39b5e | 2711 | #endif /* HAVE_ELF */ |