]>
Commit | Line | Data |
---|---|---|
1d129d19 JP |
1 | /* |
2 | * tc_bpf.c BPF common code | |
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 | * | |
9 | * Authors: Daniel Borkmann <dborkman@redhat.com> | |
10 | * Jiri Pirko <jiri@resnulli.us> | |
11c39b5e | 11 | * Alexei Starovoitov <ast@plumgrid.com> |
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> | |
1d129d19 | 23 | |
11c39b5e DB |
24 | #ifdef HAVE_ELF |
25 | #include <libelf.h> | |
26 | #include <gelf.h> | |
27 | #endif | |
28 | ||
32e93fb7 DB |
29 | #include <sys/types.h> |
30 | #include <sys/stat.h> | |
31 | #include <sys/un.h> | |
32 | #include <sys/vfs.h> | |
33 | #include <sys/mount.h> | |
34 | #include <sys/syscall.h> | |
35 | #include <sys/sendfile.h> | |
36 | #include <sys/resource.h> | |
37 | ||
38 | #include <linux/bpf.h> | |
39 | #include <linux/filter.h> | |
40 | #include <linux/if_alg.h> | |
41 | ||
8187b012 DB |
42 | #include <arpa/inet.h> |
43 | ||
1d129d19 | 44 | #include "utils.h" |
6256f8c9 DB |
45 | |
46 | #include "bpf_elf.h" | |
47 | #include "bpf_scm.h" | |
48 | ||
1d129d19 JP |
49 | #include "tc_util.h" |
50 | #include "tc_bpf.h" | |
51 | ||
67584e3a ND |
52 | #ifndef AF_ALG |
53 | #define AF_ALG 38 | |
54 | #endif | |
55 | ||
32e93fb7 DB |
56 | #ifdef HAVE_ELF |
57 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
58 | const char *sec, bool verbose); | |
59 | #else | |
60 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
61 | const char *sec, bool verbose) | |
62 | { | |
63 | fprintf(stderr, "No ELF library support compiled in.\n"); | |
64 | errno = ENOSYS; | |
65 | return -1; | |
66 | } | |
67 | #endif | |
68 | ||
69 | static inline __u64 bpf_ptr_to_u64(const void *ptr) | |
70 | { | |
71 | return (__u64)(unsigned long)ptr; | |
72 | } | |
73 | ||
74 | static int bpf(int cmd, union bpf_attr *attr, unsigned int size) | |
75 | { | |
76 | #ifdef __NR_bpf | |
77 | return syscall(__NR_bpf, cmd, attr, size); | |
78 | #else | |
79 | fprintf(stderr, "No bpf syscall, kernel headers too old?\n"); | |
80 | errno = ENOSYS; | |
81 | return -1; | |
82 | #endif | |
83 | } | |
84 | ||
91d88eeb DB |
85 | static int bpf_map_update(int fd, const void *key, const void *value, |
86 | uint64_t flags) | |
32e93fb7 | 87 | { |
67584e3a ND |
88 | union bpf_attr attr; |
89 | ||
90 | memset(&attr, 0, sizeof(attr)); | |
91 | attr.map_fd = fd; | |
92 | attr.key = bpf_ptr_to_u64(key); | |
93 | attr.value = bpf_ptr_to_u64(value); | |
94 | attr.flags = flags; | |
32e93fb7 | 95 | |
91d88eeb | 96 | return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
32e93fb7 DB |
97 | } |
98 | ||
99 | static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len, | |
100 | char **bpf_string, bool *need_release, | |
101 | const char separator) | |
1d129d19 JP |
102 | { |
103 | char sp; | |
104 | ||
105 | if (from_file) { | |
106 | size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,"); | |
107 | char *tmp_string; | |
108 | FILE *fp; | |
109 | ||
110 | tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len; | |
111 | tmp_string = malloc(tmp_len); | |
112 | if (tmp_string == NULL) | |
113 | return -ENOMEM; | |
114 | ||
115 | memset(tmp_string, 0, tmp_len); | |
116 | ||
117 | fp = fopen(arg, "r"); | |
118 | if (fp == NULL) { | |
119 | perror("Cannot fopen"); | |
120 | free(tmp_string); | |
121 | return -ENOENT; | |
122 | } | |
123 | ||
124 | if (!fgets(tmp_string, tmp_len, fp)) { | |
125 | free(tmp_string); | |
126 | fclose(fp); | |
127 | return -EIO; | |
128 | } | |
129 | ||
130 | fclose(fp); | |
131 | ||
132 | *need_release = true; | |
133 | *bpf_string = tmp_string; | |
134 | } else { | |
135 | *need_release = false; | |
136 | *bpf_string = arg; | |
137 | } | |
138 | ||
139 | if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 || | |
140 | sp != separator) { | |
141 | if (*need_release) | |
142 | free(*bpf_string); | |
143 | return -EINVAL; | |
144 | } | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
32e93fb7 DB |
149 | static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops, |
150 | bool from_file) | |
1d129d19 JP |
151 | { |
152 | char *bpf_string, *token, separator = ','; | |
153 | int ret = 0, i = 0; | |
154 | bool need_release; | |
155 | __u16 bpf_len = 0; | |
156 | ||
157 | if (argc < 1) | |
158 | return -EINVAL; | |
159 | if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string, | |
160 | &need_release, separator)) | |
161 | return -EINVAL; | |
162 | if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) { | |
163 | ret = -EINVAL; | |
164 | goto out; | |
165 | } | |
166 | ||
167 | token = bpf_string; | |
168 | while ((token = strchr(token, separator)) && (++token)[0]) { | |
169 | if (i >= bpf_len) { | |
170 | fprintf(stderr, "Real program length exceeds encoded " | |
171 | "length parameter!\n"); | |
172 | ret = -EINVAL; | |
173 | goto out; | |
174 | } | |
175 | ||
176 | if (sscanf(token, "%hu %hhu %hhu %u,", | |
177 | &bpf_ops[i].code, &bpf_ops[i].jt, | |
178 | &bpf_ops[i].jf, &bpf_ops[i].k) != 4) { | |
179 | fprintf(stderr, "Error at instruction %d!\n", i); | |
180 | ret = -EINVAL; | |
181 | goto out; | |
182 | } | |
183 | ||
184 | i++; | |
185 | } | |
186 | ||
187 | if (i != bpf_len) { | |
188 | fprintf(stderr, "Parsed program length is less than encoded" | |
189 | "length parameter!\n"); | |
190 | ret = -EINVAL; | |
191 | goto out; | |
192 | } | |
193 | ret = bpf_len; | |
1d129d19 JP |
194 | out: |
195 | if (need_release) | |
196 | free(bpf_string); | |
197 | ||
198 | return ret; | |
199 | } | |
200 | ||
201 | void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len) | |
202 | { | |
203 | struct sock_filter *ops = (struct sock_filter *) RTA_DATA(bpf_ops); | |
204 | int i; | |
205 | ||
206 | if (len == 0) | |
207 | return; | |
208 | ||
209 | fprintf(f, "bytecode \'%u,", len); | |
210 | ||
211 | for (i = 0; i < len - 1; i++) | |
212 | fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt, | |
213 | ops[i].jf, ops[i].k); | |
214 | ||
6256f8c9 | 215 | fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt, |
1d129d19 JP |
216 | ops[i].jf, ops[i].k); |
217 | } | |
11c39b5e | 218 | |
91d88eeb DB |
219 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, |
220 | int length) | |
9e607f2e DB |
221 | { |
222 | char file[PATH_MAX], buff[4096]; | |
223 | struct bpf_elf_map tmp, zero; | |
224 | unsigned int val; | |
225 | FILE *fp; | |
226 | ||
227 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
228 | ||
229 | fp = fopen(file, "r"); | |
230 | if (!fp) { | |
231 | fprintf(stderr, "No procfs support?!\n"); | |
232 | return -EIO; | |
233 | } | |
234 | ||
235 | memset(&tmp, 0, sizeof(tmp)); | |
236 | while (fgets(buff, sizeof(buff), fp)) { | |
237 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
238 | tmp.type = val; | |
239 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
240 | tmp.size_key = val; | |
241 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
242 | tmp.size_value = val; | |
243 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
244 | tmp.max_elem = val; | |
245 | } | |
246 | ||
247 | fclose(fp); | |
248 | ||
91d88eeb | 249 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
250 | return 0; |
251 | } else { | |
252 | memset(&zero, 0, sizeof(zero)); | |
253 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, | |
254 | * so just accept it. We know we do have an eBPF fd and in this | |
255 | * case, everything is 0. It is guaranteed that no such map exists | |
256 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
257 | */ | |
91d88eeb | 258 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
259 | return 0; |
260 | ||
261 | fprintf(stderr, "Map specs from pinned file differ!\n"); | |
262 | return -EINVAL; | |
263 | } | |
264 | } | |
265 | ||
91d88eeb DB |
266 | static int bpf_mnt_fs(const char *target) |
267 | { | |
268 | bool bind_done = false; | |
269 | ||
270 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
271 | if (errno != EINVAL || bind_done) { | |
272 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
273 | target, strerror(errno)); | |
274 | return -1; | |
275 | } | |
276 | ||
277 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
278 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
279 | target, target, strerror(errno)); | |
280 | return -1; | |
281 | } | |
282 | ||
283 | bind_done = true; | |
284 | } | |
285 | ||
286 | if (mount("bpf", target, "bpf", 0, NULL)) { | |
287 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", | |
288 | target, strerror(errno)); | |
289 | return -1; | |
290 | } | |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
32e93fb7 DB |
295 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
296 | { | |
297 | struct statfs st_fs; | |
298 | ||
299 | if (statfs(mnt, &st_fs) < 0) | |
300 | return -ENOENT; | |
301 | if ((unsigned long)st_fs.f_type != magic) | |
302 | return -ENOENT; | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
307 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, | |
308 | char *mnt, int len, | |
309 | const char * const *known_mnts) | |
310 | { | |
311 | const char * const *ptr; | |
312 | char type[100]; | |
313 | FILE *fp; | |
314 | ||
315 | if (known_mnts) { | |
316 | ptr = known_mnts; | |
317 | while (*ptr) { | |
318 | if (bpf_valid_mntpt(*ptr, magic) == 0) { | |
319 | strncpy(mnt, *ptr, len - 1); | |
320 | mnt[len - 1] = 0; | |
321 | return mnt; | |
322 | } | |
323 | ptr++; | |
324 | } | |
325 | } | |
326 | ||
327 | fp = fopen("/proc/mounts", "r"); | |
328 | if (fp == NULL || len != PATH_MAX) | |
329 | return NULL; | |
330 | ||
331 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
332 | mnt, type) == 2) { | |
333 | if (strcmp(type, fstype) == 0) | |
334 | break; | |
335 | } | |
336 | ||
337 | fclose(fp); | |
338 | if (strcmp(type, fstype) != 0) | |
339 | return NULL; | |
340 | ||
341 | return mnt; | |
342 | } | |
343 | ||
344 | int bpf_trace_pipe(void) | |
345 | { | |
346 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
347 | static const char * const tracefs_known_mnts[] = { | |
348 | TRACE_DIR_MNT, | |
349 | "/sys/kernel/debug/tracing", | |
350 | "/tracing", | |
351 | "/trace", | |
352 | 0, | |
353 | }; | |
354 | char tpipe[PATH_MAX]; | |
355 | const char *mnt; | |
356 | int fd; | |
357 | ||
358 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
359 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
360 | if (!mnt) { | |
361 | fprintf(stderr, "tracefs not mounted?\n"); | |
362 | return -1; | |
363 | } | |
364 | ||
365 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
366 | ||
367 | fd = open(tpipe, O_RDONLY); | |
368 | if (fd < 0) | |
369 | return -1; | |
370 | ||
371 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
372 | while (1) { | |
373 | static char buff[4096]; | |
374 | ssize_t ret; | |
375 | ||
376 | ret = read(fd, buff, sizeof(buff) - 1); | |
377 | if (ret > 0) { | |
378 | write(2, buff, ret); | |
379 | fflush(stderr); | |
380 | } | |
381 | } | |
382 | ||
383 | return 0; | |
384 | } | |
385 | ||
91d88eeb DB |
386 | static const char *bpf_get_tc_dir(void) |
387 | { | |
388 | static bool bpf_mnt_cached = false; | |
389 | static char bpf_tc_dir[PATH_MAX]; | |
390 | static const char *mnt; | |
391 | static const char * const bpf_known_mnts[] = { | |
392 | BPF_DIR_MNT, | |
393 | 0, | |
394 | }; | |
395 | char bpf_mnt[PATH_MAX] = BPF_DIR_MNT; | |
396 | char bpf_glo_dir[PATH_MAX]; | |
397 | int ret; | |
398 | ||
399 | if (bpf_mnt_cached) | |
400 | goto done; | |
401 | ||
402 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_mnt, sizeof(bpf_mnt), | |
403 | bpf_known_mnts); | |
404 | if (!mnt) { | |
405 | mnt = getenv(BPF_ENV_MNT); | |
406 | if (!mnt) | |
407 | mnt = BPF_DIR_MNT; | |
408 | ret = bpf_mnt_fs(mnt); | |
409 | if (ret) { | |
410 | mnt = NULL; | |
411 | goto out; | |
412 | } | |
413 | } | |
414 | ||
415 | snprintf(bpf_tc_dir, sizeof(bpf_tc_dir), "%s/%s", mnt, BPF_DIR_TC); | |
416 | ret = mkdir(bpf_tc_dir, S_IRWXU); | |
417 | if (ret && errno != EEXIST) { | |
418 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_tc_dir, | |
419 | strerror(errno)); | |
420 | mnt = NULL; | |
421 | goto out; | |
422 | } | |
423 | ||
424 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s", | |
425 | bpf_tc_dir, BPF_DIR_GLOBALS); | |
426 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
427 | if (ret && errno != EEXIST) { | |
428 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
429 | strerror(errno)); | |
430 | mnt = NULL; | |
431 | goto out; | |
432 | } | |
433 | ||
434 | mnt = bpf_tc_dir; | |
435 | out: | |
436 | bpf_mnt_cached = true; | |
437 | done: | |
438 | return mnt; | |
439 | } | |
440 | ||
441 | static int bpf_obj_get(const char *pathname) | |
442 | { | |
443 | union bpf_attr attr; | |
444 | char tmp[PATH_MAX]; | |
445 | ||
446 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
447 | pathname[1] == ':' && bpf_get_tc_dir()) { | |
448 | snprintf(tmp, sizeof(tmp), "%s/%s", | |
449 | bpf_get_tc_dir(), pathname + 2); | |
450 | pathname = tmp; | |
451 | } | |
452 | ||
453 | memset(&attr, 0, sizeof(attr)); | |
454 | attr.pathname = bpf_ptr_to_u64(pathname); | |
455 | ||
456 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
457 | } | |
458 | ||
6256f8c9 | 459 | const char *bpf_default_section(const enum bpf_prog_type type) |
11c39b5e DB |
460 | { |
461 | switch (type) { | |
462 | case BPF_PROG_TYPE_SCHED_CLS: | |
463 | return ELF_SECTION_CLASSIFIER; | |
6256f8c9 DB |
464 | case BPF_PROG_TYPE_SCHED_ACT: |
465 | return ELF_SECTION_ACTION; | |
11c39b5e DB |
466 | default: |
467 | return NULL; | |
468 | } | |
469 | } | |
470 | ||
91d88eeb DB |
471 | enum bpf_mode { |
472 | CBPF_BYTECODE = 0, | |
473 | CBPF_FILE, | |
474 | EBPF_OBJECT, | |
475 | EBPF_PINNED, | |
476 | __BPF_MODE_MAX, | |
477 | #define BPF_MODE_MAX __BPF_MODE_MAX | |
478 | }; | |
479 | ||
480 | static int bpf_parse(int *ptr_argc, char ***ptr_argv, const bool *opt_tbl, | |
481 | enum bpf_prog_type *type, enum bpf_mode *mode, | |
482 | const char **ptr_object, const char **ptr_section, | |
483 | const char **ptr_uds_name, struct sock_filter *opcodes) | |
32e93fb7 | 484 | { |
32e93fb7 | 485 | const char *file, *section, *uds_name; |
32e93fb7 | 486 | bool verbose = false; |
91d88eeb DB |
487 | int ret, argc; |
488 | char **argv; | |
489 | ||
490 | argv = *ptr_argv; | |
491 | argc = *ptr_argc; | |
492 | ||
493 | if (opt_tbl[CBPF_BYTECODE] && | |
494 | (matches(*argv, "bytecode") == 0 || | |
495 | strcmp(*argv, "bc") == 0)) { | |
496 | *mode = CBPF_BYTECODE; | |
497 | } else if (opt_tbl[CBPF_FILE] && | |
498 | (matches(*argv, "bytecode-file") == 0 || | |
499 | strcmp(*argv, "bcf") == 0)) { | |
500 | *mode = CBPF_FILE; | |
501 | } else if (opt_tbl[EBPF_OBJECT] && | |
502 | (matches(*argv, "object-file") == 0 || | |
503 | strcmp(*argv, "obj") == 0)) { | |
504 | *mode = EBPF_OBJECT; | |
505 | } else if (opt_tbl[EBPF_PINNED] && | |
506 | (matches(*argv, "object-pinned") == 0 || | |
507 | matches(*argv, "pinned") == 0 || | |
508 | matches(*argv, "fd") == 0)) { | |
509 | *mode = EBPF_PINNED; | |
32e93fb7 DB |
510 | } else { |
511 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
512 | return -1; | |
513 | } | |
514 | ||
515 | NEXT_ARG(); | |
516 | file = section = uds_name = NULL; | |
91d88eeb | 517 | if (*mode == EBPF_OBJECT || *mode == EBPF_PINNED) { |
32e93fb7 DB |
518 | file = *argv; |
519 | NEXT_ARG_FWD(); | |
520 | ||
91d88eeb DB |
521 | if (*type == BPF_PROG_TYPE_UNSPEC) { |
522 | if (argc > 0 && matches(*argv, "type") == 0) { | |
523 | NEXT_ARG(); | |
524 | if (matches(*argv, "cls") == 0) { | |
525 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
526 | } else if (matches(*argv, "act") == 0) { | |
527 | *type = BPF_PROG_TYPE_SCHED_ACT; | |
528 | } else { | |
529 | fprintf(stderr, "What type is \"%s\"?\n", | |
530 | *argv); | |
531 | return -1; | |
532 | } | |
533 | NEXT_ARG_FWD(); | |
534 | } else { | |
535 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
536 | } | |
537 | } | |
538 | ||
539 | section = bpf_default_section(*type); | |
32e93fb7 DB |
540 | if (argc > 0 && matches(*argv, "section") == 0) { |
541 | NEXT_ARG(); | |
542 | section = *argv; | |
543 | NEXT_ARG_FWD(); | |
544 | } | |
545 | ||
546 | uds_name = getenv(BPF_ENV_UDS); | |
547 | if (argc > 0 && !uds_name && | |
548 | matches(*argv, "export") == 0) { | |
549 | NEXT_ARG(); | |
550 | uds_name = *argv; | |
551 | NEXT_ARG_FWD(); | |
552 | } | |
553 | ||
554 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
555 | verbose = true; | |
556 | NEXT_ARG_FWD(); | |
557 | } | |
558 | ||
559 | PREV_ARG(); | |
560 | } | |
561 | ||
91d88eeb DB |
562 | if (*mode == CBPF_BYTECODE || *mode == CBPF_FILE) |
563 | ret = bpf_ops_parse(argc, argv, opcodes, *mode == CBPF_FILE); | |
564 | else if (*mode == EBPF_OBJECT) | |
565 | ret = bpf_obj_open(file, *type, section, verbose); | |
566 | else if (*mode == EBPF_PINNED) | |
32e93fb7 | 567 | ret = bpf_obj_get(file); |
91d88eeb | 568 | else |
32e93fb7 DB |
569 | return -1; |
570 | ||
91d88eeb DB |
571 | if (ptr_object) |
572 | *ptr_object = file; | |
573 | if (ptr_section) | |
574 | *ptr_section = section; | |
575 | if (ptr_uds_name) | |
576 | *ptr_uds_name = uds_name; | |
577 | ||
578 | *ptr_argc = argc; | |
579 | *ptr_argv = argv; | |
580 | ||
581 | return ret; | |
582 | } | |
583 | ||
584 | int bpf_parse_common(int *ptr_argc, char ***ptr_argv, const int *nla_tbl, | |
585 | enum bpf_prog_type type, const char **ptr_object, | |
586 | const char **ptr_uds_name, struct nlmsghdr *n) | |
587 | { | |
588 | struct sock_filter opcodes[BPF_MAXINSNS]; | |
589 | const bool opt_tbl[BPF_MODE_MAX] = { | |
590 | [CBPF_BYTECODE] = true, | |
591 | [CBPF_FILE] = true, | |
592 | [EBPF_OBJECT] = true, | |
593 | [EBPF_PINNED] = true, | |
594 | }; | |
595 | char annotation[256]; | |
596 | const char *section; | |
597 | enum bpf_mode mode; | |
598 | int ret; | |
599 | ||
600 | ret = bpf_parse(ptr_argc, ptr_argv, opt_tbl, &type, &mode, | |
601 | ptr_object, §ion, ptr_uds_name, opcodes); | |
602 | if (ret < 0) | |
603 | return ret; | |
604 | ||
32e93fb7 DB |
605 | if (mode == CBPF_BYTECODE || mode == CBPF_FILE) { |
606 | addattr16(n, MAX_MSG, nla_tbl[BPF_NLA_OPS_LEN], ret); | |
607 | addattr_l(n, MAX_MSG, nla_tbl[BPF_NLA_OPS], opcodes, | |
608 | ret * sizeof(struct sock_filter)); | |
91d88eeb DB |
609 | } |
610 | ||
611 | if (mode == EBPF_OBJECT || mode == EBPF_PINNED) { | |
32e93fb7 | 612 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
91d88eeb DB |
613 | basename(*ptr_object), mode == EBPF_PINNED ? |
614 | "*fsobj" : section); | |
32e93fb7 DB |
615 | |
616 | addattr32(n, MAX_MSG, nla_tbl[BPF_NLA_FD], ret); | |
617 | addattrstrz(n, MAX_MSG, nla_tbl[BPF_NLA_NAME], annotation); | |
618 | } | |
619 | ||
91d88eeb DB |
620 | return 0; |
621 | } | |
32e93fb7 | 622 | |
91d88eeb DB |
623 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
624 | { | |
625 | enum bpf_prog_type type = BPF_PROG_TYPE_UNSPEC; | |
626 | const bool opt_tbl[BPF_MODE_MAX] = { | |
627 | [CBPF_BYTECODE] = false, | |
628 | [CBPF_FILE] = false, | |
629 | [EBPF_OBJECT] = true, | |
630 | [EBPF_PINNED] = true, | |
631 | }; | |
632 | const struct bpf_elf_map test = { | |
633 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
634 | .size_key = sizeof(int), | |
635 | .size_value = sizeof(int), | |
636 | }; | |
637 | int ret, prog_fd, map_fd; | |
638 | const char *section; | |
639 | enum bpf_mode mode; | |
640 | uint32_t map_key; | |
641 | ||
642 | prog_fd = bpf_parse(&argc, &argv, opt_tbl, &type, &mode, | |
643 | NULL, §ion, NULL, NULL); | |
644 | if (prog_fd < 0) | |
645 | return prog_fd; | |
646 | if (key) { | |
647 | map_key = *key; | |
648 | } else { | |
649 | ret = sscanf(section, "%*i/%i", &map_key); | |
650 | if (ret != 1) { | |
651 | fprintf(stderr, "Couldn\'t infer map key from section " | |
652 | "name! Please provide \'key\' argument!\n"); | |
653 | ret = -EINVAL; | |
654 | goto out_prog; | |
655 | } | |
656 | } | |
32e93fb7 | 657 | |
91d88eeb DB |
658 | map_fd = bpf_obj_get(map_path); |
659 | if (map_fd < 0) { | |
660 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
661 | map_path, strerror(errno)); | |
662 | ret = map_fd; | |
663 | goto out_prog; | |
664 | } | |
665 | ||
666 | ret = bpf_map_selfcheck_pinned(map_fd, &test, | |
667 | offsetof(struct bpf_elf_map, max_elem)); | |
668 | if (ret < 0) { | |
669 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
670 | goto out_map; | |
671 | } | |
672 | ||
673 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
674 | if (ret < 0) | |
675 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
676 | out_map: | |
677 | close(map_fd); | |
678 | out_prog: | |
679 | close(prog_fd); | |
680 | return ret; | |
32e93fb7 DB |
681 | } |
682 | ||
6256f8c9 | 683 | #ifdef HAVE_ELF |
32e93fb7 DB |
684 | struct bpf_elf_prog { |
685 | enum bpf_prog_type type; | |
686 | const struct bpf_insn *insns; | |
687 | size_t size; | |
688 | const char *license; | |
689 | }; | |
690 | ||
f6793eec DB |
691 | struct bpf_hash_entry { |
692 | unsigned int pinning; | |
693 | const char *subpath; | |
694 | struct bpf_hash_entry *next; | |
695 | }; | |
696 | ||
32e93fb7 DB |
697 | struct bpf_elf_ctx { |
698 | Elf *elf_fd; | |
699 | GElf_Ehdr elf_hdr; | |
700 | Elf_Data *sym_tab; | |
701 | Elf_Data *str_tab; | |
702 | int obj_fd; | |
703 | int map_fds[ELF_MAX_MAPS]; | |
704 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
705 | int sym_num; | |
706 | int map_num; | |
707 | bool *sec_done; | |
708 | int sec_maps; | |
709 | char license[ELF_MAX_LICENSE_LEN]; | |
710 | enum bpf_prog_type type; | |
711 | bool verbose; | |
712 | struct bpf_elf_st stat; | |
f6793eec | 713 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
714 | char *log; |
715 | size_t log_size; | |
32e93fb7 DB |
716 | }; |
717 | ||
6256f8c9 | 718 | struct bpf_elf_sec_data { |
32e93fb7 DB |
719 | GElf_Shdr sec_hdr; |
720 | Elf_Data *sec_data; | |
721 | const char *sec_name; | |
6256f8c9 DB |
722 | }; |
723 | ||
724 | struct bpf_map_data { | |
32e93fb7 DB |
725 | int *fds; |
726 | const char *obj; | |
727 | struct bpf_elf_st *st; | |
728 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
729 | }; |
730 | ||
f31645d1 DB |
731 | static __check_format_string(2, 3) void |
732 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
733 | { |
734 | va_list vl; | |
735 | ||
736 | va_start(vl, format); | |
737 | vfprintf(stderr, format, vl); | |
738 | va_end(vl); | |
739 | ||
f31645d1 DB |
740 | if (ctx->log && ctx->log[0]) { |
741 | fprintf(stderr, "%s\n", ctx->log); | |
742 | memset(ctx->log, 0, ctx->log_size); | |
743 | } | |
744 | } | |
745 | ||
746 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
747 | { | |
748 | size_t log_size = ctx->log_size; | |
749 | void *ptr; | |
750 | ||
751 | if (!ctx->log) { | |
752 | log_size = 65536; | |
753 | } else { | |
754 | log_size <<= 1; | |
755 | if (log_size > (UINT_MAX >> 8)) | |
756 | return -EINVAL; | |
d937a74b | 757 | } |
f31645d1 DB |
758 | |
759 | ptr = realloc(ctx->log, log_size); | |
760 | if (!ptr) | |
761 | return -ENOMEM; | |
762 | ||
763 | ctx->log = ptr; | |
764 | ctx->log_size = log_size; | |
765 | ||
766 | return 0; | |
11c39b5e DB |
767 | } |
768 | ||
32e93fb7 | 769 | static int bpf_map_create(enum bpf_map_type type, unsigned int size_key, |
11c39b5e DB |
770 | unsigned int size_value, unsigned int max_elem) |
771 | { | |
67584e3a ND |
772 | union bpf_attr attr; |
773 | ||
774 | memset(&attr, 0, sizeof(attr)); | |
775 | attr.map_type = type; | |
776 | attr.key_size = size_key; | |
777 | attr.value_size = size_value; | |
778 | attr.max_entries = max_elem; | |
11c39b5e DB |
779 | |
780 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
781 | } | |
782 | ||
783 | static int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, | |
f31645d1 DB |
784 | size_t size_insns, const char *license, char *log, |
785 | size_t size_log) | |
11c39b5e | 786 | { |
67584e3a ND |
787 | union bpf_attr attr; |
788 | ||
789 | memset(&attr, 0, sizeof(attr)); | |
790 | attr.prog_type = type; | |
791 | attr.insns = bpf_ptr_to_u64(insns); | |
f31645d1 | 792 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); |
67584e3a | 793 | attr.license = bpf_ptr_to_u64(license); |
11c39b5e | 794 | |
f31645d1 DB |
795 | if (size_log > 0) { |
796 | attr.log_buf = bpf_ptr_to_u64(log); | |
797 | attr.log_size = size_log; | |
798 | attr.log_level = 1; | |
32e93fb7 DB |
799 | } |
800 | ||
11c39b5e DB |
801 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); |
802 | } | |
803 | ||
32e93fb7 | 804 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 805 | { |
67584e3a ND |
806 | union bpf_attr attr; |
807 | ||
808 | memset(&attr, 0, sizeof(attr)); | |
809 | attr.pathname = bpf_ptr_to_u64(pathname); | |
810 | attr.bpf_fd = fd; | |
32e93fb7 DB |
811 | |
812 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
813 | } | |
11c39b5e | 814 | |
32e93fb7 DB |
815 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
816 | { | |
817 | struct sockaddr_alg alg = { | |
818 | .salg_family = AF_ALG, | |
819 | .salg_type = "hash", | |
820 | .salg_name = "sha1", | |
821 | }; | |
822 | int ret, cfd, ofd, ffd; | |
823 | struct stat stbuff; | |
824 | ssize_t size; | |
825 | ||
826 | if (!object || len != 20) | |
827 | return -EINVAL; | |
828 | ||
829 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
830 | if (cfd < 0) { | |
831 | fprintf(stderr, "Cannot get AF_ALG socket: %s\n", | |
832 | strerror(errno)); | |
833 | return cfd; | |
834 | } | |
835 | ||
836 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
837 | if (ret < 0) { | |
838 | fprintf(stderr, "Error binding socket: %s\n", strerror(errno)); | |
839 | goto out_cfd; | |
840 | } | |
841 | ||
842 | ofd = accept(cfd, NULL, 0); | |
843 | if (ofd < 0) { | |
844 | fprintf(stderr, "Error accepting socket: %s\n", | |
845 | strerror(errno)); | |
846 | ret = ofd; | |
847 | goto out_cfd; | |
848 | } | |
849 | ||
850 | ffd = open(object, O_RDONLY); | |
851 | if (ffd < 0) { | |
852 | fprintf(stderr, "Error opening object %s: %s\n", | |
853 | object, strerror(errno)); | |
854 | ret = ffd; | |
855 | goto out_ofd; | |
856 | } | |
857 | ||
858 | ret = fstat(ffd, &stbuff); | |
859 | if (ret < 0) { | |
860 | fprintf(stderr, "Error doing fstat: %s\n", | |
861 | strerror(errno)); | |
862 | goto out_ffd; | |
d937a74b | 863 | } |
11c39b5e | 864 | |
32e93fb7 DB |
865 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
866 | if (size != stbuff.st_size) { | |
867 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
868 | size, stbuff.st_size, strerror(errno)); | |
869 | ret = -1; | |
870 | goto out_ffd; | |
871 | } | |
872 | ||
873 | size = read(ofd, out, len); | |
874 | if (size != len) { | |
875 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
876 | size, len, strerror(errno)); | |
877 | ret = -1; | |
878 | } else { | |
879 | ret = 0; | |
880 | } | |
881 | out_ffd: | |
882 | close(ffd); | |
883 | out_ofd: | |
884 | close(ofd); | |
885 | out_cfd: | |
886 | close(cfd); | |
887 | return ret; | |
11c39b5e DB |
888 | } |
889 | ||
32e93fb7 | 890 | static const char *bpf_get_obj_uid(const char *pathname) |
11c39b5e | 891 | { |
32e93fb7 DB |
892 | static bool bpf_uid_cached = false; |
893 | static char bpf_uid[64]; | |
894 | uint8_t tmp[20]; | |
895 | int ret; | |
11c39b5e | 896 | |
32e93fb7 DB |
897 | if (bpf_uid_cached) |
898 | goto done; | |
11c39b5e | 899 | |
32e93fb7 DB |
900 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); |
901 | if (ret) { | |
902 | fprintf(stderr, "Object hashing failed!\n"); | |
903 | return NULL; | |
904 | } | |
905 | ||
906 | hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid)); | |
907 | bpf_uid_cached = true; | |
908 | done: | |
909 | return bpf_uid; | |
11c39b5e DB |
910 | } |
911 | ||
32e93fb7 DB |
912 | static int bpf_init_env(const char *pathname) |
913 | { | |
914 | struct rlimit limit = { | |
915 | .rlim_cur = RLIM_INFINITY, | |
916 | .rlim_max = RLIM_INFINITY, | |
917 | }; | |
918 | ||
919 | /* Don't bother in case we fail! */ | |
920 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
921 | ||
922 | if (!bpf_get_tc_dir()) { | |
923 | fprintf(stderr, "Continuing without mounted eBPF fs. " | |
924 | "Too old kernel?\n"); | |
925 | return 0; | |
926 | } | |
927 | ||
928 | if (!bpf_get_obj_uid(pathname)) | |
929 | return -1; | |
930 | ||
931 | return 0; | |
6256f8c9 DB |
932 | } |
933 | ||
f6793eec DB |
934 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
935 | uint32_t pinning) | |
936 | { | |
937 | struct bpf_hash_entry *entry; | |
938 | ||
939 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
940 | while (entry && entry->pinning != pinning) | |
941 | entry = entry->next; | |
942 | ||
943 | return entry ? entry->subpath : NULL; | |
944 | } | |
945 | ||
946 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
947 | uint32_t pinning) | |
11c39b5e | 948 | { |
32e93fb7 DB |
949 | switch (pinning) { |
950 | case PIN_OBJECT_NS: | |
951 | case PIN_GLOBAL_NS: | |
952 | return false; | |
953 | case PIN_NONE: | |
32e93fb7 | 954 | return true; |
f6793eec DB |
955 | default: |
956 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
957 | } |
958 | } | |
959 | ||
960 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 961 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
962 | { |
963 | switch (pinning) { | |
964 | case PIN_OBJECT_NS: | |
965 | snprintf(pathname, len, "%s/%s/%s", bpf_get_tc_dir(), | |
966 | bpf_get_obj_uid(NULL), name); | |
967 | break; | |
968 | case PIN_GLOBAL_NS: | |
969 | snprintf(pathname, len, "%s/%s/%s", bpf_get_tc_dir(), | |
970 | BPF_DIR_GLOBALS, name); | |
971 | break; | |
f6793eec DB |
972 | default: |
973 | snprintf(pathname, len, "%s/../%s/%s", bpf_get_tc_dir(), | |
974 | bpf_custom_pinning(ctx, pinning), name); | |
975 | break; | |
32e93fb7 DB |
976 | } |
977 | } | |
978 | ||
f6793eec DB |
979 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
980 | uint32_t pinning) | |
32e93fb7 DB |
981 | { |
982 | char pathname[PATH_MAX]; | |
983 | ||
f6793eec | 984 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_tc_dir()) |
32e93fb7 DB |
985 | return 0; |
986 | ||
f6793eec | 987 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
32e93fb7 DB |
988 | return bpf_obj_get(pathname); |
989 | } | |
990 | ||
f6793eec | 991 | static int bpf_make_obj_path(void) |
32e93fb7 | 992 | { |
f6793eec | 993 | char tmp[PATH_MAX]; |
32e93fb7 DB |
994 | int ret; |
995 | ||
f6793eec DB |
996 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_tc_dir(), |
997 | bpf_get_obj_uid(NULL)); | |
998 | ||
999 | ret = mkdir(tmp, S_IRWXU); | |
1000 | if (ret && errno != EEXIST) { | |
1001 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1002 | return ret; | |
1003 | } | |
1004 | ||
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | static int bpf_make_custom_path(const char *todo) | |
1009 | { | |
1010 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1011 | int ret; | |
1012 | ||
1013 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_tc_dir()); | |
1014 | snprintf(rem, sizeof(rem), "%s/", todo); | |
1015 | sub = strtok(rem, "/"); | |
32e93fb7 | 1016 | |
f6793eec DB |
1017 | while (sub) { |
1018 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1019 | return -EINVAL; | |
1020 | ||
1021 | strcat(tmp, sub); | |
1022 | strcat(tmp, "/"); | |
32e93fb7 | 1023 | |
f6793eec | 1024 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1025 | if (ret && errno != EEXIST) { |
f6793eec | 1026 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1027 | strerror(errno)); |
1028 | return ret; | |
1029 | } | |
f6793eec DB |
1030 | |
1031 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1032 | } |
1033 | ||
f6793eec DB |
1034 | return 0; |
1035 | } | |
1036 | ||
1037 | static int bpf_place_pinned(int fd, const char *name, | |
1038 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1039 | { | |
1040 | char pathname[PATH_MAX]; | |
1041 | const char *tmp; | |
1042 | int ret = 0; | |
1043 | ||
1044 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_tc_dir()) | |
1045 | return 0; | |
1046 | ||
1047 | if (pinning == PIN_OBJECT_NS) | |
1048 | ret = bpf_make_obj_path(); | |
1049 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) | |
1050 | ret = bpf_make_custom_path(tmp); | |
1051 | if (ret < 0) | |
1052 | return ret; | |
1053 | ||
1054 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1055 | return bpf_obj_pin(fd, pathname); |
1056 | } | |
1057 | ||
f31645d1 DB |
1058 | static void bpf_prog_report(int fd, const char *section, |
1059 | const struct bpf_elf_prog *prog, | |
1060 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1061 | { |
f31645d1 DB |
1062 | fprintf(stderr, "Prog section \'%s\' %s%s (%d)!\n", section, |
1063 | fd < 0 ? "rejected: " : "loaded", | |
1064 | fd < 0 ? strerror(errno) : "", | |
1065 | fd < 0 ? errno : fd); | |
1066 | ||
1067 | fprintf(stderr, " - Type: %u\n", prog->type); | |
1068 | fprintf(stderr, " - Instructions: %zu\n", | |
1069 | prog->size / sizeof(struct bpf_insn)); | |
1070 | fprintf(stderr, " - License: %s\n\n", prog->license); | |
1071 | ||
1072 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1073 | } | |
32e93fb7 | 1074 | |
f31645d1 DB |
1075 | static int bpf_prog_attach(const char *section, |
1076 | const struct bpf_elf_prog *prog, | |
1077 | struct bpf_elf_ctx *ctx) | |
1078 | { | |
1079 | int tries = 0, fd; | |
1080 | retry: | |
32e93fb7 DB |
1081 | errno = 0; |
1082 | fd = bpf_prog_load(prog->type, prog->insns, prog->size, | |
f31645d1 DB |
1083 | prog->license, ctx->log, ctx->log_size); |
1084 | if (fd < 0 || ctx->verbose) { | |
1085 | /* The verifier log is pretty chatty, sometimes so chatty | |
1086 | * on larger programs, that we could fail to dump everything | |
1087 | * into our buffer. Still, try to give a debuggable error | |
1088 | * log for the user, so enlarge it and re-fail. | |
1089 | */ | |
1090 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
1091 | if (tries++ < 6 && !bpf_log_realloc(ctx)) | |
1092 | goto retry; | |
1093 | ||
1094 | fprintf(stderr, "Log buffer too small to dump " | |
1095 | "verifier log %zu bytes (%d tries)!\n", | |
1096 | ctx->log_size, tries); | |
1097 | return fd; | |
1098 | } | |
1099 | ||
1100 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1101 | } |
1102 | ||
1103 | return fd; | |
1104 | } | |
1105 | ||
f31645d1 DB |
1106 | static void bpf_map_report(int fd, const char *name, |
1107 | const struct bpf_elf_map *map, | |
1108 | struct bpf_elf_ctx *ctx) | |
1109 | { | |
1110 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1111 | fd < 0 ? "rejected: " : "loaded", | |
1112 | fd < 0 ? strerror(errno) : "", | |
1113 | fd < 0 ? errno : fd); | |
1114 | ||
1115 | fprintf(stderr, " - Type: %u\n", map->type); | |
1116 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1117 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1118 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
1119 | fprintf(stderr, " - Size value: %u\n", map->size_value); | |
1120 | fprintf(stderr, " - Max elems: %u\n\n", map->max_elem); | |
1121 | } | |
1122 | ||
32e93fb7 | 1123 | static int bpf_map_attach(const char *name, const struct bpf_elf_map *map, |
f31645d1 | 1124 | struct bpf_elf_ctx *ctx) |
32e93fb7 DB |
1125 | { |
1126 | int fd, ret; | |
1127 | ||
f6793eec | 1128 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1129 | if (fd > 0) { |
91d88eeb DB |
1130 | ret = bpf_map_selfcheck_pinned(fd, map, |
1131 | offsetof(struct bpf_elf_map, | |
1132 | id)); | |
9e607f2e DB |
1133 | if (ret < 0) { |
1134 | close(fd); | |
1135 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1136 | name); | |
1137 | return ret; | |
1138 | } | |
f31645d1 | 1139 | if (ctx->verbose) |
32e93fb7 DB |
1140 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1141 | name); | |
1142 | return fd; | |
1143 | } | |
1144 | ||
1145 | errno = 0; | |
1146 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
1147 | map->max_elem); | |
f31645d1 DB |
1148 | if (fd < 0 || ctx->verbose) { |
1149 | bpf_map_report(fd, name, map, ctx); | |
32e93fb7 DB |
1150 | if (fd < 0) |
1151 | return fd; | |
1152 | } | |
1153 | ||
f6793eec | 1154 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1155 | if (ret < 0 && errno != EEXIST) { |
1156 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1157 | strerror(errno)); | |
1158 | close(fd); | |
1159 | return ret; | |
1160 | } | |
1161 | ||
1162 | return fd; | |
1163 | } | |
1164 | ||
32e93fb7 DB |
1165 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1166 | const GElf_Sym *sym) | |
1167 | { | |
1168 | return ctx->str_tab->d_buf + sym->st_name; | |
1169 | } | |
1170 | ||
1171 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) | |
1172 | { | |
1173 | GElf_Sym sym; | |
11c39b5e DB |
1174 | int i; |
1175 | ||
32e93fb7 DB |
1176 | for (i = 0; i < ctx->sym_num; i++) { |
1177 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1178 | continue; | |
1179 | ||
5230a2ed DB |
1180 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
1181 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
32e93fb7 DB |
1182 | sym.st_shndx != ctx->sec_maps || |
1183 | sym.st_value / sizeof(struct bpf_elf_map) != which) | |
1184 | continue; | |
1185 | ||
1186 | return bpf_str_tab_name(ctx, &sym); | |
11c39b5e | 1187 | } |
32e93fb7 DB |
1188 | |
1189 | return NULL; | |
11c39b5e DB |
1190 | } |
1191 | ||
32e93fb7 | 1192 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1193 | { |
32e93fb7 DB |
1194 | const char *map_name; |
1195 | int i, fd; | |
11c39b5e | 1196 | |
32e93fb7 DB |
1197 | for (i = 0; i < ctx->map_num; i++) { |
1198 | map_name = bpf_map_fetch_name(ctx, i); | |
1199 | if (!map_name) | |
1200 | return -EIO; | |
11c39b5e | 1201 | |
f31645d1 | 1202 | fd = bpf_map_attach(map_name, &ctx->maps[i], ctx); |
32e93fb7 DB |
1203 | if (fd < 0) |
1204 | return fd; | |
11c39b5e | 1205 | |
32e93fb7 | 1206 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1207 | } |
1208 | ||
1209 | return 0; | |
11c39b5e DB |
1210 | } |
1211 | ||
32e93fb7 DB |
1212 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1213 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1214 | { |
32e93fb7 | 1215 | Elf_Data *sec_edata; |
11c39b5e DB |
1216 | GElf_Shdr sec_hdr; |
1217 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1218 | char *sec_name; |
1219 | ||
32e93fb7 | 1220 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1221 | |
32e93fb7 | 1222 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1223 | if (!sec_fd) |
1224 | return -EINVAL; | |
11c39b5e DB |
1225 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1226 | return -EIO; | |
1227 | ||
32e93fb7 | 1228 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1229 | sec_hdr.sh_name); |
1230 | if (!sec_name || !sec_hdr.sh_size) | |
1231 | return -ENOENT; | |
1232 | ||
1233 | sec_edata = elf_getdata(sec_fd, NULL); | |
1234 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1235 | return -EIO; | |
1236 | ||
32e93fb7 | 1237 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1238 | |
32e93fb7 DB |
1239 | data->sec_name = sec_name; |
1240 | data->sec_data = sec_edata; | |
11c39b5e DB |
1241 | return 0; |
1242 | } | |
1243 | ||
32e93fb7 DB |
1244 | static int bpf_fetch_maps(struct bpf_elf_ctx *ctx, int section, |
1245 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1246 | { |
32e93fb7 DB |
1247 | if (data->sec_data->d_size % sizeof(struct bpf_elf_map) != 0) |
1248 | return -EINVAL; | |
11c39b5e | 1249 | |
32e93fb7 DB |
1250 | ctx->map_num = data->sec_data->d_size / sizeof(struct bpf_elf_map); |
1251 | ctx->sec_maps = section; | |
1252 | ctx->sec_done[section] = true; | |
11c39b5e | 1253 | |
32e93fb7 DB |
1254 | if (ctx->map_num > ARRAY_SIZE(ctx->map_fds)) { |
1255 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); | |
1256 | return -ENOMEM; | |
1257 | } | |
11c39b5e | 1258 | |
32e93fb7 DB |
1259 | memcpy(ctx->maps, data->sec_data->d_buf, data->sec_data->d_size); |
1260 | return 0; | |
1261 | } | |
11c39b5e | 1262 | |
32e93fb7 DB |
1263 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
1264 | struct bpf_elf_sec_data *data) | |
1265 | { | |
1266 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
1267 | return -ENOMEM; | |
11c39b5e | 1268 | |
32e93fb7 DB |
1269 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
1270 | ctx->sec_done[section] = true; | |
1271 | return 0; | |
1272 | } | |
11c39b5e | 1273 | |
32e93fb7 DB |
1274 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
1275 | struct bpf_elf_sec_data *data) | |
1276 | { | |
1277 | ctx->sym_tab = data->sec_data; | |
1278 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
1279 | ctx->sec_done[section] = true; | |
11c39b5e DB |
1280 | return 0; |
1281 | } | |
1282 | ||
32e93fb7 DB |
1283 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
1284 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1285 | { |
32e93fb7 DB |
1286 | ctx->str_tab = data->sec_data; |
1287 | ctx->sec_done[section] = true; | |
1288 | return 0; | |
1289 | } | |
11c39b5e | 1290 | |
32e93fb7 DB |
1291 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx) |
1292 | { | |
1293 | struct bpf_elf_sec_data data; | |
1294 | int i, ret = -1; | |
11c39b5e | 1295 | |
32e93fb7 DB |
1296 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1297 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
1298 | if (ret < 0) |
1299 | continue; | |
1300 | ||
cce3d466 DB |
1301 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1302 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
32e93fb7 | 1303 | ret = bpf_fetch_maps(ctx, i, &data); |
cce3d466 DB |
1304 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1305 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 1306 | ret = bpf_fetch_license(ctx, i, &data); |
cce3d466 DB |
1307 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
1308 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
1309 | ret = bpf_fetch_symtab(ctx, i, &data); |
1310 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 1311 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 DB |
1312 | ret = bpf_fetch_strtab(ctx, i, &data); |
1313 | if (ret < 0) { | |
1314 | fprintf(stderr, "Error parsing section %d! Perhaps" | |
1315 | "check with readelf -a?\n", i); | |
1316 | break; | |
11c39b5e | 1317 | } |
32e93fb7 DB |
1318 | } |
1319 | ||
1320 | if (ctx->sym_tab && ctx->str_tab && ctx->sec_maps) { | |
1321 | ret = bpf_maps_attach_all(ctx); | |
1322 | if (ret < 0) { | |
1323 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
1324 | return ret; | |
11c39b5e DB |
1325 | } |
1326 | } | |
1327 | ||
1328 | return ret; | |
1329 | } | |
1330 | ||
32e93fb7 | 1331 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section) |
11c39b5e | 1332 | { |
32e93fb7 DB |
1333 | struct bpf_elf_sec_data data; |
1334 | struct bpf_elf_prog prog; | |
1335 | int ret, i, fd = -1; | |
11c39b5e | 1336 | |
32e93fb7 DB |
1337 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1338 | if (ctx->sec_done[i]) | |
11c39b5e DB |
1339 | continue; |
1340 | ||
32e93fb7 | 1341 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
1342 | if (ret < 0 || |
1343 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
1344 | data.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1345 | !strcmp(data.sec_name, section))) | |
11c39b5e DB |
1346 | continue; |
1347 | ||
32e93fb7 DB |
1348 | memset(&prog, 0, sizeof(prog)); |
1349 | prog.type = ctx->type; | |
1350 | prog.insns = data.sec_data->d_buf; | |
1351 | prog.size = data.sec_data->d_size; | |
1352 | prog.license = ctx->license; | |
11c39b5e | 1353 | |
f31645d1 | 1354 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1355 | if (fd < 0) |
11c39b5e DB |
1356 | continue; |
1357 | ||
32e93fb7 | 1358 | ctx->sec_done[i] = true; |
11c39b5e DB |
1359 | break; |
1360 | } | |
1361 | ||
32e93fb7 | 1362 | return fd; |
11c39b5e DB |
1363 | } |
1364 | ||
32e93fb7 DB |
1365 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
1366 | struct bpf_elf_sec_data *data_relo, | |
1367 | struct bpf_elf_sec_data *data_insn) | |
11c39b5e | 1368 | { |
32e93fb7 DB |
1369 | Elf_Data *idata = data_insn->sec_data; |
1370 | GElf_Shdr *rhdr = &data_relo->sec_hdr; | |
1371 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
1372 | struct bpf_insn *insns = idata->d_buf; | |
1373 | unsigned int num_insns = idata->d_size / sizeof(*insns); | |
11c39b5e | 1374 | |
32e93fb7 DB |
1375 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
1376 | unsigned int ioff, rmap; | |
1377 | GElf_Rel relo; | |
1378 | GElf_Sym sym; | |
1379 | ||
1380 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
1381 | return -EIO; | |
1382 | ||
1383 | ioff = relo.r_offset / sizeof(struct bpf_insn); | |
1384 | if (ioff >= num_insns || | |
a576c6b9 DB |
1385 | insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) { |
1386 | fprintf(stderr, "ELF contains relo data for non ld64 " | |
1387 | "instruction at offset %u! Compiler bug?!\n", | |
1388 | ioff); | |
1389 | if (ioff < num_insns && | |
1390 | insns[ioff].code == (BPF_JMP | BPF_CALL)) | |
1391 | fprintf(stderr, " - Try to annotate functions " | |
1392 | "with always_inline attribute!\n"); | |
32e93fb7 | 1393 | return -EINVAL; |
a576c6b9 | 1394 | } |
32e93fb7 DB |
1395 | |
1396 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) | |
1397 | return -EIO; | |
2486337a DB |
1398 | if (sym.st_shndx != ctx->sec_maps) { |
1399 | fprintf(stderr, "ELF contains non-map related relo data in " | |
1400 | "entry %u pointing to section %u! Compiler bug?!\n", | |
1401 | relo_ent, sym.st_shndx); | |
1402 | return -EIO; | |
1403 | } | |
32e93fb7 DB |
1404 | |
1405 | rmap = sym.st_value / sizeof(struct bpf_elf_map); | |
1406 | if (rmap >= ARRAY_SIZE(ctx->map_fds)) | |
1407 | return -EINVAL; | |
1408 | if (!ctx->map_fds[rmap]) | |
1409 | return -EINVAL; | |
1410 | ||
1411 | if (ctx->verbose) | |
1412 | fprintf(stderr, "Map \'%s\' (%d) injected into prog " | |
1413 | "section \'%s\' at offset %u!\n", | |
1414 | bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap], | |
1415 | data_insn->sec_name, ioff); | |
11c39b5e | 1416 | |
32e93fb7 DB |
1417 | insns[ioff].src_reg = BPF_PSEUDO_MAP_FD; |
1418 | insns[ioff].imm = ctx->map_fds[rmap]; | |
1419 | } | |
1420 | ||
1421 | return 0; | |
1422 | } | |
1423 | ||
1424 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section) | |
1425 | { | |
1426 | struct bpf_elf_sec_data data_relo, data_insn; | |
1427 | struct bpf_elf_prog prog; | |
1428 | int ret, idx, i, fd = -1; | |
1429 | ||
1430 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
1431 | ret = bpf_fill_section_data(ctx, i, &data_relo); | |
1432 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
1433 | continue; |
1434 | ||
32e93fb7 DB |
1435 | idx = data_relo.sec_hdr.sh_info; |
1436 | ret = bpf_fill_section_data(ctx, idx, &data_insn); | |
cce3d466 DB |
1437 | if (ret < 0 || |
1438 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
1439 | data_insn.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1440 | !strcmp(data_insn.sec_name, section))) | |
11c39b5e | 1441 | continue; |
32e93fb7 DB |
1442 | |
1443 | ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn); | |
1444 | if (ret < 0) | |
11c39b5e DB |
1445 | continue; |
1446 | ||
32e93fb7 DB |
1447 | memset(&prog, 0, sizeof(prog)); |
1448 | prog.type = ctx->type; | |
1449 | prog.insns = data_insn.sec_data->d_buf; | |
1450 | prog.size = data_insn.sec_data->d_size; | |
1451 | prog.license = ctx->license; | |
1452 | ||
f31645d1 | 1453 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1454 | if (fd < 0) |
11c39b5e DB |
1455 | continue; |
1456 | ||
32e93fb7 DB |
1457 | ctx->sec_done[i] = true; |
1458 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
1459 | break; |
1460 | } | |
1461 | ||
32e93fb7 | 1462 | return fd; |
11c39b5e DB |
1463 | } |
1464 | ||
32e93fb7 | 1465 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 DB |
1466 | { |
1467 | int ret = -1; | |
1468 | ||
32e93fb7 DB |
1469 | if (ctx->sym_tab) |
1470 | ret = bpf_fetch_prog_relo(ctx, section); | |
473d7840 | 1471 | if (ret < 0) |
32e93fb7 DB |
1472 | ret = bpf_fetch_prog(ctx, section); |
1473 | ||
473d7840 DB |
1474 | return ret; |
1475 | } | |
1476 | ||
910b543d DB |
1477 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
1478 | { | |
1479 | int i; | |
1480 | ||
1481 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
1482 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
1483 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
1484 | return i; | |
1485 | return -1; | |
1486 | } | |
1487 | ||
32e93fb7 | 1488 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 1489 | { |
32e93fb7 DB |
1490 | struct bpf_elf_sec_data data; |
1491 | uint32_t map_id, key_id; | |
910b543d | 1492 | int fd, i, ret, idx; |
473d7840 | 1493 | |
32e93fb7 DB |
1494 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1495 | if (ctx->sec_done[i]) | |
473d7840 DB |
1496 | continue; |
1497 | ||
32e93fb7 | 1498 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
1499 | if (ret < 0) |
1500 | continue; | |
1501 | ||
910b543d DB |
1502 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
1503 | if (ret != 2) | |
32e93fb7 | 1504 | continue; |
910b543d DB |
1505 | |
1506 | idx = bpf_find_map_by_id(ctx, map_id); | |
1507 | if (idx < 0) | |
473d7840 DB |
1508 | continue; |
1509 | ||
32e93fb7 DB |
1510 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
1511 | if (fd < 0) | |
473d7840 DB |
1512 | return -EIO; |
1513 | ||
910b543d DB |
1514 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
1515 | &fd, BPF_ANY); | |
473d7840 DB |
1516 | if (ret < 0) |
1517 | return -ENOENT; | |
1518 | ||
32e93fb7 | 1519 | ctx->sec_done[i] = true; |
473d7840 DB |
1520 | } |
1521 | ||
1522 | return 0; | |
1523 | } | |
1524 | ||
32e93fb7 | 1525 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 1526 | { |
32e93fb7 DB |
1527 | struct stat st; |
1528 | int ret; | |
11c39b5e | 1529 | |
32e93fb7 | 1530 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 1531 | |
32e93fb7 DB |
1532 | ret = fstat(ctx->obj_fd, &st); |
1533 | if (ret < 0) { | |
1534 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
1535 | strerror(errno)); | |
1536 | return; | |
1537 | } | |
11c39b5e | 1538 | |
32e93fb7 DB |
1539 | ctx->stat.st_dev = st.st_dev; |
1540 | ctx->stat.st_ino = st.st_ino; | |
1541 | } | |
1542 | ||
f6793eec DB |
1543 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
1544 | { | |
1545 | char buff[PATH_MAX]; | |
1546 | ||
1547 | while (fgets(buff, sizeof(buff), fp)) { | |
1548 | char *ptr = buff; | |
1549 | ||
1550 | while (*ptr == ' ' || *ptr == '\t') | |
1551 | ptr++; | |
1552 | ||
1553 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
1554 | continue; | |
1555 | ||
1556 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
1557 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
1558 | strcpy(path, ptr); | |
1559 | return -1; | |
1560 | } | |
1561 | ||
1562 | return 1; | |
1563 | } | |
1564 | ||
1565 | return 0; | |
1566 | } | |
1567 | ||
1568 | static bool bpf_pinning_reserved(uint32_t pinning) | |
1569 | { | |
1570 | switch (pinning) { | |
1571 | case PIN_NONE: | |
1572 | case PIN_OBJECT_NS: | |
1573 | case PIN_GLOBAL_NS: | |
1574 | return true; | |
1575 | default: | |
1576 | return false; | |
1577 | } | |
1578 | } | |
1579 | ||
1580 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
1581 | { | |
1582 | struct bpf_hash_entry *entry; | |
1583 | char subpath[PATH_MAX]; | |
1584 | uint32_t pinning; | |
1585 | FILE *fp; | |
1586 | int ret; | |
1587 | ||
1588 | fp = fopen(db_file, "r"); | |
1589 | if (!fp) | |
1590 | return; | |
1591 | ||
1592 | memset(subpath, 0, sizeof(subpath)); | |
1593 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { | |
1594 | if (ret == -1) { | |
1595 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
1596 | db_file, subpath); | |
1597 | fclose(fp); | |
1598 | return; | |
1599 | } | |
1600 | ||
1601 | if (bpf_pinning_reserved(pinning)) { | |
1602 | fprintf(stderr, "Database %s, id %u is reserved - " | |
1603 | "ignoring!\n", db_file, pinning); | |
1604 | continue; | |
1605 | } | |
1606 | ||
1607 | entry = malloc(sizeof(*entry)); | |
1608 | if (!entry) { | |
1609 | fprintf(stderr, "No memory left for db entry!\n"); | |
1610 | continue; | |
1611 | } | |
1612 | ||
1613 | entry->pinning = pinning; | |
1614 | entry->subpath = strdup(subpath); | |
1615 | if (!entry->subpath) { | |
1616 | fprintf(stderr, "No memory left for db entry!\n"); | |
1617 | free(entry); | |
1618 | continue; | |
1619 | } | |
1620 | ||
1621 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1622 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
1623 | } | |
1624 | ||
1625 | fclose(fp); | |
1626 | } | |
1627 | ||
1628 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
1629 | { | |
1630 | struct bpf_hash_entry *entry; | |
1631 | int i; | |
1632 | ||
1633 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
1634 | while ((entry = ctx->ht[i]) != NULL) { | |
1635 | ctx->ht[i] = entry->next; | |
1636 | free((char *)entry->subpath); | |
1637 | free(entry); | |
1638 | } | |
1639 | } | |
1640 | } | |
1641 | ||
8187b012 DB |
1642 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
1643 | { | |
1644 | if (ctx->elf_hdr.e_type != ET_REL || | |
1645 | ctx->elf_hdr.e_machine != 0 || | |
1646 | ctx->elf_hdr.e_version != EV_CURRENT) { | |
1647 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
1648 | return -EINVAL; | |
1649 | } | |
1650 | ||
1651 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
1652 | default: | |
1653 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
1654 | return -EINVAL; | |
1655 | case ELFDATA2LSB: | |
1656 | if (htons(1) == 1) { | |
1657 | fprintf(stderr, | |
1658 | "We are big endian, eBPF object is little endian!\n"); | |
1659 | return -EIO; | |
1660 | } | |
1661 | break; | |
1662 | case ELFDATA2MSB: | |
1663 | if (htons(1) != 1) { | |
1664 | fprintf(stderr, | |
1665 | "We are little endian, eBPF object is big endian!\n"); | |
1666 | return -EIO; | |
1667 | } | |
1668 | break; | |
1669 | } | |
1670 | ||
1671 | return 0; | |
1672 | } | |
1673 | ||
32e93fb7 DB |
1674 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
1675 | enum bpf_prog_type type, bool verbose) | |
1676 | { | |
1677 | int ret = -EINVAL; | |
1678 | ||
1679 | if (elf_version(EV_CURRENT) == EV_NONE || | |
1680 | bpf_init_env(pathname)) | |
1681 | return ret; | |
1682 | ||
1683 | memset(ctx, 0, sizeof(*ctx)); | |
1684 | ctx->verbose = verbose; | |
1685 | ctx->type = type; | |
1686 | ||
1687 | ctx->obj_fd = open(pathname, O_RDONLY); | |
1688 | if (ctx->obj_fd < 0) | |
1689 | return ctx->obj_fd; | |
1690 | ||
1691 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
1692 | if (!ctx->elf_fd) { | |
11c39b5e | 1693 | ret = -EINVAL; |
32e93fb7 | 1694 | goto out_fd; |
11c39b5e DB |
1695 | } |
1696 | ||
8187b012 DB |
1697 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
1698 | ret = -EINVAL; | |
1699 | goto out_fd; | |
1700 | } | |
1701 | ||
32e93fb7 DB |
1702 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
1703 | &ctx->elf_hdr) { | |
11c39b5e DB |
1704 | ret = -EIO; |
1705 | goto out_elf; | |
1706 | } | |
1707 | ||
8187b012 DB |
1708 | ret = bpf_elf_check_ehdr(ctx); |
1709 | if (ret < 0) | |
1710 | goto out_elf; | |
1711 | ||
32e93fb7 DB |
1712 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
1713 | sizeof(*(ctx->sec_done))); | |
1714 | if (!ctx->sec_done) { | |
11c39b5e DB |
1715 | ret = -ENOMEM; |
1716 | goto out_elf; | |
1717 | } | |
1718 | ||
f31645d1 DB |
1719 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
1720 | ret = -ENOMEM; | |
1721 | goto out_free; | |
1722 | } | |
1723 | ||
32e93fb7 | 1724 | bpf_save_finfo(ctx); |
f6793eec DB |
1725 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
1726 | ||
32e93fb7 | 1727 | return 0; |
f31645d1 DB |
1728 | out_free: |
1729 | free(ctx->sec_done); | |
32e93fb7 DB |
1730 | out_elf: |
1731 | elf_end(ctx->elf_fd); | |
1732 | out_fd: | |
1733 | close(ctx->obj_fd); | |
1734 | return ret; | |
1735 | } | |
d937a74b | 1736 | |
32e93fb7 DB |
1737 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
1738 | { | |
1739 | int i, count = 0; | |
11c39b5e | 1740 | |
32e93fb7 DB |
1741 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
1742 | if (!ctx->map_fds[i]) | |
1743 | break; | |
1744 | count++; | |
1745 | } | |
473d7840 | 1746 | |
32e93fb7 DB |
1747 | return count; |
1748 | } | |
6256f8c9 | 1749 | |
32e93fb7 DB |
1750 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
1751 | { | |
1752 | int i; | |
1753 | ||
1754 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
1755 | if (ctx->map_fds[i]) | |
1756 | close(ctx->map_fds[i]); | |
473d7840 | 1757 | } |
32e93fb7 DB |
1758 | } |
1759 | ||
1760 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
1761 | { | |
1762 | if (failure) | |
1763 | bpf_maps_teardown(ctx); | |
473d7840 | 1764 | |
f6793eec | 1765 | bpf_hash_destroy(ctx); |
f31645d1 | 1766 | |
32e93fb7 | 1767 | free(ctx->sec_done); |
f31645d1 DB |
1768 | free(ctx->log); |
1769 | ||
32e93fb7 DB |
1770 | elf_end(ctx->elf_fd); |
1771 | close(ctx->obj_fd); | |
1772 | } | |
6256f8c9 | 1773 | |
32e93fb7 | 1774 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 1775 | |
32e93fb7 DB |
1776 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
1777 | const char *section, bool verbose) | |
1778 | { | |
1779 | struct bpf_elf_ctx *ctx = &__ctx; | |
1780 | int fd = 0, ret; | |
6256f8c9 | 1781 | |
32e93fb7 DB |
1782 | ret = bpf_elf_ctx_init(ctx, pathname, type, verbose); |
1783 | if (ret < 0) { | |
1784 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
1785 | return ret; | |
1786 | } | |
6256f8c9 | 1787 | |
32e93fb7 DB |
1788 | ret = bpf_fetch_ancillary(ctx); |
1789 | if (ret < 0) { | |
1790 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
1791 | goto out; | |
1792 | } | |
1793 | ||
1794 | fd = bpf_fetch_prog_sec(ctx, section); | |
1795 | if (fd < 0) { | |
1796 | fprintf(stderr, "Error fetching program/map!\n"); | |
1797 | ret = fd; | |
1798 | goto out; | |
1799 | } | |
1800 | ||
1801 | ret = bpf_fill_prog_arrays(ctx); | |
1802 | if (ret < 0) | |
1803 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 1804 | out: |
32e93fb7 DB |
1805 | bpf_elf_ctx_destroy(ctx, ret < 0); |
1806 | if (ret < 0) { | |
1807 | if (fd) | |
1808 | close(fd); | |
1809 | return ret; | |
1810 | } | |
1811 | ||
1812 | return fd; | |
6256f8c9 | 1813 | } |
11c39b5e | 1814 | |
6256f8c9 | 1815 | static int |
4bd62446 DB |
1816 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
1817 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 DB |
1818 | { |
1819 | struct bpf_map_set_msg msg; | |
1820 | int *cmsg_buf, min_fd; | |
1821 | char *amsg_buf; | |
1822 | int i; | |
1823 | ||
1824 | memset(&msg, 0, sizeof(msg)); | |
1825 | ||
1826 | msg.aux.uds_ver = BPF_SCM_AUX_VER; | |
4bd62446 | 1827 | msg.aux.num_ent = entries; |
6256f8c9 DB |
1828 | |
1829 | strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); | |
1830 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); | |
1831 | ||
1832 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
1833 | amsg_buf = (char *)msg.aux.ent; | |
1834 | ||
4bd62446 | 1835 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
1836 | int ret; |
1837 | ||
4bd62446 | 1838 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
1839 | bpf_map_set_init_single(&msg, min_fd); |
1840 | ||
1841 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
1842 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
1843 | ||
1844 | ret = sendmsg(fd, &msg.hdr, 0); | |
1845 | if (ret <= 0) | |
1846 | return ret ? : -1; | |
1847 | } | |
1848 | ||
1849 | return 0; | |
11c39b5e DB |
1850 | } |
1851 | ||
4bd62446 DB |
1852 | static int |
1853 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
1854 | unsigned int entries) | |
1855 | { | |
1856 | struct bpf_map_set_msg msg; | |
1857 | int *cmsg_buf, min_fd; | |
1858 | char *amsg_buf, *mmsg_buf; | |
1859 | unsigned int needed = 1; | |
1860 | int i; | |
1861 | ||
1862 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
1863 | amsg_buf = (char *)msg.aux.ent; | |
1864 | mmsg_buf = (char *)&msg.aux; | |
1865 | ||
1866 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
1867 | struct cmsghdr *cmsg; | |
1868 | int ret; | |
1869 | ||
1870 | min_fd = min(entries, entries - i); | |
1871 | bpf_map_set_init_single(&msg, min_fd); | |
1872 | ||
1873 | ret = recvmsg(fd, &msg.hdr, 0); | |
1874 | if (ret <= 0) | |
1875 | return ret ? : -1; | |
1876 | ||
1877 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
1878 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
1879 | return -EINVAL; | |
1880 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
1881 | return -EIO; | |
1882 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
1883 | return -ENOSYS; | |
1884 | ||
1885 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
1886 | if (min_fd > entries || min_fd <= 0) | |
1887 | return -EINVAL; | |
1888 | ||
1889 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
1890 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
1891 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
1892 | ||
1893 | needed = aux->num_ent; | |
1894 | } | |
1895 | ||
1896 | return 0; | |
1897 | } | |
1898 | ||
1899 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 1900 | { |
32e93fb7 | 1901 | struct bpf_elf_ctx *ctx = &__ctx; |
6256f8c9 DB |
1902 | struct sockaddr_un addr; |
1903 | struct bpf_map_data bpf_aux; | |
1904 | int fd, ret; | |
1905 | ||
1906 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
1907 | if (fd < 0) { | |
1908 | fprintf(stderr, "Cannot open socket: %s\n", | |
1909 | strerror(errno)); | |
1910 | return -1; | |
1911 | } | |
1912 | ||
1913 | memset(&addr, 0, sizeof(addr)); | |
1914 | addr.sun_family = AF_UNIX; | |
1915 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); | |
1916 | ||
1917 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
1918 | if (ret < 0) { | |
1919 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
1920 | path, strerror(errno)); | |
1921 | return -1; | |
1922 | } | |
1923 | ||
1924 | memset(&bpf_aux, 0, sizeof(bpf_aux)); | |
1925 | ||
32e93fb7 DB |
1926 | bpf_aux.fds = ctx->map_fds; |
1927 | bpf_aux.ent = ctx->maps; | |
1928 | bpf_aux.st = &ctx->stat; | |
6256f8c9 | 1929 | bpf_aux.obj = obj; |
6256f8c9 | 1930 | |
4bd62446 | 1931 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 1932 | bpf_maps_count(ctx)); |
6256f8c9 | 1933 | if (ret < 0) |
4bd62446 DB |
1934 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
1935 | path, strerror(errno)); | |
1936 | ||
32e93fb7 | 1937 | bpf_maps_teardown(ctx); |
4bd62446 DB |
1938 | close(fd); |
1939 | return ret; | |
1940 | } | |
1941 | ||
1942 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
1943 | unsigned int entries) | |
1944 | { | |
1945 | struct sockaddr_un addr; | |
1946 | int fd, ret; | |
1947 | ||
1948 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
1949 | if (fd < 0) { | |
1950 | fprintf(stderr, "Cannot open socket: %s\n", | |
1951 | strerror(errno)); | |
1952 | return -1; | |
1953 | } | |
1954 | ||
1955 | memset(&addr, 0, sizeof(addr)); | |
1956 | addr.sun_family = AF_UNIX; | |
1957 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); | |
1958 | ||
1959 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
1960 | if (ret < 0) { | |
1961 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
1962 | strerror(errno)); | |
1963 | return -1; | |
1964 | } | |
1965 | ||
1966 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
1967 | if (ret < 0) | |
1968 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
1969 | path, strerror(errno)); |
1970 | ||
4bd62446 | 1971 | unlink(addr.sun_path); |
6256f8c9 DB |
1972 | close(fd); |
1973 | return ret; | |
1974 | } | |
11c39b5e | 1975 | #endif /* HAVE_ELF */ |