]>
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> | |
5c5a0f3d | 23 | #include <limits.h> |
1d129d19 | 24 | |
11c39b5e DB |
25 | #ifdef HAVE_ELF |
26 | #include <libelf.h> | |
27 | #include <gelf.h> | |
28 | #endif | |
29 | ||
32e93fb7 DB |
30 | #include <sys/types.h> |
31 | #include <sys/stat.h> | |
32 | #include <sys/un.h> | |
33 | #include <sys/vfs.h> | |
34 | #include <sys/mount.h> | |
35 | #include <sys/syscall.h> | |
36 | #include <sys/sendfile.h> | |
37 | #include <sys/resource.h> | |
38 | ||
39 | #include <linux/bpf.h> | |
40 | #include <linux/filter.h> | |
41 | #include <linux/if_alg.h> | |
42 | ||
8187b012 DB |
43 | #include <arpa/inet.h> |
44 | ||
1d129d19 | 45 | #include "utils.h" |
6256f8c9 DB |
46 | |
47 | #include "bpf_elf.h" | |
48 | #include "bpf_scm.h" | |
49 | ||
1d129d19 JP |
50 | #include "tc_util.h" |
51 | #include "tc_bpf.h" | |
52 | ||
67584e3a ND |
53 | #ifndef AF_ALG |
54 | #define AF_ALG 38 | |
55 | #endif | |
56 | ||
32e93fb7 DB |
57 | #ifdef HAVE_ELF |
58 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
59 | const char *sec, bool verbose); | |
60 | #else | |
61 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
62 | const char *sec, bool verbose) | |
63 | { | |
64 | fprintf(stderr, "No ELF library support compiled in.\n"); | |
65 | errno = ENOSYS; | |
66 | return -1; | |
67 | } | |
68 | #endif | |
69 | ||
70 | static inline __u64 bpf_ptr_to_u64(const void *ptr) | |
71 | { | |
72 | return (__u64)(unsigned long)ptr; | |
73 | } | |
74 | ||
75 | static int bpf(int cmd, union bpf_attr *attr, unsigned int size) | |
76 | { | |
77 | #ifdef __NR_bpf | |
78 | return syscall(__NR_bpf, cmd, attr, size); | |
79 | #else | |
80 | fprintf(stderr, "No bpf syscall, kernel headers too old?\n"); | |
81 | errno = ENOSYS; | |
82 | return -1; | |
83 | #endif | |
84 | } | |
85 | ||
91d88eeb DB |
86 | static int bpf_map_update(int fd, const void *key, const void *value, |
87 | uint64_t flags) | |
32e93fb7 | 88 | { |
67584e3a ND |
89 | union bpf_attr attr; |
90 | ||
91 | memset(&attr, 0, sizeof(attr)); | |
92 | attr.map_fd = fd; | |
93 | attr.key = bpf_ptr_to_u64(key); | |
94 | attr.value = bpf_ptr_to_u64(value); | |
95 | attr.flags = flags; | |
32e93fb7 | 96 | |
91d88eeb | 97 | return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
32e93fb7 DB |
98 | } |
99 | ||
100 | static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len, | |
101 | char **bpf_string, bool *need_release, | |
102 | const char separator) | |
1d129d19 JP |
103 | { |
104 | char sp; | |
105 | ||
106 | if (from_file) { | |
107 | size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,"); | |
108 | char *tmp_string; | |
109 | FILE *fp; | |
110 | ||
111 | tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len; | |
112 | tmp_string = malloc(tmp_len); | |
113 | if (tmp_string == NULL) | |
114 | return -ENOMEM; | |
115 | ||
116 | memset(tmp_string, 0, tmp_len); | |
117 | ||
118 | fp = fopen(arg, "r"); | |
119 | if (fp == NULL) { | |
120 | perror("Cannot fopen"); | |
121 | free(tmp_string); | |
122 | return -ENOENT; | |
123 | } | |
124 | ||
125 | if (!fgets(tmp_string, tmp_len, fp)) { | |
126 | free(tmp_string); | |
127 | fclose(fp); | |
128 | return -EIO; | |
129 | } | |
130 | ||
131 | fclose(fp); | |
132 | ||
133 | *need_release = true; | |
134 | *bpf_string = tmp_string; | |
135 | } else { | |
136 | *need_release = false; | |
137 | *bpf_string = arg; | |
138 | } | |
139 | ||
140 | if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 || | |
141 | sp != separator) { | |
142 | if (*need_release) | |
143 | free(*bpf_string); | |
144 | return -EINVAL; | |
145 | } | |
146 | ||
147 | return 0; | |
148 | } | |
149 | ||
32e93fb7 DB |
150 | static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops, |
151 | bool from_file) | |
1d129d19 JP |
152 | { |
153 | char *bpf_string, *token, separator = ','; | |
154 | int ret = 0, i = 0; | |
155 | bool need_release; | |
156 | __u16 bpf_len = 0; | |
157 | ||
158 | if (argc < 1) | |
159 | return -EINVAL; | |
160 | if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string, | |
161 | &need_release, separator)) | |
162 | return -EINVAL; | |
163 | if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) { | |
164 | ret = -EINVAL; | |
165 | goto out; | |
166 | } | |
167 | ||
168 | token = bpf_string; | |
169 | while ((token = strchr(token, separator)) && (++token)[0]) { | |
170 | if (i >= bpf_len) { | |
32a121cb | 171 | fprintf(stderr, "Real program length exceeds encoded length parameter!\n"); |
1d129d19 JP |
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) { | |
afc1a200 | 188 | fprintf(stderr, "Parsed program length is less than encoded length parameter!\n"); |
1d129d19 JP |
189 | ret = -EINVAL; |
190 | goto out; | |
191 | } | |
192 | ret = bpf_len; | |
1d129d19 JP |
193 | out: |
194 | if (need_release) | |
195 | free(bpf_string); | |
196 | ||
197 | return ret; | |
198 | } | |
199 | ||
200 | void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len) | |
201 | { | |
202 | struct sock_filter *ops = (struct sock_filter *) RTA_DATA(bpf_ops); | |
203 | int i; | |
204 | ||
205 | if (len == 0) | |
206 | return; | |
207 | ||
208 | fprintf(f, "bytecode \'%u,", len); | |
209 | ||
210 | for (i = 0; i < len - 1; i++) | |
211 | fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt, | |
212 | ops[i].jf, ops[i].k); | |
213 | ||
6256f8c9 | 214 | fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt, |
1d129d19 JP |
215 | ops[i].jf, ops[i].k); |
216 | } | |
11c39b5e | 217 | |
afc1a200 DB |
218 | static void bpf_map_pin_report(const struct bpf_elf_map *pin, |
219 | const struct bpf_elf_map *obj) | |
220 | { | |
221 | fprintf(stderr, "Map specification differs from pinned file!\n"); | |
222 | ||
223 | if (obj->type != pin->type) | |
224 | fprintf(stderr, " - Type: %u (obj) != %u (pin)\n", | |
225 | obj->type, pin->type); | |
226 | if (obj->size_key != pin->size_key) | |
227 | fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n", | |
228 | obj->size_key, pin->size_key); | |
229 | if (obj->size_value != pin->size_value) | |
230 | fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n", | |
231 | obj->size_value, pin->size_value); | |
232 | if (obj->max_elem != pin->max_elem) | |
233 | fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n", | |
234 | obj->max_elem, pin->max_elem); | |
4dd3f50a DB |
235 | if (obj->flags != pin->flags) |
236 | fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n", | |
237 | obj->flags, pin->flags); | |
afc1a200 DB |
238 | |
239 | fprintf(stderr, "\n"); | |
240 | } | |
241 | ||
91d88eeb DB |
242 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, |
243 | int length) | |
9e607f2e DB |
244 | { |
245 | char file[PATH_MAX], buff[4096]; | |
246 | struct bpf_elf_map tmp, zero; | |
247 | unsigned int val; | |
248 | FILE *fp; | |
249 | ||
250 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
251 | ||
252 | fp = fopen(file, "r"); | |
253 | if (!fp) { | |
254 | fprintf(stderr, "No procfs support?!\n"); | |
255 | return -EIO; | |
256 | } | |
257 | ||
258 | memset(&tmp, 0, sizeof(tmp)); | |
259 | while (fgets(buff, sizeof(buff), fp)) { | |
260 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
261 | tmp.type = val; | |
262 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
263 | tmp.size_key = val; | |
264 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
265 | tmp.size_value = val; | |
266 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
267 | tmp.max_elem = val; | |
4dd3f50a DB |
268 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) |
269 | tmp.flags = val; | |
9e607f2e DB |
270 | } |
271 | ||
272 | fclose(fp); | |
273 | ||
91d88eeb | 274 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
275 | return 0; |
276 | } else { | |
277 | memset(&zero, 0, sizeof(zero)); | |
278 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, | |
279 | * so just accept it. We know we do have an eBPF fd and in this | |
280 | * case, everything is 0. It is guaranteed that no such map exists | |
281 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
282 | */ | |
91d88eeb | 283 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
284 | return 0; |
285 | ||
afc1a200 | 286 | bpf_map_pin_report(&tmp, map); |
9e607f2e DB |
287 | return -EINVAL; |
288 | } | |
289 | } | |
290 | ||
91d88eeb DB |
291 | static int bpf_mnt_fs(const char *target) |
292 | { | |
293 | bool bind_done = false; | |
294 | ||
295 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
296 | if (errno != EINVAL || bind_done) { | |
297 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
298 | target, strerror(errno)); | |
299 | return -1; | |
300 | } | |
301 | ||
302 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
303 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
304 | target, target, strerror(errno)); | |
305 | return -1; | |
306 | } | |
307 | ||
308 | bind_done = true; | |
309 | } | |
310 | ||
311 | if (mount("bpf", target, "bpf", 0, NULL)) { | |
312 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", | |
313 | target, strerror(errno)); | |
314 | return -1; | |
315 | } | |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
32e93fb7 DB |
320 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
321 | { | |
322 | struct statfs st_fs; | |
323 | ||
324 | if (statfs(mnt, &st_fs) < 0) | |
325 | return -ENOENT; | |
326 | if ((unsigned long)st_fs.f_type != magic) | |
327 | return -ENOENT; | |
328 | ||
329 | return 0; | |
330 | } | |
331 | ||
332 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, | |
333 | char *mnt, int len, | |
334 | const char * const *known_mnts) | |
335 | { | |
336 | const char * const *ptr; | |
337 | char type[100]; | |
338 | FILE *fp; | |
339 | ||
340 | if (known_mnts) { | |
341 | ptr = known_mnts; | |
342 | while (*ptr) { | |
343 | if (bpf_valid_mntpt(*ptr, magic) == 0) { | |
344 | strncpy(mnt, *ptr, len - 1); | |
345 | mnt[len - 1] = 0; | |
346 | return mnt; | |
347 | } | |
348 | ptr++; | |
349 | } | |
350 | } | |
351 | ||
352 | fp = fopen("/proc/mounts", "r"); | |
353 | if (fp == NULL || len != PATH_MAX) | |
354 | return NULL; | |
355 | ||
356 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
357 | mnt, type) == 2) { | |
358 | if (strcmp(type, fstype) == 0) | |
359 | break; | |
360 | } | |
361 | ||
362 | fclose(fp); | |
363 | if (strcmp(type, fstype) != 0) | |
364 | return NULL; | |
365 | ||
366 | return mnt; | |
367 | } | |
368 | ||
369 | int bpf_trace_pipe(void) | |
370 | { | |
371 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
372 | static const char * const tracefs_known_mnts[] = { | |
373 | TRACE_DIR_MNT, | |
374 | "/sys/kernel/debug/tracing", | |
375 | "/tracing", | |
376 | "/trace", | |
377 | 0, | |
378 | }; | |
379 | char tpipe[PATH_MAX]; | |
380 | const char *mnt; | |
381 | int fd; | |
382 | ||
383 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
384 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
385 | if (!mnt) { | |
386 | fprintf(stderr, "tracefs not mounted?\n"); | |
387 | return -1; | |
388 | } | |
389 | ||
390 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
391 | ||
392 | fd = open(tpipe, O_RDONLY); | |
393 | if (fd < 0) | |
394 | return -1; | |
395 | ||
396 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
397 | while (1) { | |
398 | static char buff[4096]; | |
399 | ssize_t ret; | |
400 | ||
401 | ret = read(fd, buff, sizeof(buff) - 1); | |
402 | if (ret > 0) { | |
403 | write(2, buff, ret); | |
404 | fflush(stderr); | |
405 | } | |
406 | } | |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
91d88eeb DB |
411 | static const char *bpf_get_tc_dir(void) |
412 | { | |
32a121cb | 413 | static bool bpf_mnt_cached; |
91d88eeb DB |
414 | static char bpf_tc_dir[PATH_MAX]; |
415 | static const char *mnt; | |
416 | static const char * const bpf_known_mnts[] = { | |
417 | BPF_DIR_MNT, | |
418 | 0, | |
419 | }; | |
420 | char bpf_mnt[PATH_MAX] = BPF_DIR_MNT; | |
421 | char bpf_glo_dir[PATH_MAX]; | |
422 | int ret; | |
423 | ||
424 | if (bpf_mnt_cached) | |
425 | goto done; | |
426 | ||
427 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_mnt, sizeof(bpf_mnt), | |
428 | bpf_known_mnts); | |
429 | if (!mnt) { | |
430 | mnt = getenv(BPF_ENV_MNT); | |
431 | if (!mnt) | |
432 | mnt = BPF_DIR_MNT; | |
433 | ret = bpf_mnt_fs(mnt); | |
434 | if (ret) { | |
435 | mnt = NULL; | |
436 | goto out; | |
437 | } | |
438 | } | |
439 | ||
440 | snprintf(bpf_tc_dir, sizeof(bpf_tc_dir), "%s/%s", mnt, BPF_DIR_TC); | |
441 | ret = mkdir(bpf_tc_dir, S_IRWXU); | |
442 | if (ret && errno != EEXIST) { | |
443 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_tc_dir, | |
444 | strerror(errno)); | |
445 | mnt = NULL; | |
446 | goto out; | |
447 | } | |
448 | ||
449 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s", | |
450 | bpf_tc_dir, BPF_DIR_GLOBALS); | |
451 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
452 | if (ret && errno != EEXIST) { | |
453 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
454 | strerror(errno)); | |
455 | mnt = NULL; | |
456 | goto out; | |
457 | } | |
458 | ||
459 | mnt = bpf_tc_dir; | |
460 | out: | |
461 | bpf_mnt_cached = true; | |
462 | done: | |
463 | return mnt; | |
464 | } | |
465 | ||
466 | static int bpf_obj_get(const char *pathname) | |
467 | { | |
468 | union bpf_attr attr; | |
469 | char tmp[PATH_MAX]; | |
470 | ||
471 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
472 | pathname[1] == ':' && bpf_get_tc_dir()) { | |
473 | snprintf(tmp, sizeof(tmp), "%s/%s", | |
474 | bpf_get_tc_dir(), pathname + 2); | |
475 | pathname = tmp; | |
476 | } | |
477 | ||
478 | memset(&attr, 0, sizeof(attr)); | |
479 | attr.pathname = bpf_ptr_to_u64(pathname); | |
480 | ||
481 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
482 | } | |
483 | ||
6256f8c9 | 484 | const char *bpf_default_section(const enum bpf_prog_type type) |
11c39b5e DB |
485 | { |
486 | switch (type) { | |
487 | case BPF_PROG_TYPE_SCHED_CLS: | |
488 | return ELF_SECTION_CLASSIFIER; | |
6256f8c9 DB |
489 | case BPF_PROG_TYPE_SCHED_ACT: |
490 | return ELF_SECTION_ACTION; | |
11c39b5e DB |
491 | default: |
492 | return NULL; | |
493 | } | |
494 | } | |
495 | ||
91d88eeb DB |
496 | enum bpf_mode { |
497 | CBPF_BYTECODE = 0, | |
498 | CBPF_FILE, | |
499 | EBPF_OBJECT, | |
500 | EBPF_PINNED, | |
501 | __BPF_MODE_MAX, | |
502 | #define BPF_MODE_MAX __BPF_MODE_MAX | |
503 | }; | |
504 | ||
505 | static int bpf_parse(int *ptr_argc, char ***ptr_argv, const bool *opt_tbl, | |
506 | enum bpf_prog_type *type, enum bpf_mode *mode, | |
507 | const char **ptr_object, const char **ptr_section, | |
508 | const char **ptr_uds_name, struct sock_filter *opcodes) | |
32e93fb7 | 509 | { |
32e93fb7 | 510 | const char *file, *section, *uds_name; |
32e93fb7 | 511 | bool verbose = false; |
91d88eeb DB |
512 | int ret, argc; |
513 | char **argv; | |
514 | ||
515 | argv = *ptr_argv; | |
516 | argc = *ptr_argc; | |
517 | ||
518 | if (opt_tbl[CBPF_BYTECODE] && | |
519 | (matches(*argv, "bytecode") == 0 || | |
520 | strcmp(*argv, "bc") == 0)) { | |
521 | *mode = CBPF_BYTECODE; | |
522 | } else if (opt_tbl[CBPF_FILE] && | |
523 | (matches(*argv, "bytecode-file") == 0 || | |
524 | strcmp(*argv, "bcf") == 0)) { | |
525 | *mode = CBPF_FILE; | |
526 | } else if (opt_tbl[EBPF_OBJECT] && | |
527 | (matches(*argv, "object-file") == 0 || | |
528 | strcmp(*argv, "obj") == 0)) { | |
529 | *mode = EBPF_OBJECT; | |
530 | } else if (opt_tbl[EBPF_PINNED] && | |
531 | (matches(*argv, "object-pinned") == 0 || | |
532 | matches(*argv, "pinned") == 0 || | |
533 | matches(*argv, "fd") == 0)) { | |
534 | *mode = EBPF_PINNED; | |
32e93fb7 DB |
535 | } else { |
536 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
537 | return -1; | |
538 | } | |
539 | ||
540 | NEXT_ARG(); | |
541 | file = section = uds_name = NULL; | |
91d88eeb | 542 | if (*mode == EBPF_OBJECT || *mode == EBPF_PINNED) { |
32e93fb7 DB |
543 | file = *argv; |
544 | NEXT_ARG_FWD(); | |
545 | ||
91d88eeb DB |
546 | if (*type == BPF_PROG_TYPE_UNSPEC) { |
547 | if (argc > 0 && matches(*argv, "type") == 0) { | |
548 | NEXT_ARG(); | |
549 | if (matches(*argv, "cls") == 0) { | |
550 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
551 | } else if (matches(*argv, "act") == 0) { | |
552 | *type = BPF_PROG_TYPE_SCHED_ACT; | |
553 | } else { | |
554 | fprintf(stderr, "What type is \"%s\"?\n", | |
555 | *argv); | |
556 | return -1; | |
557 | } | |
558 | NEXT_ARG_FWD(); | |
559 | } else { | |
560 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
561 | } | |
562 | } | |
563 | ||
564 | section = bpf_default_section(*type); | |
32e93fb7 DB |
565 | if (argc > 0 && matches(*argv, "section") == 0) { |
566 | NEXT_ARG(); | |
567 | section = *argv; | |
568 | NEXT_ARG_FWD(); | |
569 | } | |
570 | ||
571 | uds_name = getenv(BPF_ENV_UDS); | |
572 | if (argc > 0 && !uds_name && | |
573 | matches(*argv, "export") == 0) { | |
574 | NEXT_ARG(); | |
575 | uds_name = *argv; | |
576 | NEXT_ARG_FWD(); | |
577 | } | |
578 | ||
579 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
580 | verbose = true; | |
581 | NEXT_ARG_FWD(); | |
582 | } | |
583 | ||
584 | PREV_ARG(); | |
585 | } | |
586 | ||
91d88eeb DB |
587 | if (*mode == CBPF_BYTECODE || *mode == CBPF_FILE) |
588 | ret = bpf_ops_parse(argc, argv, opcodes, *mode == CBPF_FILE); | |
589 | else if (*mode == EBPF_OBJECT) | |
590 | ret = bpf_obj_open(file, *type, section, verbose); | |
591 | else if (*mode == EBPF_PINNED) | |
32e93fb7 | 592 | ret = bpf_obj_get(file); |
91d88eeb | 593 | else |
32e93fb7 DB |
594 | return -1; |
595 | ||
91d88eeb DB |
596 | if (ptr_object) |
597 | *ptr_object = file; | |
598 | if (ptr_section) | |
599 | *ptr_section = section; | |
600 | if (ptr_uds_name) | |
601 | *ptr_uds_name = uds_name; | |
602 | ||
603 | *ptr_argc = argc; | |
604 | *ptr_argv = argv; | |
605 | ||
606 | return ret; | |
607 | } | |
608 | ||
609 | int bpf_parse_common(int *ptr_argc, char ***ptr_argv, const int *nla_tbl, | |
610 | enum bpf_prog_type type, const char **ptr_object, | |
611 | const char **ptr_uds_name, struct nlmsghdr *n) | |
612 | { | |
613 | struct sock_filter opcodes[BPF_MAXINSNS]; | |
614 | const bool opt_tbl[BPF_MODE_MAX] = { | |
615 | [CBPF_BYTECODE] = true, | |
616 | [CBPF_FILE] = true, | |
617 | [EBPF_OBJECT] = true, | |
618 | [EBPF_PINNED] = true, | |
619 | }; | |
620 | char annotation[256]; | |
621 | const char *section; | |
622 | enum bpf_mode mode; | |
623 | int ret; | |
624 | ||
625 | ret = bpf_parse(ptr_argc, ptr_argv, opt_tbl, &type, &mode, | |
626 | ptr_object, §ion, ptr_uds_name, opcodes); | |
627 | if (ret < 0) | |
628 | return ret; | |
629 | ||
32e93fb7 DB |
630 | if (mode == CBPF_BYTECODE || mode == CBPF_FILE) { |
631 | addattr16(n, MAX_MSG, nla_tbl[BPF_NLA_OPS_LEN], ret); | |
632 | addattr_l(n, MAX_MSG, nla_tbl[BPF_NLA_OPS], opcodes, | |
633 | ret * sizeof(struct sock_filter)); | |
91d88eeb DB |
634 | } |
635 | ||
636 | if (mode == EBPF_OBJECT || mode == EBPF_PINNED) { | |
32e93fb7 | 637 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
91d88eeb DB |
638 | basename(*ptr_object), mode == EBPF_PINNED ? |
639 | "*fsobj" : section); | |
32e93fb7 DB |
640 | |
641 | addattr32(n, MAX_MSG, nla_tbl[BPF_NLA_FD], ret); | |
642 | addattrstrz(n, MAX_MSG, nla_tbl[BPF_NLA_NAME], annotation); | |
643 | } | |
644 | ||
91d88eeb DB |
645 | return 0; |
646 | } | |
32e93fb7 | 647 | |
91d88eeb DB |
648 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
649 | { | |
650 | enum bpf_prog_type type = BPF_PROG_TYPE_UNSPEC; | |
651 | const bool opt_tbl[BPF_MODE_MAX] = { | |
652 | [CBPF_BYTECODE] = false, | |
653 | [CBPF_FILE] = false, | |
654 | [EBPF_OBJECT] = true, | |
655 | [EBPF_PINNED] = true, | |
656 | }; | |
657 | const struct bpf_elf_map test = { | |
658 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
659 | .size_key = sizeof(int), | |
660 | .size_value = sizeof(int), | |
661 | }; | |
662 | int ret, prog_fd, map_fd; | |
663 | const char *section; | |
664 | enum bpf_mode mode; | |
665 | uint32_t map_key; | |
666 | ||
667 | prog_fd = bpf_parse(&argc, &argv, opt_tbl, &type, &mode, | |
668 | NULL, §ion, NULL, NULL); | |
669 | if (prog_fd < 0) | |
670 | return prog_fd; | |
671 | if (key) { | |
672 | map_key = *key; | |
673 | } else { | |
674 | ret = sscanf(section, "%*i/%i", &map_key); | |
675 | if (ret != 1) { | |
32a121cb | 676 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
677 | ret = -EINVAL; |
678 | goto out_prog; | |
679 | } | |
680 | } | |
32e93fb7 | 681 | |
91d88eeb DB |
682 | map_fd = bpf_obj_get(map_path); |
683 | if (map_fd < 0) { | |
684 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
685 | map_path, strerror(errno)); | |
686 | ret = map_fd; | |
687 | goto out_prog; | |
688 | } | |
689 | ||
690 | ret = bpf_map_selfcheck_pinned(map_fd, &test, | |
691 | offsetof(struct bpf_elf_map, max_elem)); | |
692 | if (ret < 0) { | |
693 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
694 | goto out_map; | |
695 | } | |
696 | ||
697 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
698 | if (ret < 0) | |
699 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
700 | out_map: | |
701 | close(map_fd); | |
702 | out_prog: | |
703 | close(prog_fd); | |
704 | return ret; | |
32e93fb7 DB |
705 | } |
706 | ||
6256f8c9 | 707 | #ifdef HAVE_ELF |
32e93fb7 DB |
708 | struct bpf_elf_prog { |
709 | enum bpf_prog_type type; | |
710 | const struct bpf_insn *insns; | |
711 | size_t size; | |
712 | const char *license; | |
713 | }; | |
714 | ||
f6793eec DB |
715 | struct bpf_hash_entry { |
716 | unsigned int pinning; | |
717 | const char *subpath; | |
718 | struct bpf_hash_entry *next; | |
719 | }; | |
720 | ||
32e93fb7 DB |
721 | struct bpf_elf_ctx { |
722 | Elf *elf_fd; | |
723 | GElf_Ehdr elf_hdr; | |
724 | Elf_Data *sym_tab; | |
725 | Elf_Data *str_tab; | |
726 | int obj_fd; | |
727 | int map_fds[ELF_MAX_MAPS]; | |
728 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
729 | int sym_num; | |
730 | int map_num; | |
731 | bool *sec_done; | |
732 | int sec_maps; | |
733 | char license[ELF_MAX_LICENSE_LEN]; | |
734 | enum bpf_prog_type type; | |
735 | bool verbose; | |
736 | struct bpf_elf_st stat; | |
f6793eec | 737 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
738 | char *log; |
739 | size_t log_size; | |
32e93fb7 DB |
740 | }; |
741 | ||
6256f8c9 | 742 | struct bpf_elf_sec_data { |
32e93fb7 DB |
743 | GElf_Shdr sec_hdr; |
744 | Elf_Data *sec_data; | |
745 | const char *sec_name; | |
6256f8c9 DB |
746 | }; |
747 | ||
748 | struct bpf_map_data { | |
32e93fb7 DB |
749 | int *fds; |
750 | const char *obj; | |
751 | struct bpf_elf_st *st; | |
752 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
753 | }; |
754 | ||
f31645d1 DB |
755 | static __check_format_string(2, 3) void |
756 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
757 | { |
758 | va_list vl; | |
759 | ||
760 | va_start(vl, format); | |
761 | vfprintf(stderr, format, vl); | |
762 | va_end(vl); | |
763 | ||
f31645d1 | 764 | if (ctx->log && ctx->log[0]) { |
afc1a200 DB |
765 | if (ctx->verbose) { |
766 | fprintf(stderr, "%s\n", ctx->log); | |
767 | } else { | |
768 | unsigned int off = 0, len = strlen(ctx->log); | |
769 | ||
770 | if (len > BPF_MAX_LOG) { | |
771 | off = len - BPF_MAX_LOG; | |
772 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
773 | off); | |
774 | } | |
775 | fprintf(stderr, "%s\n", ctx->log + off); | |
776 | } | |
777 | ||
f31645d1 DB |
778 | memset(ctx->log, 0, ctx->log_size); |
779 | } | |
780 | } | |
781 | ||
782 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
783 | { | |
784 | size_t log_size = ctx->log_size; | |
785 | void *ptr; | |
786 | ||
787 | if (!ctx->log) { | |
788 | log_size = 65536; | |
789 | } else { | |
790 | log_size <<= 1; | |
791 | if (log_size > (UINT_MAX >> 8)) | |
792 | return -EINVAL; | |
d937a74b | 793 | } |
f31645d1 DB |
794 | |
795 | ptr = realloc(ctx->log, log_size); | |
796 | if (!ptr) | |
797 | return -ENOMEM; | |
798 | ||
799 | ctx->log = ptr; | |
800 | ctx->log_size = log_size; | |
801 | ||
802 | return 0; | |
11c39b5e DB |
803 | } |
804 | ||
4dd3f50a DB |
805 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
806 | uint32_t size_value, uint32_t max_elem, | |
807 | uint32_t flags) | |
11c39b5e | 808 | { |
67584e3a ND |
809 | union bpf_attr attr; |
810 | ||
811 | memset(&attr, 0, sizeof(attr)); | |
812 | attr.map_type = type; | |
813 | attr.key_size = size_key; | |
814 | attr.value_size = size_value; | |
815 | attr.max_entries = max_elem; | |
4dd3f50a | 816 | attr.map_flags = flags; |
11c39b5e DB |
817 | |
818 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
819 | } | |
820 | ||
821 | static int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, | |
f31645d1 DB |
822 | size_t size_insns, const char *license, char *log, |
823 | size_t size_log) | |
11c39b5e | 824 | { |
67584e3a ND |
825 | union bpf_attr attr; |
826 | ||
827 | memset(&attr, 0, sizeof(attr)); | |
828 | attr.prog_type = type; | |
829 | attr.insns = bpf_ptr_to_u64(insns); | |
f31645d1 | 830 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); |
67584e3a | 831 | attr.license = bpf_ptr_to_u64(license); |
11c39b5e | 832 | |
f31645d1 DB |
833 | if (size_log > 0) { |
834 | attr.log_buf = bpf_ptr_to_u64(log); | |
835 | attr.log_size = size_log; | |
836 | attr.log_level = 1; | |
32e93fb7 DB |
837 | } |
838 | ||
11c39b5e DB |
839 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); |
840 | } | |
841 | ||
32e93fb7 | 842 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 843 | { |
67584e3a ND |
844 | union bpf_attr attr; |
845 | ||
846 | memset(&attr, 0, sizeof(attr)); | |
847 | attr.pathname = bpf_ptr_to_u64(pathname); | |
848 | attr.bpf_fd = fd; | |
32e93fb7 DB |
849 | |
850 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
851 | } | |
11c39b5e | 852 | |
32e93fb7 DB |
853 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
854 | { | |
855 | struct sockaddr_alg alg = { | |
856 | .salg_family = AF_ALG, | |
857 | .salg_type = "hash", | |
858 | .salg_name = "sha1", | |
859 | }; | |
860 | int ret, cfd, ofd, ffd; | |
861 | struct stat stbuff; | |
862 | ssize_t size; | |
863 | ||
864 | if (!object || len != 20) | |
865 | return -EINVAL; | |
866 | ||
867 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
868 | if (cfd < 0) { | |
869 | fprintf(stderr, "Cannot get AF_ALG socket: %s\n", | |
870 | strerror(errno)); | |
871 | return cfd; | |
872 | } | |
873 | ||
874 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
875 | if (ret < 0) { | |
876 | fprintf(stderr, "Error binding socket: %s\n", strerror(errno)); | |
877 | goto out_cfd; | |
878 | } | |
879 | ||
880 | ofd = accept(cfd, NULL, 0); | |
881 | if (ofd < 0) { | |
882 | fprintf(stderr, "Error accepting socket: %s\n", | |
883 | strerror(errno)); | |
884 | ret = ofd; | |
885 | goto out_cfd; | |
886 | } | |
887 | ||
888 | ffd = open(object, O_RDONLY); | |
889 | if (ffd < 0) { | |
890 | fprintf(stderr, "Error opening object %s: %s\n", | |
891 | object, strerror(errno)); | |
892 | ret = ffd; | |
893 | goto out_ofd; | |
894 | } | |
895 | ||
32a121cb | 896 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
897 | if (ret < 0) { |
898 | fprintf(stderr, "Error doing fstat: %s\n", | |
899 | strerror(errno)); | |
900 | goto out_ffd; | |
d937a74b | 901 | } |
11c39b5e | 902 | |
32e93fb7 DB |
903 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
904 | if (size != stbuff.st_size) { | |
905 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
906 | size, stbuff.st_size, strerror(errno)); | |
907 | ret = -1; | |
908 | goto out_ffd; | |
909 | } | |
910 | ||
911 | size = read(ofd, out, len); | |
912 | if (size != len) { | |
913 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
914 | size, len, strerror(errno)); | |
915 | ret = -1; | |
916 | } else { | |
917 | ret = 0; | |
918 | } | |
919 | out_ffd: | |
920 | close(ffd); | |
921 | out_ofd: | |
922 | close(ofd); | |
923 | out_cfd: | |
924 | close(cfd); | |
925 | return ret; | |
11c39b5e DB |
926 | } |
927 | ||
32e93fb7 | 928 | static const char *bpf_get_obj_uid(const char *pathname) |
11c39b5e | 929 | { |
32a121cb | 930 | static bool bpf_uid_cached; |
32e93fb7 DB |
931 | static char bpf_uid[64]; |
932 | uint8_t tmp[20]; | |
933 | int ret; | |
11c39b5e | 934 | |
32e93fb7 DB |
935 | if (bpf_uid_cached) |
936 | goto done; | |
11c39b5e | 937 | |
32e93fb7 DB |
938 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); |
939 | if (ret) { | |
940 | fprintf(stderr, "Object hashing failed!\n"); | |
941 | return NULL; | |
942 | } | |
943 | ||
944 | hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid)); | |
945 | bpf_uid_cached = true; | |
946 | done: | |
947 | return bpf_uid; | |
11c39b5e DB |
948 | } |
949 | ||
32e93fb7 DB |
950 | static int bpf_init_env(const char *pathname) |
951 | { | |
952 | struct rlimit limit = { | |
953 | .rlim_cur = RLIM_INFINITY, | |
954 | .rlim_max = RLIM_INFINITY, | |
955 | }; | |
956 | ||
957 | /* Don't bother in case we fail! */ | |
958 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
959 | ||
960 | if (!bpf_get_tc_dir()) { | |
32a121cb | 961 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
32e93fb7 DB |
962 | return 0; |
963 | } | |
964 | ||
965 | if (!bpf_get_obj_uid(pathname)) | |
966 | return -1; | |
967 | ||
968 | return 0; | |
6256f8c9 DB |
969 | } |
970 | ||
f6793eec DB |
971 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
972 | uint32_t pinning) | |
973 | { | |
974 | struct bpf_hash_entry *entry; | |
975 | ||
976 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
977 | while (entry && entry->pinning != pinning) | |
978 | entry = entry->next; | |
979 | ||
980 | return entry ? entry->subpath : NULL; | |
981 | } | |
982 | ||
983 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
984 | uint32_t pinning) | |
11c39b5e | 985 | { |
32e93fb7 DB |
986 | switch (pinning) { |
987 | case PIN_OBJECT_NS: | |
988 | case PIN_GLOBAL_NS: | |
989 | return false; | |
990 | case PIN_NONE: | |
32e93fb7 | 991 | return true; |
f6793eec DB |
992 | default: |
993 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
994 | } |
995 | } | |
996 | ||
997 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 998 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
999 | { |
1000 | switch (pinning) { | |
1001 | case PIN_OBJECT_NS: | |
1002 | snprintf(pathname, len, "%s/%s/%s", bpf_get_tc_dir(), | |
1003 | bpf_get_obj_uid(NULL), name); | |
1004 | break; | |
1005 | case PIN_GLOBAL_NS: | |
1006 | snprintf(pathname, len, "%s/%s/%s", bpf_get_tc_dir(), | |
1007 | BPF_DIR_GLOBALS, name); | |
1008 | break; | |
f6793eec DB |
1009 | default: |
1010 | snprintf(pathname, len, "%s/../%s/%s", bpf_get_tc_dir(), | |
1011 | bpf_custom_pinning(ctx, pinning), name); | |
1012 | break; | |
32e93fb7 DB |
1013 | } |
1014 | } | |
1015 | ||
f6793eec DB |
1016 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1017 | uint32_t pinning) | |
32e93fb7 DB |
1018 | { |
1019 | char pathname[PATH_MAX]; | |
1020 | ||
f6793eec | 1021 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_tc_dir()) |
32e93fb7 DB |
1022 | return 0; |
1023 | ||
f6793eec | 1024 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
32e93fb7 DB |
1025 | return bpf_obj_get(pathname); |
1026 | } | |
1027 | ||
f6793eec | 1028 | static int bpf_make_obj_path(void) |
32e93fb7 | 1029 | { |
f6793eec | 1030 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1031 | int ret; |
1032 | ||
f6793eec DB |
1033 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_tc_dir(), |
1034 | bpf_get_obj_uid(NULL)); | |
1035 | ||
1036 | ret = mkdir(tmp, S_IRWXU); | |
1037 | if (ret && errno != EEXIST) { | |
1038 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1039 | return ret; | |
1040 | } | |
1041 | ||
1042 | return 0; | |
1043 | } | |
1044 | ||
1045 | static int bpf_make_custom_path(const char *todo) | |
1046 | { | |
1047 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1048 | int ret; | |
1049 | ||
1050 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_tc_dir()); | |
1051 | snprintf(rem, sizeof(rem), "%s/", todo); | |
1052 | sub = strtok(rem, "/"); | |
32e93fb7 | 1053 | |
f6793eec DB |
1054 | while (sub) { |
1055 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1056 | return -EINVAL; | |
1057 | ||
1058 | strcat(tmp, sub); | |
1059 | strcat(tmp, "/"); | |
32e93fb7 | 1060 | |
f6793eec | 1061 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1062 | if (ret && errno != EEXIST) { |
f6793eec | 1063 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1064 | strerror(errno)); |
1065 | return ret; | |
1066 | } | |
f6793eec DB |
1067 | |
1068 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1069 | } |
1070 | ||
f6793eec DB |
1071 | return 0; |
1072 | } | |
1073 | ||
1074 | static int bpf_place_pinned(int fd, const char *name, | |
1075 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1076 | { | |
1077 | char pathname[PATH_MAX]; | |
1078 | const char *tmp; | |
1079 | int ret = 0; | |
1080 | ||
1081 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_tc_dir()) | |
1082 | return 0; | |
1083 | ||
1084 | if (pinning == PIN_OBJECT_NS) | |
1085 | ret = bpf_make_obj_path(); | |
1086 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) | |
1087 | ret = bpf_make_custom_path(tmp); | |
1088 | if (ret < 0) | |
1089 | return ret; | |
1090 | ||
1091 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1092 | return bpf_obj_pin(fd, pathname); |
1093 | } | |
1094 | ||
f31645d1 DB |
1095 | static void bpf_prog_report(int fd, const char *section, |
1096 | const struct bpf_elf_prog *prog, | |
1097 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1098 | { |
afc1a200 DB |
1099 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1100 | ||
1101 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1102 | fd < 0 ? "rejected: " : "loaded", |
1103 | fd < 0 ? strerror(errno) : "", | |
1104 | fd < 0 ? errno : fd); | |
1105 | ||
1106 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1107 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1108 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1109 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1110 | ||
1111 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1112 | } | |
32e93fb7 | 1113 | |
f31645d1 DB |
1114 | static int bpf_prog_attach(const char *section, |
1115 | const struct bpf_elf_prog *prog, | |
1116 | struct bpf_elf_ctx *ctx) | |
1117 | { | |
1118 | int tries = 0, fd; | |
1119 | retry: | |
32e93fb7 DB |
1120 | errno = 0; |
1121 | fd = bpf_prog_load(prog->type, prog->insns, prog->size, | |
f31645d1 DB |
1122 | prog->license, ctx->log, ctx->log_size); |
1123 | if (fd < 0 || ctx->verbose) { | |
1124 | /* The verifier log is pretty chatty, sometimes so chatty | |
1125 | * on larger programs, that we could fail to dump everything | |
1126 | * into our buffer. Still, try to give a debuggable error | |
1127 | * log for the user, so enlarge it and re-fail. | |
1128 | */ | |
1129 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
1130 | if (tries++ < 6 && !bpf_log_realloc(ctx)) | |
1131 | goto retry; | |
1132 | ||
32a121cb | 1133 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1134 | ctx->log_size, tries); |
1135 | return fd; | |
1136 | } | |
1137 | ||
1138 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1139 | } |
1140 | ||
1141 | return fd; | |
1142 | } | |
1143 | ||
f31645d1 DB |
1144 | static void bpf_map_report(int fd, const char *name, |
1145 | const struct bpf_elf_map *map, | |
1146 | struct bpf_elf_ctx *ctx) | |
1147 | { | |
1148 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1149 | fd < 0 ? "rejected: " : "loaded", | |
1150 | fd < 0 ? strerror(errno) : "", | |
1151 | fd < 0 ? errno : fd); | |
1152 | ||
1153 | fprintf(stderr, " - Type: %u\n", map->type); | |
1154 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1155 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1156 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
1157 | fprintf(stderr, " - Size value: %u\n", map->size_value); | |
4dd3f50a DB |
1158 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1159 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1160 | } |
1161 | ||
32e93fb7 | 1162 | static int bpf_map_attach(const char *name, const struct bpf_elf_map *map, |
f31645d1 | 1163 | struct bpf_elf_ctx *ctx) |
32e93fb7 DB |
1164 | { |
1165 | int fd, ret; | |
1166 | ||
f6793eec | 1167 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1168 | if (fd > 0) { |
91d88eeb DB |
1169 | ret = bpf_map_selfcheck_pinned(fd, map, |
1170 | offsetof(struct bpf_elf_map, | |
1171 | id)); | |
9e607f2e DB |
1172 | if (ret < 0) { |
1173 | close(fd); | |
1174 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1175 | name); | |
1176 | return ret; | |
1177 | } | |
f31645d1 | 1178 | if (ctx->verbose) |
32e93fb7 DB |
1179 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1180 | name); | |
1181 | return fd; | |
1182 | } | |
1183 | ||
1184 | errno = 0; | |
1185 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
4dd3f50a | 1186 | map->max_elem, map->flags); |
f31645d1 DB |
1187 | if (fd < 0 || ctx->verbose) { |
1188 | bpf_map_report(fd, name, map, ctx); | |
32e93fb7 DB |
1189 | if (fd < 0) |
1190 | return fd; | |
1191 | } | |
1192 | ||
f6793eec | 1193 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1194 | if (ret < 0 && errno != EEXIST) { |
1195 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1196 | strerror(errno)); | |
1197 | close(fd); | |
1198 | return ret; | |
1199 | } | |
1200 | ||
1201 | return fd; | |
1202 | } | |
1203 | ||
32e93fb7 DB |
1204 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1205 | const GElf_Sym *sym) | |
1206 | { | |
1207 | return ctx->str_tab->d_buf + sym->st_name; | |
1208 | } | |
1209 | ||
1210 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) | |
1211 | { | |
1212 | GElf_Sym sym; | |
11c39b5e DB |
1213 | int i; |
1214 | ||
32e93fb7 DB |
1215 | for (i = 0; i < ctx->sym_num; i++) { |
1216 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1217 | continue; | |
1218 | ||
5230a2ed DB |
1219 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
1220 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
32e93fb7 DB |
1221 | sym.st_shndx != ctx->sec_maps || |
1222 | sym.st_value / sizeof(struct bpf_elf_map) != which) | |
1223 | continue; | |
1224 | ||
1225 | return bpf_str_tab_name(ctx, &sym); | |
11c39b5e | 1226 | } |
32e93fb7 DB |
1227 | |
1228 | return NULL; | |
11c39b5e DB |
1229 | } |
1230 | ||
32e93fb7 | 1231 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1232 | { |
32e93fb7 DB |
1233 | const char *map_name; |
1234 | int i, fd; | |
11c39b5e | 1235 | |
32e93fb7 DB |
1236 | for (i = 0; i < ctx->map_num; i++) { |
1237 | map_name = bpf_map_fetch_name(ctx, i); | |
1238 | if (!map_name) | |
1239 | return -EIO; | |
11c39b5e | 1240 | |
f31645d1 | 1241 | fd = bpf_map_attach(map_name, &ctx->maps[i], ctx); |
32e93fb7 DB |
1242 | if (fd < 0) |
1243 | return fd; | |
11c39b5e | 1244 | |
32e93fb7 | 1245 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1246 | } |
1247 | ||
1248 | return 0; | |
11c39b5e DB |
1249 | } |
1250 | ||
32e93fb7 DB |
1251 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1252 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1253 | { |
32e93fb7 | 1254 | Elf_Data *sec_edata; |
11c39b5e DB |
1255 | GElf_Shdr sec_hdr; |
1256 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1257 | char *sec_name; |
1258 | ||
32e93fb7 | 1259 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1260 | |
32e93fb7 | 1261 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1262 | if (!sec_fd) |
1263 | return -EINVAL; | |
11c39b5e DB |
1264 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1265 | return -EIO; | |
1266 | ||
32e93fb7 | 1267 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1268 | sec_hdr.sh_name); |
1269 | if (!sec_name || !sec_hdr.sh_size) | |
1270 | return -ENOENT; | |
1271 | ||
1272 | sec_edata = elf_getdata(sec_fd, NULL); | |
1273 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1274 | return -EIO; | |
1275 | ||
32e93fb7 | 1276 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1277 | |
32e93fb7 DB |
1278 | data->sec_name = sec_name; |
1279 | data->sec_data = sec_edata; | |
11c39b5e DB |
1280 | return 0; |
1281 | } | |
1282 | ||
32e93fb7 DB |
1283 | static int bpf_fetch_maps(struct bpf_elf_ctx *ctx, int section, |
1284 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1285 | { |
32e93fb7 DB |
1286 | if (data->sec_data->d_size % sizeof(struct bpf_elf_map) != 0) |
1287 | return -EINVAL; | |
11c39b5e | 1288 | |
32e93fb7 DB |
1289 | ctx->map_num = data->sec_data->d_size / sizeof(struct bpf_elf_map); |
1290 | ctx->sec_maps = section; | |
1291 | ctx->sec_done[section] = true; | |
11c39b5e | 1292 | |
32e93fb7 DB |
1293 | if (ctx->map_num > ARRAY_SIZE(ctx->map_fds)) { |
1294 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); | |
1295 | return -ENOMEM; | |
1296 | } | |
11c39b5e | 1297 | |
32e93fb7 DB |
1298 | memcpy(ctx->maps, data->sec_data->d_buf, data->sec_data->d_size); |
1299 | return 0; | |
1300 | } | |
11c39b5e | 1301 | |
32e93fb7 DB |
1302 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
1303 | struct bpf_elf_sec_data *data) | |
1304 | { | |
1305 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
1306 | return -ENOMEM; | |
11c39b5e | 1307 | |
32e93fb7 DB |
1308 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
1309 | ctx->sec_done[section] = true; | |
1310 | return 0; | |
1311 | } | |
11c39b5e | 1312 | |
32e93fb7 DB |
1313 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
1314 | struct bpf_elf_sec_data *data) | |
1315 | { | |
1316 | ctx->sym_tab = data->sec_data; | |
1317 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
1318 | ctx->sec_done[section] = true; | |
11c39b5e DB |
1319 | return 0; |
1320 | } | |
1321 | ||
32e93fb7 DB |
1322 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
1323 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1324 | { |
32e93fb7 DB |
1325 | ctx->str_tab = data->sec_data; |
1326 | ctx->sec_done[section] = true; | |
1327 | return 0; | |
1328 | } | |
11c39b5e | 1329 | |
afc1a200 DB |
1330 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
1331 | { | |
1332 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
1333 | } | |
1334 | ||
32e93fb7 DB |
1335 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx) |
1336 | { | |
1337 | struct bpf_elf_sec_data data; | |
1338 | int i, ret = -1; | |
11c39b5e | 1339 | |
32e93fb7 DB |
1340 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1341 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
1342 | if (ret < 0) |
1343 | continue; | |
1344 | ||
cce3d466 DB |
1345 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1346 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
32e93fb7 | 1347 | ret = bpf_fetch_maps(ctx, i, &data); |
cce3d466 DB |
1348 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1349 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 1350 | ret = bpf_fetch_license(ctx, i, &data); |
cce3d466 DB |
1351 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
1352 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
1353 | ret = bpf_fetch_symtab(ctx, i, &data); |
1354 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 1355 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 DB |
1356 | ret = bpf_fetch_strtab(ctx, i, &data); |
1357 | if (ret < 0) { | |
afc1a200 | 1358 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 1359 | i); |
32e93fb7 | 1360 | break; |
11c39b5e | 1361 | } |
32e93fb7 DB |
1362 | } |
1363 | ||
afc1a200 | 1364 | if (bpf_has_map_data(ctx)) { |
32e93fb7 DB |
1365 | ret = bpf_maps_attach_all(ctx); |
1366 | if (ret < 0) { | |
1367 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
1368 | return ret; | |
11c39b5e DB |
1369 | } |
1370 | } | |
1371 | ||
1372 | return ret; | |
1373 | } | |
1374 | ||
32e93fb7 | 1375 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section) |
11c39b5e | 1376 | { |
32e93fb7 DB |
1377 | struct bpf_elf_sec_data data; |
1378 | struct bpf_elf_prog prog; | |
1379 | int ret, i, fd = -1; | |
11c39b5e | 1380 | |
32e93fb7 DB |
1381 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1382 | if (ctx->sec_done[i]) | |
11c39b5e DB |
1383 | continue; |
1384 | ||
32e93fb7 | 1385 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
1386 | if (ret < 0 || |
1387 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
1388 | data.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1389 | !strcmp(data.sec_name, section))) | |
11c39b5e DB |
1390 | continue; |
1391 | ||
32e93fb7 DB |
1392 | memset(&prog, 0, sizeof(prog)); |
1393 | prog.type = ctx->type; | |
1394 | prog.insns = data.sec_data->d_buf; | |
1395 | prog.size = data.sec_data->d_size; | |
1396 | prog.license = ctx->license; | |
11c39b5e | 1397 | |
f31645d1 | 1398 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1399 | if (fd < 0) |
afc1a200 | 1400 | break; |
11c39b5e | 1401 | |
32e93fb7 | 1402 | ctx->sec_done[i] = true; |
11c39b5e DB |
1403 | break; |
1404 | } | |
1405 | ||
32e93fb7 | 1406 | return fd; |
11c39b5e DB |
1407 | } |
1408 | ||
32e93fb7 DB |
1409 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
1410 | struct bpf_elf_sec_data *data_relo, | |
1411 | struct bpf_elf_sec_data *data_insn) | |
11c39b5e | 1412 | { |
32e93fb7 DB |
1413 | Elf_Data *idata = data_insn->sec_data; |
1414 | GElf_Shdr *rhdr = &data_relo->sec_hdr; | |
1415 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
1416 | struct bpf_insn *insns = idata->d_buf; | |
1417 | unsigned int num_insns = idata->d_size / sizeof(*insns); | |
11c39b5e | 1418 | |
32e93fb7 DB |
1419 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
1420 | unsigned int ioff, rmap; | |
1421 | GElf_Rel relo; | |
1422 | GElf_Sym sym; | |
1423 | ||
1424 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
1425 | return -EIO; | |
1426 | ||
1427 | ioff = relo.r_offset / sizeof(struct bpf_insn); | |
1428 | if (ioff >= num_insns || | |
a576c6b9 | 1429 | insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) { |
32a121cb | 1430 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", |
a576c6b9 DB |
1431 | ioff); |
1432 | if (ioff < num_insns && | |
1433 | insns[ioff].code == (BPF_JMP | BPF_CALL)) | |
32a121cb | 1434 | fprintf(stderr, " - Try to annotate functions with always_inline attribute!\n"); |
32e93fb7 | 1435 | return -EINVAL; |
a576c6b9 | 1436 | } |
32e93fb7 DB |
1437 | |
1438 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) | |
1439 | return -EIO; | |
2486337a | 1440 | if (sym.st_shndx != ctx->sec_maps) { |
32a121cb | 1441 | fprintf(stderr, "ELF contains non-map related relo data in entry %u pointing to section %u! Compiler bug?!\n", |
2486337a DB |
1442 | relo_ent, sym.st_shndx); |
1443 | return -EIO; | |
1444 | } | |
32e93fb7 DB |
1445 | |
1446 | rmap = sym.st_value / sizeof(struct bpf_elf_map); | |
1447 | if (rmap >= ARRAY_SIZE(ctx->map_fds)) | |
1448 | return -EINVAL; | |
1449 | if (!ctx->map_fds[rmap]) | |
1450 | return -EINVAL; | |
1451 | ||
1452 | if (ctx->verbose) | |
32a121cb | 1453 | fprintf(stderr, "Map \'%s\' (%d) injected into prog section \'%s\' at offset %u!\n", |
32e93fb7 DB |
1454 | bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap], |
1455 | data_insn->sec_name, ioff); | |
11c39b5e | 1456 | |
32e93fb7 DB |
1457 | insns[ioff].src_reg = BPF_PSEUDO_MAP_FD; |
1458 | insns[ioff].imm = ctx->map_fds[rmap]; | |
1459 | } | |
1460 | ||
1461 | return 0; | |
1462 | } | |
1463 | ||
afc1a200 DB |
1464 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
1465 | bool *lderr) | |
32e93fb7 DB |
1466 | { |
1467 | struct bpf_elf_sec_data data_relo, data_insn; | |
1468 | struct bpf_elf_prog prog; | |
1469 | int ret, idx, i, fd = -1; | |
1470 | ||
1471 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
1472 | ret = bpf_fill_section_data(ctx, i, &data_relo); | |
1473 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
1474 | continue; |
1475 | ||
32e93fb7 DB |
1476 | idx = data_relo.sec_hdr.sh_info; |
1477 | ret = bpf_fill_section_data(ctx, idx, &data_insn); | |
cce3d466 DB |
1478 | if (ret < 0 || |
1479 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
1480 | data_insn.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1481 | !strcmp(data_insn.sec_name, section))) | |
11c39b5e | 1482 | continue; |
32e93fb7 DB |
1483 | |
1484 | ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn); | |
1485 | if (ret < 0) | |
11c39b5e DB |
1486 | continue; |
1487 | ||
32e93fb7 DB |
1488 | memset(&prog, 0, sizeof(prog)); |
1489 | prog.type = ctx->type; | |
1490 | prog.insns = data_insn.sec_data->d_buf; | |
1491 | prog.size = data_insn.sec_data->d_size; | |
1492 | prog.license = ctx->license; | |
1493 | ||
f31645d1 | 1494 | fd = bpf_prog_attach(section, &prog, ctx); |
afc1a200 DB |
1495 | if (fd < 0) { |
1496 | *lderr = true; | |
1497 | break; | |
1498 | } | |
11c39b5e | 1499 | |
32e93fb7 DB |
1500 | ctx->sec_done[i] = true; |
1501 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
1502 | break; |
1503 | } | |
1504 | ||
32e93fb7 | 1505 | return fd; |
11c39b5e DB |
1506 | } |
1507 | ||
32e93fb7 | 1508 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 1509 | { |
afc1a200 | 1510 | bool lderr = false; |
473d7840 DB |
1511 | int ret = -1; |
1512 | ||
afc1a200 DB |
1513 | if (bpf_has_map_data(ctx)) |
1514 | ret = bpf_fetch_prog_relo(ctx, section, &lderr); | |
1515 | if (ret < 0 && !lderr) | |
32e93fb7 DB |
1516 | ret = bpf_fetch_prog(ctx, section); |
1517 | ||
473d7840 DB |
1518 | return ret; |
1519 | } | |
1520 | ||
910b543d DB |
1521 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
1522 | { | |
1523 | int i; | |
1524 | ||
1525 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
1526 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
1527 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
1528 | return i; | |
1529 | return -1; | |
1530 | } | |
1531 | ||
32e93fb7 | 1532 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 1533 | { |
32e93fb7 DB |
1534 | struct bpf_elf_sec_data data; |
1535 | uint32_t map_id, key_id; | |
910b543d | 1536 | int fd, i, ret, idx; |
473d7840 | 1537 | |
32e93fb7 DB |
1538 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1539 | if (ctx->sec_done[i]) | |
473d7840 DB |
1540 | continue; |
1541 | ||
32e93fb7 | 1542 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
1543 | if (ret < 0) |
1544 | continue; | |
1545 | ||
910b543d DB |
1546 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
1547 | if (ret != 2) | |
32e93fb7 | 1548 | continue; |
910b543d DB |
1549 | |
1550 | idx = bpf_find_map_by_id(ctx, map_id); | |
1551 | if (idx < 0) | |
473d7840 DB |
1552 | continue; |
1553 | ||
32e93fb7 DB |
1554 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
1555 | if (fd < 0) | |
473d7840 DB |
1556 | return -EIO; |
1557 | ||
910b543d DB |
1558 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
1559 | &fd, BPF_ANY); | |
afc1a200 DB |
1560 | if (ret < 0) { |
1561 | if (errno == E2BIG) | |
1562 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", | |
1563 | key_id, map_id); | |
1564 | return -errno; | |
1565 | } | |
473d7840 | 1566 | |
32e93fb7 | 1567 | ctx->sec_done[i] = true; |
473d7840 DB |
1568 | } |
1569 | ||
1570 | return 0; | |
1571 | } | |
1572 | ||
32e93fb7 | 1573 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 1574 | { |
32e93fb7 DB |
1575 | struct stat st; |
1576 | int ret; | |
11c39b5e | 1577 | |
32e93fb7 | 1578 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 1579 | |
32e93fb7 DB |
1580 | ret = fstat(ctx->obj_fd, &st); |
1581 | if (ret < 0) { | |
1582 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
1583 | strerror(errno)); | |
1584 | return; | |
1585 | } | |
11c39b5e | 1586 | |
32e93fb7 DB |
1587 | ctx->stat.st_dev = st.st_dev; |
1588 | ctx->stat.st_ino = st.st_ino; | |
1589 | } | |
1590 | ||
f6793eec DB |
1591 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
1592 | { | |
1593 | char buff[PATH_MAX]; | |
1594 | ||
1595 | while (fgets(buff, sizeof(buff), fp)) { | |
1596 | char *ptr = buff; | |
1597 | ||
1598 | while (*ptr == ' ' || *ptr == '\t') | |
1599 | ptr++; | |
1600 | ||
1601 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
1602 | continue; | |
1603 | ||
1604 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
1605 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
1606 | strcpy(path, ptr); | |
1607 | return -1; | |
1608 | } | |
1609 | ||
1610 | return 1; | |
1611 | } | |
1612 | ||
1613 | return 0; | |
1614 | } | |
1615 | ||
1616 | static bool bpf_pinning_reserved(uint32_t pinning) | |
1617 | { | |
1618 | switch (pinning) { | |
1619 | case PIN_NONE: | |
1620 | case PIN_OBJECT_NS: | |
1621 | case PIN_GLOBAL_NS: | |
1622 | return true; | |
1623 | default: | |
1624 | return false; | |
1625 | } | |
1626 | } | |
1627 | ||
1628 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
1629 | { | |
1630 | struct bpf_hash_entry *entry; | |
1631 | char subpath[PATH_MAX]; | |
1632 | uint32_t pinning; | |
1633 | FILE *fp; | |
1634 | int ret; | |
1635 | ||
1636 | fp = fopen(db_file, "r"); | |
1637 | if (!fp) | |
1638 | return; | |
1639 | ||
1640 | memset(subpath, 0, sizeof(subpath)); | |
1641 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { | |
1642 | if (ret == -1) { | |
1643 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
1644 | db_file, subpath); | |
1645 | fclose(fp); | |
1646 | return; | |
1647 | } | |
1648 | ||
1649 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
1650 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
1651 | db_file, pinning); | |
f6793eec DB |
1652 | continue; |
1653 | } | |
1654 | ||
1655 | entry = malloc(sizeof(*entry)); | |
1656 | if (!entry) { | |
1657 | fprintf(stderr, "No memory left for db entry!\n"); | |
1658 | continue; | |
1659 | } | |
1660 | ||
1661 | entry->pinning = pinning; | |
1662 | entry->subpath = strdup(subpath); | |
1663 | if (!entry->subpath) { | |
1664 | fprintf(stderr, "No memory left for db entry!\n"); | |
1665 | free(entry); | |
1666 | continue; | |
1667 | } | |
1668 | ||
1669 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1670 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
1671 | } | |
1672 | ||
1673 | fclose(fp); | |
1674 | } | |
1675 | ||
1676 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
1677 | { | |
1678 | struct bpf_hash_entry *entry; | |
1679 | int i; | |
1680 | ||
1681 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
1682 | while ((entry = ctx->ht[i]) != NULL) { | |
1683 | ctx->ht[i] = entry->next; | |
1684 | free((char *)entry->subpath); | |
1685 | free(entry); | |
1686 | } | |
1687 | } | |
1688 | } | |
1689 | ||
8187b012 DB |
1690 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
1691 | { | |
1692 | if (ctx->elf_hdr.e_type != ET_REL || | |
1693 | ctx->elf_hdr.e_machine != 0 || | |
1694 | ctx->elf_hdr.e_version != EV_CURRENT) { | |
1695 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
1696 | return -EINVAL; | |
1697 | } | |
1698 | ||
1699 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
1700 | default: | |
1701 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
1702 | return -EINVAL; | |
1703 | case ELFDATA2LSB: | |
1704 | if (htons(1) == 1) { | |
1705 | fprintf(stderr, | |
1706 | "We are big endian, eBPF object is little endian!\n"); | |
1707 | return -EIO; | |
1708 | } | |
1709 | break; | |
1710 | case ELFDATA2MSB: | |
1711 | if (htons(1) != 1) { | |
1712 | fprintf(stderr, | |
1713 | "We are little endian, eBPF object is big endian!\n"); | |
1714 | return -EIO; | |
1715 | } | |
1716 | break; | |
1717 | } | |
1718 | ||
1719 | return 0; | |
1720 | } | |
1721 | ||
32e93fb7 DB |
1722 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
1723 | enum bpf_prog_type type, bool verbose) | |
1724 | { | |
1725 | int ret = -EINVAL; | |
1726 | ||
1727 | if (elf_version(EV_CURRENT) == EV_NONE || | |
1728 | bpf_init_env(pathname)) | |
1729 | return ret; | |
1730 | ||
1731 | memset(ctx, 0, sizeof(*ctx)); | |
1732 | ctx->verbose = verbose; | |
1733 | ctx->type = type; | |
1734 | ||
1735 | ctx->obj_fd = open(pathname, O_RDONLY); | |
1736 | if (ctx->obj_fd < 0) | |
1737 | return ctx->obj_fd; | |
1738 | ||
1739 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
1740 | if (!ctx->elf_fd) { | |
11c39b5e | 1741 | ret = -EINVAL; |
32e93fb7 | 1742 | goto out_fd; |
11c39b5e DB |
1743 | } |
1744 | ||
8187b012 DB |
1745 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
1746 | ret = -EINVAL; | |
1747 | goto out_fd; | |
1748 | } | |
1749 | ||
32e93fb7 DB |
1750 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
1751 | &ctx->elf_hdr) { | |
11c39b5e DB |
1752 | ret = -EIO; |
1753 | goto out_elf; | |
1754 | } | |
1755 | ||
8187b012 DB |
1756 | ret = bpf_elf_check_ehdr(ctx); |
1757 | if (ret < 0) | |
1758 | goto out_elf; | |
1759 | ||
32e93fb7 DB |
1760 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
1761 | sizeof(*(ctx->sec_done))); | |
1762 | if (!ctx->sec_done) { | |
11c39b5e DB |
1763 | ret = -ENOMEM; |
1764 | goto out_elf; | |
1765 | } | |
1766 | ||
f31645d1 DB |
1767 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
1768 | ret = -ENOMEM; | |
1769 | goto out_free; | |
1770 | } | |
1771 | ||
32e93fb7 | 1772 | bpf_save_finfo(ctx); |
f6793eec DB |
1773 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
1774 | ||
32e93fb7 | 1775 | return 0; |
f31645d1 DB |
1776 | out_free: |
1777 | free(ctx->sec_done); | |
32e93fb7 DB |
1778 | out_elf: |
1779 | elf_end(ctx->elf_fd); | |
1780 | out_fd: | |
1781 | close(ctx->obj_fd); | |
1782 | return ret; | |
1783 | } | |
d937a74b | 1784 | |
32e93fb7 DB |
1785 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
1786 | { | |
1787 | int i, count = 0; | |
11c39b5e | 1788 | |
32e93fb7 DB |
1789 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
1790 | if (!ctx->map_fds[i]) | |
1791 | break; | |
1792 | count++; | |
1793 | } | |
473d7840 | 1794 | |
32e93fb7 DB |
1795 | return count; |
1796 | } | |
6256f8c9 | 1797 | |
32e93fb7 DB |
1798 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
1799 | { | |
1800 | int i; | |
1801 | ||
1802 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
1803 | if (ctx->map_fds[i]) | |
1804 | close(ctx->map_fds[i]); | |
473d7840 | 1805 | } |
32e93fb7 DB |
1806 | } |
1807 | ||
1808 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
1809 | { | |
1810 | if (failure) | |
1811 | bpf_maps_teardown(ctx); | |
473d7840 | 1812 | |
f6793eec | 1813 | bpf_hash_destroy(ctx); |
f31645d1 | 1814 | |
32e93fb7 | 1815 | free(ctx->sec_done); |
f31645d1 DB |
1816 | free(ctx->log); |
1817 | ||
32e93fb7 DB |
1818 | elf_end(ctx->elf_fd); |
1819 | close(ctx->obj_fd); | |
1820 | } | |
6256f8c9 | 1821 | |
32e93fb7 | 1822 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 1823 | |
32e93fb7 DB |
1824 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
1825 | const char *section, bool verbose) | |
1826 | { | |
1827 | struct bpf_elf_ctx *ctx = &__ctx; | |
1828 | int fd = 0, ret; | |
6256f8c9 | 1829 | |
32e93fb7 DB |
1830 | ret = bpf_elf_ctx_init(ctx, pathname, type, verbose); |
1831 | if (ret < 0) { | |
1832 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
1833 | return ret; | |
1834 | } | |
6256f8c9 | 1835 | |
32e93fb7 DB |
1836 | ret = bpf_fetch_ancillary(ctx); |
1837 | if (ret < 0) { | |
1838 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
1839 | goto out; | |
1840 | } | |
1841 | ||
1842 | fd = bpf_fetch_prog_sec(ctx, section); | |
1843 | if (fd < 0) { | |
1844 | fprintf(stderr, "Error fetching program/map!\n"); | |
1845 | ret = fd; | |
1846 | goto out; | |
1847 | } | |
1848 | ||
1849 | ret = bpf_fill_prog_arrays(ctx); | |
1850 | if (ret < 0) | |
1851 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 1852 | out: |
32e93fb7 DB |
1853 | bpf_elf_ctx_destroy(ctx, ret < 0); |
1854 | if (ret < 0) { | |
1855 | if (fd) | |
1856 | close(fd); | |
1857 | return ret; | |
1858 | } | |
1859 | ||
1860 | return fd; | |
6256f8c9 | 1861 | } |
11c39b5e | 1862 | |
6256f8c9 | 1863 | static int |
4bd62446 DB |
1864 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
1865 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 DB |
1866 | { |
1867 | struct bpf_map_set_msg msg; | |
1868 | int *cmsg_buf, min_fd; | |
1869 | char *amsg_buf; | |
1870 | int i; | |
1871 | ||
1872 | memset(&msg, 0, sizeof(msg)); | |
1873 | ||
1874 | msg.aux.uds_ver = BPF_SCM_AUX_VER; | |
4bd62446 | 1875 | msg.aux.num_ent = entries; |
6256f8c9 DB |
1876 | |
1877 | strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); | |
1878 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); | |
1879 | ||
1880 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
1881 | amsg_buf = (char *)msg.aux.ent; | |
1882 | ||
4bd62446 | 1883 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
1884 | int ret; |
1885 | ||
4bd62446 | 1886 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
1887 | bpf_map_set_init_single(&msg, min_fd); |
1888 | ||
1889 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
1890 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
1891 | ||
1892 | ret = sendmsg(fd, &msg.hdr, 0); | |
1893 | if (ret <= 0) | |
1894 | return ret ? : -1; | |
1895 | } | |
1896 | ||
1897 | return 0; | |
11c39b5e DB |
1898 | } |
1899 | ||
4bd62446 DB |
1900 | static int |
1901 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
1902 | unsigned int entries) | |
1903 | { | |
1904 | struct bpf_map_set_msg msg; | |
1905 | int *cmsg_buf, min_fd; | |
1906 | char *amsg_buf, *mmsg_buf; | |
1907 | unsigned int needed = 1; | |
1908 | int i; | |
1909 | ||
1910 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
1911 | amsg_buf = (char *)msg.aux.ent; | |
1912 | mmsg_buf = (char *)&msg.aux; | |
1913 | ||
1914 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
1915 | struct cmsghdr *cmsg; | |
1916 | int ret; | |
1917 | ||
1918 | min_fd = min(entries, entries - i); | |
1919 | bpf_map_set_init_single(&msg, min_fd); | |
1920 | ||
1921 | ret = recvmsg(fd, &msg.hdr, 0); | |
1922 | if (ret <= 0) | |
1923 | return ret ? : -1; | |
1924 | ||
1925 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
1926 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
1927 | return -EINVAL; | |
1928 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
1929 | return -EIO; | |
1930 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
1931 | return -ENOSYS; | |
1932 | ||
1933 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
1934 | if (min_fd > entries || min_fd <= 0) | |
1935 | return -EINVAL; | |
1936 | ||
1937 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
1938 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
1939 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
1940 | ||
1941 | needed = aux->num_ent; | |
1942 | } | |
1943 | ||
1944 | return 0; | |
1945 | } | |
1946 | ||
1947 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 1948 | { |
32e93fb7 | 1949 | struct bpf_elf_ctx *ctx = &__ctx; |
6256f8c9 DB |
1950 | struct sockaddr_un addr; |
1951 | struct bpf_map_data bpf_aux; | |
1952 | int fd, ret; | |
1953 | ||
1954 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
1955 | if (fd < 0) { | |
1956 | fprintf(stderr, "Cannot open socket: %s\n", | |
1957 | strerror(errno)); | |
1958 | return -1; | |
1959 | } | |
1960 | ||
1961 | memset(&addr, 0, sizeof(addr)); | |
1962 | addr.sun_family = AF_UNIX; | |
1963 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); | |
1964 | ||
1965 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
1966 | if (ret < 0) { | |
1967 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
1968 | path, strerror(errno)); | |
1969 | return -1; | |
1970 | } | |
1971 | ||
1972 | memset(&bpf_aux, 0, sizeof(bpf_aux)); | |
1973 | ||
32e93fb7 DB |
1974 | bpf_aux.fds = ctx->map_fds; |
1975 | bpf_aux.ent = ctx->maps; | |
1976 | bpf_aux.st = &ctx->stat; | |
6256f8c9 | 1977 | bpf_aux.obj = obj; |
6256f8c9 | 1978 | |
4bd62446 | 1979 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 1980 | bpf_maps_count(ctx)); |
6256f8c9 | 1981 | if (ret < 0) |
4bd62446 DB |
1982 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
1983 | path, strerror(errno)); | |
1984 | ||
32e93fb7 | 1985 | bpf_maps_teardown(ctx); |
4bd62446 DB |
1986 | close(fd); |
1987 | return ret; | |
1988 | } | |
1989 | ||
1990 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
1991 | unsigned int entries) | |
1992 | { | |
1993 | struct sockaddr_un addr; | |
1994 | int fd, ret; | |
1995 | ||
1996 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
1997 | if (fd < 0) { | |
1998 | fprintf(stderr, "Cannot open socket: %s\n", | |
1999 | strerror(errno)); | |
2000 | return -1; | |
2001 | } | |
2002 | ||
2003 | memset(&addr, 0, sizeof(addr)); | |
2004 | addr.sun_family = AF_UNIX; | |
2005 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); | |
2006 | ||
2007 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2008 | if (ret < 0) { | |
2009 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
2010 | strerror(errno)); | |
2011 | return -1; | |
2012 | } | |
2013 | ||
2014 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
2015 | if (ret < 0) | |
2016 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
2017 | path, strerror(errno)); |
2018 | ||
4bd62446 | 2019 | unlink(addr.sun_path); |
6256f8c9 DB |
2020 | close(fd); |
2021 | return ret; | |
2022 | } | |
11c39b5e | 2023 | #endif /* HAVE_ELF */ |