]> git.proxmox.com Git - mirror_iproute2.git/blob - lib/bpf.c
projects / mirror_iproute2.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
bpf: allow loading programs for a specific ifindex
[mirror_iproute2.git] / lib / bpf.c
1 /*
2 * 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 <daniel@iogearbox.net>
10 * Jiri Pirko <jiri@resnulli.us>
11 * Alexei Starovoitov <ast@kernel.org>
12 */
13
14 #include <stdio.h>
15 #include <stdlib.h>
16 #include <unistd.h>
17 #include <string.h>
18 #include <stdbool.h>
19 #include <stdint.h>
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <stdarg.h>
23 #include <limits.h>
24 #include <assert.h>
25
26 #ifdef HAVE_ELF
27 #include <libelf.h>
28 #include <gelf.h>
29 #endif
30
31 #include <sys/types.h>
32 #include <sys/stat.h>
33 #include <sys/un.h>
34 #include <sys/vfs.h>
35 #include <sys/mount.h>
36 #include <sys/syscall.h>
37 #include <sys/sendfile.h>
38 #include <sys/resource.h>
39
40 #include <arpa/inet.h>
41
42 #include "utils.h"
43 #include "json_print.h"
44
45 #include "bpf_util.h"
46 #include "bpf_elf.h"
47 #include "bpf_scm.h"
48
49 struct bpf_prog_meta {
50 const char *type;
51 const char *subdir;
52 const char *section;
53 bool may_uds_export;
54 };
55
56 static const enum bpf_prog_type __bpf_types[] = {
57 BPF_PROG_TYPE_SCHED_CLS,
58 BPF_PROG_TYPE_SCHED_ACT,
59 BPF_PROG_TYPE_XDP,
60 BPF_PROG_TYPE_LWT_IN,
61 BPF_PROG_TYPE_LWT_OUT,
62 BPF_PROG_TYPE_LWT_XMIT,
63 };
64
65 static const struct bpf_prog_meta __bpf_prog_meta[] = {
66 [BPF_PROG_TYPE_SCHED_CLS] = {
67 .type = "cls",
68 .subdir = "tc",
69 .section = ELF_SECTION_CLASSIFIER,
70 .may_uds_export = true,
71 },
72 [BPF_PROG_TYPE_SCHED_ACT] = {
73 .type = "act",
74 .subdir = "tc",
75 .section = ELF_SECTION_ACTION,
76 .may_uds_export = true,
77 },
78 [BPF_PROG_TYPE_XDP] = {
79 .type = "xdp",
80 .subdir = "xdp",
81 .section = ELF_SECTION_PROG,
82 },
83 [BPF_PROG_TYPE_LWT_IN] = {
84 .type = "lwt_in",
85 .subdir = "ip",
86 .section = ELF_SECTION_PROG,
87 },
88 [BPF_PROG_TYPE_LWT_OUT] = {
89 .type = "lwt_out",
90 .subdir = "ip",
91 .section = ELF_SECTION_PROG,
92 },
93 [BPF_PROG_TYPE_LWT_XMIT] = {
94 .type = "lwt_xmit",
95 .subdir = "ip",
96 .section = ELF_SECTION_PROG,
97 },
98 };
99
100 static const char *bpf_prog_to_subdir(enum bpf_prog_type type)
101 {
102 assert(type < ARRAY_SIZE(__bpf_prog_meta) &&
103 __bpf_prog_meta[type].subdir);
104 return __bpf_prog_meta[type].subdir;
105 }
106
107 const char *bpf_prog_to_default_section(enum bpf_prog_type type)
108 {
109 assert(type < ARRAY_SIZE(__bpf_prog_meta) &&
110 __bpf_prog_meta[type].section);
111 return __bpf_prog_meta[type].section;
112 }
113
114 #ifdef HAVE_ELF
115 static int bpf_obj_open(const char *path, enum bpf_prog_type type,
116 const char *sec, __u32 ifindex, bool verbose);
117 #else
118 static int bpf_obj_open(const char *path, enum bpf_prog_type type,
119 const char *sec, __u32 ifindex, bool verbose)
120 {
121 fprintf(stderr, "No ELF library support compiled in.\n");
122 errno = ENOSYS;
123 return -1;
124 }
125 #endif
126
127 static inline __u64 bpf_ptr_to_u64(const void *ptr)
128 {
129 return (__u64)(unsigned long)ptr;
130 }
131
132 static int bpf(int cmd, union bpf_attr *attr, unsigned int size)
133 {
134 #ifdef __NR_bpf
135 return syscall(__NR_bpf, cmd, attr, size);
136 #else
137 fprintf(stderr, "No bpf syscall, kernel headers too old?\n");
138 errno = ENOSYS;
139 return -1;
140 #endif
141 }
142
143 static int bpf_map_update(int fd, const void *key, const void *value,
144 uint64_t flags)
145 {
146 union bpf_attr attr = {};
147
148 attr.map_fd = fd;
149 attr.key = bpf_ptr_to_u64(key);
150 attr.value = bpf_ptr_to_u64(value);
151 attr.flags = flags;
152
153 return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr));
154 }
155
156 static int bpf_prog_fd_by_id(uint32_t id)
157 {
158 union bpf_attr attr = {};
159
160 attr.prog_id = id;
161
162 return bpf(BPF_PROG_GET_FD_BY_ID, &attr, sizeof(attr));
163 }
164
165 static int bpf_prog_info_by_fd(int fd, struct bpf_prog_info *info,
166 uint32_t *info_len)
167 {
168 union bpf_attr attr = {};
169 int ret;
170
171 attr.info.bpf_fd = fd;
172 attr.info.info = bpf_ptr_to_u64(info);
173 attr.info.info_len = *info_len;
174
175 *info_len = 0;
176 ret = bpf(BPF_OBJ_GET_INFO_BY_FD, &attr, sizeof(attr));
177 if (!ret)
178 *info_len = attr.info.info_len;
179
180 return ret;
181 }
182
183 int bpf_dump_prog_info(FILE *f, uint32_t id)
184 {
185 struct bpf_prog_info info = {};
186 uint32_t len = sizeof(info);
187 int fd, ret, dump_ok = 0;
188 SPRINT_BUF(tmp);
189
190 open_json_object("prog");
191 print_uint(PRINT_ANY, "id", "id %u ", id);
192
193 fd = bpf_prog_fd_by_id(id);
194 if (fd < 0)
195 goto out;
196
197 ret = bpf_prog_info_by_fd(fd, &info, &len);
198 if (!ret && len) {
199 int jited = !!info.jited_prog_len;
200
201 print_string(PRINT_ANY, "tag", "tag %s ",
202 hexstring_n2a(info.tag, sizeof(info.tag),
203 tmp, sizeof(tmp)));
204 print_uint(PRINT_JSON, "jited", NULL, jited);
205 if (jited && !is_json_context())
206 fprintf(f, "jited ");
207 dump_ok = 1;
208 }
209
210 close(fd);
211 out:
212 close_json_object();
213 return dump_ok;
214 }
215
216 static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len,
217 char **bpf_string, bool *need_release,
218 const char separator)
219 {
220 char sp;
221
222 if (from_file) {
223 size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,");
224 char *tmp_string, *pos, c_prev = ' ';
225 FILE *fp;
226 int c;
227
228 tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len;
229 tmp_string = pos = calloc(1, tmp_len);
230 if (tmp_string == NULL)
231 return -ENOMEM;
232
233 fp = fopen(arg, "r");
234 if (fp == NULL) {
235 perror("Cannot fopen");
236 free(tmp_string);
237 return -ENOENT;
238 }
239
240 while ((c = fgetc(fp)) != EOF) {
241 switch (c) {
242 case '\n':
243 if (c_prev != ',')
244 *(pos++) = ',';
245 c_prev = ',';
246 break;
247 case ' ':
248 case '\t':
249 if (c_prev != ' ')
250 *(pos++) = c;
251 c_prev = ' ';
252 break;
253 default:
254 *(pos++) = c;
255 c_prev = c;
256 }
257 if (pos - tmp_string == tmp_len)
258 break;
259 }
260
261 if (!feof(fp)) {
262 free(tmp_string);
263 fclose(fp);
264 return -E2BIG;
265 }
266
267 fclose(fp);
268 *pos = 0;
269
270 *need_release = true;
271 *bpf_string = tmp_string;
272 } else {
273 *need_release = false;
274 *bpf_string = arg;
275 }
276
277 if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 ||
278 sp != separator) {
279 if (*need_release)
280 free(*bpf_string);
281 return -EINVAL;
282 }
283
284 return 0;
285 }
286
287 static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops,
288 bool from_file)
289 {
290 char *bpf_string, *token, separator = ',';
291 int ret = 0, i = 0;
292 bool need_release;
293 __u16 bpf_len = 0;
294
295 if (argc < 1)
296 return -EINVAL;
297 if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string,
298 &need_release, separator))
299 return -EINVAL;
300 if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) {
301 ret = -EINVAL;
302 goto out;
303 }
304
305 token = bpf_string;
306 while ((token = strchr(token, separator)) && (++token)[0]) {
307 if (i >= bpf_len) {
308 fprintf(stderr, "Real program length exceeds encoded length parameter!\n");
309 ret = -EINVAL;
310 goto out;
311 }
312
313 if (sscanf(token, "%hu %hhu %hhu %u,",
314 &bpf_ops[i].code, &bpf_ops[i].jt,
315 &bpf_ops[i].jf, &bpf_ops[i].k) != 4) {
316 fprintf(stderr, "Error at instruction %d!\n", i);
317 ret = -EINVAL;
318 goto out;
319 }
320
321 i++;
322 }
323
324 if (i != bpf_len) {
325 fprintf(stderr, "Parsed program length is less than encoded length parameter!\n");
326 ret = -EINVAL;
327 goto out;
328 }
329 ret = bpf_len;
330 out:
331 if (need_release)
332 free(bpf_string);
333
334 return ret;
335 }
336
337 void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len)
338 {
339 struct sock_filter *ops = RTA_DATA(bpf_ops);
340 int i;
341
342 if (len == 0)
343 return;
344
345 fprintf(f, "bytecode \'%u,", len);
346
347 for (i = 0; i < len - 1; i++)
348 fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt,
349 ops[i].jf, ops[i].k);
350
351 fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt,
352 ops[i].jf, ops[i].k);
353 }
354
355 static void bpf_map_pin_report(const struct bpf_elf_map *pin,
356 const struct bpf_elf_map *obj)
357 {
358 fprintf(stderr, "Map specification differs from pinned file!\n");
359
360 if (obj->type != pin->type)
361 fprintf(stderr, " - Type: %u (obj) != %u (pin)\n",
362 obj->type, pin->type);
363 if (obj->size_key != pin->size_key)
364 fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n",
365 obj->size_key, pin->size_key);
366 if (obj->size_value != pin->size_value)
367 fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n",
368 obj->size_value, pin->size_value);
369 if (obj->max_elem != pin->max_elem)
370 fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n",
371 obj->max_elem, pin->max_elem);
372 if (obj->flags != pin->flags)
373 fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n",
374 obj->flags, pin->flags);
375
376 fprintf(stderr, "\n");
377 }
378
379 struct bpf_prog_data {
380 unsigned int type;
381 unsigned int jited;
382 };
383
384 struct bpf_map_ext {
385 struct bpf_prog_data owner;
386 };
387
388 static int bpf_derive_elf_map_from_fdinfo(int fd, struct bpf_elf_map *map,
389 struct bpf_map_ext *ext)
390 {
391 unsigned int val, owner_type = 0, owner_jited = 0;
392 char file[PATH_MAX], buff[4096];
393 FILE *fp;
394
395 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
396 memset(map, 0, sizeof(*map));
397
398 fp = fopen(file, "r");
399 if (!fp) {
400 fprintf(stderr, "No procfs support?!\n");
401 return -EIO;
402 }
403
404 while (fgets(buff, sizeof(buff), fp)) {
405 if (sscanf(buff, "map_type:\t%u", &val) == 1)
406 map->type = val;
407 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
408 map->size_key = val;
409 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
410 map->size_value = val;
411 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
412 map->max_elem = val;
413 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
414 map->flags = val;
415 else if (sscanf(buff, "owner_prog_type:\t%i", &val) == 1)
416 owner_type = val;
417 else if (sscanf(buff, "owner_jited:\t%i", &val) == 1)
418 owner_jited = val;
419 }
420
421 fclose(fp);
422 if (ext) {
423 memset(ext, 0, sizeof(*ext));
424 ext->owner.type = owner_type;
425 ext->owner.jited = owner_jited;
426 }
427
428 return 0;
429 }
430
431 static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map,
432 struct bpf_map_ext *ext, int length,
433 enum bpf_prog_type type)
434 {
435 struct bpf_elf_map tmp, zero = {};
436 int ret;
437
438 ret = bpf_derive_elf_map_from_fdinfo(fd, &tmp, ext);
439 if (ret < 0)
440 return ret;
441
442 /* The decision to reject this is on kernel side eventually, but
443 * at least give the user a chance to know what's wrong.
444 */
445 if (ext->owner.type && ext->owner.type != type)
446 fprintf(stderr, "Program array map owner types differ: %u (obj) != %u (pin)\n",
447 type, ext->owner.type);
448
449 if (!memcmp(&tmp, map, length)) {
450 return 0;
451 } else {
452 /* If kernel doesn't have eBPF-related fdinfo, we cannot do much,
453 * so just accept it. We know we do have an eBPF fd and in this
454 * case, everything is 0. It is guaranteed that no such map exists
455 * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC.
456 */
457 if (!memcmp(&tmp, &zero, length))
458 return 0;
459
460 bpf_map_pin_report(&tmp, map);
461 return -EINVAL;
462 }
463 }
464
465 static int bpf_mnt_fs(const char *target)
466 {
467 bool bind_done = false;
468
469 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) {
470 if (errno != EINVAL || bind_done) {
471 fprintf(stderr, "mount --make-private %s failed: %s\n",
472 target, strerror(errno));
473 return -1;
474 }
475
476 if (mount(target, target, "none", MS_BIND, NULL)) {
477 fprintf(stderr, "mount --bind %s %s failed: %s\n",
478 target, target, strerror(errno));
479 return -1;
480 }
481
482 bind_done = true;
483 }
484
485 if (mount("bpf", target, "bpf", 0, "mode=0700")) {
486 fprintf(stderr, "mount -t bpf bpf %s failed: %s\n",
487 target, strerror(errno));
488 return -1;
489 }
490
491 return 0;
492 }
493
494 static int bpf_mnt_check_target(const char *target)
495 {
496 struct stat sb = {};
497 int ret;
498
499 ret = stat(target, &sb);
500 if (ret) {
501 ret = mkdir(target, S_IRWXU);
502 if (ret) {
503 fprintf(stderr, "mkdir %s failed: %s\n", target,
504 strerror(errno));
505 return ret;
506 }
507 }
508
509 return 0;
510 }
511
512 static int bpf_valid_mntpt(const char *mnt, unsigned long magic)
513 {
514 struct statfs st_fs;
515
516 if (statfs(mnt, &st_fs) < 0)
517 return -ENOENT;
518 if ((unsigned long)st_fs.f_type != magic)
519 return -ENOENT;
520
521 return 0;
522 }
523
524 static const char *bpf_find_mntpt_single(unsigned long magic, char *mnt,
525 int len, const char *mntpt)
526 {
527 int ret;
528
529 ret = bpf_valid_mntpt(mntpt, magic);
530 if (!ret) {
531 strlcpy(mnt, mntpt, len);
532 return mnt;
533 }
534
535 return NULL;
536 }
537
538 static const char *bpf_find_mntpt(const char *fstype, unsigned long magic,
539 char *mnt, int len,
540 const char * const *known_mnts)
541 {
542 const char * const *ptr;
543 char type[100];
544 FILE *fp;
545
546 if (known_mnts) {
547 ptr = known_mnts;
548 while (*ptr) {
549 if (bpf_find_mntpt_single(magic, mnt, len, *ptr))
550 return mnt;
551 ptr++;
552 }
553 }
554
555 if (len != PATH_MAX)
556 return NULL;
557
558 fp = fopen("/proc/mounts", "r");
559 if (fp == NULL)
560 return NULL;
561
562 while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n",
563 mnt, type) == 2) {
564 if (strcmp(type, fstype) == 0)
565 break;
566 }
567
568 fclose(fp);
569 if (strcmp(type, fstype) != 0)
570 return NULL;
571
572 return mnt;
573 }
574
575 int bpf_trace_pipe(void)
576 {
577 char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT;
578 static const char * const tracefs_known_mnts[] = {
579 TRACE_DIR_MNT,
580 "/sys/kernel/debug/tracing",
581 "/tracing",
582 "/trace",
583 0,
584 };
585 int fd_in, fd_out = STDERR_FILENO;
586 char tpipe[PATH_MAX];
587 const char *mnt;
588
589 mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt,
590 sizeof(tracefs_mnt), tracefs_known_mnts);
591 if (!mnt) {
592 fprintf(stderr, "tracefs not mounted?\n");
593 return -1;
594 }
595
596 snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt);
597
598 fd_in = open(tpipe, O_RDONLY);
599 if (fd_in < 0)
600 return -1;
601
602 fprintf(stderr, "Running! Hang up with ^C!\n\n");
603 while (1) {
604 static char buff[4096];
605 ssize_t ret;
606
607 ret = read(fd_in, buff, sizeof(buff));
608 if (ret > 0 && write(fd_out, buff, ret) == ret)
609 continue;
610 break;
611 }
612
613 close(fd_in);
614 return -1;
615 }
616
617 static int bpf_gen_global(const char *bpf_sub_dir)
618 {
619 char bpf_glo_dir[PATH_MAX];
620 int ret;
621
622 snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/",
623 bpf_sub_dir, BPF_DIR_GLOBALS);
624
625 ret = mkdir(bpf_glo_dir, S_IRWXU);
626 if (ret && errno != EEXIST) {
627 fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir,
628 strerror(errno));
629 return ret;
630 }
631
632 return 0;
633 }
634
635 static int bpf_gen_master(const char *base, const char *name)
636 {
637 char bpf_sub_dir[PATH_MAX];
638 int ret;
639
640 snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name);
641
642 ret = mkdir(bpf_sub_dir, S_IRWXU);
643 if (ret && errno != EEXIST) {
644 fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir,
645 strerror(errno));
646 return ret;
647 }
648
649 return bpf_gen_global(bpf_sub_dir);
650 }
651
652 static int bpf_slave_via_bind_mnt(const char *full_name,
653 const char *full_link)
654 {
655 int ret;
656
657 ret = mkdir(full_name, S_IRWXU);
658 if (ret) {
659 assert(errno != EEXIST);
660 fprintf(stderr, "mkdir %s failed: %s\n", full_name,
661 strerror(errno));
662 return ret;
663 }
664
665 ret = mount(full_link, full_name, "none", MS_BIND, NULL);
666 if (ret) {
667 rmdir(full_name);
668 fprintf(stderr, "mount --bind %s %s failed: %s\n",
669 full_link, full_name, strerror(errno));
670 }
671
672 return ret;
673 }
674
675 static int bpf_gen_slave(const char *base, const char *name,
676 const char *link)
677 {
678 char bpf_lnk_dir[PATH_MAX];
679 char bpf_sub_dir[PATH_MAX];
680 struct stat sb = {};
681 int ret;
682
683 snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link);
684 snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name);
685
686 ret = symlink(bpf_lnk_dir, bpf_sub_dir);
687 if (ret) {
688 if (errno != EEXIST) {
689 if (errno != EPERM) {
690 fprintf(stderr, "symlink %s failed: %s\n",
691 bpf_sub_dir, strerror(errno));
692 return ret;
693 }
694
695 return bpf_slave_via_bind_mnt(bpf_sub_dir,
696 bpf_lnk_dir);
697 }
698
699 ret = lstat(bpf_sub_dir, &sb);
700 if (ret) {
701 fprintf(stderr, "lstat %s failed: %s\n",
702 bpf_sub_dir, strerror(errno));
703 return ret;
704 }
705
706 if ((sb.st_mode & S_IFMT) != S_IFLNK)
707 return bpf_gen_global(bpf_sub_dir);
708 }
709
710 return 0;
711 }
712
713 static int bpf_gen_hierarchy(const char *base)
714 {
715 int ret, i;
716
717 ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0]));
718 for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++)
719 ret = bpf_gen_slave(base,
720 bpf_prog_to_subdir(__bpf_types[i]),
721 bpf_prog_to_subdir(__bpf_types[0]));
722 return ret;
723 }
724
725 static const char *bpf_get_work_dir(enum bpf_prog_type type)
726 {
727 static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT;
728 static char bpf_wrk_dir[PATH_MAX];
729 static const char *mnt;
730 static bool bpf_mnt_cached;
731 const char *mnt_env = getenv(BPF_ENV_MNT);
732 static const char * const bpf_known_mnts[] = {
733 BPF_DIR_MNT,
734 "/bpf",
735 0,
736 };
737 int ret;
738
739 if (bpf_mnt_cached) {
740 const char *out = mnt;
741
742 if (out && type) {
743 snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/",
744 out, bpf_prog_to_subdir(type));
745 out = bpf_tmp;
746 }
747 return out;
748 }
749
750 if (mnt_env)
751 mnt = bpf_find_mntpt_single(BPF_FS_MAGIC, bpf_tmp,
752 sizeof(bpf_tmp), mnt_env);
753 else
754 mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp,
755 sizeof(bpf_tmp), bpf_known_mnts);
756 if (!mnt) {
757 mnt = mnt_env ? : BPF_DIR_MNT;
758 ret = bpf_mnt_check_target(mnt);
759 if (!ret)
760 ret = bpf_mnt_fs(mnt);
761 if (ret) {
762 mnt = NULL;
763 goto out;
764 }
765 }
766
767 snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt);
768
769 ret = bpf_gen_hierarchy(bpf_wrk_dir);
770 if (ret) {
771 mnt = NULL;
772 goto out;
773 }
774
775 mnt = bpf_wrk_dir;
776 out:
777 bpf_mnt_cached = true;
778 return mnt;
779 }
780
781 static int bpf_obj_get(const char *pathname, enum bpf_prog_type type)
782 {
783 union bpf_attr attr = {};
784 char tmp[PATH_MAX];
785
786 if (strlen(pathname) > 2 && pathname[0] == 'm' &&
787 pathname[1] == ':' && bpf_get_work_dir(type)) {
788 snprintf(tmp, sizeof(tmp), "%s/%s",
789 bpf_get_work_dir(type), pathname + 2);
790 pathname = tmp;
791 }
792
793 attr.pathname = bpf_ptr_to_u64(pathname);
794
795 return bpf(BPF_OBJ_GET, &attr, sizeof(attr));
796 }
797
798 static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type)
799 {
800 int prog_fd = bpf_obj_get(pathname, type);
801
802 if (prog_fd < 0)
803 fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n",
804 pathname, strerror(errno));
805 return prog_fd;
806 }
807
808 static int bpf_do_parse(struct bpf_cfg_in *cfg, const bool *opt_tbl)
809 {
810 const char *file, *section, *uds_name;
811 bool verbose = false;
812 int i, ret, argc;
813 char **argv;
814
815 argv = cfg->argv;
816 argc = cfg->argc;
817
818 if (opt_tbl[CBPF_BYTECODE] &&
819 (matches(*argv, "bytecode") == 0 ||
820 strcmp(*argv, "bc") == 0)) {
821 cfg->mode = CBPF_BYTECODE;
822 } else if (opt_tbl[CBPF_FILE] &&
823 (matches(*argv, "bytecode-file") == 0 ||
824 strcmp(*argv, "bcf") == 0)) {
825 cfg->mode = CBPF_FILE;
826 } else if (opt_tbl[EBPF_OBJECT] &&
827 (matches(*argv, "object-file") == 0 ||
828 strcmp(*argv, "obj") == 0)) {
829 cfg->mode = EBPF_OBJECT;
830 } else if (opt_tbl[EBPF_PINNED] &&
831 (matches(*argv, "object-pinned") == 0 ||
832 matches(*argv, "pinned") == 0 ||
833 matches(*argv, "fd") == 0)) {
834 cfg->mode = EBPF_PINNED;
835 } else {
836 fprintf(stderr, "What mode is \"%s\"?\n", *argv);
837 return -1;
838 }
839
840 NEXT_ARG();
841 file = section = uds_name = NULL;
842 if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) {
843 file = *argv;
844 NEXT_ARG_FWD();
845
846 if (cfg->type == BPF_PROG_TYPE_UNSPEC) {
847 if (argc > 0 && matches(*argv, "type") == 0) {
848 NEXT_ARG();
849 for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta);
850 i++) {
851 if (!__bpf_prog_meta[i].type)
852 continue;
853 if (!matches(*argv,
854 __bpf_prog_meta[i].type)) {
855 cfg->type = i;
856 break;
857 }
858 }
859
860 if (cfg->type == BPF_PROG_TYPE_UNSPEC) {
861 fprintf(stderr, "What type is \"%s\"?\n",
862 *argv);
863 return -1;
864 }
865 NEXT_ARG_FWD();
866 } else {
867 cfg->type = BPF_PROG_TYPE_SCHED_CLS;
868 }
869 }
870
871 section = bpf_prog_to_default_section(cfg->type);
872 if (argc > 0 && matches(*argv, "section") == 0) {
873 NEXT_ARG();
874 section = *argv;
875 NEXT_ARG_FWD();
876 }
877
878 if (__bpf_prog_meta[cfg->type].may_uds_export) {
879 uds_name = getenv(BPF_ENV_UDS);
880 if (argc > 0 && !uds_name &&
881 matches(*argv, "export") == 0) {
882 NEXT_ARG();
883 uds_name = *argv;
884 NEXT_ARG_FWD();
885 }
886 }
887
888 if (argc > 0 && matches(*argv, "verbose") == 0) {
889 verbose = true;
890 NEXT_ARG_FWD();
891 }
892
893 PREV_ARG();
894 }
895
896 if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE) {
897 ret = bpf_ops_parse(argc, argv, cfg->opcodes,
898 cfg->mode == CBPF_FILE);
899 cfg->n_opcodes = ret;
900 } else if (cfg->mode == EBPF_OBJECT) {
901 ret = 0; /* program will be loaded by load stage */
902 } else if (cfg->mode == EBPF_PINNED) {
903 ret = bpf_obj_pinned(file, cfg->type);
904 cfg->prog_fd = ret;
905 } else {
906 return -1;
907 }
908
909 cfg->object = file;
910 cfg->section = section;
911 cfg->uds = uds_name;
912 cfg->argc = argc;
913 cfg->argv = argv;
914 cfg->verbose = verbose;
915
916 return ret;
917 }
918
919 static int bpf_do_load(struct bpf_cfg_in *cfg)
920 {
921 if (cfg->mode == EBPF_OBJECT) {
922 cfg->prog_fd = bpf_obj_open(cfg->object, cfg->type,
923 cfg->section, cfg->ifindex,
924 cfg->verbose);
925 return cfg->prog_fd;
926 }
927 return 0;
928 }
929
930 int bpf_load_common(struct bpf_cfg_in *cfg, const struct bpf_cfg_ops *ops,
931 void *nl)
932 {
933 char annotation[256];
934 int ret;
935
936 ret = bpf_do_load(cfg);
937 if (ret < 0)
938 return ret;
939
940 if (cfg->mode == CBPF_BYTECODE || cfg->mode == CBPF_FILE)
941 ops->cbpf_cb(nl, cfg->opcodes, cfg->n_opcodes);
942 if (cfg->mode == EBPF_OBJECT || cfg->mode == EBPF_PINNED) {
943 snprintf(annotation, sizeof(annotation), "%s:[%s]",
944 basename(cfg->object), cfg->mode == EBPF_PINNED ?
945 "*fsobj" : cfg->section);
946 ops->ebpf_cb(nl, cfg->prog_fd, annotation);
947 }
948
949 return 0;
950 }
951
952 int bpf_parse_common(struct bpf_cfg_in *cfg, const struct bpf_cfg_ops *ops)
953 {
954 bool opt_tbl[BPF_MODE_MAX] = {};
955
956 if (ops->cbpf_cb) {
957 opt_tbl[CBPF_BYTECODE] = true;
958 opt_tbl[CBPF_FILE] = true;
959 }
960
961 if (ops->ebpf_cb) {
962 opt_tbl[EBPF_OBJECT] = true;
963 opt_tbl[EBPF_PINNED] = true;
964 }
965
966 return bpf_do_parse(cfg, opt_tbl);
967 }
968
969 int bpf_parse_and_load_common(struct bpf_cfg_in *cfg,
970 const struct bpf_cfg_ops *ops, void *nl)
971 {
972 int ret;
973
974 ret = bpf_parse_common(cfg, ops);
975 if (ret < 0)
976 return ret;
977
978 return bpf_load_common(cfg, ops, nl);
979 }
980
981 int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv)
982 {
983 const bool opt_tbl[BPF_MODE_MAX] = {
984 [EBPF_OBJECT] = true,
985 [EBPF_PINNED] = true,
986 };
987 const struct bpf_elf_map test = {
988 .type = BPF_MAP_TYPE_PROG_ARRAY,
989 .size_key = sizeof(int),
990 .size_value = sizeof(int),
991 };
992 struct bpf_cfg_in cfg = {
993 .type = BPF_PROG_TYPE_UNSPEC,
994 .argc = argc,
995 .argv = argv,
996 };
997 struct bpf_map_ext ext = {};
998 int ret, prog_fd, map_fd;
999 uint32_t map_key;
1000
1001 ret = bpf_do_parse(&cfg, opt_tbl);
1002 if (ret < 0)
1003 return ret;
1004
1005 ret = bpf_do_load(&cfg);
1006 if (ret < 0)
1007 return ret;
1008
1009 prog_fd = cfg.prog_fd;
1010
1011 if (key) {
1012 map_key = *key;
1013 } else {
1014 ret = sscanf(cfg.section, "%*i/%i", &map_key);
1015 if (ret != 1) {
1016 fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n");
1017 ret = -EINVAL;
1018 goto out_prog;
1019 }
1020 }
1021
1022 map_fd = bpf_obj_get(map_path, cfg.type);
1023 if (map_fd < 0) {
1024 fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n",
1025 map_path, strerror(errno));
1026 ret = map_fd;
1027 goto out_prog;
1028 }
1029
1030 ret = bpf_map_selfcheck_pinned(map_fd, &test, &ext,
1031 offsetof(struct bpf_elf_map, max_elem),
1032 cfg.type);
1033 if (ret < 0) {
1034 fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path);
1035 goto out_map;
1036 }
1037
1038 ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY);
1039 if (ret < 0)
1040 fprintf(stderr, "Map update failed: %s\n", strerror(errno));
1041 out_map:
1042 close(map_fd);
1043 out_prog:
1044 close(prog_fd);
1045 return ret;
1046 }
1047
1048 int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type)
1049 {
1050 union bpf_attr attr = {};
1051
1052 attr.target_fd = target_fd;
1053 attr.attach_bpf_fd = prog_fd;
1054 attr.attach_type = type;
1055
1056 return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr));
1057 }
1058
1059 int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type)
1060 {
1061 union bpf_attr attr = {};
1062
1063 attr.target_fd = target_fd;
1064 attr.attach_type = type;
1065
1066 return bpf(BPF_PROG_DETACH, &attr, sizeof(attr));
1067 }
1068
1069 static int bpf_prog_load_dev(enum bpf_prog_type type,
1070 const struct bpf_insn *insns, size_t size_insns,
1071 const char *license, __u32 ifindex,
1072 char *log, size_t size_log)
1073 {
1074 union bpf_attr attr = {};
1075
1076 attr.prog_type = type;
1077 attr.insns = bpf_ptr_to_u64(insns);
1078 attr.insn_cnt = size_insns / sizeof(struct bpf_insn);
1079 attr.license = bpf_ptr_to_u64(license);
1080 attr.prog_ifindex = ifindex;
1081
1082 if (size_log > 0) {
1083 attr.log_buf = bpf_ptr_to_u64(log);
1084 attr.log_size = size_log;
1085 attr.log_level = 1;
1086 }
1087
1088 return bpf(BPF_PROG_LOAD, &attr, sizeof(attr));
1089 }
1090
1091 int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns,
1092 size_t size_insns, const char *license, char *log,
1093 size_t size_log)
1094 {
1095 return bpf_prog_load_dev(type, insns, size_insns, license, 0,
1096 log, size_log);
1097 }
1098
1099 #ifdef HAVE_ELF
1100 struct bpf_elf_prog {
1101 enum bpf_prog_type type;
1102 const struct bpf_insn *insns;
1103 size_t size;
1104 const char *license;
1105 };
1106
1107 struct bpf_hash_entry {
1108 unsigned int pinning;
1109 const char *subpath;
1110 struct bpf_hash_entry *next;
1111 };
1112
1113 struct bpf_config {
1114 unsigned int jit_enabled;
1115 };
1116
1117 struct bpf_elf_ctx {
1118 struct bpf_config cfg;
1119 Elf *elf_fd;
1120 GElf_Ehdr elf_hdr;
1121 Elf_Data *sym_tab;
1122 Elf_Data *str_tab;
1123 int obj_fd;
1124 int map_fds[ELF_MAX_MAPS];
1125 struct bpf_elf_map maps[ELF_MAX_MAPS];
1126 struct bpf_map_ext maps_ext[ELF_MAX_MAPS];
1127 int sym_num;
1128 int map_num;
1129 int map_len;
1130 bool *sec_done;
1131 int sec_maps;
1132 char license[ELF_MAX_LICENSE_LEN];
1133 enum bpf_prog_type type;
1134 __u32 ifindex;
1135 bool verbose;
1136 struct bpf_elf_st stat;
1137 struct bpf_hash_entry *ht[256];
1138 char *log;
1139 size_t log_size;
1140 };
1141
1142 struct bpf_elf_sec_data {
1143 GElf_Shdr sec_hdr;
1144 Elf_Data *sec_data;
1145 const char *sec_name;
1146 };
1147
1148 struct bpf_map_data {
1149 int *fds;
1150 const char *obj;
1151 struct bpf_elf_st *st;
1152 struct bpf_elf_map *ent;
1153 };
1154
1155 static __check_format_string(2, 3) void
1156 bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...)
1157 {
1158 va_list vl;
1159
1160 va_start(vl, format);
1161 vfprintf(stderr, format, vl);
1162 va_end(vl);
1163
1164 if (ctx->log && ctx->log[0]) {
1165 if (ctx->verbose) {
1166 fprintf(stderr, "%s\n", ctx->log);
1167 } else {
1168 unsigned int off = 0, len = strlen(ctx->log);
1169
1170 if (len > BPF_MAX_LOG) {
1171 off = len - BPF_MAX_LOG;
1172 fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n",
1173 off);
1174 }
1175 fprintf(stderr, "%s\n", ctx->log + off);
1176 }
1177
1178 memset(ctx->log, 0, ctx->log_size);
1179 }
1180 }
1181
1182 static int bpf_log_realloc(struct bpf_elf_ctx *ctx)
1183 {
1184 const size_t log_max = UINT_MAX >> 8;
1185 size_t log_size = ctx->log_size;
1186 char *ptr;
1187
1188 if (!ctx->log) {
1189 log_size = 65536;
1190 } else if (log_size < log_max) {
1191 log_size <<= 1;
1192 if (log_size > log_max)
1193 log_size = log_max;
1194 } else {
1195 return -EINVAL;
1196 }
1197
1198 ptr = realloc(ctx->log, log_size);
1199 if (!ptr)
1200 return -ENOMEM;
1201
1202 ptr[0] = 0;
1203 ctx->log = ptr;
1204 ctx->log_size = log_size;
1205
1206 return 0;
1207 }
1208
1209 static int bpf_map_create(enum bpf_map_type type, uint32_t size_key,
1210 uint32_t size_value, uint32_t max_elem,
1211 uint32_t flags, int inner_fd)
1212 {
1213 union bpf_attr attr = {};
1214
1215 attr.map_type = type;
1216 attr.key_size = size_key;
1217 attr.value_size = inner_fd ? sizeof(int) : size_value;
1218 attr.max_entries = max_elem;
1219 attr.map_flags = flags;
1220 attr.inner_map_fd = inner_fd;
1221
1222 return bpf(BPF_MAP_CREATE, &attr, sizeof(attr));
1223 }
1224
1225 static int bpf_obj_pin(int fd, const char *pathname)
1226 {
1227 union bpf_attr attr = {};
1228
1229 attr.pathname = bpf_ptr_to_u64(pathname);
1230 attr.bpf_fd = fd;
1231
1232 return bpf(BPF_OBJ_PIN, &attr, sizeof(attr));
1233 }
1234
1235 static int bpf_obj_hash(const char *object, uint8_t *out, size_t len)
1236 {
1237 struct sockaddr_alg alg = {
1238 .salg_family = AF_ALG,
1239 .salg_type = "hash",
1240 .salg_name = "sha1",
1241 };
1242 int ret, cfd, ofd, ffd;
1243 struct stat stbuff;
1244 ssize_t size;
1245
1246 if (!object || len != 20)
1247 return -EINVAL;
1248
1249 cfd = socket(AF_ALG, SOCK_SEQPACKET, 0);
1250 if (cfd < 0) {
1251 fprintf(stderr, "Cannot get AF_ALG socket: %s\n",
1252 strerror(errno));
1253 return cfd;
1254 }
1255
1256 ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg));
1257 if (ret < 0) {
1258 fprintf(stderr, "Error binding socket: %s\n", strerror(errno));
1259 goto out_cfd;
1260 }
1261
1262 ofd = accept(cfd, NULL, 0);
1263 if (ofd < 0) {
1264 fprintf(stderr, "Error accepting socket: %s\n",
1265 strerror(errno));
1266 ret = ofd;
1267 goto out_cfd;
1268 }
1269
1270 ffd = open(object, O_RDONLY);
1271 if (ffd < 0) {
1272 fprintf(stderr, "Error opening object %s: %s\n",
1273 object, strerror(errno));
1274 ret = ffd;
1275 goto out_ofd;
1276 }
1277
1278 ret = fstat(ffd, &stbuff);
1279 if (ret < 0) {
1280 fprintf(stderr, "Error doing fstat: %s\n",
1281 strerror(errno));
1282 goto out_ffd;
1283 }
1284
1285 size = sendfile(ofd, ffd, NULL, stbuff.st_size);
1286 if (size != stbuff.st_size) {
1287 fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n",
1288 size, stbuff.st_size, strerror(errno));
1289 ret = -1;
1290 goto out_ffd;
1291 }
1292
1293 size = read(ofd, out, len);
1294 if (size != len) {
1295 fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n",
1296 size, len, strerror(errno));
1297 ret = -1;
1298 } else {
1299 ret = 0;
1300 }
1301 out_ffd:
1302 close(ffd);
1303 out_ofd:
1304 close(ofd);
1305 out_cfd:
1306 close(cfd);
1307 return ret;
1308 }
1309
1310 static const char *bpf_get_obj_uid(const char *pathname)
1311 {
1312 static bool bpf_uid_cached;
1313 static char bpf_uid[64];
1314 uint8_t tmp[20];
1315 int ret;
1316
1317 if (bpf_uid_cached)
1318 goto done;
1319
1320 ret = bpf_obj_hash(pathname, tmp, sizeof(tmp));
1321 if (ret) {
1322 fprintf(stderr, "Object hashing failed!\n");
1323 return NULL;
1324 }
1325
1326 hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid));
1327 bpf_uid_cached = true;
1328 done:
1329 return bpf_uid;
1330 }
1331
1332 static int bpf_init_env(const char *pathname)
1333 {
1334 struct rlimit limit = {
1335 .rlim_cur = RLIM_INFINITY,
1336 .rlim_max = RLIM_INFINITY,
1337 };
1338
1339 /* Don't bother in case we fail! */
1340 setrlimit(RLIMIT_MEMLOCK, &limit);
1341
1342 if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) {
1343 fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n");
1344 return 0;
1345 }
1346
1347 if (!bpf_get_obj_uid(pathname))
1348 return -1;
1349
1350 return 0;
1351 }
1352
1353 static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx,
1354 uint32_t pinning)
1355 {
1356 struct bpf_hash_entry *entry;
1357
1358 entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)];
1359 while (entry && entry->pinning != pinning)
1360 entry = entry->next;
1361
1362 return entry ? entry->subpath : NULL;
1363 }
1364
1365 static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx,
1366 uint32_t pinning)
1367 {
1368 switch (pinning) {
1369 case PIN_OBJECT_NS:
1370 case PIN_GLOBAL_NS:
1371 return false;
1372 case PIN_NONE:
1373 return true;
1374 default:
1375 return !bpf_custom_pinning(ctx, pinning);
1376 }
1377 }
1378
1379 static void bpf_make_pathname(char *pathname, size_t len, const char *name,
1380 const struct bpf_elf_ctx *ctx, uint32_t pinning)
1381 {
1382 switch (pinning) {
1383 case PIN_OBJECT_NS:
1384 snprintf(pathname, len, "%s/%s/%s",
1385 bpf_get_work_dir(ctx->type),
1386 bpf_get_obj_uid(NULL), name);
1387 break;
1388 case PIN_GLOBAL_NS:
1389 snprintf(pathname, len, "%s/%s/%s",
1390 bpf_get_work_dir(ctx->type),
1391 BPF_DIR_GLOBALS, name);
1392 break;
1393 default:
1394 snprintf(pathname, len, "%s/../%s/%s",
1395 bpf_get_work_dir(ctx->type),
1396 bpf_custom_pinning(ctx, pinning), name);
1397 break;
1398 }
1399 }
1400
1401 static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx,
1402 uint32_t pinning)
1403 {
1404 char pathname[PATH_MAX];
1405
1406 if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type))
1407 return 0;
1408
1409 bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning);
1410 return bpf_obj_get(pathname, ctx->type);
1411 }
1412
1413 static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx)
1414 {
1415 char tmp[PATH_MAX];
1416 int ret;
1417
1418 snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type),
1419 bpf_get_obj_uid(NULL));
1420
1421 ret = mkdir(tmp, S_IRWXU);
1422 if (ret && errno != EEXIST) {
1423 fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno));
1424 return ret;
1425 }
1426
1427 return 0;
1428 }
1429
1430 static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx,
1431 const char *todo)
1432 {
1433 char tmp[PATH_MAX], rem[PATH_MAX], *sub;
1434 int ret;
1435
1436 snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type));
1437 snprintf(rem, sizeof(rem), "%s/", todo);
1438 sub = strtok(rem, "/");
1439
1440 while (sub) {
1441 if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX)
1442 return -EINVAL;
1443
1444 strcat(tmp, sub);
1445 strcat(tmp, "/");
1446
1447 ret = mkdir(tmp, S_IRWXU);
1448 if (ret && errno != EEXIST) {
1449 fprintf(stderr, "mkdir %s failed: %s\n", tmp,
1450 strerror(errno));
1451 return ret;
1452 }
1453
1454 sub = strtok(NULL, "/");
1455 }
1456
1457 return 0;
1458 }
1459
1460 static int bpf_place_pinned(int fd, const char *name,
1461 const struct bpf_elf_ctx *ctx, uint32_t pinning)
1462 {
1463 char pathname[PATH_MAX];
1464 const char *tmp;
1465 int ret = 0;
1466
1467 if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type))
1468 return 0;
1469
1470 if (pinning == PIN_OBJECT_NS)
1471 ret = bpf_make_obj_path(ctx);
1472 else if ((tmp = bpf_custom_pinning(ctx, pinning)))
1473 ret = bpf_make_custom_path(ctx, tmp);
1474 if (ret < 0)
1475 return ret;
1476
1477 bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning);
1478 return bpf_obj_pin(fd, pathname);
1479 }
1480
1481 static void bpf_prog_report(int fd, const char *section,
1482 const struct bpf_elf_prog *prog,
1483 struct bpf_elf_ctx *ctx)
1484 {
1485 unsigned int insns = prog->size / sizeof(struct bpf_insn);
1486
1487 fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section,
1488 fd < 0 ? "rejected: " : "loaded",
1489 fd < 0 ? strerror(errno) : "",
1490 fd < 0 ? errno : fd);
1491
1492 fprintf(stderr, " - Type: %u\n", prog->type);
1493 fprintf(stderr, " - Instructions: %u (%u over limit)\n",
1494 insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0);
1495 fprintf(stderr, " - License: %s\n\n", prog->license);
1496
1497 bpf_dump_error(ctx, "Verifier analysis:\n\n");
1498 }
1499
1500 static int bpf_prog_attach(const char *section,
1501 const struct bpf_elf_prog *prog,
1502 struct bpf_elf_ctx *ctx)
1503 {
1504 int tries = 0, fd;
1505 retry:
1506 errno = 0;
1507 fd = bpf_prog_load_dev(prog->type, prog->insns, prog->size,
1508 prog->license, ctx->ifindex,
1509 ctx->log, ctx->log_size);
1510 if (fd < 0 || ctx->verbose) {
1511 /* The verifier log is pretty chatty, sometimes so chatty
1512 * on larger programs, that we could fail to dump everything
1513 * into our buffer. Still, try to give a debuggable error
1514 * log for the user, so enlarge it and re-fail.
1515 */
1516 if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) {
1517 if (tries++ < 10 && !bpf_log_realloc(ctx))
1518 goto retry;
1519
1520 fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n",
1521 ctx->log_size, tries);
1522 return fd;
1523 }
1524
1525 bpf_prog_report(fd, section, prog, ctx);
1526 }
1527
1528 return fd;
1529 }
1530
1531 static void bpf_map_report(int fd, const char *name,
1532 const struct bpf_elf_map *map,
1533 struct bpf_elf_ctx *ctx, int inner_fd)
1534 {
1535 fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name,
1536 fd < 0 ? "rejected: " : "loaded",
1537 fd < 0 ? strerror(errno) : "",
1538 fd < 0 ? errno : fd);
1539
1540 fprintf(stderr, " - Type: %u\n", map->type);
1541 fprintf(stderr, " - Identifier: %u\n", map->id);
1542 fprintf(stderr, " - Pinning: %u\n", map->pinning);
1543 fprintf(stderr, " - Size key: %u\n", map->size_key);
1544 fprintf(stderr, " - Size value: %u\n",
1545 inner_fd ? (int)sizeof(int) : map->size_value);
1546 fprintf(stderr, " - Max elems: %u\n", map->max_elem);
1547 fprintf(stderr, " - Flags: %#x\n\n", map->flags);
1548 }
1549
1550 static int bpf_find_map_id(const struct bpf_elf_ctx *ctx, uint32_t id)
1551 {
1552 int i;
1553
1554 for (i = 0; i < ctx->map_num; i++) {
1555 if (ctx->maps[i].id != id)
1556 continue;
1557 if (ctx->map_fds[i] < 0)
1558 return -EINVAL;
1559
1560 return ctx->map_fds[i];
1561 }
1562
1563 return -ENOENT;
1564 }
1565
1566 static void bpf_report_map_in_map(int outer_fd, uint32_t idx)
1567 {
1568 struct bpf_elf_map outer_map;
1569 int ret;
1570
1571 fprintf(stderr, "Cannot insert map into map! ");
1572
1573 ret = bpf_derive_elf_map_from_fdinfo(outer_fd, &outer_map, NULL);
1574 if (!ret) {
1575 if (idx >= outer_map.max_elem &&
1576 outer_map.type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
1577 fprintf(stderr, "Outer map has %u elements, index %u is invalid!\n",
1578 outer_map.max_elem, idx);
1579 return;
1580 }
1581 }
1582
1583 fprintf(stderr, "Different map specs used for outer and inner map?\n");
1584 }
1585
1586 static bool bpf_is_map_in_map_type(const struct bpf_elf_map *map)
1587 {
1588 return map->type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1589 map->type == BPF_MAP_TYPE_HASH_OF_MAPS;
1590 }
1591
1592 static int bpf_map_attach(const char *name, struct bpf_elf_ctx *ctx,
1593 const struct bpf_elf_map *map, struct bpf_map_ext *ext,
1594 int *have_map_in_map)
1595 {
1596 int fd, ret, map_inner_fd = 0;
1597
1598 fd = bpf_probe_pinned(name, ctx, map->pinning);
1599 if (fd > 0) {
1600 ret = bpf_map_selfcheck_pinned(fd, map, ext,
1601 offsetof(struct bpf_elf_map,
1602 id), ctx->type);
1603 if (ret < 0) {
1604 close(fd);
1605 fprintf(stderr, "Map \'%s\' self-check failed!\n",
1606 name);
1607 return ret;
1608 }
1609 if (ctx->verbose)
1610 fprintf(stderr, "Map \'%s\' loaded as pinned!\n",
1611 name);
1612 return fd;
1613 }
1614
1615 if (have_map_in_map && bpf_is_map_in_map_type(map)) {
1616 (*have_map_in_map)++;
1617 if (map->inner_id)
1618 return 0;
1619 fprintf(stderr, "Map \'%s\' cannot be created since no inner map ID defined!\n",
1620 name);
1621 return -EINVAL;
1622 }
1623
1624 if (!have_map_in_map && bpf_is_map_in_map_type(map)) {
1625 map_inner_fd = bpf_find_map_id(ctx, map->inner_id);
1626 if (map_inner_fd < 0) {
1627 fprintf(stderr, "Map \'%s\' cannot be loaded. Inner map with ID %u not found!\n",
1628 name, map->inner_id);
1629 return -EINVAL;
1630 }
1631 }
1632
1633 errno = 0;
1634 fd = bpf_map_create(map->type, map->size_key, map->size_value,
1635 map->max_elem, map->flags, map_inner_fd);
1636 if (fd < 0 || ctx->verbose) {
1637 bpf_map_report(fd, name, map, ctx, map_inner_fd);
1638 if (fd < 0)
1639 return fd;
1640 }
1641
1642 ret = bpf_place_pinned(fd, name, ctx, map->pinning);
1643 if (ret < 0 && errno != EEXIST) {
1644 fprintf(stderr, "Could not pin %s map: %s\n", name,
1645 strerror(errno));
1646 close(fd);
1647 return ret;
1648 }
1649
1650 return fd;
1651 }
1652
1653 static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx,
1654 const GElf_Sym *sym)
1655 {
1656 return ctx->str_tab->d_buf + sym->st_name;
1657 }
1658
1659 static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which)
1660 {
1661 GElf_Sym sym;
1662 int i;
1663
1664 for (i = 0; i < ctx->sym_num; i++) {
1665 if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym)
1666 continue;
1667
1668 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1669 GELF_ST_TYPE(sym.st_info) != STT_NOTYPE ||
1670 sym.st_shndx != ctx->sec_maps ||
1671 sym.st_value / ctx->map_len != which)
1672 continue;
1673
1674 return bpf_str_tab_name(ctx, &sym);
1675 }
1676
1677 return NULL;
1678 }
1679
1680 static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx)
1681 {
1682 int i, j, ret, fd, inner_fd, inner_idx, have_map_in_map = 0;
1683 const char *map_name;
1684
1685 for (i = 0; i < ctx->map_num; i++) {
1686 map_name = bpf_map_fetch_name(ctx, i);
1687 if (!map_name)
1688 return -EIO;
1689
1690 fd = bpf_map_attach(map_name, ctx, &ctx->maps[i],
1691 &ctx->maps_ext[i], &have_map_in_map);
1692 if (fd < 0)
1693 return fd;
1694
1695 ctx->map_fds[i] = !fd ? -1 : fd;
1696 }
1697
1698 for (i = 0; have_map_in_map && i < ctx->map_num; i++) {
1699 if (ctx->map_fds[i] >= 0)
1700 continue;
1701
1702 map_name = bpf_map_fetch_name(ctx, i);
1703 if (!map_name)
1704 return -EIO;
1705
1706 fd = bpf_map_attach(map_name, ctx, &ctx->maps[i],
1707 &ctx->maps_ext[i], NULL);
1708 if (fd < 0)
1709 return fd;
1710
1711 ctx->map_fds[i] = fd;
1712 }
1713
1714 for (i = 0; have_map_in_map && i < ctx->map_num; i++) {
1715 if (!ctx->maps[i].id ||
1716 ctx->maps[i].inner_id ||
1717 ctx->maps[i].inner_idx == -1)
1718 continue;
1719
1720 inner_fd = ctx->map_fds[i];
1721 inner_idx = ctx->maps[i].inner_idx;
1722
1723 for (j = 0; j < ctx->map_num; j++) {
1724 if (!bpf_is_map_in_map_type(&ctx->maps[j]))
1725 continue;
1726 if (ctx->maps[j].inner_id != ctx->maps[i].id)
1727 continue;
1728
1729 ret = bpf_map_update(ctx->map_fds[j], &inner_idx,
1730 &inner_fd, BPF_ANY);
1731 if (ret < 0) {
1732 bpf_report_map_in_map(ctx->map_fds[j],
1733 inner_idx);
1734 return ret;
1735 }
1736 }
1737 }
1738
1739 return 0;
1740 }
1741
1742 static int bpf_map_num_sym(struct bpf_elf_ctx *ctx)
1743 {
1744 int i, num = 0;
1745 GElf_Sym sym;
1746
1747 for (i = 0; i < ctx->sym_num; i++) {
1748 if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym)
1749 continue;
1750
1751 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1752 GELF_ST_TYPE(sym.st_info) != STT_NOTYPE ||
1753 sym.st_shndx != ctx->sec_maps)
1754 continue;
1755 num++;
1756 }
1757
1758 return num;
1759 }
1760
1761 static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section,
1762 struct bpf_elf_sec_data *data)
1763 {
1764 Elf_Data *sec_edata;
1765 GElf_Shdr sec_hdr;
1766 Elf_Scn *sec_fd;
1767 char *sec_name;
1768
1769 memset(data, 0, sizeof(*data));
1770
1771 sec_fd = elf_getscn(ctx->elf_fd, section);
1772 if (!sec_fd)
1773 return -EINVAL;
1774 if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr)
1775 return -EIO;
1776
1777 sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx,
1778 sec_hdr.sh_name);
1779 if (!sec_name || !sec_hdr.sh_size)
1780 return -ENOENT;
1781
1782 sec_edata = elf_getdata(sec_fd, NULL);
1783 if (!sec_edata || elf_getdata(sec_fd, sec_edata))
1784 return -EIO;
1785
1786 memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr));
1787
1788 data->sec_name = sec_name;
1789 data->sec_data = sec_edata;
1790 return 0;
1791 }
1792
1793 struct bpf_elf_map_min {
1794 __u32 type;
1795 __u32 size_key;
1796 __u32 size_value;
1797 __u32 max_elem;
1798 };
1799
1800 static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section,
1801 struct bpf_elf_sec_data *data)
1802 {
1803 ctx->map_num = data->sec_data->d_size;
1804 ctx->sec_maps = section;
1805 ctx->sec_done[section] = true;
1806
1807 if (ctx->map_num > sizeof(ctx->maps)) {
1808 fprintf(stderr, "Too many BPF maps in ELF section!\n");
1809 return -ENOMEM;
1810 }
1811
1812 memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num);
1813 return 0;
1814 }
1815
1816 static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end)
1817 {
1818 GElf_Sym sym;
1819 int off, i;
1820
1821 for (off = 0; off < end; off += ctx->map_len) {
1822 /* Order doesn't need to be linear here, hence we walk
1823 * the table again.
1824 */
1825 for (i = 0; i < ctx->sym_num; i++) {
1826 if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym)
1827 continue;
1828 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1829 GELF_ST_TYPE(sym.st_info) != STT_NOTYPE ||
1830 sym.st_shndx != ctx->sec_maps)
1831 continue;
1832 if (sym.st_value == off)
1833 break;
1834 if (i == ctx->sym_num - 1)
1835 return -1;
1836 }
1837 }
1838
1839 return off == end ? 0 : -1;
1840 }
1841
1842 static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx)
1843 {
1844 struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {};
1845 int i, sym_num = bpf_map_num_sym(ctx);
1846 __u8 *buff;
1847
1848 if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) {
1849 fprintf(stderr, "%u maps not supported in current map section!\n",
1850 sym_num);
1851 return -EINVAL;
1852 }
1853
1854 if (ctx->map_num % sym_num != 0 ||
1855 ctx->map_num % sizeof(__u32) != 0) {
1856 fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n");
1857 return -EINVAL;
1858 }
1859
1860 ctx->map_len = ctx->map_num / sym_num;
1861 if (bpf_map_verify_all_offs(ctx, ctx->map_num)) {
1862 fprintf(stderr, "Different struct bpf_elf_map in use!\n");
1863 return -EINVAL;
1864 }
1865
1866 if (ctx->map_len == sizeof(struct bpf_elf_map)) {
1867 ctx->map_num = sym_num;
1868 return 0;
1869 } else if (ctx->map_len > sizeof(struct bpf_elf_map)) {
1870 fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n");
1871 return -EINVAL;
1872 } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) {
1873 fprintf(stderr, "struct bpf_elf_map too small, not supported!\n");
1874 return -EINVAL;
1875 }
1876
1877 ctx->map_num = sym_num;
1878 for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num;
1879 i++, buff += ctx->map_len) {
1880 /* The fixup leaves the rest of the members as zero, which
1881 * is fine currently, but option exist to set some other
1882 * default value as well when needed in future.
1883 */
1884 memcpy(&fixup[i], buff, ctx->map_len);
1885 }
1886
1887 memcpy(ctx->maps, fixup, sizeof(fixup));
1888
1889 printf("Note: %zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n",
1890 sizeof(struct bpf_elf_map) - ctx->map_len);
1891 return 0;
1892 }
1893
1894 static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section,
1895 struct bpf_elf_sec_data *data)
1896 {
1897 if (data->sec_data->d_size > sizeof(ctx->license))
1898 return -ENOMEM;
1899
1900 memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size);
1901 ctx->sec_done[section] = true;
1902 return 0;
1903 }
1904
1905 static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section,
1906 struct bpf_elf_sec_data *data)
1907 {
1908 ctx->sym_tab = data->sec_data;
1909 ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize;
1910 ctx->sec_done[section] = true;
1911 return 0;
1912 }
1913
1914 static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section,
1915 struct bpf_elf_sec_data *data)
1916 {
1917 ctx->str_tab = data->sec_data;
1918 ctx->sec_done[section] = true;
1919 return 0;
1920 }
1921
1922 static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx)
1923 {
1924 return ctx->sym_tab && ctx->str_tab && ctx->sec_maps;
1925 }
1926
1927 static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx)
1928 {
1929 struct bpf_elf_sec_data data;
1930 int i, ret = -1;
1931
1932 for (i = 1; i < ctx->elf_hdr.e_shnum; i++) {
1933 ret = bpf_fill_section_data(ctx, i, &data);
1934 if (ret < 0)
1935 continue;
1936
1937 if (data.sec_hdr.sh_type == SHT_PROGBITS &&
1938 !strcmp(data.sec_name, ELF_SECTION_MAPS))
1939 ret = bpf_fetch_maps_begin(ctx, i, &data);
1940 else if (data.sec_hdr.sh_type == SHT_PROGBITS &&
1941 !strcmp(data.sec_name, ELF_SECTION_LICENSE))
1942 ret = bpf_fetch_license(ctx, i, &data);
1943 else if (data.sec_hdr.sh_type == SHT_SYMTAB &&
1944 !strcmp(data.sec_name, ".symtab"))
1945 ret = bpf_fetch_symtab(ctx, i, &data);
1946 else if (data.sec_hdr.sh_type == SHT_STRTAB &&
1947 !strcmp(data.sec_name, ".strtab"))
1948 ret = bpf_fetch_strtab(ctx, i, &data);
1949 if (ret < 0) {
1950 fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n",
1951 i);
1952 return ret;
1953 }
1954 }
1955
1956 if (bpf_has_map_data(ctx)) {
1957 ret = bpf_fetch_maps_end(ctx);
1958 if (ret < 0) {
1959 fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n");
1960 return ret;
1961 }
1962
1963 ret = bpf_maps_attach_all(ctx);
1964 if (ret < 0) {
1965 fprintf(stderr, "Error loading maps into kernel!\n");
1966 return ret;
1967 }
1968 }
1969
1970 return ret;
1971 }
1972
1973 static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section,
1974 bool *sseen)
1975 {
1976 struct bpf_elf_sec_data data;
1977 struct bpf_elf_prog prog;
1978 int ret, i, fd = -1;
1979
1980 for (i = 1; i < ctx->elf_hdr.e_shnum; i++) {
1981 if (ctx->sec_done[i])
1982 continue;
1983
1984 ret = bpf_fill_section_data(ctx, i, &data);
1985 if (ret < 0 ||
1986 !(data.sec_hdr.sh_type == SHT_PROGBITS &&
1987 data.sec_hdr.sh_flags & SHF_EXECINSTR &&
1988 !strcmp(data.sec_name, section)))
1989 continue;
1990
1991 *sseen = true;
1992
1993 memset(&prog, 0, sizeof(prog));
1994 prog.type = ctx->type;
1995 prog.insns = data.sec_data->d_buf;
1996 prog.size = data.sec_data->d_size;
1997 prog.license = ctx->license;
1998
1999 fd = bpf_prog_attach(section, &prog, ctx);
2000 if (fd < 0)
2001 return fd;
2002
2003 ctx->sec_done[i] = true;
2004 break;
2005 }
2006
2007 return fd;
2008 }
2009
2010 struct bpf_tail_call_props {
2011 unsigned int total;
2012 unsigned int jited;
2013 };
2014
2015 static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx,
2016 struct bpf_elf_sec_data *data_relo,
2017 struct bpf_elf_sec_data *data_insn,
2018 struct bpf_tail_call_props *props)
2019 {
2020 Elf_Data *idata = data_insn->sec_data;
2021 GElf_Shdr *rhdr = &data_relo->sec_hdr;
2022 int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize;
2023 struct bpf_insn *insns = idata->d_buf;
2024 unsigned int num_insns = idata->d_size / sizeof(*insns);
2025
2026 for (relo_ent = 0; relo_ent < relo_num; relo_ent++) {
2027 unsigned int ioff, rmap;
2028 GElf_Rel relo;
2029 GElf_Sym sym;
2030
2031 if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo)
2032 return -EIO;
2033
2034 ioff = relo.r_offset / sizeof(struct bpf_insn);
2035 if (ioff >= num_insns ||
2036 insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) {
2037 fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n",
2038 ioff);
2039 if (ioff < num_insns &&
2040 insns[ioff].code == (BPF_JMP | BPF_CALL))
2041 fprintf(stderr, " - Try to annotate functions with always_inline attribute!\n");
2042 return -EINVAL;
2043 }
2044
2045 if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym)
2046 return -EIO;
2047 if (sym.st_shndx != ctx->sec_maps) {
2048 fprintf(stderr, "ELF contains non-map related relo data in entry %u pointing to section %u! Compiler bug?!\n",
2049 relo_ent, sym.st_shndx);
2050 return -EIO;
2051 }
2052
2053 rmap = sym.st_value / ctx->map_len;
2054 if (rmap >= ARRAY_SIZE(ctx->map_fds))
2055 return -EINVAL;
2056 if (!ctx->map_fds[rmap])
2057 return -EINVAL;
2058 if (ctx->maps[rmap].type == BPF_MAP_TYPE_PROG_ARRAY) {
2059 props->total++;
2060 if (ctx->maps_ext[rmap].owner.jited ||
2061 (ctx->maps_ext[rmap].owner.type == 0 &&
2062 ctx->cfg.jit_enabled))
2063 props->jited++;
2064 }
2065
2066 if (ctx->verbose)
2067 fprintf(stderr, "Map \'%s\' (%d) injected into prog section \'%s\' at offset %u!\n",
2068 bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap],
2069 data_insn->sec_name, ioff);
2070
2071 insns[ioff].src_reg = BPF_PSEUDO_MAP_FD;
2072 insns[ioff].imm = ctx->map_fds[rmap];
2073 }
2074
2075 return 0;
2076 }
2077
2078 static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section,
2079 bool *lderr, bool *sseen)
2080 {
2081 struct bpf_elf_sec_data data_relo, data_insn;
2082 struct bpf_elf_prog prog;
2083 int ret, idx, i, fd = -1;
2084
2085 for (i = 1; i < ctx->elf_hdr.e_shnum; i++) {
2086 struct bpf_tail_call_props props = {};
2087
2088 ret = bpf_fill_section_data(ctx, i, &data_relo);
2089 if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL)
2090 continue;
2091
2092 idx = data_relo.sec_hdr.sh_info;
2093
2094 ret = bpf_fill_section_data(ctx, idx, &data_insn);
2095 if (ret < 0 ||
2096 !(data_insn.sec_hdr.sh_type == SHT_PROGBITS &&
2097 data_insn.sec_hdr.sh_flags & SHF_EXECINSTR &&
2098 !strcmp(data_insn.sec_name, section)))
2099 continue;
2100
2101 *sseen = true;
2102
2103 ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn, &props);
2104 if (ret < 0) {
2105 *lderr = true;
2106 return ret;
2107 }
2108
2109 memset(&prog, 0, sizeof(prog));
2110 prog.type = ctx->type;
2111 prog.insns = data_insn.sec_data->d_buf;
2112 prog.size = data_insn.sec_data->d_size;
2113 prog.license = ctx->license;
2114
2115 fd = bpf_prog_attach(section, &prog, ctx);
2116 if (fd < 0) {
2117 *lderr = true;
2118 if (props.total) {
2119 if (ctx->cfg.jit_enabled &&
2120 props.total != props.jited)
2121 fprintf(stderr, "JIT enabled, but only %u/%u tail call maps in the program have JITed owner!\n",
2122 props.jited, props.total);
2123 if (!ctx->cfg.jit_enabled &&
2124 props.jited)
2125 fprintf(stderr, "JIT disabled, but %u/%u tail call maps in the program have JITed owner!\n",
2126 props.jited, props.total);
2127 }
2128 return fd;
2129 }
2130
2131 ctx->sec_done[i] = true;
2132 ctx->sec_done[idx] = true;
2133 break;
2134 }
2135
2136 return fd;
2137 }
2138
2139 static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section)
2140 {
2141 bool lderr = false, sseen = false;
2142 int ret = -1;
2143
2144 if (bpf_has_map_data(ctx))
2145 ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen);
2146 if (ret < 0 && !lderr)
2147 ret = bpf_fetch_prog(ctx, section, &sseen);
2148 if (ret < 0 && !sseen)
2149 fprintf(stderr, "Program section \'%s\' not found in ELF file!\n",
2150 section);
2151 return ret;
2152 }
2153
2154 static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id)
2155 {
2156 int i;
2157
2158 for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++)
2159 if (ctx->map_fds[i] && ctx->maps[i].id == id &&
2160 ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY)
2161 return i;
2162 return -1;
2163 }
2164
2165 struct bpf_jited_aux {
2166 int prog_fd;
2167 int map_fd;
2168 struct bpf_prog_data prog;
2169 struct bpf_map_ext map;
2170 };
2171
2172 static int bpf_derive_prog_from_fdinfo(int fd, struct bpf_prog_data *prog)
2173 {
2174 char file[PATH_MAX], buff[4096];
2175 unsigned int val;
2176 FILE *fp;
2177
2178 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
2179 memset(prog, 0, sizeof(*prog));
2180
2181 fp = fopen(file, "r");
2182 if (!fp) {
2183 fprintf(stderr, "No procfs support?!\n");
2184 return -EIO;
2185 }
2186
2187 while (fgets(buff, sizeof(buff), fp)) {
2188 if (sscanf(buff, "prog_type:\t%u", &val) == 1)
2189 prog->type = val;
2190 else if (sscanf(buff, "prog_jited:\t%u", &val) == 1)
2191 prog->jited = val;
2192 }
2193
2194 fclose(fp);
2195 return 0;
2196 }
2197
2198 static int bpf_tail_call_get_aux(struct bpf_jited_aux *aux)
2199 {
2200 struct bpf_elf_map tmp;
2201 int ret;
2202
2203 ret = bpf_derive_elf_map_from_fdinfo(aux->map_fd, &tmp, &aux->map);
2204 if (!ret)
2205 ret = bpf_derive_prog_from_fdinfo(aux->prog_fd, &aux->prog);
2206
2207 return ret;
2208 }
2209
2210 static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx)
2211 {
2212 struct bpf_elf_sec_data data;
2213 uint32_t map_id, key_id;
2214 int fd, i, ret, idx;
2215
2216 for (i = 1; i < ctx->elf_hdr.e_shnum; i++) {
2217 if (ctx->sec_done[i])
2218 continue;
2219
2220 ret = bpf_fill_section_data(ctx, i, &data);
2221 if (ret < 0)
2222 continue;
2223
2224 ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id);
2225 if (ret != 2)
2226 continue;
2227
2228 idx = bpf_find_map_by_id(ctx, map_id);
2229 if (idx < 0)
2230 continue;
2231
2232 fd = bpf_fetch_prog_sec(ctx, data.sec_name);
2233 if (fd < 0)
2234 return -EIO;
2235
2236 ret = bpf_map_update(ctx->map_fds[idx], &key_id,
2237 &fd, BPF_ANY);
2238 if (ret < 0) {
2239 struct bpf_jited_aux aux = {};
2240
2241 ret = -errno;
2242 if (errno == E2BIG) {
2243 fprintf(stderr, "Tail call key %u for map %u out of bounds?\n",
2244 key_id, map_id);
2245 return ret;
2246 }
2247
2248 aux.map_fd = ctx->map_fds[idx];
2249 aux.prog_fd = fd;
2250
2251 if (bpf_tail_call_get_aux(&aux))
2252 return ret;
2253 if (!aux.map.owner.type)
2254 return ret;
2255
2256 if (aux.prog.type != aux.map.owner.type)
2257 fprintf(stderr, "Tail call map owned by prog type %u, but prog type is %u!\n",
2258 aux.map.owner.type, aux.prog.type);
2259 if (aux.prog.jited != aux.map.owner.jited)
2260 fprintf(stderr, "Tail call map %s jited, but prog %s!\n",
2261 aux.map.owner.jited ? "is" : "not",
2262 aux.prog.jited ? "is" : "not");
2263 return ret;
2264 }
2265
2266 ctx->sec_done[i] = true;
2267 }
2268
2269 return 0;
2270 }
2271
2272 static void bpf_save_finfo(struct bpf_elf_ctx *ctx)
2273 {
2274 struct stat st;
2275 int ret;
2276
2277 memset(&ctx->stat, 0, sizeof(ctx->stat));
2278
2279 ret = fstat(ctx->obj_fd, &st);
2280 if (ret < 0) {
2281 fprintf(stderr, "Stat of elf file failed: %s\n",
2282 strerror(errno));
2283 return;
2284 }
2285
2286 ctx->stat.st_dev = st.st_dev;
2287 ctx->stat.st_ino = st.st_ino;
2288 }
2289
2290 static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path)
2291 {
2292 char buff[PATH_MAX];
2293
2294 while (fgets(buff, sizeof(buff), fp)) {
2295 char *ptr = buff;
2296
2297 while (*ptr == ' ' || *ptr == '\t')
2298 ptr++;
2299
2300 if (*ptr == '#' || *ptr == '\n' || *ptr == 0)
2301 continue;
2302
2303 if (sscanf(ptr, "%i %s\n", id, path) != 2 &&
2304 sscanf(ptr, "%i %s #", id, path) != 2) {
2305 strcpy(path, ptr);
2306 return -1;
2307 }
2308
2309 return 1;
2310 }
2311
2312 return 0;
2313 }
2314
2315 static bool bpf_pinning_reserved(uint32_t pinning)
2316 {
2317 switch (pinning) {
2318 case PIN_NONE:
2319 case PIN_OBJECT_NS:
2320 case PIN_GLOBAL_NS:
2321 return true;
2322 default:
2323 return false;
2324 }
2325 }
2326
2327 static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file)
2328 {
2329 struct bpf_hash_entry *entry;
2330 char subpath[PATH_MAX] = {};
2331 uint32_t pinning;
2332 FILE *fp;
2333 int ret;
2334
2335 fp = fopen(db_file, "r");
2336 if (!fp)
2337 return;
2338
2339 while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) {
2340 if (ret == -1) {
2341 fprintf(stderr, "Database %s is corrupted at: %s\n",
2342 db_file, subpath);
2343 fclose(fp);
2344 return;
2345 }
2346
2347 if (bpf_pinning_reserved(pinning)) {
2348 fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n",
2349 db_file, pinning);
2350 continue;
2351 }
2352
2353 entry = malloc(sizeof(*entry));
2354 if (!entry) {
2355 fprintf(stderr, "No memory left for db entry!\n");
2356 continue;
2357 }
2358
2359 entry->pinning = pinning;
2360 entry->subpath = strdup(subpath);
2361 if (!entry->subpath) {
2362 fprintf(stderr, "No memory left for db entry!\n");
2363 free(entry);
2364 continue;
2365 }
2366
2367 entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)];
2368 ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry;
2369 }
2370
2371 fclose(fp);
2372 }
2373
2374 static void bpf_hash_destroy(struct bpf_elf_ctx *ctx)
2375 {
2376 struct bpf_hash_entry *entry;
2377 int i;
2378
2379 for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) {
2380 while ((entry = ctx->ht[i]) != NULL) {
2381 ctx->ht[i] = entry->next;
2382 free((char *)entry->subpath);
2383 free(entry);
2384 }
2385 }
2386 }
2387
2388 static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx)
2389 {
2390 if (ctx->elf_hdr.e_type != ET_REL ||
2391 (ctx->elf_hdr.e_machine != EM_NONE &&
2392 ctx->elf_hdr.e_machine != EM_BPF) ||
2393 ctx->elf_hdr.e_version != EV_CURRENT) {
2394 fprintf(stderr, "ELF format error, ELF file not for eBPF?\n");
2395 return -EINVAL;
2396 }
2397
2398 switch (ctx->elf_hdr.e_ident[EI_DATA]) {
2399 default:
2400 fprintf(stderr, "ELF format error, wrong endianness info?\n");
2401 return -EINVAL;
2402 case ELFDATA2LSB:
2403 if (htons(1) == 1) {
2404 fprintf(stderr,
2405 "We are big endian, eBPF object is little endian!\n");
2406 return -EIO;
2407 }
2408 break;
2409 case ELFDATA2MSB:
2410 if (htons(1) != 1) {
2411 fprintf(stderr,
2412 "We are little endian, eBPF object is big endian!\n");
2413 return -EIO;
2414 }
2415 break;
2416 }
2417
2418 return 0;
2419 }
2420
2421 static void bpf_get_cfg(struct bpf_elf_ctx *ctx)
2422 {
2423 static const char *path_jit = "/proc/sys/net/core/bpf_jit_enable";
2424 int fd;
2425
2426 fd = open(path_jit, O_RDONLY);
2427 if (fd > 0) {
2428 char tmp[16] = {};
2429
2430 if (read(fd, tmp, sizeof(tmp)) > 0)
2431 ctx->cfg.jit_enabled = atoi(tmp);
2432 close(fd);
2433 }
2434 }
2435
2436 static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname,
2437 enum bpf_prog_type type, __u32 ifindex,
2438 bool verbose)
2439 {
2440 int ret = -EINVAL;
2441
2442 if (elf_version(EV_CURRENT) == EV_NONE ||
2443 bpf_init_env(pathname))
2444 return ret;
2445
2446 memset(ctx, 0, sizeof(*ctx));
2447 bpf_get_cfg(ctx);
2448 ctx->verbose = verbose;
2449 ctx->type = type;
2450 ctx->ifindex = ifindex;
2451
2452 ctx->obj_fd = open(pathname, O_RDONLY);
2453 if (ctx->obj_fd < 0)
2454 return ctx->obj_fd;
2455
2456 ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL);
2457 if (!ctx->elf_fd) {
2458 ret = -EINVAL;
2459 goto out_fd;
2460 }
2461
2462 if (elf_kind(ctx->elf_fd) != ELF_K_ELF) {
2463 ret = -EINVAL;
2464 goto out_fd;
2465 }
2466
2467 if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) !=
2468 &ctx->elf_hdr) {
2469 ret = -EIO;
2470 goto out_elf;
2471 }
2472
2473 ret = bpf_elf_check_ehdr(ctx);
2474 if (ret < 0)
2475 goto out_elf;
2476
2477 ctx->sec_done = calloc(ctx->elf_hdr.e_shnum,
2478 sizeof(*(ctx->sec_done)));
2479 if (!ctx->sec_done) {
2480 ret = -ENOMEM;
2481 goto out_elf;
2482 }
2483
2484 if (ctx->verbose && bpf_log_realloc(ctx)) {
2485 ret = -ENOMEM;
2486 goto out_free;
2487 }
2488
2489 bpf_save_finfo(ctx);
2490 bpf_hash_init(ctx, CONFDIR "/bpf_pinning");
2491
2492 return 0;
2493 out_free:
2494 free(ctx->sec_done);
2495 out_elf:
2496 elf_end(ctx->elf_fd);
2497 out_fd:
2498 close(ctx->obj_fd);
2499 return ret;
2500 }
2501
2502 static int bpf_maps_count(struct bpf_elf_ctx *ctx)
2503 {
2504 int i, count = 0;
2505
2506 for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) {
2507 if (!ctx->map_fds[i])
2508 break;
2509 count++;
2510 }
2511
2512 return count;
2513 }
2514
2515 static void bpf_maps_teardown(struct bpf_elf_ctx *ctx)
2516 {
2517 int i;
2518
2519 for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) {
2520 if (ctx->map_fds[i])
2521 close(ctx->map_fds[i]);
2522 }
2523 }
2524
2525 static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure)
2526 {
2527 if (failure)
2528 bpf_maps_teardown(ctx);
2529
2530 bpf_hash_destroy(ctx);
2531
2532 free(ctx->sec_done);
2533 free(ctx->log);
2534
2535 elf_end(ctx->elf_fd);
2536 close(ctx->obj_fd);
2537 }
2538
2539 static struct bpf_elf_ctx __ctx;
2540
2541 static int bpf_obj_open(const char *pathname, enum bpf_prog_type type,
2542 const char *section, __u32 ifindex, bool verbose)
2543 {
2544 struct bpf_elf_ctx *ctx = &__ctx;
2545 int fd = 0, ret;
2546
2547 ret = bpf_elf_ctx_init(ctx, pathname, type, ifindex, verbose);
2548 if (ret < 0) {
2549 fprintf(stderr, "Cannot initialize ELF context!\n");
2550 return ret;
2551 }
2552
2553 ret = bpf_fetch_ancillary(ctx);
2554 if (ret < 0) {
2555 fprintf(stderr, "Error fetching ELF ancillary data!\n");
2556 goto out;
2557 }
2558
2559 fd = bpf_fetch_prog_sec(ctx, section);
2560 if (fd < 0) {
2561 fprintf(stderr, "Error fetching program/map!\n");
2562 ret = fd;
2563 goto out;
2564 }
2565
2566 ret = bpf_fill_prog_arrays(ctx);
2567 if (ret < 0)
2568 fprintf(stderr, "Error filling program arrays!\n");
2569 out:
2570 bpf_elf_ctx_destroy(ctx, ret < 0);
2571 if (ret < 0) {
2572 if (fd)
2573 close(fd);
2574 return ret;
2575 }
2576
2577 return fd;
2578 }
2579
2580 static int
2581 bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len,
2582 const struct bpf_map_data *aux, unsigned int entries)
2583 {
2584 struct bpf_map_set_msg msg = {
2585 .aux.uds_ver = BPF_SCM_AUX_VER,
2586 .aux.num_ent = entries,
2587 };
2588 int *cmsg_buf, min_fd;
2589 char *amsg_buf;
2590 int i;
2591
2592 strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name));
2593 memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st));
2594
2595 cmsg_buf = bpf_map_set_init(&msg, addr, addr_len);
2596 amsg_buf = (char *)msg.aux.ent;
2597
2598 for (i = 0; i < entries; i += min_fd) {
2599 int ret;
2600
2601 min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i);
2602 bpf_map_set_init_single(&msg, min_fd);
2603
2604 memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd);
2605 memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd);
2606
2607 ret = sendmsg(fd, &msg.hdr, 0);
2608 if (ret <= 0)
2609 return ret ? : -1;
2610 }
2611
2612 return 0;
2613 }
2614
2615 static int
2616 bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux,
2617 unsigned int entries)
2618 {
2619 struct bpf_map_set_msg msg;
2620 int *cmsg_buf, min_fd;
2621 char *amsg_buf, *mmsg_buf;
2622 unsigned int needed = 1;
2623 int i;
2624
2625 cmsg_buf = bpf_map_set_init(&msg, NULL, 0);
2626 amsg_buf = (char *)msg.aux.ent;
2627 mmsg_buf = (char *)&msg.aux;
2628
2629 for (i = 0; i < min(entries, needed); i += min_fd) {
2630 struct cmsghdr *cmsg;
2631 int ret;
2632
2633 min_fd = min(entries, entries - i);
2634 bpf_map_set_init_single(&msg, min_fd);
2635
2636 ret = recvmsg(fd, &msg.hdr, 0);
2637 if (ret <= 0)
2638 return ret ? : -1;
2639
2640 cmsg = CMSG_FIRSTHDR(&msg.hdr);
2641 if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS)
2642 return -EINVAL;
2643 if (msg.hdr.msg_flags & MSG_CTRUNC)
2644 return -EIO;
2645 if (msg.aux.uds_ver != BPF_SCM_AUX_VER)
2646 return -ENOSYS;
2647
2648 min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd);
2649 if (min_fd > entries || min_fd <= 0)
2650 return -EINVAL;
2651
2652 memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd);
2653 memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd);
2654 memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent));
2655
2656 needed = aux->num_ent;
2657 }
2658
2659 return 0;
2660 }
2661
2662 int bpf_send_map_fds(const char *path, const char *obj)
2663 {
2664 struct bpf_elf_ctx *ctx = &__ctx;
2665 struct sockaddr_un addr = { .sun_family = AF_UNIX };
2666 struct bpf_map_data bpf_aux = {
2667 .fds = ctx->map_fds,
2668 .ent = ctx->maps,
2669 .st = &ctx->stat,
2670 .obj = obj,
2671 };
2672 int fd, ret;
2673
2674 fd = socket(AF_UNIX, SOCK_DGRAM, 0);
2675 if (fd < 0) {
2676 fprintf(stderr, "Cannot open socket: %s\n",
2677 strerror(errno));
2678 return -1;
2679 }
2680
2681 strncpy(addr.sun_path, path, sizeof(addr.sun_path));
2682
2683 ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr));
2684 if (ret < 0) {
2685 fprintf(stderr, "Cannot connect to %s: %s\n",
2686 path, strerror(errno));
2687 return -1;
2688 }
2689
2690 ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux,
2691 bpf_maps_count(ctx));
2692 if (ret < 0)
2693 fprintf(stderr, "Cannot send fds to %s: %s\n",
2694 path, strerror(errno));
2695
2696 bpf_maps_teardown(ctx);
2697 close(fd);
2698 return ret;
2699 }
2700
2701 int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux,
2702 unsigned int entries)
2703 {
2704 struct sockaddr_un addr = { .sun_family = AF_UNIX };
2705 int fd, ret;
2706
2707 fd = socket(AF_UNIX, SOCK_DGRAM, 0);
2708 if (fd < 0) {
2709 fprintf(stderr, "Cannot open socket: %s\n",
2710 strerror(errno));
2711 return -1;
2712 }
2713
2714 strncpy(addr.sun_path, path, sizeof(addr.sun_path));
2715
2716 ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr));
2717 if (ret < 0) {
2718 fprintf(stderr, "Cannot bind to socket: %s\n",
2719 strerror(errno));
2720 return -1;
2721 }
2722
2723 ret = bpf_map_set_recv(fd, fds, aux, entries);
2724 if (ret < 0)
2725 fprintf(stderr, "Cannot recv fds from %s: %s\n",
2726 path, strerror(errno));
2727
2728 unlink(addr.sun_path);
2729 close(fd);
2730 return ret;
2731 }
2732 #endif /* HAVE_ELF */
Upstream mirror of iproute2