]>
Commit | Line | Data |
---|---|---|
1d129d19 | 1 | /* |
e4225669 | 2 | * bpf.c BPF common code |
1d129d19 JP |
3 | * |
4 | * This program is free software; you can distribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
e4225669 | 9 | * Authors: Daniel Borkmann <daniel@iogearbox.net> |
1d129d19 | 10 | * Jiri Pirko <jiri@resnulli.us> |
e4225669 | 11 | * Alexei Starovoitov <ast@kernel.org> |
1d129d19 JP |
12 | */ |
13 | ||
14 | #include <stdio.h> | |
15 | #include <stdlib.h> | |
16 | #include <unistd.h> | |
17 | #include <string.h> | |
18 | #include <stdbool.h> | |
473d7840 | 19 | #include <stdint.h> |
1d129d19 | 20 | #include <errno.h> |
11c39b5e DB |
21 | #include <fcntl.h> |
22 | #include <stdarg.h> | |
5c5a0f3d | 23 | #include <limits.h> |
e4225669 | 24 | #include <assert.h> |
1d129d19 | 25 | |
11c39b5e DB |
26 | #ifdef HAVE_ELF |
27 | #include <libelf.h> | |
28 | #include <gelf.h> | |
29 | #endif | |
30 | ||
32e93fb7 DB |
31 | #include <sys/types.h> |
32 | #include <sys/stat.h> | |
33 | #include <sys/un.h> | |
34 | #include <sys/vfs.h> | |
35 | #include <sys/mount.h> | |
36 | #include <sys/syscall.h> | |
37 | #include <sys/sendfile.h> | |
38 | #include <sys/resource.h> | |
39 | ||
8187b012 DB |
40 | #include <arpa/inet.h> |
41 | ||
1d129d19 | 42 | #include "utils.h" |
6256f8c9 | 43 | |
e4225669 | 44 | #include "bpf_util.h" |
6256f8c9 DB |
45 | #include "bpf_elf.h" |
46 | #include "bpf_scm.h" | |
47 | ||
e4225669 DB |
48 | struct bpf_prog_meta { |
49 | const char *type; | |
50 | const char *subdir; | |
51 | const char *section; | |
52 | bool may_uds_export; | |
53 | }; | |
1d129d19 | 54 | |
e4225669 DB |
55 | static const enum bpf_prog_type __bpf_types[] = { |
56 | BPF_PROG_TYPE_SCHED_CLS, | |
57 | BPF_PROG_TYPE_SCHED_ACT, | |
c7272ca7 | 58 | BPF_PROG_TYPE_XDP, |
b15f440e TG |
59 | BPF_PROG_TYPE_LWT_IN, |
60 | BPF_PROG_TYPE_LWT_OUT, | |
61 | BPF_PROG_TYPE_LWT_XMIT, | |
e4225669 | 62 | }; |
67584e3a | 63 | |
e4225669 DB |
64 | static const struct bpf_prog_meta __bpf_prog_meta[] = { |
65 | [BPF_PROG_TYPE_SCHED_CLS] = { | |
66 | .type = "cls", | |
67 | .subdir = "tc", | |
68 | .section = ELF_SECTION_CLASSIFIER, | |
69 | .may_uds_export = true, | |
70 | }, | |
71 | [BPF_PROG_TYPE_SCHED_ACT] = { | |
72 | .type = "act", | |
73 | .subdir = "tc", | |
74 | .section = ELF_SECTION_ACTION, | |
75 | .may_uds_export = true, | |
76 | }, | |
c7272ca7 DB |
77 | [BPF_PROG_TYPE_XDP] = { |
78 | .type = "xdp", | |
79 | .subdir = "xdp", | |
80 | .section = ELF_SECTION_PROG, | |
81 | }, | |
b15f440e TG |
82 | [BPF_PROG_TYPE_LWT_IN] = { |
83 | .type = "lwt_in", | |
84 | .subdir = "ip", | |
85 | .section = ELF_SECTION_PROG, | |
86 | }, | |
87 | [BPF_PROG_TYPE_LWT_OUT] = { | |
88 | .type = "lwt_out", | |
89 | .subdir = "ip", | |
90 | .section = ELF_SECTION_PROG, | |
91 | }, | |
92 | [BPF_PROG_TYPE_LWT_XMIT] = { | |
93 | .type = "lwt_xmit", | |
94 | .subdir = "ip", | |
95 | .section = ELF_SECTION_PROG, | |
96 | }, | |
e4225669 DB |
97 | }; |
98 | ||
99 | static const char *bpf_prog_to_subdir(enum bpf_prog_type type) | |
100 | { | |
101 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
102 | __bpf_prog_meta[type].subdir); | |
103 | return __bpf_prog_meta[type].subdir; | |
104 | } | |
105 | ||
106 | const char *bpf_prog_to_default_section(enum bpf_prog_type type) | |
107 | { | |
108 | assert(type < ARRAY_SIZE(__bpf_prog_meta) && | |
109 | __bpf_prog_meta[type].section); | |
110 | return __bpf_prog_meta[type].section; | |
111 | } | |
e77fa41d | 112 | |
32e93fb7 DB |
113 | #ifdef HAVE_ELF |
114 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
115 | const char *sec, bool verbose); | |
116 | #else | |
117 | static int bpf_obj_open(const char *path, enum bpf_prog_type type, | |
118 | const char *sec, bool verbose) | |
119 | { | |
120 | fprintf(stderr, "No ELF library support compiled in.\n"); | |
121 | errno = ENOSYS; | |
122 | return -1; | |
123 | } | |
124 | #endif | |
125 | ||
126 | static inline __u64 bpf_ptr_to_u64(const void *ptr) | |
127 | { | |
128 | return (__u64)(unsigned long)ptr; | |
129 | } | |
130 | ||
131 | static int bpf(int cmd, union bpf_attr *attr, unsigned int size) | |
132 | { | |
133 | #ifdef __NR_bpf | |
134 | return syscall(__NR_bpf, cmd, attr, size); | |
135 | #else | |
136 | fprintf(stderr, "No bpf syscall, kernel headers too old?\n"); | |
137 | errno = ENOSYS; | |
138 | return -1; | |
139 | #endif | |
140 | } | |
141 | ||
91d88eeb DB |
142 | static int bpf_map_update(int fd, const void *key, const void *value, |
143 | uint64_t flags) | |
32e93fb7 | 144 | { |
d17b136f | 145 | union bpf_attr attr = {}; |
67584e3a | 146 | |
67584e3a ND |
147 | attr.map_fd = fd; |
148 | attr.key = bpf_ptr_to_u64(key); | |
149 | attr.value = bpf_ptr_to_u64(value); | |
150 | attr.flags = flags; | |
32e93fb7 | 151 | |
91d88eeb | 152 | return bpf(BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
32e93fb7 DB |
153 | } |
154 | ||
155 | static int bpf_parse_string(char *arg, bool from_file, __u16 *bpf_len, | |
156 | char **bpf_string, bool *need_release, | |
157 | const char separator) | |
1d129d19 JP |
158 | { |
159 | char sp; | |
160 | ||
161 | if (from_file) { | |
162 | size_t tmp_len, op_len = sizeof("65535 255 255 4294967295,"); | |
e4225669 | 163 | char *tmp_string, *last; |
1d129d19 JP |
164 | FILE *fp; |
165 | ||
166 | tmp_len = sizeof("4096,") + BPF_MAXINSNS * op_len; | |
f89bb021 | 167 | tmp_string = calloc(1, tmp_len); |
1d129d19 JP |
168 | if (tmp_string == NULL) |
169 | return -ENOMEM; | |
170 | ||
1d129d19 JP |
171 | fp = fopen(arg, "r"); |
172 | if (fp == NULL) { | |
173 | perror("Cannot fopen"); | |
174 | free(tmp_string); | |
175 | return -ENOENT; | |
176 | } | |
177 | ||
178 | if (!fgets(tmp_string, tmp_len, fp)) { | |
179 | free(tmp_string); | |
180 | fclose(fp); | |
181 | return -EIO; | |
182 | } | |
183 | ||
184 | fclose(fp); | |
185 | ||
e4225669 DB |
186 | last = &tmp_string[strlen(tmp_string) - 1]; |
187 | if (*last == '\n') | |
188 | *last = 0; | |
189 | ||
1d129d19 JP |
190 | *need_release = true; |
191 | *bpf_string = tmp_string; | |
192 | } else { | |
193 | *need_release = false; | |
194 | *bpf_string = arg; | |
195 | } | |
196 | ||
197 | if (sscanf(*bpf_string, "%hu%c", bpf_len, &sp) != 2 || | |
198 | sp != separator) { | |
199 | if (*need_release) | |
200 | free(*bpf_string); | |
201 | return -EINVAL; | |
202 | } | |
203 | ||
204 | return 0; | |
205 | } | |
206 | ||
32e93fb7 DB |
207 | static int bpf_ops_parse(int argc, char **argv, struct sock_filter *bpf_ops, |
208 | bool from_file) | |
1d129d19 JP |
209 | { |
210 | char *bpf_string, *token, separator = ','; | |
211 | int ret = 0, i = 0; | |
212 | bool need_release; | |
213 | __u16 bpf_len = 0; | |
214 | ||
215 | if (argc < 1) | |
216 | return -EINVAL; | |
217 | if (bpf_parse_string(argv[0], from_file, &bpf_len, &bpf_string, | |
218 | &need_release, separator)) | |
219 | return -EINVAL; | |
220 | if (bpf_len == 0 || bpf_len > BPF_MAXINSNS) { | |
221 | ret = -EINVAL; | |
222 | goto out; | |
223 | } | |
224 | ||
225 | token = bpf_string; | |
226 | while ((token = strchr(token, separator)) && (++token)[0]) { | |
227 | if (i >= bpf_len) { | |
32a121cb | 228 | fprintf(stderr, "Real program length exceeds encoded length parameter!\n"); |
1d129d19 JP |
229 | ret = -EINVAL; |
230 | goto out; | |
231 | } | |
232 | ||
233 | if (sscanf(token, "%hu %hhu %hhu %u,", | |
234 | &bpf_ops[i].code, &bpf_ops[i].jt, | |
235 | &bpf_ops[i].jf, &bpf_ops[i].k) != 4) { | |
236 | fprintf(stderr, "Error at instruction %d!\n", i); | |
237 | ret = -EINVAL; | |
238 | goto out; | |
239 | } | |
240 | ||
241 | i++; | |
242 | } | |
243 | ||
244 | if (i != bpf_len) { | |
afc1a200 | 245 | fprintf(stderr, "Parsed program length is less than encoded length parameter!\n"); |
1d129d19 JP |
246 | ret = -EINVAL; |
247 | goto out; | |
248 | } | |
249 | ret = bpf_len; | |
1d129d19 JP |
250 | out: |
251 | if (need_release) | |
252 | free(bpf_string); | |
253 | ||
254 | return ret; | |
255 | } | |
256 | ||
257 | void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len) | |
258 | { | |
d896797c | 259 | struct sock_filter *ops = RTA_DATA(bpf_ops); |
1d129d19 JP |
260 | int i; |
261 | ||
262 | if (len == 0) | |
263 | return; | |
264 | ||
265 | fprintf(f, "bytecode \'%u,", len); | |
266 | ||
267 | for (i = 0; i < len - 1; i++) | |
268 | fprintf(f, "%hu %hhu %hhu %u,", ops[i].code, ops[i].jt, | |
269 | ops[i].jf, ops[i].k); | |
270 | ||
6256f8c9 | 271 | fprintf(f, "%hu %hhu %hhu %u\'", ops[i].code, ops[i].jt, |
1d129d19 JP |
272 | ops[i].jf, ops[i].k); |
273 | } | |
11c39b5e | 274 | |
afc1a200 DB |
275 | static void bpf_map_pin_report(const struct bpf_elf_map *pin, |
276 | const struct bpf_elf_map *obj) | |
277 | { | |
278 | fprintf(stderr, "Map specification differs from pinned file!\n"); | |
279 | ||
280 | if (obj->type != pin->type) | |
281 | fprintf(stderr, " - Type: %u (obj) != %u (pin)\n", | |
282 | obj->type, pin->type); | |
283 | if (obj->size_key != pin->size_key) | |
284 | fprintf(stderr, " - Size key: %u (obj) != %u (pin)\n", | |
285 | obj->size_key, pin->size_key); | |
286 | if (obj->size_value != pin->size_value) | |
287 | fprintf(stderr, " - Size value: %u (obj) != %u (pin)\n", | |
288 | obj->size_value, pin->size_value); | |
289 | if (obj->max_elem != pin->max_elem) | |
290 | fprintf(stderr, " - Max elems: %u (obj) != %u (pin)\n", | |
291 | obj->max_elem, pin->max_elem); | |
4dd3f50a DB |
292 | if (obj->flags != pin->flags) |
293 | fprintf(stderr, " - Flags: %#x (obj) != %#x (pin)\n", | |
294 | obj->flags, pin->flags); | |
afc1a200 DB |
295 | |
296 | fprintf(stderr, "\n"); | |
297 | } | |
298 | ||
91d88eeb | 299 | static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map, |
fb24802b | 300 | int length, enum bpf_prog_type type) |
9e607f2e DB |
301 | { |
302 | char file[PATH_MAX], buff[4096]; | |
d17b136f | 303 | struct bpf_elf_map tmp = {}, zero = {}; |
fb24802b | 304 | unsigned int val, owner_type = 0; |
9e607f2e DB |
305 | FILE *fp; |
306 | ||
307 | snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); | |
308 | ||
309 | fp = fopen(file, "r"); | |
310 | if (!fp) { | |
311 | fprintf(stderr, "No procfs support?!\n"); | |
312 | return -EIO; | |
313 | } | |
314 | ||
9e607f2e DB |
315 | while (fgets(buff, sizeof(buff), fp)) { |
316 | if (sscanf(buff, "map_type:\t%u", &val) == 1) | |
317 | tmp.type = val; | |
318 | else if (sscanf(buff, "key_size:\t%u", &val) == 1) | |
319 | tmp.size_key = val; | |
320 | else if (sscanf(buff, "value_size:\t%u", &val) == 1) | |
321 | tmp.size_value = val; | |
322 | else if (sscanf(buff, "max_entries:\t%u", &val) == 1) | |
323 | tmp.max_elem = val; | |
4dd3f50a DB |
324 | else if (sscanf(buff, "map_flags:\t%i", &val) == 1) |
325 | tmp.flags = val; | |
fb24802b DB |
326 | else if (sscanf(buff, "owner_prog_type:\t%i", &val) == 1) |
327 | owner_type = val; | |
9e607f2e DB |
328 | } |
329 | ||
330 | fclose(fp); | |
331 | ||
fb24802b DB |
332 | /* The decision to reject this is on kernel side eventually, but |
333 | * at least give the user a chance to know what's wrong. | |
334 | */ | |
335 | if (owner_type && owner_type != type) | |
336 | fprintf(stderr, "Program array map owner types differ: %u (obj) != %u (pin)\n", | |
337 | type, owner_type); | |
338 | ||
91d88eeb | 339 | if (!memcmp(&tmp, map, length)) { |
9e607f2e DB |
340 | return 0; |
341 | } else { | |
9e607f2e DB |
342 | /* If kernel doesn't have eBPF-related fdinfo, we cannot do much, |
343 | * so just accept it. We know we do have an eBPF fd and in this | |
344 | * case, everything is 0. It is guaranteed that no such map exists | |
345 | * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC. | |
346 | */ | |
91d88eeb | 347 | if (!memcmp(&tmp, &zero, length)) |
9e607f2e DB |
348 | return 0; |
349 | ||
afc1a200 | 350 | bpf_map_pin_report(&tmp, map); |
9e607f2e DB |
351 | return -EINVAL; |
352 | } | |
353 | } | |
354 | ||
91d88eeb DB |
355 | static int bpf_mnt_fs(const char *target) |
356 | { | |
357 | bool bind_done = false; | |
358 | ||
359 | while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { | |
360 | if (errno != EINVAL || bind_done) { | |
361 | fprintf(stderr, "mount --make-private %s failed: %s\n", | |
362 | target, strerror(errno)); | |
363 | return -1; | |
364 | } | |
365 | ||
366 | if (mount(target, target, "none", MS_BIND, NULL)) { | |
367 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
368 | target, target, strerror(errno)); | |
369 | return -1; | |
370 | } | |
371 | ||
372 | bind_done = true; | |
373 | } | |
374 | ||
e4225669 | 375 | if (mount("bpf", target, "bpf", 0, "mode=0700")) { |
91d88eeb DB |
376 | fprintf(stderr, "mount -t bpf bpf %s failed: %s\n", |
377 | target, strerror(errno)); | |
378 | return -1; | |
379 | } | |
380 | ||
381 | return 0; | |
382 | } | |
383 | ||
32e93fb7 DB |
384 | static int bpf_valid_mntpt(const char *mnt, unsigned long magic) |
385 | { | |
386 | struct statfs st_fs; | |
387 | ||
388 | if (statfs(mnt, &st_fs) < 0) | |
389 | return -ENOENT; | |
390 | if ((unsigned long)st_fs.f_type != magic) | |
391 | return -ENOENT; | |
392 | ||
393 | return 0; | |
394 | } | |
395 | ||
396 | static const char *bpf_find_mntpt(const char *fstype, unsigned long magic, | |
397 | char *mnt, int len, | |
398 | const char * const *known_mnts) | |
399 | { | |
400 | const char * const *ptr; | |
401 | char type[100]; | |
402 | FILE *fp; | |
403 | ||
404 | if (known_mnts) { | |
405 | ptr = known_mnts; | |
406 | while (*ptr) { | |
407 | if (bpf_valid_mntpt(*ptr, magic) == 0) { | |
408 | strncpy(mnt, *ptr, len - 1); | |
409 | mnt[len - 1] = 0; | |
410 | return mnt; | |
411 | } | |
412 | ptr++; | |
413 | } | |
414 | } | |
415 | ||
416 | fp = fopen("/proc/mounts", "r"); | |
417 | if (fp == NULL || len != PATH_MAX) | |
418 | return NULL; | |
419 | ||
420 | while (fscanf(fp, "%*s %" textify(PATH_MAX) "s %99s %*s %*d %*d\n", | |
421 | mnt, type) == 2) { | |
422 | if (strcmp(type, fstype) == 0) | |
423 | break; | |
424 | } | |
425 | ||
426 | fclose(fp); | |
427 | if (strcmp(type, fstype) != 0) | |
428 | return NULL; | |
429 | ||
430 | return mnt; | |
431 | } | |
432 | ||
433 | int bpf_trace_pipe(void) | |
434 | { | |
435 | char tracefs_mnt[PATH_MAX] = TRACE_DIR_MNT; | |
436 | static const char * const tracefs_known_mnts[] = { | |
437 | TRACE_DIR_MNT, | |
438 | "/sys/kernel/debug/tracing", | |
439 | "/tracing", | |
440 | "/trace", | |
441 | 0, | |
442 | }; | |
443 | char tpipe[PATH_MAX]; | |
444 | const char *mnt; | |
445 | int fd; | |
446 | ||
447 | mnt = bpf_find_mntpt("tracefs", TRACEFS_MAGIC, tracefs_mnt, | |
448 | sizeof(tracefs_mnt), tracefs_known_mnts); | |
449 | if (!mnt) { | |
450 | fprintf(stderr, "tracefs not mounted?\n"); | |
451 | return -1; | |
452 | } | |
453 | ||
454 | snprintf(tpipe, sizeof(tpipe), "%s/trace_pipe", mnt); | |
455 | ||
456 | fd = open(tpipe, O_RDONLY); | |
457 | if (fd < 0) | |
458 | return -1; | |
459 | ||
460 | fprintf(stderr, "Running! Hang up with ^C!\n\n"); | |
461 | while (1) { | |
462 | static char buff[4096]; | |
463 | ssize_t ret; | |
464 | ||
465 | ret = read(fd, buff, sizeof(buff) - 1); | |
466 | if (ret > 0) { | |
467 | write(2, buff, ret); | |
468 | fflush(stderr); | |
469 | } | |
470 | } | |
471 | ||
472 | return 0; | |
473 | } | |
474 | ||
e4225669 | 475 | static int bpf_gen_global(const char *bpf_sub_dir) |
91d88eeb | 476 | { |
e4225669 DB |
477 | char bpf_glo_dir[PATH_MAX]; |
478 | int ret; | |
479 | ||
480 | snprintf(bpf_glo_dir, sizeof(bpf_glo_dir), "%s/%s/", | |
481 | bpf_sub_dir, BPF_DIR_GLOBALS); | |
482 | ||
483 | ret = mkdir(bpf_glo_dir, S_IRWXU); | |
484 | if (ret && errno != EEXIST) { | |
485 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_glo_dir, | |
486 | strerror(errno)); | |
487 | return ret; | |
488 | } | |
489 | ||
490 | return 0; | |
491 | } | |
492 | ||
493 | static int bpf_gen_master(const char *base, const char *name) | |
494 | { | |
495 | char bpf_sub_dir[PATH_MAX]; | |
496 | int ret; | |
497 | ||
498 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s/", base, name); | |
499 | ||
500 | ret = mkdir(bpf_sub_dir, S_IRWXU); | |
501 | if (ret && errno != EEXIST) { | |
502 | fprintf(stderr, "mkdir %s failed: %s\n", bpf_sub_dir, | |
503 | strerror(errno)); | |
504 | return ret; | |
505 | } | |
506 | ||
507 | return bpf_gen_global(bpf_sub_dir); | |
508 | } | |
509 | ||
510 | static int bpf_slave_via_bind_mnt(const char *full_name, | |
511 | const char *full_link) | |
512 | { | |
513 | int ret; | |
514 | ||
515 | ret = mkdir(full_name, S_IRWXU); | |
516 | if (ret) { | |
517 | assert(errno != EEXIST); | |
518 | fprintf(stderr, "mkdir %s failed: %s\n", full_name, | |
519 | strerror(errno)); | |
520 | return ret; | |
521 | } | |
522 | ||
523 | ret = mount(full_link, full_name, "none", MS_BIND, NULL); | |
524 | if (ret) { | |
525 | rmdir(full_name); | |
526 | fprintf(stderr, "mount --bind %s %s failed: %s\n", | |
527 | full_link, full_name, strerror(errno)); | |
528 | } | |
529 | ||
530 | return ret; | |
531 | } | |
532 | ||
533 | static int bpf_gen_slave(const char *base, const char *name, | |
534 | const char *link) | |
535 | { | |
536 | char bpf_lnk_dir[PATH_MAX]; | |
537 | char bpf_sub_dir[PATH_MAX]; | |
538 | struct stat sb = {}; | |
539 | int ret; | |
540 | ||
541 | snprintf(bpf_lnk_dir, sizeof(bpf_lnk_dir), "%s%s/", base, link); | |
542 | snprintf(bpf_sub_dir, sizeof(bpf_sub_dir), "%s%s", base, name); | |
543 | ||
544 | ret = symlink(bpf_lnk_dir, bpf_sub_dir); | |
545 | if (ret) { | |
546 | if (errno != EEXIST) { | |
547 | if (errno != EPERM) { | |
548 | fprintf(stderr, "symlink %s failed: %s\n", | |
549 | bpf_sub_dir, strerror(errno)); | |
550 | return ret; | |
551 | } | |
552 | ||
553 | return bpf_slave_via_bind_mnt(bpf_sub_dir, | |
554 | bpf_lnk_dir); | |
555 | } | |
556 | ||
557 | ret = lstat(bpf_sub_dir, &sb); | |
558 | if (ret) { | |
559 | fprintf(stderr, "lstat %s failed: %s\n", | |
560 | bpf_sub_dir, strerror(errno)); | |
561 | return ret; | |
562 | } | |
563 | ||
564 | if ((sb.st_mode & S_IFMT) != S_IFLNK) | |
565 | return bpf_gen_global(bpf_sub_dir); | |
566 | } | |
567 | ||
568 | return 0; | |
569 | } | |
570 | ||
571 | static int bpf_gen_hierarchy(const char *base) | |
572 | { | |
573 | int ret, i; | |
574 | ||
575 | ret = bpf_gen_master(base, bpf_prog_to_subdir(__bpf_types[0])); | |
576 | for (i = 1; i < ARRAY_SIZE(__bpf_types) && !ret; i++) | |
577 | ret = bpf_gen_slave(base, | |
578 | bpf_prog_to_subdir(__bpf_types[i]), | |
579 | bpf_prog_to_subdir(__bpf_types[0])); | |
580 | return ret; | |
581 | } | |
582 | ||
583 | static const char *bpf_get_work_dir(enum bpf_prog_type type) | |
584 | { | |
585 | static char bpf_tmp[PATH_MAX] = BPF_DIR_MNT; | |
586 | static char bpf_wrk_dir[PATH_MAX]; | |
91d88eeb | 587 | static const char *mnt; |
e4225669 | 588 | static bool bpf_mnt_cached; |
91d88eeb DB |
589 | static const char * const bpf_known_mnts[] = { |
590 | BPF_DIR_MNT, | |
e4225669 | 591 | "/bpf", |
91d88eeb DB |
592 | 0, |
593 | }; | |
91d88eeb DB |
594 | int ret; |
595 | ||
e4225669 DB |
596 | if (bpf_mnt_cached) { |
597 | const char *out = mnt; | |
598 | ||
51361a9f | 599 | if (out && type) { |
e4225669 DB |
600 | snprintf(bpf_tmp, sizeof(bpf_tmp), "%s%s/", |
601 | out, bpf_prog_to_subdir(type)); | |
602 | out = bpf_tmp; | |
603 | } | |
604 | return out; | |
605 | } | |
91d88eeb | 606 | |
e4225669 | 607 | mnt = bpf_find_mntpt("bpf", BPF_FS_MAGIC, bpf_tmp, sizeof(bpf_tmp), |
91d88eeb DB |
608 | bpf_known_mnts); |
609 | if (!mnt) { | |
610 | mnt = getenv(BPF_ENV_MNT); | |
611 | if (!mnt) | |
612 | mnt = BPF_DIR_MNT; | |
613 | ret = bpf_mnt_fs(mnt); | |
614 | if (ret) { | |
615 | mnt = NULL; | |
616 | goto out; | |
617 | } | |
618 | } | |
619 | ||
e4225669 | 620 | snprintf(bpf_wrk_dir, sizeof(bpf_wrk_dir), "%s/", mnt); |
91d88eeb | 621 | |
e4225669 DB |
622 | ret = bpf_gen_hierarchy(bpf_wrk_dir); |
623 | if (ret) { | |
91d88eeb DB |
624 | mnt = NULL; |
625 | goto out; | |
626 | } | |
627 | ||
e4225669 | 628 | mnt = bpf_wrk_dir; |
91d88eeb DB |
629 | out: |
630 | bpf_mnt_cached = true; | |
91d88eeb DB |
631 | return mnt; |
632 | } | |
633 | ||
e4225669 | 634 | static int bpf_obj_get(const char *pathname, enum bpf_prog_type type) |
91d88eeb | 635 | { |
d17b136f | 636 | union bpf_attr attr = {}; |
91d88eeb DB |
637 | char tmp[PATH_MAX]; |
638 | ||
639 | if (strlen(pathname) > 2 && pathname[0] == 'm' && | |
e4225669 | 640 | pathname[1] == ':' && bpf_get_work_dir(type)) { |
91d88eeb | 641 | snprintf(tmp, sizeof(tmp), "%s/%s", |
e4225669 | 642 | bpf_get_work_dir(type), pathname + 2); |
91d88eeb DB |
643 | pathname = tmp; |
644 | } | |
645 | ||
91d88eeb DB |
646 | attr.pathname = bpf_ptr_to_u64(pathname); |
647 | ||
648 | return bpf(BPF_OBJ_GET, &attr, sizeof(attr)); | |
649 | } | |
650 | ||
21856018 DB |
651 | static int bpf_obj_pinned(const char *pathname, enum bpf_prog_type type) |
652 | { | |
653 | int prog_fd = bpf_obj_get(pathname, type); | |
654 | ||
655 | if (prog_fd < 0) | |
656 | fprintf(stderr, "Couldn\'t retrieve pinned program \'%s\': %s\n", | |
657 | pathname, strerror(errno)); | |
658 | return prog_fd; | |
659 | } | |
660 | ||
91d88eeb | 661 | enum bpf_mode { |
e4225669 | 662 | CBPF_BYTECODE, |
91d88eeb DB |
663 | CBPF_FILE, |
664 | EBPF_OBJECT, | |
665 | EBPF_PINNED, | |
e4225669 | 666 | BPF_MODE_MAX, |
91d88eeb DB |
667 | }; |
668 | ||
e4225669 DB |
669 | static int bpf_parse(enum bpf_prog_type *type, enum bpf_mode *mode, |
670 | struct bpf_cfg_in *cfg, const bool *opt_tbl) | |
32e93fb7 | 671 | { |
32e93fb7 | 672 | const char *file, *section, *uds_name; |
32e93fb7 | 673 | bool verbose = false; |
e4225669 | 674 | int i, ret, argc; |
91d88eeb DB |
675 | char **argv; |
676 | ||
e4225669 DB |
677 | argv = cfg->argv; |
678 | argc = cfg->argc; | |
91d88eeb DB |
679 | |
680 | if (opt_tbl[CBPF_BYTECODE] && | |
681 | (matches(*argv, "bytecode") == 0 || | |
682 | strcmp(*argv, "bc") == 0)) { | |
683 | *mode = CBPF_BYTECODE; | |
684 | } else if (opt_tbl[CBPF_FILE] && | |
685 | (matches(*argv, "bytecode-file") == 0 || | |
686 | strcmp(*argv, "bcf") == 0)) { | |
687 | *mode = CBPF_FILE; | |
688 | } else if (opt_tbl[EBPF_OBJECT] && | |
689 | (matches(*argv, "object-file") == 0 || | |
690 | strcmp(*argv, "obj") == 0)) { | |
691 | *mode = EBPF_OBJECT; | |
692 | } else if (opt_tbl[EBPF_PINNED] && | |
693 | (matches(*argv, "object-pinned") == 0 || | |
694 | matches(*argv, "pinned") == 0 || | |
695 | matches(*argv, "fd") == 0)) { | |
696 | *mode = EBPF_PINNED; | |
32e93fb7 DB |
697 | } else { |
698 | fprintf(stderr, "What mode is \"%s\"?\n", *argv); | |
699 | return -1; | |
700 | } | |
701 | ||
702 | NEXT_ARG(); | |
703 | file = section = uds_name = NULL; | |
91d88eeb | 704 | if (*mode == EBPF_OBJECT || *mode == EBPF_PINNED) { |
32e93fb7 DB |
705 | file = *argv; |
706 | NEXT_ARG_FWD(); | |
707 | ||
91d88eeb DB |
708 | if (*type == BPF_PROG_TYPE_UNSPEC) { |
709 | if (argc > 0 && matches(*argv, "type") == 0) { | |
710 | NEXT_ARG(); | |
e4225669 DB |
711 | for (i = 0; i < ARRAY_SIZE(__bpf_prog_meta); |
712 | i++) { | |
713 | if (!__bpf_prog_meta[i].type) | |
714 | continue; | |
715 | if (!matches(*argv, | |
716 | __bpf_prog_meta[i].type)) { | |
717 | *type = i; | |
718 | break; | |
719 | } | |
720 | } | |
721 | ||
722 | if (*type == BPF_PROG_TYPE_UNSPEC) { | |
91d88eeb DB |
723 | fprintf(stderr, "What type is \"%s\"?\n", |
724 | *argv); | |
725 | return -1; | |
726 | } | |
727 | NEXT_ARG_FWD(); | |
728 | } else { | |
729 | *type = BPF_PROG_TYPE_SCHED_CLS; | |
730 | } | |
731 | } | |
732 | ||
e4225669 | 733 | section = bpf_prog_to_default_section(*type); |
32e93fb7 DB |
734 | if (argc > 0 && matches(*argv, "section") == 0) { |
735 | NEXT_ARG(); | |
736 | section = *argv; | |
737 | NEXT_ARG_FWD(); | |
738 | } | |
739 | ||
e4225669 DB |
740 | if (__bpf_prog_meta[*type].may_uds_export) { |
741 | uds_name = getenv(BPF_ENV_UDS); | |
742 | if (argc > 0 && !uds_name && | |
743 | matches(*argv, "export") == 0) { | |
744 | NEXT_ARG(); | |
745 | uds_name = *argv; | |
746 | NEXT_ARG_FWD(); | |
747 | } | |
32e93fb7 DB |
748 | } |
749 | ||
750 | if (argc > 0 && matches(*argv, "verbose") == 0) { | |
751 | verbose = true; | |
752 | NEXT_ARG_FWD(); | |
753 | } | |
754 | ||
755 | PREV_ARG(); | |
756 | } | |
757 | ||
91d88eeb | 758 | if (*mode == CBPF_BYTECODE || *mode == CBPF_FILE) |
e4225669 | 759 | ret = bpf_ops_parse(argc, argv, cfg->ops, *mode == CBPF_FILE); |
91d88eeb DB |
760 | else if (*mode == EBPF_OBJECT) |
761 | ret = bpf_obj_open(file, *type, section, verbose); | |
762 | else if (*mode == EBPF_PINNED) | |
21856018 | 763 | ret = bpf_obj_pinned(file, *type); |
91d88eeb | 764 | else |
32e93fb7 DB |
765 | return -1; |
766 | ||
e4225669 DB |
767 | cfg->object = file; |
768 | cfg->section = section; | |
769 | cfg->uds = uds_name; | |
770 | cfg->argc = argc; | |
771 | cfg->argv = argv; | |
91d88eeb DB |
772 | |
773 | return ret; | |
774 | } | |
775 | ||
e4225669 DB |
776 | static int bpf_parse_opt_tbl(enum bpf_prog_type type, struct bpf_cfg_in *cfg, |
777 | const struct bpf_cfg_ops *ops, void *nl, | |
778 | const bool *opt_tbl) | |
91d88eeb DB |
779 | { |
780 | struct sock_filter opcodes[BPF_MAXINSNS]; | |
91d88eeb | 781 | char annotation[256]; |
91d88eeb DB |
782 | enum bpf_mode mode; |
783 | int ret; | |
784 | ||
e4225669 DB |
785 | cfg->ops = opcodes; |
786 | ret = bpf_parse(&type, &mode, cfg, opt_tbl); | |
787 | cfg->ops = NULL; | |
91d88eeb DB |
788 | if (ret < 0) |
789 | return ret; | |
790 | ||
e4225669 DB |
791 | if (mode == CBPF_BYTECODE || mode == CBPF_FILE) |
792 | ops->cbpf_cb(nl, opcodes, ret); | |
91d88eeb | 793 | if (mode == EBPF_OBJECT || mode == EBPF_PINNED) { |
32e93fb7 | 794 | snprintf(annotation, sizeof(annotation), "%s:[%s]", |
e4225669 DB |
795 | basename(cfg->object), mode == EBPF_PINNED ? |
796 | "*fsobj" : cfg->section); | |
797 | ops->ebpf_cb(nl, ret, annotation); | |
32e93fb7 DB |
798 | } |
799 | ||
91d88eeb DB |
800 | return 0; |
801 | } | |
32e93fb7 | 802 | |
e4225669 DB |
803 | int bpf_parse_common(enum bpf_prog_type type, struct bpf_cfg_in *cfg, |
804 | const struct bpf_cfg_ops *ops, void *nl) | |
805 | { | |
806 | bool opt_tbl[BPF_MODE_MAX] = {}; | |
807 | ||
808 | if (ops->cbpf_cb) { | |
809 | opt_tbl[CBPF_BYTECODE] = true; | |
810 | opt_tbl[CBPF_FILE] = true; | |
811 | } | |
812 | ||
813 | if (ops->ebpf_cb) { | |
814 | opt_tbl[EBPF_OBJECT] = true; | |
815 | opt_tbl[EBPF_PINNED] = true; | |
816 | } | |
817 | ||
818 | return bpf_parse_opt_tbl(type, cfg, ops, nl, opt_tbl); | |
819 | } | |
820 | ||
91d88eeb DB |
821 | int bpf_graft_map(const char *map_path, uint32_t *key, int argc, char **argv) |
822 | { | |
823 | enum bpf_prog_type type = BPF_PROG_TYPE_UNSPEC; | |
824 | const bool opt_tbl[BPF_MODE_MAX] = { | |
91d88eeb DB |
825 | [EBPF_OBJECT] = true, |
826 | [EBPF_PINNED] = true, | |
827 | }; | |
828 | const struct bpf_elf_map test = { | |
829 | .type = BPF_MAP_TYPE_PROG_ARRAY, | |
830 | .size_key = sizeof(int), | |
831 | .size_value = sizeof(int), | |
832 | }; | |
e4225669 DB |
833 | struct bpf_cfg_in cfg = { |
834 | .argc = argc, | |
835 | .argv = argv, | |
836 | }; | |
91d88eeb | 837 | int ret, prog_fd, map_fd; |
91d88eeb DB |
838 | enum bpf_mode mode; |
839 | uint32_t map_key; | |
840 | ||
e4225669 | 841 | prog_fd = bpf_parse(&type, &mode, &cfg, opt_tbl); |
91d88eeb DB |
842 | if (prog_fd < 0) |
843 | return prog_fd; | |
844 | if (key) { | |
845 | map_key = *key; | |
846 | } else { | |
e4225669 | 847 | ret = sscanf(cfg.section, "%*i/%i", &map_key); |
91d88eeb | 848 | if (ret != 1) { |
32a121cb | 849 | fprintf(stderr, "Couldn\'t infer map key from section name! Please provide \'key\' argument!\n"); |
91d88eeb DB |
850 | ret = -EINVAL; |
851 | goto out_prog; | |
852 | } | |
853 | } | |
32e93fb7 | 854 | |
e4225669 | 855 | map_fd = bpf_obj_get(map_path, type); |
91d88eeb DB |
856 | if (map_fd < 0) { |
857 | fprintf(stderr, "Couldn\'t retrieve pinned map \'%s\': %s\n", | |
858 | map_path, strerror(errno)); | |
859 | ret = map_fd; | |
860 | goto out_prog; | |
861 | } | |
862 | ||
863 | ret = bpf_map_selfcheck_pinned(map_fd, &test, | |
fb24802b DB |
864 | offsetof(struct bpf_elf_map, max_elem), |
865 | type); | |
91d88eeb DB |
866 | if (ret < 0) { |
867 | fprintf(stderr, "Map \'%s\' self-check failed!\n", map_path); | |
868 | goto out_map; | |
869 | } | |
870 | ||
871 | ret = bpf_map_update(map_fd, &map_key, &prog_fd, BPF_ANY); | |
872 | if (ret < 0) | |
873 | fprintf(stderr, "Map update failed: %s\n", strerror(errno)); | |
874 | out_map: | |
875 | close(map_fd); | |
876 | out_prog: | |
877 | close(prog_fd); | |
878 | return ret; | |
32e93fb7 DB |
879 | } |
880 | ||
fc4ccce0 DA |
881 | int bpf_prog_attach_fd(int prog_fd, int target_fd, enum bpf_attach_type type) |
882 | { | |
883 | union bpf_attr attr = {}; | |
884 | ||
885 | attr.target_fd = target_fd; | |
886 | attr.attach_bpf_fd = prog_fd; | |
887 | attr.attach_type = type; | |
888 | ||
889 | return bpf(BPF_PROG_ATTACH, &attr, sizeof(attr)); | |
890 | } | |
891 | ||
892 | int bpf_prog_detach_fd(int target_fd, enum bpf_attach_type type) | |
893 | { | |
894 | union bpf_attr attr = {}; | |
895 | ||
896 | attr.target_fd = target_fd; | |
897 | attr.attach_type = type; | |
898 | ||
899 | return bpf(BPF_PROG_DETACH, &attr, sizeof(attr)); | |
900 | } | |
901 | ||
869d889e DA |
902 | int bpf_prog_load(enum bpf_prog_type type, const struct bpf_insn *insns, |
903 | size_t size_insns, const char *license, char *log, | |
904 | size_t size_log) | |
905 | { | |
906 | union bpf_attr attr = {}; | |
907 | ||
908 | attr.prog_type = type; | |
909 | attr.insns = bpf_ptr_to_u64(insns); | |
910 | attr.insn_cnt = size_insns / sizeof(struct bpf_insn); | |
911 | attr.license = bpf_ptr_to_u64(license); | |
912 | ||
913 | if (size_log > 0) { | |
914 | attr.log_buf = bpf_ptr_to_u64(log); | |
915 | attr.log_size = size_log; | |
916 | attr.log_level = 1; | |
917 | } | |
918 | ||
919 | return bpf(BPF_PROG_LOAD, &attr, sizeof(attr)); | |
920 | } | |
921 | ||
6256f8c9 | 922 | #ifdef HAVE_ELF |
32e93fb7 DB |
923 | struct bpf_elf_prog { |
924 | enum bpf_prog_type type; | |
925 | const struct bpf_insn *insns; | |
926 | size_t size; | |
927 | const char *license; | |
928 | }; | |
929 | ||
f6793eec DB |
930 | struct bpf_hash_entry { |
931 | unsigned int pinning; | |
932 | const char *subpath; | |
933 | struct bpf_hash_entry *next; | |
934 | }; | |
935 | ||
32e93fb7 DB |
936 | struct bpf_elf_ctx { |
937 | Elf *elf_fd; | |
938 | GElf_Ehdr elf_hdr; | |
939 | Elf_Data *sym_tab; | |
940 | Elf_Data *str_tab; | |
941 | int obj_fd; | |
942 | int map_fds[ELF_MAX_MAPS]; | |
943 | struct bpf_elf_map maps[ELF_MAX_MAPS]; | |
944 | int sym_num; | |
945 | int map_num; | |
e4225669 | 946 | int map_len; |
32e93fb7 DB |
947 | bool *sec_done; |
948 | int sec_maps; | |
949 | char license[ELF_MAX_LICENSE_LEN]; | |
950 | enum bpf_prog_type type; | |
951 | bool verbose; | |
952 | struct bpf_elf_st stat; | |
f6793eec | 953 | struct bpf_hash_entry *ht[256]; |
f31645d1 DB |
954 | char *log; |
955 | size_t log_size; | |
32e93fb7 DB |
956 | }; |
957 | ||
6256f8c9 | 958 | struct bpf_elf_sec_data { |
32e93fb7 DB |
959 | GElf_Shdr sec_hdr; |
960 | Elf_Data *sec_data; | |
961 | const char *sec_name; | |
6256f8c9 DB |
962 | }; |
963 | ||
964 | struct bpf_map_data { | |
32e93fb7 DB |
965 | int *fds; |
966 | const char *obj; | |
967 | struct bpf_elf_st *st; | |
968 | struct bpf_elf_map *ent; | |
6256f8c9 DB |
969 | }; |
970 | ||
f31645d1 DB |
971 | static __check_format_string(2, 3) void |
972 | bpf_dump_error(struct bpf_elf_ctx *ctx, const char *format, ...) | |
11c39b5e DB |
973 | { |
974 | va_list vl; | |
975 | ||
976 | va_start(vl, format); | |
977 | vfprintf(stderr, format, vl); | |
978 | va_end(vl); | |
979 | ||
f31645d1 | 980 | if (ctx->log && ctx->log[0]) { |
afc1a200 DB |
981 | if (ctx->verbose) { |
982 | fprintf(stderr, "%s\n", ctx->log); | |
983 | } else { | |
984 | unsigned int off = 0, len = strlen(ctx->log); | |
985 | ||
986 | if (len > BPF_MAX_LOG) { | |
987 | off = len - BPF_MAX_LOG; | |
988 | fprintf(stderr, "Skipped %u bytes, use \'verb\' option for the full verbose log.\n[...]\n", | |
989 | off); | |
990 | } | |
991 | fprintf(stderr, "%s\n", ctx->log + off); | |
992 | } | |
993 | ||
f31645d1 DB |
994 | memset(ctx->log, 0, ctx->log_size); |
995 | } | |
996 | } | |
997 | ||
998 | static int bpf_log_realloc(struct bpf_elf_ctx *ctx) | |
999 | { | |
0f74d0f3 | 1000 | const size_t log_max = UINT_MAX >> 8; |
f31645d1 DB |
1001 | size_t log_size = ctx->log_size; |
1002 | void *ptr; | |
1003 | ||
1004 | if (!ctx->log) { | |
1005 | log_size = 65536; | |
0f74d0f3 | 1006 | } else if (log_size < log_max) { |
f31645d1 | 1007 | log_size <<= 1; |
0f74d0f3 TG |
1008 | if (log_size > log_max) |
1009 | log_size = log_max; | |
1010 | } else { | |
1011 | return -EINVAL; | |
d937a74b | 1012 | } |
f31645d1 DB |
1013 | |
1014 | ptr = realloc(ctx->log, log_size); | |
1015 | if (!ptr) | |
1016 | return -ENOMEM; | |
1017 | ||
1018 | ctx->log = ptr; | |
1019 | ctx->log_size = log_size; | |
1020 | ||
1021 | return 0; | |
11c39b5e DB |
1022 | } |
1023 | ||
4dd3f50a DB |
1024 | static int bpf_map_create(enum bpf_map_type type, uint32_t size_key, |
1025 | uint32_t size_value, uint32_t max_elem, | |
1026 | uint32_t flags) | |
11c39b5e | 1027 | { |
d17b136f | 1028 | union bpf_attr attr = {}; |
67584e3a | 1029 | |
67584e3a ND |
1030 | attr.map_type = type; |
1031 | attr.key_size = size_key; | |
1032 | attr.value_size = size_value; | |
1033 | attr.max_entries = max_elem; | |
4dd3f50a | 1034 | attr.map_flags = flags; |
11c39b5e DB |
1035 | |
1036 | return bpf(BPF_MAP_CREATE, &attr, sizeof(attr)); | |
1037 | } | |
11c39b5e | 1038 | |
32e93fb7 | 1039 | static int bpf_obj_pin(int fd, const char *pathname) |
11c39b5e | 1040 | { |
d17b136f | 1041 | union bpf_attr attr = {}; |
67584e3a | 1042 | |
67584e3a ND |
1043 | attr.pathname = bpf_ptr_to_u64(pathname); |
1044 | attr.bpf_fd = fd; | |
32e93fb7 DB |
1045 | |
1046 | return bpf(BPF_OBJ_PIN, &attr, sizeof(attr)); | |
1047 | } | |
11c39b5e | 1048 | |
32e93fb7 DB |
1049 | static int bpf_obj_hash(const char *object, uint8_t *out, size_t len) |
1050 | { | |
1051 | struct sockaddr_alg alg = { | |
1052 | .salg_family = AF_ALG, | |
1053 | .salg_type = "hash", | |
1054 | .salg_name = "sha1", | |
1055 | }; | |
1056 | int ret, cfd, ofd, ffd; | |
1057 | struct stat stbuff; | |
1058 | ssize_t size; | |
1059 | ||
1060 | if (!object || len != 20) | |
1061 | return -EINVAL; | |
1062 | ||
1063 | cfd = socket(AF_ALG, SOCK_SEQPACKET, 0); | |
1064 | if (cfd < 0) { | |
1065 | fprintf(stderr, "Cannot get AF_ALG socket: %s\n", | |
1066 | strerror(errno)); | |
1067 | return cfd; | |
1068 | } | |
1069 | ||
1070 | ret = bind(cfd, (struct sockaddr *)&alg, sizeof(alg)); | |
1071 | if (ret < 0) { | |
1072 | fprintf(stderr, "Error binding socket: %s\n", strerror(errno)); | |
1073 | goto out_cfd; | |
1074 | } | |
1075 | ||
1076 | ofd = accept(cfd, NULL, 0); | |
1077 | if (ofd < 0) { | |
1078 | fprintf(stderr, "Error accepting socket: %s\n", | |
1079 | strerror(errno)); | |
1080 | ret = ofd; | |
1081 | goto out_cfd; | |
1082 | } | |
1083 | ||
1084 | ffd = open(object, O_RDONLY); | |
1085 | if (ffd < 0) { | |
1086 | fprintf(stderr, "Error opening object %s: %s\n", | |
1087 | object, strerror(errno)); | |
1088 | ret = ffd; | |
1089 | goto out_ofd; | |
1090 | } | |
1091 | ||
32a121cb | 1092 | ret = fstat(ffd, &stbuff); |
32e93fb7 DB |
1093 | if (ret < 0) { |
1094 | fprintf(stderr, "Error doing fstat: %s\n", | |
1095 | strerror(errno)); | |
1096 | goto out_ffd; | |
d937a74b | 1097 | } |
11c39b5e | 1098 | |
32e93fb7 DB |
1099 | size = sendfile(ofd, ffd, NULL, stbuff.st_size); |
1100 | if (size != stbuff.st_size) { | |
1101 | fprintf(stderr, "Error from sendfile (%zd vs %zu bytes): %s\n", | |
1102 | size, stbuff.st_size, strerror(errno)); | |
1103 | ret = -1; | |
1104 | goto out_ffd; | |
1105 | } | |
1106 | ||
1107 | size = read(ofd, out, len); | |
1108 | if (size != len) { | |
1109 | fprintf(stderr, "Error from read (%zd vs %zu bytes): %s\n", | |
1110 | size, len, strerror(errno)); | |
1111 | ret = -1; | |
1112 | } else { | |
1113 | ret = 0; | |
1114 | } | |
1115 | out_ffd: | |
1116 | close(ffd); | |
1117 | out_ofd: | |
1118 | close(ofd); | |
1119 | out_cfd: | |
1120 | close(cfd); | |
1121 | return ret; | |
11c39b5e DB |
1122 | } |
1123 | ||
32e93fb7 | 1124 | static const char *bpf_get_obj_uid(const char *pathname) |
11c39b5e | 1125 | { |
32a121cb | 1126 | static bool bpf_uid_cached; |
32e93fb7 DB |
1127 | static char bpf_uid[64]; |
1128 | uint8_t tmp[20]; | |
1129 | int ret; | |
11c39b5e | 1130 | |
32e93fb7 DB |
1131 | if (bpf_uid_cached) |
1132 | goto done; | |
11c39b5e | 1133 | |
32e93fb7 DB |
1134 | ret = bpf_obj_hash(pathname, tmp, sizeof(tmp)); |
1135 | if (ret) { | |
1136 | fprintf(stderr, "Object hashing failed!\n"); | |
1137 | return NULL; | |
1138 | } | |
1139 | ||
1140 | hexstring_n2a(tmp, sizeof(tmp), bpf_uid, sizeof(bpf_uid)); | |
1141 | bpf_uid_cached = true; | |
1142 | done: | |
1143 | return bpf_uid; | |
11c39b5e DB |
1144 | } |
1145 | ||
32e93fb7 DB |
1146 | static int bpf_init_env(const char *pathname) |
1147 | { | |
1148 | struct rlimit limit = { | |
1149 | .rlim_cur = RLIM_INFINITY, | |
1150 | .rlim_max = RLIM_INFINITY, | |
1151 | }; | |
1152 | ||
1153 | /* Don't bother in case we fail! */ | |
1154 | setrlimit(RLIMIT_MEMLOCK, &limit); | |
1155 | ||
e4225669 | 1156 | if (!bpf_get_work_dir(BPF_PROG_TYPE_UNSPEC)) { |
32a121cb | 1157 | fprintf(stderr, "Continuing without mounted eBPF fs. Too old kernel?\n"); |
32e93fb7 DB |
1158 | return 0; |
1159 | } | |
1160 | ||
1161 | if (!bpf_get_obj_uid(pathname)) | |
1162 | return -1; | |
1163 | ||
1164 | return 0; | |
6256f8c9 DB |
1165 | } |
1166 | ||
f6793eec DB |
1167 | static const char *bpf_custom_pinning(const struct bpf_elf_ctx *ctx, |
1168 | uint32_t pinning) | |
1169 | { | |
1170 | struct bpf_hash_entry *entry; | |
1171 | ||
1172 | entry = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1173 | while (entry && entry->pinning != pinning) | |
1174 | entry = entry->next; | |
1175 | ||
1176 | return entry ? entry->subpath : NULL; | |
1177 | } | |
1178 | ||
1179 | static bool bpf_no_pinning(const struct bpf_elf_ctx *ctx, | |
1180 | uint32_t pinning) | |
11c39b5e | 1181 | { |
32e93fb7 DB |
1182 | switch (pinning) { |
1183 | case PIN_OBJECT_NS: | |
1184 | case PIN_GLOBAL_NS: | |
1185 | return false; | |
1186 | case PIN_NONE: | |
32e93fb7 | 1187 | return true; |
f6793eec DB |
1188 | default: |
1189 | return !bpf_custom_pinning(ctx, pinning); | |
32e93fb7 DB |
1190 | } |
1191 | } | |
1192 | ||
1193 | static void bpf_make_pathname(char *pathname, size_t len, const char *name, | |
f6793eec | 1194 | const struct bpf_elf_ctx *ctx, uint32_t pinning) |
32e93fb7 DB |
1195 | { |
1196 | switch (pinning) { | |
1197 | case PIN_OBJECT_NS: | |
e4225669 DB |
1198 | snprintf(pathname, len, "%s/%s/%s", |
1199 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1200 | bpf_get_obj_uid(NULL), name); |
1201 | break; | |
1202 | case PIN_GLOBAL_NS: | |
e4225669 DB |
1203 | snprintf(pathname, len, "%s/%s/%s", |
1204 | bpf_get_work_dir(ctx->type), | |
32e93fb7 DB |
1205 | BPF_DIR_GLOBALS, name); |
1206 | break; | |
f6793eec | 1207 | default: |
e4225669 DB |
1208 | snprintf(pathname, len, "%s/../%s/%s", |
1209 | bpf_get_work_dir(ctx->type), | |
f6793eec DB |
1210 | bpf_custom_pinning(ctx, pinning), name); |
1211 | break; | |
32e93fb7 DB |
1212 | } |
1213 | } | |
1214 | ||
f6793eec DB |
1215 | static int bpf_probe_pinned(const char *name, const struct bpf_elf_ctx *ctx, |
1216 | uint32_t pinning) | |
32e93fb7 DB |
1217 | { |
1218 | char pathname[PATH_MAX]; | |
1219 | ||
e4225669 | 1220 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
32e93fb7 DB |
1221 | return 0; |
1222 | ||
f6793eec | 1223 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); |
e4225669 | 1224 | return bpf_obj_get(pathname, ctx->type); |
32e93fb7 DB |
1225 | } |
1226 | ||
e4225669 | 1227 | static int bpf_make_obj_path(const struct bpf_elf_ctx *ctx) |
32e93fb7 | 1228 | { |
f6793eec | 1229 | char tmp[PATH_MAX]; |
32e93fb7 DB |
1230 | int ret; |
1231 | ||
e4225669 | 1232 | snprintf(tmp, sizeof(tmp), "%s/%s", bpf_get_work_dir(ctx->type), |
f6793eec DB |
1233 | bpf_get_obj_uid(NULL)); |
1234 | ||
1235 | ret = mkdir(tmp, S_IRWXU); | |
1236 | if (ret && errno != EEXIST) { | |
1237 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, strerror(errno)); | |
1238 | return ret; | |
1239 | } | |
1240 | ||
1241 | return 0; | |
1242 | } | |
1243 | ||
e4225669 DB |
1244 | static int bpf_make_custom_path(const struct bpf_elf_ctx *ctx, |
1245 | const char *todo) | |
f6793eec DB |
1246 | { |
1247 | char tmp[PATH_MAX], rem[PATH_MAX], *sub; | |
1248 | int ret; | |
1249 | ||
e4225669 | 1250 | snprintf(tmp, sizeof(tmp), "%s/../", bpf_get_work_dir(ctx->type)); |
f6793eec DB |
1251 | snprintf(rem, sizeof(rem), "%s/", todo); |
1252 | sub = strtok(rem, "/"); | |
32e93fb7 | 1253 | |
f6793eec DB |
1254 | while (sub) { |
1255 | if (strlen(tmp) + strlen(sub) + 2 > PATH_MAX) | |
1256 | return -EINVAL; | |
1257 | ||
1258 | strcat(tmp, sub); | |
1259 | strcat(tmp, "/"); | |
32e93fb7 | 1260 | |
f6793eec | 1261 | ret = mkdir(tmp, S_IRWXU); |
32e93fb7 | 1262 | if (ret && errno != EEXIST) { |
f6793eec | 1263 | fprintf(stderr, "mkdir %s failed: %s\n", tmp, |
32e93fb7 DB |
1264 | strerror(errno)); |
1265 | return ret; | |
1266 | } | |
f6793eec DB |
1267 | |
1268 | sub = strtok(NULL, "/"); | |
32e93fb7 DB |
1269 | } |
1270 | ||
f6793eec DB |
1271 | return 0; |
1272 | } | |
1273 | ||
1274 | static int bpf_place_pinned(int fd, const char *name, | |
1275 | const struct bpf_elf_ctx *ctx, uint32_t pinning) | |
1276 | { | |
1277 | char pathname[PATH_MAX]; | |
1278 | const char *tmp; | |
1279 | int ret = 0; | |
1280 | ||
e4225669 | 1281 | if (bpf_no_pinning(ctx, pinning) || !bpf_get_work_dir(ctx->type)) |
f6793eec DB |
1282 | return 0; |
1283 | ||
1284 | if (pinning == PIN_OBJECT_NS) | |
e4225669 | 1285 | ret = bpf_make_obj_path(ctx); |
f6793eec | 1286 | else if ((tmp = bpf_custom_pinning(ctx, pinning))) |
e4225669 | 1287 | ret = bpf_make_custom_path(ctx, tmp); |
f6793eec DB |
1288 | if (ret < 0) |
1289 | return ret; | |
1290 | ||
1291 | bpf_make_pathname(pathname, sizeof(pathname), name, ctx, pinning); | |
32e93fb7 DB |
1292 | return bpf_obj_pin(fd, pathname); |
1293 | } | |
1294 | ||
f31645d1 DB |
1295 | static void bpf_prog_report(int fd, const char *section, |
1296 | const struct bpf_elf_prog *prog, | |
1297 | struct bpf_elf_ctx *ctx) | |
32e93fb7 | 1298 | { |
afc1a200 DB |
1299 | unsigned int insns = prog->size / sizeof(struct bpf_insn); |
1300 | ||
1301 | fprintf(stderr, "\nProg section \'%s\' %s%s (%d)!\n", section, | |
f31645d1 DB |
1302 | fd < 0 ? "rejected: " : "loaded", |
1303 | fd < 0 ? strerror(errno) : "", | |
1304 | fd < 0 ? errno : fd); | |
1305 | ||
1306 | fprintf(stderr, " - Type: %u\n", prog->type); | |
afc1a200 DB |
1307 | fprintf(stderr, " - Instructions: %u (%u over limit)\n", |
1308 | insns, insns > BPF_MAXINSNS ? insns - BPF_MAXINSNS : 0); | |
f31645d1 DB |
1309 | fprintf(stderr, " - License: %s\n\n", prog->license); |
1310 | ||
1311 | bpf_dump_error(ctx, "Verifier analysis:\n\n"); | |
1312 | } | |
32e93fb7 | 1313 | |
f31645d1 DB |
1314 | static int bpf_prog_attach(const char *section, |
1315 | const struct bpf_elf_prog *prog, | |
1316 | struct bpf_elf_ctx *ctx) | |
1317 | { | |
1318 | int tries = 0, fd; | |
1319 | retry: | |
32e93fb7 DB |
1320 | errno = 0; |
1321 | fd = bpf_prog_load(prog->type, prog->insns, prog->size, | |
f31645d1 DB |
1322 | prog->license, ctx->log, ctx->log_size); |
1323 | if (fd < 0 || ctx->verbose) { | |
1324 | /* The verifier log is pretty chatty, sometimes so chatty | |
1325 | * on larger programs, that we could fail to dump everything | |
1326 | * into our buffer. Still, try to give a debuggable error | |
1327 | * log for the user, so enlarge it and re-fail. | |
1328 | */ | |
1329 | if (fd < 0 && (errno == ENOSPC || !ctx->log_size)) { | |
0f74d0f3 | 1330 | if (tries++ < 10 && !bpf_log_realloc(ctx)) |
f31645d1 DB |
1331 | goto retry; |
1332 | ||
32a121cb | 1333 | fprintf(stderr, "Log buffer too small to dump verifier log %zu bytes (%d tries)!\n", |
f31645d1 DB |
1334 | ctx->log_size, tries); |
1335 | return fd; | |
1336 | } | |
1337 | ||
1338 | bpf_prog_report(fd, section, prog, ctx); | |
32e93fb7 DB |
1339 | } |
1340 | ||
1341 | return fd; | |
1342 | } | |
1343 | ||
f31645d1 DB |
1344 | static void bpf_map_report(int fd, const char *name, |
1345 | const struct bpf_elf_map *map, | |
1346 | struct bpf_elf_ctx *ctx) | |
1347 | { | |
1348 | fprintf(stderr, "Map object \'%s\' %s%s (%d)!\n", name, | |
1349 | fd < 0 ? "rejected: " : "loaded", | |
1350 | fd < 0 ? strerror(errno) : "", | |
1351 | fd < 0 ? errno : fd); | |
1352 | ||
1353 | fprintf(stderr, " - Type: %u\n", map->type); | |
1354 | fprintf(stderr, " - Identifier: %u\n", map->id); | |
1355 | fprintf(stderr, " - Pinning: %u\n", map->pinning); | |
1356 | fprintf(stderr, " - Size key: %u\n", map->size_key); | |
1357 | fprintf(stderr, " - Size value: %u\n", map->size_value); | |
4dd3f50a DB |
1358 | fprintf(stderr, " - Max elems: %u\n", map->max_elem); |
1359 | fprintf(stderr, " - Flags: %#x\n\n", map->flags); | |
f31645d1 DB |
1360 | } |
1361 | ||
32e93fb7 | 1362 | static int bpf_map_attach(const char *name, const struct bpf_elf_map *map, |
f31645d1 | 1363 | struct bpf_elf_ctx *ctx) |
32e93fb7 DB |
1364 | { |
1365 | int fd, ret; | |
1366 | ||
f6793eec | 1367 | fd = bpf_probe_pinned(name, ctx, map->pinning); |
32e93fb7 | 1368 | if (fd > 0) { |
91d88eeb DB |
1369 | ret = bpf_map_selfcheck_pinned(fd, map, |
1370 | offsetof(struct bpf_elf_map, | |
fb24802b | 1371 | id), ctx->type); |
9e607f2e DB |
1372 | if (ret < 0) { |
1373 | close(fd); | |
1374 | fprintf(stderr, "Map \'%s\' self-check failed!\n", | |
1375 | name); | |
1376 | return ret; | |
1377 | } | |
f31645d1 | 1378 | if (ctx->verbose) |
32e93fb7 DB |
1379 | fprintf(stderr, "Map \'%s\' loaded as pinned!\n", |
1380 | name); | |
1381 | return fd; | |
1382 | } | |
1383 | ||
1384 | errno = 0; | |
1385 | fd = bpf_map_create(map->type, map->size_key, map->size_value, | |
4dd3f50a | 1386 | map->max_elem, map->flags); |
f31645d1 DB |
1387 | if (fd < 0 || ctx->verbose) { |
1388 | bpf_map_report(fd, name, map, ctx); | |
32e93fb7 DB |
1389 | if (fd < 0) |
1390 | return fd; | |
1391 | } | |
1392 | ||
f6793eec | 1393 | ret = bpf_place_pinned(fd, name, ctx, map->pinning); |
32e93fb7 DB |
1394 | if (ret < 0 && errno != EEXIST) { |
1395 | fprintf(stderr, "Could not pin %s map: %s\n", name, | |
1396 | strerror(errno)); | |
1397 | close(fd); | |
1398 | return ret; | |
1399 | } | |
1400 | ||
1401 | return fd; | |
1402 | } | |
1403 | ||
32e93fb7 DB |
1404 | static const char *bpf_str_tab_name(const struct bpf_elf_ctx *ctx, |
1405 | const GElf_Sym *sym) | |
1406 | { | |
1407 | return ctx->str_tab->d_buf + sym->st_name; | |
1408 | } | |
1409 | ||
1410 | static const char *bpf_map_fetch_name(struct bpf_elf_ctx *ctx, int which) | |
1411 | { | |
1412 | GElf_Sym sym; | |
11c39b5e DB |
1413 | int i; |
1414 | ||
32e93fb7 DB |
1415 | for (i = 0; i < ctx->sym_num; i++) { |
1416 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1417 | continue; | |
1418 | ||
5230a2ed DB |
1419 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || |
1420 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
32e93fb7 | 1421 | sym.st_shndx != ctx->sec_maps || |
e4225669 | 1422 | sym.st_value / ctx->map_len != which) |
32e93fb7 DB |
1423 | continue; |
1424 | ||
1425 | return bpf_str_tab_name(ctx, &sym); | |
11c39b5e | 1426 | } |
32e93fb7 DB |
1427 | |
1428 | return NULL; | |
11c39b5e DB |
1429 | } |
1430 | ||
32e93fb7 | 1431 | static int bpf_maps_attach_all(struct bpf_elf_ctx *ctx) |
11c39b5e | 1432 | { |
32e93fb7 DB |
1433 | const char *map_name; |
1434 | int i, fd; | |
11c39b5e | 1435 | |
32e93fb7 DB |
1436 | for (i = 0; i < ctx->map_num; i++) { |
1437 | map_name = bpf_map_fetch_name(ctx, i); | |
1438 | if (!map_name) | |
1439 | return -EIO; | |
11c39b5e | 1440 | |
f31645d1 | 1441 | fd = bpf_map_attach(map_name, &ctx->maps[i], ctx); |
32e93fb7 DB |
1442 | if (fd < 0) |
1443 | return fd; | |
11c39b5e | 1444 | |
32e93fb7 | 1445 | ctx->map_fds[i] = fd; |
11c39b5e DB |
1446 | } |
1447 | ||
1448 | return 0; | |
11c39b5e DB |
1449 | } |
1450 | ||
e4225669 DB |
1451 | static int bpf_map_num_sym(struct bpf_elf_ctx *ctx) |
1452 | { | |
1453 | int i, num = 0; | |
1454 | GElf_Sym sym; | |
1455 | ||
1456 | for (i = 0; i < ctx->sym_num; i++) { | |
1457 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1458 | continue; | |
1459 | ||
1460 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1461 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1462 | sym.st_shndx != ctx->sec_maps) | |
1463 | continue; | |
1464 | num++; | |
1465 | } | |
1466 | ||
1467 | return num; | |
1468 | } | |
1469 | ||
32e93fb7 DB |
1470 | static int bpf_fill_section_data(struct bpf_elf_ctx *ctx, int section, |
1471 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1472 | { |
32e93fb7 | 1473 | Elf_Data *sec_edata; |
11c39b5e DB |
1474 | GElf_Shdr sec_hdr; |
1475 | Elf_Scn *sec_fd; | |
11c39b5e DB |
1476 | char *sec_name; |
1477 | ||
32e93fb7 | 1478 | memset(data, 0, sizeof(*data)); |
11c39b5e | 1479 | |
32e93fb7 | 1480 | sec_fd = elf_getscn(ctx->elf_fd, section); |
11c39b5e DB |
1481 | if (!sec_fd) |
1482 | return -EINVAL; | |
11c39b5e DB |
1483 | if (gelf_getshdr(sec_fd, &sec_hdr) != &sec_hdr) |
1484 | return -EIO; | |
1485 | ||
32e93fb7 | 1486 | sec_name = elf_strptr(ctx->elf_fd, ctx->elf_hdr.e_shstrndx, |
11c39b5e DB |
1487 | sec_hdr.sh_name); |
1488 | if (!sec_name || !sec_hdr.sh_size) | |
1489 | return -ENOENT; | |
1490 | ||
1491 | sec_edata = elf_getdata(sec_fd, NULL); | |
1492 | if (!sec_edata || elf_getdata(sec_fd, sec_edata)) | |
1493 | return -EIO; | |
1494 | ||
32e93fb7 | 1495 | memcpy(&data->sec_hdr, &sec_hdr, sizeof(sec_hdr)); |
11c39b5e | 1496 | |
32e93fb7 DB |
1497 | data->sec_name = sec_name; |
1498 | data->sec_data = sec_edata; | |
11c39b5e DB |
1499 | return 0; |
1500 | } | |
1501 | ||
e4225669 DB |
1502 | struct bpf_elf_map_min { |
1503 | __u32 type; | |
1504 | __u32 size_key; | |
1505 | __u32 size_value; | |
1506 | __u32 max_elem; | |
1507 | }; | |
11c39b5e | 1508 | |
e4225669 DB |
1509 | static int bpf_fetch_maps_begin(struct bpf_elf_ctx *ctx, int section, |
1510 | struct bpf_elf_sec_data *data) | |
1511 | { | |
1512 | ctx->map_num = data->sec_data->d_size; | |
32e93fb7 DB |
1513 | ctx->sec_maps = section; |
1514 | ctx->sec_done[section] = true; | |
11c39b5e | 1515 | |
e4225669 | 1516 | if (ctx->map_num > sizeof(ctx->maps)) { |
32e93fb7 DB |
1517 | fprintf(stderr, "Too many BPF maps in ELF section!\n"); |
1518 | return -ENOMEM; | |
1519 | } | |
11c39b5e | 1520 | |
e4225669 DB |
1521 | memcpy(ctx->maps, data->sec_data->d_buf, ctx->map_num); |
1522 | return 0; | |
1523 | } | |
1524 | ||
1525 | static int bpf_map_verify_all_offs(struct bpf_elf_ctx *ctx, int end) | |
1526 | { | |
1527 | GElf_Sym sym; | |
1528 | int off, i; | |
1529 | ||
1530 | for (off = 0; off < end; off += ctx->map_len) { | |
1531 | /* Order doesn't need to be linear here, hence we walk | |
1532 | * the table again. | |
1533 | */ | |
1534 | for (i = 0; i < ctx->sym_num; i++) { | |
1535 | if (gelf_getsym(ctx->sym_tab, i, &sym) != &sym) | |
1536 | continue; | |
1537 | if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL || | |
1538 | GELF_ST_TYPE(sym.st_info) != STT_NOTYPE || | |
1539 | sym.st_shndx != ctx->sec_maps) | |
1540 | continue; | |
1541 | if (sym.st_value == off) | |
1542 | break; | |
1543 | if (i == ctx->sym_num - 1) | |
1544 | return -1; | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | return off == end ? 0 : -1; | |
1549 | } | |
1550 | ||
1551 | static int bpf_fetch_maps_end(struct bpf_elf_ctx *ctx) | |
1552 | { | |
1553 | struct bpf_elf_map fixup[ARRAY_SIZE(ctx->maps)] = {}; | |
1554 | int i, sym_num = bpf_map_num_sym(ctx); | |
1555 | __u8 *buff; | |
1556 | ||
1557 | if (sym_num == 0 || sym_num > ARRAY_SIZE(ctx->maps)) { | |
1558 | fprintf(stderr, "%u maps not supported in current map section!\n", | |
1559 | sym_num); | |
1560 | return -EINVAL; | |
1561 | } | |
1562 | ||
1563 | if (ctx->map_num % sym_num != 0 || | |
1564 | ctx->map_num % sizeof(__u32) != 0) { | |
1565 | fprintf(stderr, "Number BPF map symbols are not multiple of struct bpf_elf_map!\n"); | |
1566 | return -EINVAL; | |
1567 | } | |
1568 | ||
1569 | ctx->map_len = ctx->map_num / sym_num; | |
1570 | if (bpf_map_verify_all_offs(ctx, ctx->map_num)) { | |
1571 | fprintf(stderr, "Different struct bpf_elf_map in use!\n"); | |
1572 | return -EINVAL; | |
1573 | } | |
1574 | ||
1575 | if (ctx->map_len == sizeof(struct bpf_elf_map)) { | |
1576 | ctx->map_num = sym_num; | |
1577 | return 0; | |
1578 | } else if (ctx->map_len > sizeof(struct bpf_elf_map)) { | |
1579 | fprintf(stderr, "struct bpf_elf_map not supported, coming from future version?\n"); | |
1580 | return -EINVAL; | |
1581 | } else if (ctx->map_len < sizeof(struct bpf_elf_map_min)) { | |
1582 | fprintf(stderr, "struct bpf_elf_map too small, not supported!\n"); | |
1583 | return -EINVAL; | |
1584 | } | |
1585 | ||
1586 | ctx->map_num = sym_num; | |
1587 | for (i = 0, buff = (void *)ctx->maps; i < ctx->map_num; | |
1588 | i++, buff += ctx->map_len) { | |
1589 | /* The fixup leaves the rest of the members as zero, which | |
1590 | * is fine currently, but option exist to set some other | |
1591 | * default value as well when needed in future. | |
1592 | */ | |
1593 | memcpy(&fixup[i], buff, ctx->map_len); | |
1594 | } | |
1595 | ||
1596 | memcpy(ctx->maps, fixup, sizeof(fixup)); | |
1597 | ||
1598 | printf("Note: %zu bytes struct bpf_elf_map fixup performed due to size mismatch!\n", | |
1599 | sizeof(struct bpf_elf_map) - ctx->map_len); | |
32e93fb7 DB |
1600 | return 0; |
1601 | } | |
11c39b5e | 1602 | |
32e93fb7 DB |
1603 | static int bpf_fetch_license(struct bpf_elf_ctx *ctx, int section, |
1604 | struct bpf_elf_sec_data *data) | |
1605 | { | |
1606 | if (data->sec_data->d_size > sizeof(ctx->license)) | |
1607 | return -ENOMEM; | |
11c39b5e | 1608 | |
32e93fb7 DB |
1609 | memcpy(ctx->license, data->sec_data->d_buf, data->sec_data->d_size); |
1610 | ctx->sec_done[section] = true; | |
1611 | return 0; | |
1612 | } | |
11c39b5e | 1613 | |
32e93fb7 DB |
1614 | static int bpf_fetch_symtab(struct bpf_elf_ctx *ctx, int section, |
1615 | struct bpf_elf_sec_data *data) | |
1616 | { | |
1617 | ctx->sym_tab = data->sec_data; | |
1618 | ctx->sym_num = data->sec_hdr.sh_size / data->sec_hdr.sh_entsize; | |
1619 | ctx->sec_done[section] = true; | |
11c39b5e DB |
1620 | return 0; |
1621 | } | |
1622 | ||
32e93fb7 DB |
1623 | static int bpf_fetch_strtab(struct bpf_elf_ctx *ctx, int section, |
1624 | struct bpf_elf_sec_data *data) | |
11c39b5e | 1625 | { |
32e93fb7 DB |
1626 | ctx->str_tab = data->sec_data; |
1627 | ctx->sec_done[section] = true; | |
1628 | return 0; | |
1629 | } | |
11c39b5e | 1630 | |
afc1a200 DB |
1631 | static bool bpf_has_map_data(const struct bpf_elf_ctx *ctx) |
1632 | { | |
1633 | return ctx->sym_tab && ctx->str_tab && ctx->sec_maps; | |
1634 | } | |
1635 | ||
32e93fb7 DB |
1636 | static int bpf_fetch_ancillary(struct bpf_elf_ctx *ctx) |
1637 | { | |
1638 | struct bpf_elf_sec_data data; | |
1639 | int i, ret = -1; | |
11c39b5e | 1640 | |
32e93fb7 DB |
1641 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1642 | ret = bpf_fill_section_data(ctx, i, &data); | |
11c39b5e DB |
1643 | if (ret < 0) |
1644 | continue; | |
1645 | ||
cce3d466 DB |
1646 | if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1647 | !strcmp(data.sec_name, ELF_SECTION_MAPS)) | |
e4225669 | 1648 | ret = bpf_fetch_maps_begin(ctx, i, &data); |
cce3d466 DB |
1649 | else if (data.sec_hdr.sh_type == SHT_PROGBITS && |
1650 | !strcmp(data.sec_name, ELF_SECTION_LICENSE)) | |
32e93fb7 | 1651 | ret = bpf_fetch_license(ctx, i, &data); |
cce3d466 DB |
1652 | else if (data.sec_hdr.sh_type == SHT_SYMTAB && |
1653 | !strcmp(data.sec_name, ".symtab")) | |
32e93fb7 DB |
1654 | ret = bpf_fetch_symtab(ctx, i, &data); |
1655 | else if (data.sec_hdr.sh_type == SHT_STRTAB && | |
cce3d466 | 1656 | !strcmp(data.sec_name, ".strtab")) |
32e93fb7 DB |
1657 | ret = bpf_fetch_strtab(ctx, i, &data); |
1658 | if (ret < 0) { | |
afc1a200 | 1659 | fprintf(stderr, "Error parsing section %d! Perhaps check with readelf -a?\n", |
32a121cb | 1660 | i); |
e4225669 | 1661 | return ret; |
11c39b5e | 1662 | } |
32e93fb7 DB |
1663 | } |
1664 | ||
afc1a200 | 1665 | if (bpf_has_map_data(ctx)) { |
e4225669 DB |
1666 | ret = bpf_fetch_maps_end(ctx); |
1667 | if (ret < 0) { | |
1668 | fprintf(stderr, "Error fixing up map structure, incompatible struct bpf_elf_map used?\n"); | |
1669 | return ret; | |
1670 | } | |
1671 | ||
32e93fb7 DB |
1672 | ret = bpf_maps_attach_all(ctx); |
1673 | if (ret < 0) { | |
1674 | fprintf(stderr, "Error loading maps into kernel!\n"); | |
1675 | return ret; | |
11c39b5e DB |
1676 | } |
1677 | } | |
1678 | ||
1679 | return ret; | |
1680 | } | |
1681 | ||
e4225669 DB |
1682 | static int bpf_fetch_prog(struct bpf_elf_ctx *ctx, const char *section, |
1683 | bool *sseen) | |
11c39b5e | 1684 | { |
32e93fb7 DB |
1685 | struct bpf_elf_sec_data data; |
1686 | struct bpf_elf_prog prog; | |
1687 | int ret, i, fd = -1; | |
11c39b5e | 1688 | |
32e93fb7 DB |
1689 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1690 | if (ctx->sec_done[i]) | |
11c39b5e DB |
1691 | continue; |
1692 | ||
32e93fb7 | 1693 | ret = bpf_fill_section_data(ctx, i, &data); |
cce3d466 DB |
1694 | if (ret < 0 || |
1695 | !(data.sec_hdr.sh_type == SHT_PROGBITS && | |
1696 | data.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1697 | !strcmp(data.sec_name, section))) | |
11c39b5e DB |
1698 | continue; |
1699 | ||
e4225669 DB |
1700 | *sseen = true; |
1701 | ||
32e93fb7 DB |
1702 | memset(&prog, 0, sizeof(prog)); |
1703 | prog.type = ctx->type; | |
1704 | prog.insns = data.sec_data->d_buf; | |
1705 | prog.size = data.sec_data->d_size; | |
1706 | prog.license = ctx->license; | |
11c39b5e | 1707 | |
f31645d1 | 1708 | fd = bpf_prog_attach(section, &prog, ctx); |
32e93fb7 | 1709 | if (fd < 0) |
e4225669 | 1710 | return fd; |
11c39b5e | 1711 | |
32e93fb7 | 1712 | ctx->sec_done[i] = true; |
11c39b5e DB |
1713 | break; |
1714 | } | |
1715 | ||
32e93fb7 | 1716 | return fd; |
11c39b5e DB |
1717 | } |
1718 | ||
32e93fb7 DB |
1719 | static int bpf_apply_relo_data(struct bpf_elf_ctx *ctx, |
1720 | struct bpf_elf_sec_data *data_relo, | |
1721 | struct bpf_elf_sec_data *data_insn) | |
11c39b5e | 1722 | { |
32e93fb7 DB |
1723 | Elf_Data *idata = data_insn->sec_data; |
1724 | GElf_Shdr *rhdr = &data_relo->sec_hdr; | |
1725 | int relo_ent, relo_num = rhdr->sh_size / rhdr->sh_entsize; | |
1726 | struct bpf_insn *insns = idata->d_buf; | |
1727 | unsigned int num_insns = idata->d_size / sizeof(*insns); | |
11c39b5e | 1728 | |
32e93fb7 DB |
1729 | for (relo_ent = 0; relo_ent < relo_num; relo_ent++) { |
1730 | unsigned int ioff, rmap; | |
1731 | GElf_Rel relo; | |
1732 | GElf_Sym sym; | |
1733 | ||
1734 | if (gelf_getrel(data_relo->sec_data, relo_ent, &relo) != &relo) | |
1735 | return -EIO; | |
1736 | ||
1737 | ioff = relo.r_offset / sizeof(struct bpf_insn); | |
1738 | if (ioff >= num_insns || | |
a576c6b9 | 1739 | insns[ioff].code != (BPF_LD | BPF_IMM | BPF_DW)) { |
32a121cb | 1740 | fprintf(stderr, "ELF contains relo data for non ld64 instruction at offset %u! Compiler bug?!\n", |
a576c6b9 DB |
1741 | ioff); |
1742 | if (ioff < num_insns && | |
1743 | insns[ioff].code == (BPF_JMP | BPF_CALL)) | |
32a121cb | 1744 | fprintf(stderr, " - Try to annotate functions with always_inline attribute!\n"); |
32e93fb7 | 1745 | return -EINVAL; |
a576c6b9 | 1746 | } |
32e93fb7 DB |
1747 | |
1748 | if (gelf_getsym(ctx->sym_tab, GELF_R_SYM(relo.r_info), &sym) != &sym) | |
1749 | return -EIO; | |
2486337a | 1750 | if (sym.st_shndx != ctx->sec_maps) { |
32a121cb | 1751 | fprintf(stderr, "ELF contains non-map related relo data in entry %u pointing to section %u! Compiler bug?!\n", |
2486337a DB |
1752 | relo_ent, sym.st_shndx); |
1753 | return -EIO; | |
1754 | } | |
32e93fb7 | 1755 | |
e4225669 | 1756 | rmap = sym.st_value / ctx->map_len; |
32e93fb7 DB |
1757 | if (rmap >= ARRAY_SIZE(ctx->map_fds)) |
1758 | return -EINVAL; | |
1759 | if (!ctx->map_fds[rmap]) | |
1760 | return -EINVAL; | |
1761 | ||
1762 | if (ctx->verbose) | |
32a121cb | 1763 | fprintf(stderr, "Map \'%s\' (%d) injected into prog section \'%s\' at offset %u!\n", |
32e93fb7 DB |
1764 | bpf_str_tab_name(ctx, &sym), ctx->map_fds[rmap], |
1765 | data_insn->sec_name, ioff); | |
11c39b5e | 1766 | |
32e93fb7 DB |
1767 | insns[ioff].src_reg = BPF_PSEUDO_MAP_FD; |
1768 | insns[ioff].imm = ctx->map_fds[rmap]; | |
1769 | } | |
1770 | ||
1771 | return 0; | |
1772 | } | |
1773 | ||
afc1a200 | 1774 | static int bpf_fetch_prog_relo(struct bpf_elf_ctx *ctx, const char *section, |
e4225669 | 1775 | bool *lderr, bool *sseen) |
32e93fb7 DB |
1776 | { |
1777 | struct bpf_elf_sec_data data_relo, data_insn; | |
1778 | struct bpf_elf_prog prog; | |
1779 | int ret, idx, i, fd = -1; | |
1780 | ||
1781 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { | |
1782 | ret = bpf_fill_section_data(ctx, i, &data_relo); | |
1783 | if (ret < 0 || data_relo.sec_hdr.sh_type != SHT_REL) | |
11c39b5e DB |
1784 | continue; |
1785 | ||
32e93fb7 | 1786 | idx = data_relo.sec_hdr.sh_info; |
e4225669 | 1787 | |
32e93fb7 | 1788 | ret = bpf_fill_section_data(ctx, idx, &data_insn); |
cce3d466 DB |
1789 | if (ret < 0 || |
1790 | !(data_insn.sec_hdr.sh_type == SHT_PROGBITS && | |
1791 | data_insn.sec_hdr.sh_flags & SHF_EXECINSTR && | |
1792 | !strcmp(data_insn.sec_name, section))) | |
11c39b5e | 1793 | continue; |
32e93fb7 | 1794 | |
e4225669 DB |
1795 | *sseen = true; |
1796 | ||
32e93fb7 DB |
1797 | ret = bpf_apply_relo_data(ctx, &data_relo, &data_insn); |
1798 | if (ret < 0) | |
e4225669 | 1799 | return ret; |
11c39b5e | 1800 | |
32e93fb7 DB |
1801 | memset(&prog, 0, sizeof(prog)); |
1802 | prog.type = ctx->type; | |
1803 | prog.insns = data_insn.sec_data->d_buf; | |
1804 | prog.size = data_insn.sec_data->d_size; | |
1805 | prog.license = ctx->license; | |
1806 | ||
f31645d1 | 1807 | fd = bpf_prog_attach(section, &prog, ctx); |
afc1a200 DB |
1808 | if (fd < 0) { |
1809 | *lderr = true; | |
e4225669 | 1810 | return fd; |
afc1a200 | 1811 | } |
11c39b5e | 1812 | |
32e93fb7 DB |
1813 | ctx->sec_done[i] = true; |
1814 | ctx->sec_done[idx] = true; | |
11c39b5e DB |
1815 | break; |
1816 | } | |
1817 | ||
32e93fb7 | 1818 | return fd; |
11c39b5e DB |
1819 | } |
1820 | ||
32e93fb7 | 1821 | static int bpf_fetch_prog_sec(struct bpf_elf_ctx *ctx, const char *section) |
473d7840 | 1822 | { |
e4225669 | 1823 | bool lderr = false, sseen = false; |
473d7840 DB |
1824 | int ret = -1; |
1825 | ||
afc1a200 | 1826 | if (bpf_has_map_data(ctx)) |
e4225669 | 1827 | ret = bpf_fetch_prog_relo(ctx, section, &lderr, &sseen); |
afc1a200 | 1828 | if (ret < 0 && !lderr) |
e4225669 DB |
1829 | ret = bpf_fetch_prog(ctx, section, &sseen); |
1830 | if (ret < 0 && !sseen) | |
1831 | fprintf(stderr, "Program section \'%s\' not found in ELF file!\n", | |
1832 | section); | |
473d7840 DB |
1833 | return ret; |
1834 | } | |
1835 | ||
910b543d DB |
1836 | static int bpf_find_map_by_id(struct bpf_elf_ctx *ctx, uint32_t id) |
1837 | { | |
1838 | int i; | |
1839 | ||
1840 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) | |
1841 | if (ctx->map_fds[i] && ctx->maps[i].id == id && | |
1842 | ctx->maps[i].type == BPF_MAP_TYPE_PROG_ARRAY) | |
1843 | return i; | |
1844 | return -1; | |
1845 | } | |
1846 | ||
32e93fb7 | 1847 | static int bpf_fill_prog_arrays(struct bpf_elf_ctx *ctx) |
473d7840 | 1848 | { |
32e93fb7 DB |
1849 | struct bpf_elf_sec_data data; |
1850 | uint32_t map_id, key_id; | |
910b543d | 1851 | int fd, i, ret, idx; |
473d7840 | 1852 | |
32e93fb7 DB |
1853 | for (i = 1; i < ctx->elf_hdr.e_shnum; i++) { |
1854 | if (ctx->sec_done[i]) | |
473d7840 DB |
1855 | continue; |
1856 | ||
32e93fb7 | 1857 | ret = bpf_fill_section_data(ctx, i, &data); |
473d7840 DB |
1858 | if (ret < 0) |
1859 | continue; | |
1860 | ||
910b543d DB |
1861 | ret = sscanf(data.sec_name, "%i/%i", &map_id, &key_id); |
1862 | if (ret != 2) | |
32e93fb7 | 1863 | continue; |
910b543d DB |
1864 | |
1865 | idx = bpf_find_map_by_id(ctx, map_id); | |
1866 | if (idx < 0) | |
473d7840 DB |
1867 | continue; |
1868 | ||
32e93fb7 DB |
1869 | fd = bpf_fetch_prog_sec(ctx, data.sec_name); |
1870 | if (fd < 0) | |
473d7840 DB |
1871 | return -EIO; |
1872 | ||
910b543d DB |
1873 | ret = bpf_map_update(ctx->map_fds[idx], &key_id, |
1874 | &fd, BPF_ANY); | |
afc1a200 DB |
1875 | if (ret < 0) { |
1876 | if (errno == E2BIG) | |
1877 | fprintf(stderr, "Tail call key %u for map %u out of bounds?\n", | |
1878 | key_id, map_id); | |
1879 | return -errno; | |
1880 | } | |
473d7840 | 1881 | |
32e93fb7 | 1882 | ctx->sec_done[i] = true; |
473d7840 DB |
1883 | } |
1884 | ||
1885 | return 0; | |
1886 | } | |
1887 | ||
32e93fb7 | 1888 | static void bpf_save_finfo(struct bpf_elf_ctx *ctx) |
11c39b5e | 1889 | { |
32e93fb7 DB |
1890 | struct stat st; |
1891 | int ret; | |
11c39b5e | 1892 | |
32e93fb7 | 1893 | memset(&ctx->stat, 0, sizeof(ctx->stat)); |
11c39b5e | 1894 | |
32e93fb7 DB |
1895 | ret = fstat(ctx->obj_fd, &st); |
1896 | if (ret < 0) { | |
1897 | fprintf(stderr, "Stat of elf file failed: %s\n", | |
1898 | strerror(errno)); | |
1899 | return; | |
1900 | } | |
11c39b5e | 1901 | |
32e93fb7 DB |
1902 | ctx->stat.st_dev = st.st_dev; |
1903 | ctx->stat.st_ino = st.st_ino; | |
1904 | } | |
1905 | ||
f6793eec DB |
1906 | static int bpf_read_pin_mapping(FILE *fp, uint32_t *id, char *path) |
1907 | { | |
1908 | char buff[PATH_MAX]; | |
1909 | ||
1910 | while (fgets(buff, sizeof(buff), fp)) { | |
1911 | char *ptr = buff; | |
1912 | ||
1913 | while (*ptr == ' ' || *ptr == '\t') | |
1914 | ptr++; | |
1915 | ||
1916 | if (*ptr == '#' || *ptr == '\n' || *ptr == 0) | |
1917 | continue; | |
1918 | ||
1919 | if (sscanf(ptr, "%i %s\n", id, path) != 2 && | |
1920 | sscanf(ptr, "%i %s #", id, path) != 2) { | |
1921 | strcpy(path, ptr); | |
1922 | return -1; | |
1923 | } | |
1924 | ||
1925 | return 1; | |
1926 | } | |
1927 | ||
1928 | return 0; | |
1929 | } | |
1930 | ||
1931 | static bool bpf_pinning_reserved(uint32_t pinning) | |
1932 | { | |
1933 | switch (pinning) { | |
1934 | case PIN_NONE: | |
1935 | case PIN_OBJECT_NS: | |
1936 | case PIN_GLOBAL_NS: | |
1937 | return true; | |
1938 | default: | |
1939 | return false; | |
1940 | } | |
1941 | } | |
1942 | ||
1943 | static void bpf_hash_init(struct bpf_elf_ctx *ctx, const char *db_file) | |
1944 | { | |
1945 | struct bpf_hash_entry *entry; | |
d17b136f | 1946 | char subpath[PATH_MAX] = {}; |
f6793eec DB |
1947 | uint32_t pinning; |
1948 | FILE *fp; | |
1949 | int ret; | |
1950 | ||
1951 | fp = fopen(db_file, "r"); | |
1952 | if (!fp) | |
1953 | return; | |
1954 | ||
f6793eec DB |
1955 | while ((ret = bpf_read_pin_mapping(fp, &pinning, subpath))) { |
1956 | if (ret == -1) { | |
1957 | fprintf(stderr, "Database %s is corrupted at: %s\n", | |
1958 | db_file, subpath); | |
1959 | fclose(fp); | |
1960 | return; | |
1961 | } | |
1962 | ||
1963 | if (bpf_pinning_reserved(pinning)) { | |
32a121cb SH |
1964 | fprintf(stderr, "Database %s, id %u is reserved - ignoring!\n", |
1965 | db_file, pinning); | |
f6793eec DB |
1966 | continue; |
1967 | } | |
1968 | ||
1969 | entry = malloc(sizeof(*entry)); | |
1970 | if (!entry) { | |
1971 | fprintf(stderr, "No memory left for db entry!\n"); | |
1972 | continue; | |
1973 | } | |
1974 | ||
1975 | entry->pinning = pinning; | |
1976 | entry->subpath = strdup(subpath); | |
1977 | if (!entry->subpath) { | |
1978 | fprintf(stderr, "No memory left for db entry!\n"); | |
1979 | free(entry); | |
1980 | continue; | |
1981 | } | |
1982 | ||
1983 | entry->next = ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)]; | |
1984 | ctx->ht[pinning & (ARRAY_SIZE(ctx->ht) - 1)] = entry; | |
1985 | } | |
1986 | ||
1987 | fclose(fp); | |
1988 | } | |
1989 | ||
1990 | static void bpf_hash_destroy(struct bpf_elf_ctx *ctx) | |
1991 | { | |
1992 | struct bpf_hash_entry *entry; | |
1993 | int i; | |
1994 | ||
1995 | for (i = 0; i < ARRAY_SIZE(ctx->ht); i++) { | |
1996 | while ((entry = ctx->ht[i]) != NULL) { | |
1997 | ctx->ht[i] = entry->next; | |
1998 | free((char *)entry->subpath); | |
1999 | free(entry); | |
2000 | } | |
2001 | } | |
2002 | } | |
2003 | ||
8187b012 DB |
2004 | static int bpf_elf_check_ehdr(const struct bpf_elf_ctx *ctx) |
2005 | { | |
2006 | if (ctx->elf_hdr.e_type != ET_REL || | |
e77fa41d DB |
2007 | (ctx->elf_hdr.e_machine != EM_NONE && |
2008 | ctx->elf_hdr.e_machine != EM_BPF) || | |
8187b012 DB |
2009 | ctx->elf_hdr.e_version != EV_CURRENT) { |
2010 | fprintf(stderr, "ELF format error, ELF file not for eBPF?\n"); | |
2011 | return -EINVAL; | |
2012 | } | |
2013 | ||
2014 | switch (ctx->elf_hdr.e_ident[EI_DATA]) { | |
2015 | default: | |
2016 | fprintf(stderr, "ELF format error, wrong endianness info?\n"); | |
2017 | return -EINVAL; | |
2018 | case ELFDATA2LSB: | |
2019 | if (htons(1) == 1) { | |
2020 | fprintf(stderr, | |
2021 | "We are big endian, eBPF object is little endian!\n"); | |
2022 | return -EIO; | |
2023 | } | |
2024 | break; | |
2025 | case ELFDATA2MSB: | |
2026 | if (htons(1) != 1) { | |
2027 | fprintf(stderr, | |
2028 | "We are little endian, eBPF object is big endian!\n"); | |
2029 | return -EIO; | |
2030 | } | |
2031 | break; | |
2032 | } | |
2033 | ||
2034 | return 0; | |
2035 | } | |
2036 | ||
32e93fb7 DB |
2037 | static int bpf_elf_ctx_init(struct bpf_elf_ctx *ctx, const char *pathname, |
2038 | enum bpf_prog_type type, bool verbose) | |
2039 | { | |
2040 | int ret = -EINVAL; | |
2041 | ||
2042 | if (elf_version(EV_CURRENT) == EV_NONE || | |
2043 | bpf_init_env(pathname)) | |
2044 | return ret; | |
2045 | ||
2046 | memset(ctx, 0, sizeof(*ctx)); | |
2047 | ctx->verbose = verbose; | |
2048 | ctx->type = type; | |
2049 | ||
2050 | ctx->obj_fd = open(pathname, O_RDONLY); | |
2051 | if (ctx->obj_fd < 0) | |
2052 | return ctx->obj_fd; | |
2053 | ||
2054 | ctx->elf_fd = elf_begin(ctx->obj_fd, ELF_C_READ, NULL); | |
2055 | if (!ctx->elf_fd) { | |
11c39b5e | 2056 | ret = -EINVAL; |
32e93fb7 | 2057 | goto out_fd; |
11c39b5e DB |
2058 | } |
2059 | ||
8187b012 DB |
2060 | if (elf_kind(ctx->elf_fd) != ELF_K_ELF) { |
2061 | ret = -EINVAL; | |
2062 | goto out_fd; | |
2063 | } | |
2064 | ||
32e93fb7 DB |
2065 | if (gelf_getehdr(ctx->elf_fd, &ctx->elf_hdr) != |
2066 | &ctx->elf_hdr) { | |
11c39b5e DB |
2067 | ret = -EIO; |
2068 | goto out_elf; | |
2069 | } | |
2070 | ||
8187b012 DB |
2071 | ret = bpf_elf_check_ehdr(ctx); |
2072 | if (ret < 0) | |
2073 | goto out_elf; | |
2074 | ||
32e93fb7 DB |
2075 | ctx->sec_done = calloc(ctx->elf_hdr.e_shnum, |
2076 | sizeof(*(ctx->sec_done))); | |
2077 | if (!ctx->sec_done) { | |
11c39b5e DB |
2078 | ret = -ENOMEM; |
2079 | goto out_elf; | |
2080 | } | |
2081 | ||
f31645d1 DB |
2082 | if (ctx->verbose && bpf_log_realloc(ctx)) { |
2083 | ret = -ENOMEM; | |
2084 | goto out_free; | |
2085 | } | |
2086 | ||
32e93fb7 | 2087 | bpf_save_finfo(ctx); |
f6793eec DB |
2088 | bpf_hash_init(ctx, CONFDIR "/bpf_pinning"); |
2089 | ||
32e93fb7 | 2090 | return 0; |
f31645d1 DB |
2091 | out_free: |
2092 | free(ctx->sec_done); | |
32e93fb7 DB |
2093 | out_elf: |
2094 | elf_end(ctx->elf_fd); | |
2095 | out_fd: | |
2096 | close(ctx->obj_fd); | |
2097 | return ret; | |
2098 | } | |
d937a74b | 2099 | |
32e93fb7 DB |
2100 | static int bpf_maps_count(struct bpf_elf_ctx *ctx) |
2101 | { | |
2102 | int i, count = 0; | |
11c39b5e | 2103 | |
32e93fb7 DB |
2104 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { |
2105 | if (!ctx->map_fds[i]) | |
2106 | break; | |
2107 | count++; | |
2108 | } | |
473d7840 | 2109 | |
32e93fb7 DB |
2110 | return count; |
2111 | } | |
6256f8c9 | 2112 | |
32e93fb7 DB |
2113 | static void bpf_maps_teardown(struct bpf_elf_ctx *ctx) |
2114 | { | |
2115 | int i; | |
2116 | ||
2117 | for (i = 0; i < ARRAY_SIZE(ctx->map_fds); i++) { | |
2118 | if (ctx->map_fds[i]) | |
2119 | close(ctx->map_fds[i]); | |
473d7840 | 2120 | } |
32e93fb7 DB |
2121 | } |
2122 | ||
2123 | static void bpf_elf_ctx_destroy(struct bpf_elf_ctx *ctx, bool failure) | |
2124 | { | |
2125 | if (failure) | |
2126 | bpf_maps_teardown(ctx); | |
473d7840 | 2127 | |
f6793eec | 2128 | bpf_hash_destroy(ctx); |
f31645d1 | 2129 | |
32e93fb7 | 2130 | free(ctx->sec_done); |
f31645d1 DB |
2131 | free(ctx->log); |
2132 | ||
32e93fb7 DB |
2133 | elf_end(ctx->elf_fd); |
2134 | close(ctx->obj_fd); | |
2135 | } | |
6256f8c9 | 2136 | |
32e93fb7 | 2137 | static struct bpf_elf_ctx __ctx; |
6256f8c9 | 2138 | |
32e93fb7 DB |
2139 | static int bpf_obj_open(const char *pathname, enum bpf_prog_type type, |
2140 | const char *section, bool verbose) | |
2141 | { | |
2142 | struct bpf_elf_ctx *ctx = &__ctx; | |
2143 | int fd = 0, ret; | |
6256f8c9 | 2144 | |
32e93fb7 DB |
2145 | ret = bpf_elf_ctx_init(ctx, pathname, type, verbose); |
2146 | if (ret < 0) { | |
2147 | fprintf(stderr, "Cannot initialize ELF context!\n"); | |
2148 | return ret; | |
2149 | } | |
6256f8c9 | 2150 | |
32e93fb7 DB |
2151 | ret = bpf_fetch_ancillary(ctx); |
2152 | if (ret < 0) { | |
2153 | fprintf(stderr, "Error fetching ELF ancillary data!\n"); | |
2154 | goto out; | |
2155 | } | |
2156 | ||
2157 | fd = bpf_fetch_prog_sec(ctx, section); | |
2158 | if (fd < 0) { | |
2159 | fprintf(stderr, "Error fetching program/map!\n"); | |
2160 | ret = fd; | |
2161 | goto out; | |
2162 | } | |
2163 | ||
2164 | ret = bpf_fill_prog_arrays(ctx); | |
2165 | if (ret < 0) | |
2166 | fprintf(stderr, "Error filling program arrays!\n"); | |
11c39b5e | 2167 | out: |
32e93fb7 DB |
2168 | bpf_elf_ctx_destroy(ctx, ret < 0); |
2169 | if (ret < 0) { | |
2170 | if (fd) | |
2171 | close(fd); | |
2172 | return ret; | |
2173 | } | |
2174 | ||
2175 | return fd; | |
6256f8c9 | 2176 | } |
11c39b5e | 2177 | |
6256f8c9 | 2178 | static int |
4bd62446 DB |
2179 | bpf_map_set_send(int fd, struct sockaddr_un *addr, unsigned int addr_len, |
2180 | const struct bpf_map_data *aux, unsigned int entries) | |
6256f8c9 | 2181 | { |
d17b136f PS |
2182 | struct bpf_map_set_msg msg = { |
2183 | .aux.uds_ver = BPF_SCM_AUX_VER, | |
2184 | .aux.num_ent = entries, | |
2185 | }; | |
6256f8c9 DB |
2186 | int *cmsg_buf, min_fd; |
2187 | char *amsg_buf; | |
2188 | int i; | |
2189 | ||
6256f8c9 DB |
2190 | strncpy(msg.aux.obj_name, aux->obj, sizeof(msg.aux.obj_name)); |
2191 | memcpy(&msg.aux.obj_st, aux->st, sizeof(msg.aux.obj_st)); | |
2192 | ||
2193 | cmsg_buf = bpf_map_set_init(&msg, addr, addr_len); | |
2194 | amsg_buf = (char *)msg.aux.ent; | |
2195 | ||
4bd62446 | 2196 | for (i = 0; i < entries; i += min_fd) { |
6256f8c9 DB |
2197 | int ret; |
2198 | ||
4bd62446 | 2199 | min_fd = min(BPF_SCM_MAX_FDS * 1U, entries - i); |
6256f8c9 DB |
2200 | bpf_map_set_init_single(&msg, min_fd); |
2201 | ||
2202 | memcpy(cmsg_buf, &aux->fds[i], sizeof(aux->fds[0]) * min_fd); | |
2203 | memcpy(amsg_buf, &aux->ent[i], sizeof(aux->ent[0]) * min_fd); | |
2204 | ||
2205 | ret = sendmsg(fd, &msg.hdr, 0); | |
2206 | if (ret <= 0) | |
2207 | return ret ? : -1; | |
2208 | } | |
2209 | ||
2210 | return 0; | |
11c39b5e DB |
2211 | } |
2212 | ||
4bd62446 DB |
2213 | static int |
2214 | bpf_map_set_recv(int fd, int *fds, struct bpf_map_aux *aux, | |
2215 | unsigned int entries) | |
2216 | { | |
2217 | struct bpf_map_set_msg msg; | |
2218 | int *cmsg_buf, min_fd; | |
2219 | char *amsg_buf, *mmsg_buf; | |
2220 | unsigned int needed = 1; | |
2221 | int i; | |
2222 | ||
2223 | cmsg_buf = bpf_map_set_init(&msg, NULL, 0); | |
2224 | amsg_buf = (char *)msg.aux.ent; | |
2225 | mmsg_buf = (char *)&msg.aux; | |
2226 | ||
2227 | for (i = 0; i < min(entries, needed); i += min_fd) { | |
2228 | struct cmsghdr *cmsg; | |
2229 | int ret; | |
2230 | ||
2231 | min_fd = min(entries, entries - i); | |
2232 | bpf_map_set_init_single(&msg, min_fd); | |
2233 | ||
2234 | ret = recvmsg(fd, &msg.hdr, 0); | |
2235 | if (ret <= 0) | |
2236 | return ret ? : -1; | |
2237 | ||
2238 | cmsg = CMSG_FIRSTHDR(&msg.hdr); | |
2239 | if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) | |
2240 | return -EINVAL; | |
2241 | if (msg.hdr.msg_flags & MSG_CTRUNC) | |
2242 | return -EIO; | |
2243 | if (msg.aux.uds_ver != BPF_SCM_AUX_VER) | |
2244 | return -ENOSYS; | |
2245 | ||
2246 | min_fd = (cmsg->cmsg_len - sizeof(*cmsg)) / sizeof(fd); | |
2247 | if (min_fd > entries || min_fd <= 0) | |
2248 | return -EINVAL; | |
2249 | ||
2250 | memcpy(&fds[i], cmsg_buf, sizeof(fds[0]) * min_fd); | |
2251 | memcpy(&aux->ent[i], amsg_buf, sizeof(aux->ent[0]) * min_fd); | |
2252 | memcpy(aux, mmsg_buf, offsetof(struct bpf_map_aux, ent)); | |
2253 | ||
2254 | needed = aux->num_ent; | |
2255 | } | |
2256 | ||
2257 | return 0; | |
2258 | } | |
2259 | ||
2260 | int bpf_send_map_fds(const char *path, const char *obj) | |
6256f8c9 | 2261 | { |
32e93fb7 | 2262 | struct bpf_elf_ctx *ctx = &__ctx; |
d17b136f PS |
2263 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
2264 | struct bpf_map_data bpf_aux = { | |
2265 | .fds = ctx->map_fds, | |
2266 | .ent = ctx->maps, | |
2267 | .st = &ctx->stat, | |
2268 | .obj = obj, | |
2269 | }; | |
6256f8c9 DB |
2270 | int fd, ret; |
2271 | ||
2272 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2273 | if (fd < 0) { | |
2274 | fprintf(stderr, "Cannot open socket: %s\n", | |
2275 | strerror(errno)); | |
2276 | return -1; | |
2277 | } | |
2278 | ||
6256f8c9 DB |
2279 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2280 | ||
2281 | ret = connect(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2282 | if (ret < 0) { | |
2283 | fprintf(stderr, "Cannot connect to %s: %s\n", | |
2284 | path, strerror(errno)); | |
2285 | return -1; | |
2286 | } | |
2287 | ||
4bd62446 | 2288 | ret = bpf_map_set_send(fd, &addr, sizeof(addr), &bpf_aux, |
32e93fb7 | 2289 | bpf_maps_count(ctx)); |
6256f8c9 | 2290 | if (ret < 0) |
4bd62446 DB |
2291 | fprintf(stderr, "Cannot send fds to %s: %s\n", |
2292 | path, strerror(errno)); | |
2293 | ||
32e93fb7 | 2294 | bpf_maps_teardown(ctx); |
4bd62446 DB |
2295 | close(fd); |
2296 | return ret; | |
2297 | } | |
2298 | ||
2299 | int bpf_recv_map_fds(const char *path, int *fds, struct bpf_map_aux *aux, | |
2300 | unsigned int entries) | |
2301 | { | |
d17b136f | 2302 | struct sockaddr_un addr = { .sun_family = AF_UNIX }; |
4bd62446 DB |
2303 | int fd, ret; |
2304 | ||
2305 | fd = socket(AF_UNIX, SOCK_DGRAM, 0); | |
2306 | if (fd < 0) { | |
2307 | fprintf(stderr, "Cannot open socket: %s\n", | |
2308 | strerror(errno)); | |
2309 | return -1; | |
2310 | } | |
2311 | ||
4bd62446 DB |
2312 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); |
2313 | ||
2314 | ret = bind(fd, (struct sockaddr *)&addr, sizeof(addr)); | |
2315 | if (ret < 0) { | |
2316 | fprintf(stderr, "Cannot bind to socket: %s\n", | |
2317 | strerror(errno)); | |
2318 | return -1; | |
2319 | } | |
2320 | ||
2321 | ret = bpf_map_set_recv(fd, fds, aux, entries); | |
2322 | if (ret < 0) | |
2323 | fprintf(stderr, "Cannot recv fds from %s: %s\n", | |
6256f8c9 DB |
2324 | path, strerror(errno)); |
2325 | ||
4bd62446 | 2326 | unlink(addr.sun_path); |
6256f8c9 DB |
2327 | close(fd); |
2328 | return ret; | |
2329 | } | |
11c39b5e | 2330 | #endif /* HAVE_ELF */ |