1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) 2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */ 3 4 #ifndef _GNU_SOURCE 5 #define _GNU_SOURCE 6 #endif 7 #include <assert.h> 8 #include <ctype.h> 9 #include <errno.h> 10 #include <fcntl.h> 11 #include <ftw.h> 12 #include <libgen.h> 13 #include <mntent.h> 14 #include <stdbool.h> 15 #include <stdio.h> 16 #include <stdlib.h> 17 #include <string.h> 18 #include <unistd.h> 19 #include <net/if.h> 20 #include <sys/mount.h> 21 #include <sys/resource.h> 22 #include <sys/stat.h> 23 #include <sys/vfs.h> 24 #include <sys/utsname.h> 25 26 #include <linux/filter.h> 27 #include <linux/limits.h> 28 #include <linux/magic.h> 29 #include <linux/unistd.h> 30 31 #include <bpf/bpf.h> 32 #include <bpf/hashmap.h> 33 #include <bpf/libbpf.h> /* libbpf_num_possible_cpus */ 34 #include <bpf/libbpf_internal.h> 35 #include <bpf/btf.h> 36 #include <zlib.h> 37 38 #include "main.h" 39 40 #ifndef BPF_FS_MAGIC 41 #define BPF_FS_MAGIC 0xcafe4a11 42 #endif 43 44 void p_err(const char *fmt, ...) 45 { 46 va_list ap; 47 48 va_start(ap, fmt); 49 if (json_output) { 50 jsonw_start_object(json_wtr); 51 jsonw_name(json_wtr, "error"); 52 jsonw_vprintf_enquote(json_wtr, fmt, ap); 53 jsonw_end_object(json_wtr); 54 } else { 55 fprintf(stderr, "Error: "); 56 vfprintf(stderr, fmt, ap); 57 fprintf(stderr, "\n"); 58 } 59 va_end(ap); 60 } 61 62 void p_info(const char *fmt, ...) 63 { 64 va_list ap; 65 66 if (json_output) 67 return; 68 69 va_start(ap, fmt); 70 vfprintf(stderr, fmt, ap); 71 fprintf(stderr, "\n"); 72 va_end(ap); 73 } 74 75 /* 76 * Like strerror(), but also knows ENOTSUPP (524), which the kernel returns 77 * to user space from some BPF commands although it is not a userspace errno. 78 */ 79 const char *bpf_strerror(int err) 80 { 81 if (err == 524 /* ENOTSUPP */) 82 return "Operation not supported"; 83 return strerror(err); 84 } 85 86 static bool is_bpffs(const char *path) 87 { 88 struct statfs st_fs; 89 90 if (statfs(path, &st_fs) < 0) 91 return false; 92 93 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC; 94 } 95 96 /* Probe whether kernel switched from memlock-based (RLIMIT_MEMLOCK) to 97 * memcg-based memory accounting for BPF maps and programs. This was done in 98 * commit 97306be45fbe ("Merge branch 'switch to memcg-based memory 99 * accounting'"), in Linux 5.11. 100 * 101 * Libbpf also offers to probe for memcg-based accounting vs rlimit, but does 102 * so by checking for the availability of a given BPF helper and this has 103 * failed on some kernels with backports in the past, see commit 6b4384ff1088 104 * ("Revert "bpftool: Use libbpf 1.0 API mode instead of RLIMIT_MEMLOCK""). 105 * Instead, we can probe by lowering the process-based rlimit to 0, trying to 106 * load a BPF object, and resetting the rlimit. If the load succeeds then 107 * memcg-based accounting is supported. 108 * 109 * This would be too dangerous to do in the library, because multithreaded 110 * applications might attempt to load items while the rlimit is at 0. Given 111 * that bpftool is single-threaded, this is fine to do here. 112 */ 113 static bool known_to_need_rlimit(void) 114 { 115 struct rlimit rlim_init, rlim_cur_zero = {}; 116 struct bpf_insn insns[] = { 117 BPF_MOV64_IMM(BPF_REG_0, 0), 118 BPF_EXIT_INSN(), 119 }; 120 size_t insn_cnt = ARRAY_SIZE(insns); 121 union bpf_attr attr; 122 int prog_fd, err; 123 124 memset(&attr, 0, sizeof(attr)); 125 attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER; 126 attr.insns = ptr_to_u64(insns); 127 attr.insn_cnt = insn_cnt; 128 attr.license = ptr_to_u64("GPL"); 129 130 if (getrlimit(RLIMIT_MEMLOCK, &rlim_init)) 131 return false; 132 133 /* Drop the soft limit to zero. We maintain the hard limit to its 134 * current value, because lowering it would be a permanent operation 135 * for unprivileged users. 136 */ 137 rlim_cur_zero.rlim_max = rlim_init.rlim_max; 138 if (setrlimit(RLIMIT_MEMLOCK, &rlim_cur_zero)) 139 return false; 140 141 /* Do not use bpf_prog_load() from libbpf here, because it calls 142 * bump_rlimit_memlock(), interfering with the current probe. 143 */ 144 prog_fd = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr)); 145 err = errno; 146 147 /* reset soft rlimit to its initial value */ 148 setrlimit(RLIMIT_MEMLOCK, &rlim_init); 149 150 if (prog_fd < 0) 151 return err == EPERM; 152 153 close(prog_fd); 154 return false; 155 } 156 157 void set_max_rlimit(void) 158 { 159 struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY }; 160 161 if (known_to_need_rlimit()) 162 setrlimit(RLIMIT_MEMLOCK, &rinf); 163 } 164 165 static int 166 mnt_fs(const char *target, const char *type, char *buff, size_t bufflen) 167 { 168 bool bind_done = false; 169 170 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { 171 if (errno != EINVAL || bind_done) { 172 snprintf(buff, bufflen, 173 "mount --make-private %s failed: %s", 174 target, strerror(errno)); 175 return -1; 176 } 177 178 if (mount(target, target, "none", MS_BIND, NULL)) { 179 snprintf(buff, bufflen, 180 "mount --bind %s %s failed: %s", 181 target, target, strerror(errno)); 182 return -1; 183 } 184 185 bind_done = true; 186 } 187 188 if (mount(type, target, type, 0, "mode=0700")) { 189 snprintf(buff, bufflen, "mount -t %s %s %s failed: %s", 190 type, type, target, strerror(errno)); 191 return -1; 192 } 193 194 return 0; 195 } 196 197 int mount_tracefs(const char *target) 198 { 199 char err_str[ERR_MAX_LEN]; 200 int err; 201 202 err = mnt_fs(target, "tracefs", err_str, ERR_MAX_LEN); 203 if (err) { 204 err_str[ERR_MAX_LEN - 1] = '\0'; 205 p_err("can't mount tracefs: %s", err_str); 206 } 207 208 return err; 209 } 210 211 int open_obj_pinned(const char *path, bool quiet, 212 const struct bpf_obj_get_opts *opts) 213 { 214 char *pname; 215 int fd = -1; 216 217 pname = strdup(path); 218 if (!pname) { 219 if (!quiet) 220 p_err("mem alloc failed"); 221 goto out_ret; 222 } 223 224 fd = bpf_obj_get_opts(pname, opts); 225 if (fd < 0) { 226 if (!quiet) 227 p_err("bpf obj get (%s): %s", pname, 228 errno == EACCES && !is_bpffs(dirname(pname)) ? 229 "directory not in bpf file system (bpffs)" : 230 strerror(errno)); 231 goto out_free; 232 } 233 234 out_free: 235 free(pname); 236 out_ret: 237 return fd; 238 } 239 240 int open_obj_pinned_any(const char *path, enum bpf_obj_type exp_type, 241 const struct bpf_obj_get_opts *opts) 242 { 243 enum bpf_obj_type type; 244 int fd; 245 246 fd = open_obj_pinned(path, false, opts); 247 if (fd < 0) 248 return -1; 249 250 type = get_fd_type(fd); 251 if (type < 0) { 252 close(fd); 253 return type; 254 } 255 if (type != exp_type) { 256 p_err("incorrect object type: %s", get_fd_type_name(type)); 257 close(fd); 258 return -1; 259 } 260 261 return fd; 262 } 263 264 int create_and_mount_bpffs_dir(const char *dir_name) 265 { 266 char err_str[ERR_MAX_LEN]; 267 bool dir_exists; 268 int err = 0; 269 270 if (is_bpffs(dir_name)) 271 return err; 272 273 dir_exists = access(dir_name, F_OK) == 0; 274 275 if (!dir_exists) { 276 char *temp_name; 277 char *parent_name; 278 279 temp_name = strdup(dir_name); 280 if (!temp_name) { 281 p_err("mem alloc failed"); 282 return -1; 283 } 284 285 parent_name = dirname(temp_name); 286 287 if (is_bpffs(parent_name)) { 288 /* nothing to do if already mounted */ 289 free(temp_name); 290 return err; 291 } 292 293 if (access(parent_name, F_OK) == -1) { 294 p_err("can't create dir '%s' to pin BPF object: parent dir '%s' doesn't exist", 295 dir_name, parent_name); 296 free(temp_name); 297 return -1; 298 } 299 300 free(temp_name); 301 } 302 303 if (block_mount) { 304 p_err("no BPF file system found, not mounting it due to --nomount option"); 305 return -1; 306 } 307 308 if (!dir_exists) { 309 err = mkdir(dir_name, S_IRWXU); 310 if (err) { 311 p_err("failed to create dir '%s': %s", dir_name, strerror(errno)); 312 return err; 313 } 314 } 315 316 err = mnt_fs(dir_name, "bpf", err_str, ERR_MAX_LEN); 317 if (err) { 318 err_str[ERR_MAX_LEN - 1] = '\0'; 319 p_err("can't mount BPF file system on given dir '%s': %s", 320 dir_name, err_str); 321 322 if (!dir_exists) 323 rmdir(dir_name); 324 } 325 326 return err; 327 } 328 329 int mount_bpffs_for_file(const char *file_name) 330 { 331 char err_str[ERR_MAX_LEN]; 332 char *temp_name; 333 char *dir; 334 int err = 0; 335 336 if (access(file_name, F_OK) != -1) { 337 p_err("can't pin BPF object: path '%s' already exists", file_name); 338 return -1; 339 } 340 341 temp_name = strdup(file_name); 342 if (!temp_name) { 343 p_err("mem alloc failed"); 344 return -1; 345 } 346 347 dir = dirname(temp_name); 348 349 if (is_bpffs(dir)) 350 /* nothing to do if already mounted */ 351 goto out_free; 352 353 if (access(dir, F_OK) == -1) { 354 p_err("can't pin BPF object: dir '%s' doesn't exist", dir); 355 err = -1; 356 goto out_free; 357 } 358 359 if (block_mount) { 360 p_err("no BPF file system found, not mounting it due to --nomount option"); 361 err = -1; 362 goto out_free; 363 } 364 365 err = mnt_fs(dir, "bpf", err_str, ERR_MAX_LEN); 366 if (err) { 367 err_str[ERR_MAX_LEN - 1] = '\0'; 368 p_err("can't mount BPF file system to pin the object '%s': %s", 369 file_name, err_str); 370 } 371 372 out_free: 373 free(temp_name); 374 return err; 375 } 376 377 int do_pin_fd(int fd, const char *name) 378 { 379 int err; 380 381 err = mount_bpffs_for_file(name); 382 if (err) 383 return err; 384 385 err = bpf_obj_pin(fd, name); 386 if (err) 387 p_err("can't pin the object (%s): %s", name, strerror(errno)); 388 389 return err; 390 } 391 392 int do_pin_any(int argc, char **argv, int (*get_fd)(int *, char ***)) 393 { 394 int err; 395 int fd; 396 397 if (!REQ_ARGS(3)) 398 return -EINVAL; 399 400 fd = get_fd(&argc, &argv); 401 if (fd < 0) 402 return fd; 403 404 err = do_pin_fd(fd, *argv); 405 406 close(fd); 407 return err; 408 } 409 410 const char *get_fd_type_name(enum bpf_obj_type type) 411 { 412 static const char * const names[] = { 413 [BPF_OBJ_UNKNOWN] = "unknown", 414 [BPF_OBJ_PROG] = "prog", 415 [BPF_OBJ_MAP] = "map", 416 [BPF_OBJ_LINK] = "link", 417 }; 418 419 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type]) 420 return names[BPF_OBJ_UNKNOWN]; 421 422 return names[type]; 423 } 424 425 void get_prog_full_name(const struct bpf_prog_info *prog_info, int prog_fd, 426 char *name_buff, size_t buff_len) 427 { 428 const char *prog_name = prog_info->name; 429 const struct btf_type *func_type; 430 struct bpf_func_info finfo = {}; 431 struct bpf_prog_info info = {}; 432 __u32 info_len = sizeof(info); 433 struct btf *prog_btf = NULL; 434 435 if (buff_len <= BPF_OBJ_NAME_LEN || 436 strlen(prog_info->name) < BPF_OBJ_NAME_LEN - 1) 437 goto copy_name; 438 439 if (!prog_info->btf_id || prog_info->nr_func_info == 0) 440 goto copy_name; 441 442 info.nr_func_info = 1; 443 info.func_info_rec_size = prog_info->func_info_rec_size; 444 if (info.func_info_rec_size > sizeof(finfo)) 445 info.func_info_rec_size = sizeof(finfo); 446 info.func_info = ptr_to_u64(&finfo); 447 448 if (bpf_prog_get_info_by_fd(prog_fd, &info, &info_len)) 449 goto copy_name; 450 451 prog_btf = btf__load_from_kernel_by_id(info.btf_id); 452 if (!prog_btf) 453 goto copy_name; 454 455 func_type = btf__type_by_id(prog_btf, finfo.type_id); 456 if (!func_type || !btf_is_func(func_type)) 457 goto copy_name; 458 459 prog_name = btf__name_by_offset(prog_btf, func_type->name_off); 460 461 copy_name: 462 snprintf(name_buff, buff_len, "%s", prog_name); 463 464 if (prog_btf) 465 btf__free(prog_btf); 466 } 467 468 int get_fd_type(int fd) 469 { 470 char path[PATH_MAX]; 471 char buf[512]; 472 ssize_t n; 473 474 snprintf(path, sizeof(path), "/proc/self/fd/%d", fd); 475 476 n = readlink(path, buf, sizeof(buf)); 477 if (n < 0) { 478 p_err("can't read link type: %s", strerror(errno)); 479 return -1; 480 } 481 if (n == sizeof(buf)) { 482 p_err("can't read link type: path too long!"); 483 return -1; 484 } 485 buf[n] = '\0'; 486 487 if (strstr(buf, "bpf-map")) 488 return BPF_OBJ_MAP; 489 else if (strstr(buf, "bpf-prog")) 490 return BPF_OBJ_PROG; 491 else if (strstr(buf, "bpf-link")) 492 return BPF_OBJ_LINK; 493 494 return BPF_OBJ_UNKNOWN; 495 } 496 497 char *get_fdinfo(int fd, const char *key) 498 { 499 char path[PATH_MAX]; 500 char *line = NULL; 501 size_t line_n = 0; 502 ssize_t n; 503 FILE *fdi; 504 505 snprintf(path, sizeof(path), "/proc/self/fdinfo/%d", fd); 506 507 fdi = fopen(path, "r"); 508 if (!fdi) 509 return NULL; 510 511 while ((n = getline(&line, &line_n, fdi)) > 0) { 512 char *value; 513 int len; 514 515 if (!strstr(line, key)) 516 continue; 517 518 fclose(fdi); 519 520 value = strchr(line, '\t'); 521 if (!value || !value[1]) { 522 free(line); 523 return NULL; 524 } 525 value++; 526 527 len = strlen(value); 528 memmove(line, value, len); 529 line[len - 1] = '\0'; 530 531 return line; 532 } 533 534 free(line); 535 fclose(fdi); 536 return NULL; 537 } 538 539 void print_data_json(uint8_t *data, size_t len) 540 { 541 unsigned int i; 542 543 jsonw_start_array(json_wtr); 544 for (i = 0; i < len; i++) 545 jsonw_printf(json_wtr, "%d", data[i]); 546 jsonw_end_array(json_wtr); 547 } 548 549 void print_hex_data_json(uint8_t *data, size_t len) 550 { 551 unsigned int i; 552 553 jsonw_start_array(json_wtr); 554 for (i = 0; i < len; i++) 555 jsonw_printf(json_wtr, "\"0x%02hhx\"", data[i]); 556 jsonw_end_array(json_wtr); 557 } 558 559 /* extra params for nftw cb */ 560 static struct hashmap *build_fn_table; 561 static enum bpf_obj_type build_fn_type; 562 563 static int do_build_table_cb(const char *fpath, const struct stat *sb, 564 int typeflag, struct FTW *ftwbuf) 565 { 566 struct bpf_prog_info pinned_info; 567 __u32 len = sizeof(pinned_info); 568 enum bpf_obj_type objtype; 569 int fd, err = 0; 570 char *path; 571 572 if (typeflag != FTW_F) 573 goto out_ret; 574 575 fd = open_obj_pinned(fpath, true, NULL); 576 if (fd < 0) 577 goto out_ret; 578 579 objtype = get_fd_type(fd); 580 if (objtype != build_fn_type) 581 goto out_close; 582 583 memset(&pinned_info, 0, sizeof(pinned_info)); 584 if (bpf_prog_get_info_by_fd(fd, &pinned_info, &len)) 585 goto out_close; 586 587 path = strdup(fpath); 588 if (!path) { 589 err = -1; 590 goto out_close; 591 } 592 593 err = hashmap__append(build_fn_table, pinned_info.id, path); 594 if (err) { 595 p_err("failed to append entry to hashmap for ID %u, path '%s': %s", 596 pinned_info.id, path, strerror(errno)); 597 free(path); 598 goto out_close; 599 } 600 601 out_close: 602 close(fd); 603 out_ret: 604 return err; 605 } 606 607 int build_pinned_obj_table(struct hashmap *tab, 608 enum bpf_obj_type type) 609 { 610 struct mntent *mntent = NULL; 611 FILE *mntfile = NULL; 612 int flags = FTW_PHYS; 613 int nopenfd = 16; 614 int err = 0; 615 616 mntfile = setmntent("/proc/mounts", "r"); 617 if (!mntfile) 618 return -1; 619 620 build_fn_table = tab; 621 build_fn_type = type; 622 623 while ((mntent = getmntent(mntfile))) { 624 char *path = mntent->mnt_dir; 625 626 if (strncmp(mntent->mnt_type, "bpf", 3) != 0) 627 continue; 628 err = nftw(path, do_build_table_cb, nopenfd, flags); 629 if (err) 630 break; 631 } 632 fclose(mntfile); 633 return err; 634 } 635 636 void delete_pinned_obj_table(struct hashmap *map) 637 { 638 struct hashmap_entry *entry; 639 size_t bkt; 640 641 if (!map) 642 return; 643 644 hashmap__for_each_entry(map, entry, bkt) 645 free(entry->pvalue); 646 647 hashmap__free(map); 648 } 649 650 unsigned int get_page_size(void) 651 { 652 static int result; 653 654 if (!result) 655 result = getpagesize(); 656 return result; 657 } 658 659 unsigned int get_possible_cpus(void) 660 { 661 int cpus = libbpf_num_possible_cpus(); 662 663 if (cpus < 0) { 664 p_err("Can't get # of possible cpus: %s", strerror(-cpus)); 665 exit(-1); 666 } 667 return cpus; 668 } 669 670 int get_possible_cpu_ids(int **cpu_ids) 671 { 672 const char *possible_cpus_file = "/sys/devices/system/cpu/possible"; 673 bool *mask = NULL; 674 int mask_sz, nr_cpus = 0; 675 int *ids = NULL; 676 int i, res; 677 678 *cpu_ids = NULL; 679 680 res = parse_cpu_mask_file(possible_cpus_file, &mask, &mask_sz); 681 if (res) { 682 p_err("failed to parse possible CPU mask: %s", strerror(-res)); 683 return res; 684 } 685 686 for (i = 0; i < mask_sz; i++) 687 nr_cpus += mask[i]; 688 689 ids = calloc(nr_cpus, sizeof(*ids)); 690 if (!ids) { 691 p_err("mem alloc failed"); 692 res = -ENOMEM; 693 goto out; 694 } 695 696 for (i = 0, nr_cpus = 0; i < mask_sz; i++) { 697 if (mask[i]) 698 ids[nr_cpus++] = i; 699 } 700 *cpu_ids = ids; 701 ids = NULL; 702 res = nr_cpus; 703 out: 704 free(ids); 705 free(mask); 706 return res; 707 } 708 709 static char * 710 ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf) 711 { 712 struct stat st; 713 int err; 714 715 err = stat("/proc/self/ns/net", &st); 716 if (err) { 717 p_err("Can't stat /proc/self: %s", strerror(errno)); 718 return NULL; 719 } 720 721 if (st.st_dev != ns_dev || st.st_ino != ns_ino) 722 return NULL; 723 724 return if_indextoname(ifindex, buf); 725 } 726 727 static int read_sysfs_hex_int(char *path) 728 { 729 char vendor_id_buf[8]; 730 int len; 731 int fd; 732 733 fd = open(path, O_RDONLY); 734 if (fd < 0) { 735 p_err("Can't open %s: %s", path, strerror(errno)); 736 return -1; 737 } 738 739 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf)); 740 close(fd); 741 if (len < 0) { 742 p_err("Can't read %s: %s", path, strerror(errno)); 743 return -1; 744 } 745 if (len >= (int)sizeof(vendor_id_buf)) { 746 p_err("Value in %s too long", path); 747 return -1; 748 } 749 750 vendor_id_buf[len] = 0; 751 752 return strtol(vendor_id_buf, NULL, 0); 753 } 754 755 static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name) 756 { 757 char full_path[64]; 758 759 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s", 760 devname, entry_name); 761 762 return read_sysfs_hex_int(full_path); 763 } 764 765 const char * 766 ifindex_to_arch(__u32 ifindex, __u64 ns_dev, __u64 ns_ino, const char **opt) 767 { 768 __maybe_unused int device_id; 769 char devname[IF_NAMESIZE]; 770 int vendor_id; 771 772 if (!ifindex_to_name_ns(ifindex, ns_dev, ns_ino, devname)) { 773 p_err("Can't get net device name for ifindex %u: %s", ifindex, 774 strerror(errno)); 775 return NULL; 776 } 777 778 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor"); 779 if (vendor_id < 0) { 780 p_err("Can't get device vendor id for %s", devname); 781 return NULL; 782 } 783 784 switch (vendor_id) { 785 #ifdef HAVE_LIBBFD_SUPPORT 786 case 0x19ee: 787 device_id = read_sysfs_netdev_hex_int(devname, "device"); 788 if (device_id != 0x4000 && 789 device_id != 0x6000 && 790 device_id != 0x6003) 791 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch"); 792 *opt = "ctx4"; 793 return "NFP-6xxx"; 794 #endif /* HAVE_LIBBFD_SUPPORT */ 795 /* No NFP support in LLVM, we have no valid triple to return. */ 796 default: 797 p_err("Can't get arch name for device vendor id 0x%04x", 798 (unsigned int)vendor_id); 799 return NULL; 800 } 801 } 802 803 void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode) 804 { 805 char name[IF_NAMESIZE]; 806 807 if (!ifindex) 808 return; 809 810 printf(" offloaded_to "); 811 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name)) 812 printf("%s", name); 813 else 814 printf("ifindex %u ns_dev %llu ns_ino %llu", 815 ifindex, ns_dev, ns_inode); 816 } 817 818 void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode) 819 { 820 char name[IF_NAMESIZE]; 821 822 if (!ifindex) 823 return; 824 825 jsonw_name(json_wtr, "dev"); 826 jsonw_start_object(json_wtr); 827 jsonw_uint_field(json_wtr, "ifindex", ifindex); 828 jsonw_uint_field(json_wtr, "ns_dev", ns_dev); 829 jsonw_uint_field(json_wtr, "ns_inode", ns_inode); 830 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name)) 831 jsonw_string_field(json_wtr, "ifname", name); 832 jsonw_end_object(json_wtr); 833 } 834 835 int parse_u32_arg(int *argc, char ***argv, __u32 *val, const char *what) 836 { 837 char *endptr; 838 839 NEXT_ARGP(); 840 841 if (*val) { 842 p_err("%s already specified", what); 843 return -1; 844 } 845 846 *val = strtoul(**argv, &endptr, 0); 847 if (*endptr) { 848 p_err("can't parse %s as %s", **argv, what); 849 return -1; 850 } 851 NEXT_ARGP(); 852 853 return 0; 854 } 855 856 int __printf(2, 0) 857 print_all_levels(__maybe_unused enum libbpf_print_level level, 858 const char *format, va_list args) 859 { 860 return vfprintf(stderr, format, args); 861 } 862 863 static int prog_fd_by_nametag(void *nametag, int **fds, bool tag) 864 { 865 char prog_name[MAX_PROG_FULL_NAME]; 866 unsigned int id = 0; 867 int fd, nb_fds = 0; 868 void *tmp; 869 int err; 870 871 while (true) { 872 struct bpf_prog_info info = {}; 873 __u32 len = sizeof(info); 874 875 err = bpf_prog_get_next_id(id, &id); 876 if (err) { 877 if (errno != ENOENT) { 878 p_err("%s", strerror(errno)); 879 goto err_close_fds; 880 } 881 return nb_fds; 882 } 883 884 fd = bpf_prog_get_fd_by_id(id); 885 if (fd < 0) { 886 if (errno == ENOENT) 887 continue; 888 p_err("can't get prog by id (%u): %s", 889 id, strerror(errno)); 890 goto err_close_fds; 891 } 892 893 err = bpf_prog_get_info_by_fd(fd, &info, &len); 894 if (err) { 895 p_err("can't get prog info (%u): %s", 896 id, strerror(errno)); 897 goto err_close_fd; 898 } 899 900 if (tag && memcmp(nametag, info.tag, BPF_TAG_SIZE)) { 901 close(fd); 902 continue; 903 } 904 905 if (!tag) { 906 get_prog_full_name(&info, fd, prog_name, 907 sizeof(prog_name)); 908 if (strncmp(nametag, prog_name, sizeof(prog_name))) { 909 close(fd); 910 continue; 911 } 912 } 913 914 if (nb_fds > 0) { 915 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int)); 916 if (!tmp) { 917 p_err("failed to realloc"); 918 goto err_close_fd; 919 } 920 *fds = tmp; 921 } 922 (*fds)[nb_fds++] = fd; 923 } 924 925 err_close_fd: 926 close(fd); 927 err_close_fds: 928 while (--nb_fds >= 0) 929 close((*fds)[nb_fds]); 930 return -1; 931 } 932 933 int prog_parse_fds(int *argc, char ***argv, int **fds) 934 { 935 if (is_prefix(**argv, "id")) { 936 unsigned int id; 937 char *endptr; 938 939 NEXT_ARGP(); 940 941 id = strtoul(**argv, &endptr, 0); 942 if (*endptr) { 943 p_err("can't parse %s as ID", **argv); 944 return -1; 945 } 946 NEXT_ARGP(); 947 948 (*fds)[0] = bpf_prog_get_fd_by_id(id); 949 if ((*fds)[0] < 0) { 950 p_err("get by id (%u): %s", id, strerror(errno)); 951 return -1; 952 } 953 return 1; 954 } else if (is_prefix(**argv, "tag")) { 955 unsigned char tag[BPF_TAG_SIZE]; 956 957 NEXT_ARGP(); 958 959 if (sscanf(**argv, BPF_TAG_FMT, tag, tag + 1, tag + 2, 960 tag + 3, tag + 4, tag + 5, tag + 6, tag + 7) 961 != BPF_TAG_SIZE) { 962 p_err("can't parse tag"); 963 return -1; 964 } 965 NEXT_ARGP(); 966 967 return prog_fd_by_nametag(tag, fds, true); 968 } else if (is_prefix(**argv, "name")) { 969 char *name; 970 971 NEXT_ARGP(); 972 973 name = **argv; 974 if (strlen(name) > MAX_PROG_FULL_NAME - 1) { 975 p_err("can't parse name"); 976 return -1; 977 } 978 NEXT_ARGP(); 979 980 return prog_fd_by_nametag(name, fds, false); 981 } else if (is_prefix(**argv, "pinned")) { 982 char *path; 983 984 NEXT_ARGP(); 985 986 path = **argv; 987 NEXT_ARGP(); 988 989 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_PROG, NULL); 990 if ((*fds)[0] < 0) 991 return -1; 992 return 1; 993 } 994 995 p_err("expected 'id', 'tag', 'name' or 'pinned', got: '%s'?", **argv); 996 return -1; 997 } 998 999 int prog_parse_fd(int *argc, char ***argv) 1000 { 1001 int *fds = NULL; 1002 int nb_fds, fd; 1003 1004 fds = malloc(sizeof(int)); 1005 if (!fds) { 1006 p_err("mem alloc failed"); 1007 return -1; 1008 } 1009 nb_fds = prog_parse_fds(argc, argv, &fds); 1010 if (nb_fds != 1) { 1011 if (nb_fds > 1) { 1012 p_err("several programs match this handle"); 1013 while (nb_fds--) 1014 close(fds[nb_fds]); 1015 } 1016 fd = -1; 1017 goto exit_free; 1018 } 1019 1020 fd = fds[0]; 1021 exit_free: 1022 free(fds); 1023 return fd; 1024 } 1025 1026 static int map_fd_by_name(char *name, int **fds, 1027 const struct bpf_get_fd_by_id_opts *opts) 1028 { 1029 unsigned int id = 0; 1030 int fd, nb_fds = 0; 1031 void *tmp; 1032 int err; 1033 1034 while (true) { 1035 LIBBPF_OPTS(bpf_get_fd_by_id_opts, opts_ro); 1036 struct bpf_map_info info = {}; 1037 __u32 len = sizeof(info); 1038 1039 err = bpf_map_get_next_id(id, &id); 1040 if (err) { 1041 if (errno != ENOENT) { 1042 p_err("%s", strerror(errno)); 1043 goto err_close_fds; 1044 } 1045 return nb_fds; 1046 } 1047 1048 /* Request a read-only fd to query the map info */ 1049 opts_ro.open_flags = BPF_F_RDONLY; 1050 fd = bpf_map_get_fd_by_id_opts(id, &opts_ro); 1051 if (fd < 0) { 1052 if (errno == ENOENT) 1053 continue; 1054 p_err("can't get map by id (%u): %s", 1055 id, strerror(errno)); 1056 goto err_close_fds; 1057 } 1058 1059 err = bpf_map_get_info_by_fd(fd, &info, &len); 1060 if (err) { 1061 p_err("can't get map info (%u): %s", 1062 id, strerror(errno)); 1063 goto err_close_fd; 1064 } 1065 1066 if (strncmp(name, info.name, BPF_OBJ_NAME_LEN)) { 1067 close(fd); 1068 continue; 1069 } 1070 1071 /* Get an fd with the requested options, if they differ 1072 * from the read-only options used to get the fd above. 1073 */ 1074 if (memcmp(opts, &opts_ro, sizeof(opts_ro))) { 1075 close(fd); 1076 fd = bpf_map_get_fd_by_id_opts(id, opts); 1077 if (fd < 0) { 1078 p_err("can't get map by id (%u): %s", id, 1079 strerror(errno)); 1080 goto err_close_fds; 1081 } 1082 } 1083 1084 if (nb_fds > 0) { 1085 tmp = realloc(*fds, (nb_fds + 1) * sizeof(int)); 1086 if (!tmp) { 1087 p_err("failed to realloc"); 1088 goto err_close_fd; 1089 } 1090 *fds = tmp; 1091 } 1092 (*fds)[nb_fds++] = fd; 1093 } 1094 1095 err_close_fd: 1096 close(fd); 1097 err_close_fds: 1098 while (--nb_fds >= 0) 1099 close((*fds)[nb_fds]); 1100 return -1; 1101 } 1102 1103 int map_parse_fds(int *argc, char ***argv, int **fds, __u32 open_flags) 1104 { 1105 LIBBPF_OPTS(bpf_get_fd_by_id_opts, opts); 1106 1107 assert((open_flags & ~BPF_F_RDONLY) == 0); 1108 opts.open_flags = open_flags; 1109 1110 if (is_prefix(**argv, "id")) { 1111 unsigned int id; 1112 char *endptr; 1113 1114 NEXT_ARGP(); 1115 1116 id = strtoul(**argv, &endptr, 0); 1117 if (*endptr) { 1118 p_err("can't parse %s as ID", **argv); 1119 return -1; 1120 } 1121 NEXT_ARGP(); 1122 1123 (*fds)[0] = bpf_map_get_fd_by_id_opts(id, &opts); 1124 if ((*fds)[0] < 0) { 1125 p_err("get map by id (%u): %s", id, strerror(errno)); 1126 return -1; 1127 } 1128 return 1; 1129 } else if (is_prefix(**argv, "name")) { 1130 char *name; 1131 1132 NEXT_ARGP(); 1133 1134 name = **argv; 1135 if (strlen(name) > BPF_OBJ_NAME_LEN - 1) { 1136 p_err("can't parse name"); 1137 return -1; 1138 } 1139 NEXT_ARGP(); 1140 1141 return map_fd_by_name(name, fds, &opts); 1142 } else if (is_prefix(**argv, "pinned")) { 1143 char *path; 1144 LIBBPF_OPTS(bpf_obj_get_opts, get_opts); 1145 get_opts.file_flags = open_flags; 1146 1147 NEXT_ARGP(); 1148 1149 path = **argv; 1150 NEXT_ARGP(); 1151 1152 (*fds)[0] = open_obj_pinned_any(path, BPF_OBJ_MAP, &get_opts); 1153 if ((*fds)[0] < 0) 1154 return -1; 1155 return 1; 1156 } 1157 1158 p_err("expected 'id', 'name' or 'pinned', got: '%s'?", **argv); 1159 return -1; 1160 } 1161 1162 int map_parse_fd(int *argc, char ***argv, __u32 open_flags) 1163 { 1164 int *fds = NULL; 1165 int nb_fds, fd; 1166 1167 fds = malloc(sizeof(int)); 1168 if (!fds) { 1169 p_err("mem alloc failed"); 1170 return -1; 1171 } 1172 nb_fds = map_parse_fds(argc, argv, &fds, open_flags); 1173 if (nb_fds != 1) { 1174 if (nb_fds > 1) { 1175 p_err("several maps match this handle"); 1176 while (nb_fds--) 1177 close(fds[nb_fds]); 1178 } 1179 fd = -1; 1180 goto exit_free; 1181 } 1182 1183 fd = fds[0]; 1184 exit_free: 1185 free(fds); 1186 return fd; 1187 } 1188 1189 int map_parse_fd_and_info(int *argc, char ***argv, struct bpf_map_info *info, 1190 __u32 *info_len, __u32 open_flags) 1191 { 1192 int err; 1193 int fd; 1194 1195 fd = map_parse_fd(argc, argv, open_flags); 1196 if (fd < 0) 1197 return -1; 1198 1199 err = bpf_map_get_info_by_fd(fd, info, info_len); 1200 if (err) { 1201 p_err("can't get map info: %s", strerror(errno)); 1202 close(fd); 1203 return err; 1204 } 1205 1206 return fd; 1207 } 1208 1209 size_t hash_fn_for_key_as_id(long key, void *ctx) 1210 { 1211 return key; 1212 } 1213 1214 bool equal_fn_for_key_as_id(long k1, long k2, void *ctx) 1215 { 1216 return k1 == k2; 1217 } 1218 1219 const char *bpf_attach_type_input_str(enum bpf_attach_type t) 1220 { 1221 switch (t) { 1222 case BPF_CGROUP_INET_INGRESS: return "ingress"; 1223 case BPF_CGROUP_INET_EGRESS: return "egress"; 1224 case BPF_CGROUP_INET_SOCK_CREATE: return "sock_create"; 1225 case BPF_CGROUP_INET_SOCK_RELEASE: return "sock_release"; 1226 case BPF_CGROUP_SOCK_OPS: return "sock_ops"; 1227 case BPF_CGROUP_DEVICE: return "device"; 1228 case BPF_CGROUP_INET4_BIND: return "bind4"; 1229 case BPF_CGROUP_INET6_BIND: return "bind6"; 1230 case BPF_CGROUP_INET4_CONNECT: return "connect4"; 1231 case BPF_CGROUP_INET6_CONNECT: return "connect6"; 1232 case BPF_CGROUP_INET4_POST_BIND: return "post_bind4"; 1233 case BPF_CGROUP_INET6_POST_BIND: return "post_bind6"; 1234 case BPF_CGROUP_INET4_GETPEERNAME: return "getpeername4"; 1235 case BPF_CGROUP_INET6_GETPEERNAME: return "getpeername6"; 1236 case BPF_CGROUP_INET4_GETSOCKNAME: return "getsockname4"; 1237 case BPF_CGROUP_INET6_GETSOCKNAME: return "getsockname6"; 1238 case BPF_CGROUP_UDP4_SENDMSG: return "sendmsg4"; 1239 case BPF_CGROUP_UDP6_SENDMSG: return "sendmsg6"; 1240 case BPF_CGROUP_SYSCTL: return "sysctl"; 1241 case BPF_CGROUP_UDP4_RECVMSG: return "recvmsg4"; 1242 case BPF_CGROUP_UDP6_RECVMSG: return "recvmsg6"; 1243 case BPF_CGROUP_GETSOCKOPT: return "getsockopt"; 1244 case BPF_CGROUP_SETSOCKOPT: return "setsockopt"; 1245 case BPF_TRACE_RAW_TP: return "raw_tp"; 1246 case BPF_TRACE_FENTRY: return "fentry"; 1247 case BPF_TRACE_FEXIT: return "fexit"; 1248 case BPF_MODIFY_RETURN: return "mod_ret"; 1249 case BPF_TRACE_FSESSION: return "fsession"; 1250 case BPF_SK_REUSEPORT_SELECT: return "sk_skb_reuseport_select"; 1251 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE: return "sk_skb_reuseport_select_or_migrate"; 1252 default: return libbpf_bpf_attach_type_str(t); 1253 } 1254 } 1255 1256 int pathname_concat(char *buf, int buf_sz, const char *path, 1257 const char *name) 1258 { 1259 int len; 1260 1261 len = snprintf(buf, buf_sz, "%s/%s", path, name); 1262 if (len < 0) 1263 return -EINVAL; 1264 if (len >= buf_sz) 1265 return -ENAMETOOLONG; 1266 1267 return 0; 1268 } 1269 1270 static bool read_next_kernel_config_option(gzFile file, char *buf, size_t n, 1271 char **value) 1272 { 1273 char *sep; 1274 1275 while (gzgets(file, buf, n)) { 1276 if (strncmp(buf, "CONFIG_", 7)) 1277 continue; 1278 1279 sep = strchr(buf, '='); 1280 if (!sep) 1281 continue; 1282 1283 /* Trim ending '\n' */ 1284 buf[strlen(buf) - 1] = '\0'; 1285 1286 /* Split on '=' and ensure that a value is present. */ 1287 *sep = '\0'; 1288 if (!sep[1]) 1289 continue; 1290 1291 *value = sep + 1; 1292 return true; 1293 } 1294 1295 return false; 1296 } 1297 1298 int read_kernel_config(const struct kernel_config_option *requested_options, 1299 size_t num_options, char **out_values, 1300 const char *define_prefix) 1301 { 1302 struct utsname utsn; 1303 char path[PATH_MAX]; 1304 gzFile file = NULL; 1305 char buf[4096]; 1306 char *value; 1307 size_t i; 1308 int ret = 0; 1309 1310 if (!requested_options || !out_values || num_options == 0) 1311 return -1; 1312 1313 if (!uname(&utsn)) { 1314 snprintf(path, sizeof(path), "/boot/config-%s", utsn.release); 1315 1316 /* gzopen also accepts uncompressed files. */ 1317 file = gzopen(path, "r"); 1318 } 1319 1320 if (!file) { 1321 /* Some distributions build with CONFIG_IKCONFIG=y and put the 1322 * config file at /proc/config.gz. 1323 */ 1324 file = gzopen("/proc/config.gz", "r"); 1325 } 1326 1327 if (!file) { 1328 p_info("skipping kernel config, can't open file: %s", 1329 strerror(errno)); 1330 return -1; 1331 } 1332 1333 if (!gzgets(file, buf, sizeof(buf)) || !gzgets(file, buf, sizeof(buf))) { 1334 p_info("skipping kernel config, can't read from file: %s", 1335 strerror(errno)); 1336 ret = -1; 1337 goto end_parse; 1338 } 1339 1340 if (strcmp(buf, "# Automatically generated file; DO NOT EDIT.\n")) { 1341 p_info("skipping kernel config, can't find correct file"); 1342 ret = -1; 1343 goto end_parse; 1344 } 1345 1346 while (read_next_kernel_config_option(file, buf, sizeof(buf), &value)) { 1347 for (i = 0; i < num_options; i++) { 1348 if ((define_prefix && !requested_options[i].macro_dump) || 1349 out_values[i] || strcmp(buf, requested_options[i].name)) 1350 continue; 1351 1352 out_values[i] = strdup(value); 1353 } 1354 } 1355 1356 end_parse: 1357 gzclose(file); 1358 return ret; 1359 } 1360