1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) 2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */ 3 4 #include <ctype.h> 5 #include <errno.h> 6 #include <fcntl.h> 7 #include <fts.h> 8 #include <libgen.h> 9 #include <mntent.h> 10 #include <stdbool.h> 11 #include <stdio.h> 12 #include <stdlib.h> 13 #include <string.h> 14 #include <unistd.h> 15 #include <linux/limits.h> 16 #include <linux/magic.h> 17 #include <net/if.h> 18 #include <sys/mount.h> 19 #include <sys/resource.h> 20 #include <sys/stat.h> 21 #include <sys/vfs.h> 22 23 #include <bpf.h> 24 25 #include "main.h" 26 27 #ifndef BPF_FS_MAGIC 28 #define BPF_FS_MAGIC 0xcafe4a11 29 #endif 30 31 void __printf(1, 2) p_err(const char *fmt, ...) 32 { 33 va_list ap; 34 35 va_start(ap, fmt); 36 if (json_output) { 37 jsonw_start_object(json_wtr); 38 jsonw_name(json_wtr, "error"); 39 jsonw_vprintf_enquote(json_wtr, fmt, ap); 40 jsonw_end_object(json_wtr); 41 } else { 42 fprintf(stderr, "Error: "); 43 vfprintf(stderr, fmt, ap); 44 fprintf(stderr, "\n"); 45 } 46 va_end(ap); 47 } 48 49 void __printf(1, 2) p_info(const char *fmt, ...) 50 { 51 va_list ap; 52 53 if (json_output) 54 return; 55 56 va_start(ap, fmt); 57 vfprintf(stderr, fmt, ap); 58 fprintf(stderr, "\n"); 59 va_end(ap); 60 } 61 62 static bool is_bpffs(char *path) 63 { 64 struct statfs st_fs; 65 66 if (statfs(path, &st_fs) < 0) 67 return false; 68 69 return (unsigned long)st_fs.f_type == BPF_FS_MAGIC; 70 } 71 72 void set_max_rlimit(void) 73 { 74 struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY }; 75 76 setrlimit(RLIMIT_MEMLOCK, &rinf); 77 } 78 79 static int 80 mnt_fs(const char *target, const char *type, char *buff, size_t bufflen) 81 { 82 bool bind_done = false; 83 84 while (mount("", target, "none", MS_PRIVATE | MS_REC, NULL)) { 85 if (errno != EINVAL || bind_done) { 86 snprintf(buff, bufflen, 87 "mount --make-private %s failed: %s", 88 target, strerror(errno)); 89 return -1; 90 } 91 92 if (mount(target, target, "none", MS_BIND, NULL)) { 93 snprintf(buff, bufflen, 94 "mount --bind %s %s failed: %s", 95 target, target, strerror(errno)); 96 return -1; 97 } 98 99 bind_done = true; 100 } 101 102 if (mount(type, target, type, 0, "mode=0700")) { 103 snprintf(buff, bufflen, "mount -t %s %s %s failed: %s", 104 type, type, target, strerror(errno)); 105 return -1; 106 } 107 108 return 0; 109 } 110 111 int mount_tracefs(const char *target) 112 { 113 char err_str[ERR_MAX_LEN]; 114 int err; 115 116 err = mnt_fs(target, "tracefs", err_str, ERR_MAX_LEN); 117 if (err) { 118 err_str[ERR_MAX_LEN - 1] = '\0'; 119 p_err("can't mount tracefs: %s", err_str); 120 } 121 122 return err; 123 } 124 125 int open_obj_pinned(char *path, bool quiet) 126 { 127 int fd; 128 129 fd = bpf_obj_get(path); 130 if (fd < 0) { 131 if (!quiet) 132 p_err("bpf obj get (%s): %s", path, 133 errno == EACCES && !is_bpffs(dirname(path)) ? 134 "directory not in bpf file system (bpffs)" : 135 strerror(errno)); 136 return -1; 137 } 138 139 return fd; 140 } 141 142 int open_obj_pinned_any(char *path, enum bpf_obj_type exp_type) 143 { 144 enum bpf_obj_type type; 145 int fd; 146 147 fd = open_obj_pinned(path, false); 148 if (fd < 0) 149 return -1; 150 151 type = get_fd_type(fd); 152 if (type < 0) { 153 close(fd); 154 return type; 155 } 156 if (type != exp_type) { 157 p_err("incorrect object type: %s", get_fd_type_name(type)); 158 close(fd); 159 return -1; 160 } 161 162 return fd; 163 } 164 165 int mount_bpffs_for_pin(const char *name) 166 { 167 char err_str[ERR_MAX_LEN]; 168 char *file; 169 char *dir; 170 int err = 0; 171 172 file = malloc(strlen(name) + 1); 173 strcpy(file, name); 174 dir = dirname(file); 175 176 if (is_bpffs(dir)) 177 /* nothing to do if already mounted */ 178 goto out_free; 179 180 if (block_mount) { 181 p_err("no BPF file system found, not mounting it due to --nomount option"); 182 err = -1; 183 goto out_free; 184 } 185 186 err = mnt_fs(dir, "bpf", err_str, ERR_MAX_LEN); 187 if (err) { 188 err_str[ERR_MAX_LEN - 1] = '\0'; 189 p_err("can't mount BPF file system to pin the object (%s): %s", 190 name, err_str); 191 } 192 193 out_free: 194 free(file); 195 return err; 196 } 197 198 int do_pin_fd(int fd, const char *name) 199 { 200 int err; 201 202 err = mount_bpffs_for_pin(name); 203 if (err) 204 return err; 205 206 return bpf_obj_pin(fd, name); 207 } 208 209 int do_pin_any(int argc, char **argv, int (*get_fd_by_id)(__u32)) 210 { 211 unsigned int id; 212 char *endptr; 213 int err; 214 int fd; 215 216 if (argc < 3) { 217 p_err("too few arguments, id ID and FILE path is required"); 218 return -1; 219 } else if (argc > 3) { 220 p_err("too many arguments"); 221 return -1; 222 } 223 224 if (!is_prefix(*argv, "id")) { 225 p_err("expected 'id' got %s", *argv); 226 return -1; 227 } 228 NEXT_ARG(); 229 230 id = strtoul(*argv, &endptr, 0); 231 if (*endptr) { 232 p_err("can't parse %s as ID", *argv); 233 return -1; 234 } 235 NEXT_ARG(); 236 237 fd = get_fd_by_id(id); 238 if (fd < 0) { 239 p_err("can't get prog by id (%u): %s", id, strerror(errno)); 240 return -1; 241 } 242 243 err = do_pin_fd(fd, *argv); 244 245 close(fd); 246 return err; 247 } 248 249 const char *get_fd_type_name(enum bpf_obj_type type) 250 { 251 static const char * const names[] = { 252 [BPF_OBJ_UNKNOWN] = "unknown", 253 [BPF_OBJ_PROG] = "prog", 254 [BPF_OBJ_MAP] = "map", 255 }; 256 257 if (type < 0 || type >= ARRAY_SIZE(names) || !names[type]) 258 return names[BPF_OBJ_UNKNOWN]; 259 260 return names[type]; 261 } 262 263 int get_fd_type(int fd) 264 { 265 char path[PATH_MAX]; 266 char buf[512]; 267 ssize_t n; 268 269 snprintf(path, sizeof(path), "/proc/self/fd/%d", fd); 270 271 n = readlink(path, buf, sizeof(buf)); 272 if (n < 0) { 273 p_err("can't read link type: %s", strerror(errno)); 274 return -1; 275 } 276 if (n == sizeof(path)) { 277 p_err("can't read link type: path too long!"); 278 return -1; 279 } 280 281 if (strstr(buf, "bpf-map")) 282 return BPF_OBJ_MAP; 283 else if (strstr(buf, "bpf-prog")) 284 return BPF_OBJ_PROG; 285 286 return BPF_OBJ_UNKNOWN; 287 } 288 289 char *get_fdinfo(int fd, const char *key) 290 { 291 char path[PATH_MAX]; 292 char *line = NULL; 293 size_t line_n = 0; 294 ssize_t n; 295 FILE *fdi; 296 297 snprintf(path, sizeof(path), "/proc/self/fdinfo/%d", fd); 298 299 fdi = fopen(path, "r"); 300 if (!fdi) { 301 p_err("can't open fdinfo: %s", strerror(errno)); 302 return NULL; 303 } 304 305 while ((n = getline(&line, &line_n, fdi)) > 0) { 306 char *value; 307 int len; 308 309 if (!strstr(line, key)) 310 continue; 311 312 fclose(fdi); 313 314 value = strchr(line, '\t'); 315 if (!value || !value[1]) { 316 p_err("malformed fdinfo!?"); 317 free(line); 318 return NULL; 319 } 320 value++; 321 322 len = strlen(value); 323 memmove(line, value, len); 324 line[len - 1] = '\0'; 325 326 return line; 327 } 328 329 p_err("key '%s' not found in fdinfo", key); 330 free(line); 331 fclose(fdi); 332 return NULL; 333 } 334 335 void print_data_json(uint8_t *data, size_t len) 336 { 337 unsigned int i; 338 339 jsonw_start_array(json_wtr); 340 for (i = 0; i < len; i++) 341 jsonw_printf(json_wtr, "%d", data[i]); 342 jsonw_end_array(json_wtr); 343 } 344 345 void print_hex_data_json(uint8_t *data, size_t len) 346 { 347 unsigned int i; 348 349 jsonw_start_array(json_wtr); 350 for (i = 0; i < len; i++) 351 jsonw_printf(json_wtr, "\"0x%02hhx\"", data[i]); 352 jsonw_end_array(json_wtr); 353 } 354 355 int build_pinned_obj_table(struct pinned_obj_table *tab, 356 enum bpf_obj_type type) 357 { 358 struct bpf_prog_info pinned_info = {}; 359 struct pinned_obj *obj_node = NULL; 360 __u32 len = sizeof(pinned_info); 361 struct mntent *mntent = NULL; 362 enum bpf_obj_type objtype; 363 FILE *mntfile = NULL; 364 FTSENT *ftse = NULL; 365 FTS *fts = NULL; 366 int fd, err; 367 368 mntfile = setmntent("/proc/mounts", "r"); 369 if (!mntfile) 370 return -1; 371 372 while ((mntent = getmntent(mntfile))) { 373 char *path[] = { mntent->mnt_dir, NULL }; 374 375 if (strncmp(mntent->mnt_type, "bpf", 3) != 0) 376 continue; 377 378 fts = fts_open(path, 0, NULL); 379 if (!fts) 380 continue; 381 382 while ((ftse = fts_read(fts))) { 383 if (!(ftse->fts_info & FTS_F)) 384 continue; 385 fd = open_obj_pinned(ftse->fts_path, true); 386 if (fd < 0) 387 continue; 388 389 objtype = get_fd_type(fd); 390 if (objtype != type) { 391 close(fd); 392 continue; 393 } 394 memset(&pinned_info, 0, sizeof(pinned_info)); 395 err = bpf_obj_get_info_by_fd(fd, &pinned_info, &len); 396 if (err) { 397 close(fd); 398 continue; 399 } 400 401 obj_node = malloc(sizeof(*obj_node)); 402 if (!obj_node) { 403 close(fd); 404 fts_close(fts); 405 fclose(mntfile); 406 return -1; 407 } 408 409 memset(obj_node, 0, sizeof(*obj_node)); 410 obj_node->id = pinned_info.id; 411 obj_node->path = strdup(ftse->fts_path); 412 hash_add(tab->table, &obj_node->hash, obj_node->id); 413 414 close(fd); 415 } 416 fts_close(fts); 417 } 418 fclose(mntfile); 419 return 0; 420 } 421 422 void delete_pinned_obj_table(struct pinned_obj_table *tab) 423 { 424 struct pinned_obj *obj; 425 struct hlist_node *tmp; 426 unsigned int bkt; 427 428 hash_for_each_safe(tab->table, bkt, tmp, obj, hash) { 429 hash_del(&obj->hash); 430 free(obj->path); 431 free(obj); 432 } 433 } 434 435 unsigned int get_page_size(void) 436 { 437 static int result; 438 439 if (!result) 440 result = getpagesize(); 441 return result; 442 } 443 444 unsigned int get_possible_cpus(void) 445 { 446 static unsigned int result; 447 char buf[128]; 448 long int n; 449 char *ptr; 450 int fd; 451 452 if (result) 453 return result; 454 455 fd = open("/sys/devices/system/cpu/possible", O_RDONLY); 456 if (fd < 0) { 457 p_err("can't open sysfs possible cpus"); 458 exit(-1); 459 } 460 461 n = read(fd, buf, sizeof(buf)); 462 if (n < 2) { 463 p_err("can't read sysfs possible cpus"); 464 exit(-1); 465 } 466 close(fd); 467 468 if (n == sizeof(buf)) { 469 p_err("read sysfs possible cpus overflow"); 470 exit(-1); 471 } 472 473 ptr = buf; 474 n = 0; 475 while (*ptr && *ptr != '\n') { 476 unsigned int a, b; 477 478 if (sscanf(ptr, "%u-%u", &a, &b) == 2) { 479 n += b - a + 1; 480 481 ptr = strchr(ptr, '-') + 1; 482 } else if (sscanf(ptr, "%u", &a) == 1) { 483 n++; 484 } else { 485 assert(0); 486 } 487 488 while (isdigit(*ptr)) 489 ptr++; 490 if (*ptr == ',') 491 ptr++; 492 } 493 494 result = n; 495 496 return result; 497 } 498 499 static char * 500 ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf) 501 { 502 struct stat st; 503 int err; 504 505 err = stat("/proc/self/ns/net", &st); 506 if (err) { 507 p_err("Can't stat /proc/self: %s", strerror(errno)); 508 return NULL; 509 } 510 511 if (st.st_dev != ns_dev || st.st_ino != ns_ino) 512 return NULL; 513 514 return if_indextoname(ifindex, buf); 515 } 516 517 static int read_sysfs_hex_int(char *path) 518 { 519 char vendor_id_buf[8]; 520 int len; 521 int fd; 522 523 fd = open(path, O_RDONLY); 524 if (fd < 0) { 525 p_err("Can't open %s: %s", path, strerror(errno)); 526 return -1; 527 } 528 529 len = read(fd, vendor_id_buf, sizeof(vendor_id_buf)); 530 close(fd); 531 if (len < 0) { 532 p_err("Can't read %s: %s", path, strerror(errno)); 533 return -1; 534 } 535 if (len >= (int)sizeof(vendor_id_buf)) { 536 p_err("Value in %s too long", path); 537 return -1; 538 } 539 540 vendor_id_buf[len] = 0; 541 542 return strtol(vendor_id_buf, NULL, 0); 543 } 544 545 static int read_sysfs_netdev_hex_int(char *devname, const char *entry_name) 546 { 547 char full_path[64]; 548 549 snprintf(full_path, sizeof(full_path), "/sys/class/net/%s/device/%s", 550 devname, entry_name); 551 552 return read_sysfs_hex_int(full_path); 553 } 554 555 const char * 556 ifindex_to_bfd_params(__u32 ifindex, __u64 ns_dev, __u64 ns_ino, 557 const char **opt) 558 { 559 char devname[IF_NAMESIZE]; 560 int vendor_id; 561 int device_id; 562 563 if (!ifindex_to_name_ns(ifindex, ns_dev, ns_ino, devname)) { 564 p_err("Can't get net device name for ifindex %d: %s", ifindex, 565 strerror(errno)); 566 return NULL; 567 } 568 569 vendor_id = read_sysfs_netdev_hex_int(devname, "vendor"); 570 if (vendor_id < 0) { 571 p_err("Can't get device vendor id for %s", devname); 572 return NULL; 573 } 574 575 switch (vendor_id) { 576 case 0x19ee: 577 device_id = read_sysfs_netdev_hex_int(devname, "device"); 578 if (device_id != 0x4000 && 579 device_id != 0x6000 && 580 device_id != 0x6003) 581 p_info("Unknown NFP device ID, assuming it is NFP-6xxx arch"); 582 *opt = "ctx4"; 583 return "NFP-6xxx"; 584 default: 585 p_err("Can't get bfd arch name for device vendor id 0x%04x", 586 vendor_id); 587 return NULL; 588 } 589 } 590 591 void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode) 592 { 593 char name[IF_NAMESIZE]; 594 595 if (!ifindex) 596 return; 597 598 printf(" offloaded_to "); 599 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name)) 600 printf("%s", name); 601 else 602 printf("ifindex %u ns_dev %llu ns_ino %llu", 603 ifindex, ns_dev, ns_inode); 604 } 605 606 void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode) 607 { 608 char name[IF_NAMESIZE]; 609 610 if (!ifindex) 611 return; 612 613 jsonw_name(json_wtr, "dev"); 614 jsonw_start_object(json_wtr); 615 jsonw_uint_field(json_wtr, "ifindex", ifindex); 616 jsonw_uint_field(json_wtr, "ns_dev", ns_dev); 617 jsonw_uint_field(json_wtr, "ns_inode", ns_inode); 618 if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name)) 619 jsonw_string_field(json_wtr, "ifname", name); 620 jsonw_end_object(json_wtr); 621 } 622 623 int parse_u32_arg(int *argc, char ***argv, __u32 *val, const char *what) 624 { 625 char *endptr; 626 627 NEXT_ARGP(); 628 629 if (*val) { 630 p_err("%s already specified", what); 631 return -1; 632 } 633 634 *val = strtoul(**argv, &endptr, 0); 635 if (*endptr) { 636 p_err("can't parse %s as %s", **argv, what); 637 return -1; 638 } 639 NEXT_ARGP(); 640 641 return 0; 642 } 643