1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Copyright (c) 2017 Facebook 3 */ 4 #include <linux/bpf.h> 5 #include <linux/btf_ids.h> 6 #include <linux/slab.h> 7 #include <linux/vmalloc.h> 8 #include <linux/etherdevice.h> 9 #include <linux/filter.h> 10 #include <linux/rcupdate_trace.h> 11 #include <linux/sched/signal.h> 12 #include <net/bpf_sk_storage.h> 13 #include <net/sock.h> 14 #include <net/tcp.h> 15 #include <net/net_namespace.h> 16 #include <linux/error-injection.h> 17 #include <linux/smp.h> 18 #include <linux/sock_diag.h> 19 #include <net/xdp.h> 20 21 #define CREATE_TRACE_POINTS 22 #include <trace/events/bpf_test_run.h> 23 24 struct bpf_test_timer { 25 enum { NO_PREEMPT, NO_MIGRATE } mode; 26 u32 i; 27 u64 time_start, time_spent; 28 }; 29 30 static void bpf_test_timer_enter(struct bpf_test_timer *t) 31 __acquires(rcu) 32 { 33 rcu_read_lock(); 34 if (t->mode == NO_PREEMPT) 35 preempt_disable(); 36 else 37 migrate_disable(); 38 39 t->time_start = ktime_get_ns(); 40 } 41 42 static void bpf_test_timer_leave(struct bpf_test_timer *t) 43 __releases(rcu) 44 { 45 t->time_start = 0; 46 47 if (t->mode == NO_PREEMPT) 48 preempt_enable(); 49 else 50 migrate_enable(); 51 rcu_read_unlock(); 52 } 53 54 static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration) 55 __must_hold(rcu) 56 { 57 t->i++; 58 if (t->i >= repeat) { 59 /* We're done. */ 60 t->time_spent += ktime_get_ns() - t->time_start; 61 do_div(t->time_spent, t->i); 62 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent; 63 *err = 0; 64 goto reset; 65 } 66 67 if (signal_pending(current)) { 68 /* During iteration: we've been cancelled, abort. */ 69 *err = -EINTR; 70 goto reset; 71 } 72 73 if (need_resched()) { 74 /* During iteration: we need to reschedule between runs. */ 75 t->time_spent += ktime_get_ns() - t->time_start; 76 bpf_test_timer_leave(t); 77 cond_resched(); 78 bpf_test_timer_enter(t); 79 } 80 81 /* Do another round. */ 82 return true; 83 84 reset: 85 t->i = 0; 86 return false; 87 } 88 89 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat, 90 u32 *retval, u32 *time, bool xdp) 91 { 92 struct bpf_prog_array_item item = {.prog = prog}; 93 struct bpf_run_ctx *old_ctx; 94 struct bpf_cg_run_ctx run_ctx; 95 struct bpf_test_timer t = { NO_MIGRATE }; 96 enum bpf_cgroup_storage_type stype; 97 int ret; 98 99 for_each_cgroup_storage_type(stype) { 100 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype); 101 if (IS_ERR(item.cgroup_storage[stype])) { 102 item.cgroup_storage[stype] = NULL; 103 for_each_cgroup_storage_type(stype) 104 bpf_cgroup_storage_free(item.cgroup_storage[stype]); 105 return -ENOMEM; 106 } 107 } 108 109 if (!repeat) 110 repeat = 1; 111 112 bpf_test_timer_enter(&t); 113 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 114 do { 115 run_ctx.prog_item = &item; 116 if (xdp) 117 *retval = bpf_prog_run_xdp(prog, ctx); 118 else 119 *retval = bpf_prog_run(prog, ctx); 120 } while (bpf_test_timer_continue(&t, repeat, &ret, time)); 121 bpf_reset_run_ctx(old_ctx); 122 bpf_test_timer_leave(&t); 123 124 for_each_cgroup_storage_type(stype) 125 bpf_cgroup_storage_free(item.cgroup_storage[stype]); 126 127 return ret; 128 } 129 130 static int bpf_test_finish(const union bpf_attr *kattr, 131 union bpf_attr __user *uattr, const void *data, 132 u32 size, u32 retval, u32 duration) 133 { 134 void __user *data_out = u64_to_user_ptr(kattr->test.data_out); 135 int err = -EFAULT; 136 u32 copy_size = size; 137 138 /* Clamp copy if the user has provided a size hint, but copy the full 139 * buffer if not to retain old behaviour. 140 */ 141 if (kattr->test.data_size_out && 142 copy_size > kattr->test.data_size_out) { 143 copy_size = kattr->test.data_size_out; 144 err = -ENOSPC; 145 } 146 147 if (data_out && copy_to_user(data_out, data, copy_size)) 148 goto out; 149 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size))) 150 goto out; 151 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval))) 152 goto out; 153 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration))) 154 goto out; 155 if (err != -ENOSPC) 156 err = 0; 157 out: 158 trace_bpf_test_finish(&err); 159 return err; 160 } 161 162 /* Integer types of various sizes and pointer combinations cover variety of 163 * architecture dependent calling conventions. 7+ can be supported in the 164 * future. 165 */ 166 __diag_push(); 167 __diag_ignore(GCC, 8, "-Wmissing-prototypes", 168 "Global functions as their definitions will be in vmlinux BTF"); 169 int noinline bpf_fentry_test1(int a) 170 { 171 return a + 1; 172 } 173 174 int noinline bpf_fentry_test2(int a, u64 b) 175 { 176 return a + b; 177 } 178 179 int noinline bpf_fentry_test3(char a, int b, u64 c) 180 { 181 return a + b + c; 182 } 183 184 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d) 185 { 186 return (long)a + b + c + d; 187 } 188 189 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e) 190 { 191 return a + (long)b + c + d + e; 192 } 193 194 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f) 195 { 196 return a + (long)b + c + d + (long)e + f; 197 } 198 199 struct bpf_fentry_test_t { 200 struct bpf_fentry_test_t *a; 201 }; 202 203 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg) 204 { 205 return (long)arg; 206 } 207 208 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg) 209 { 210 return (long)arg->a; 211 } 212 213 int noinline bpf_modify_return_test(int a, int *b) 214 { 215 *b += 1; 216 return a + *b; 217 } 218 219 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d) 220 { 221 return a + b + c + d; 222 } 223 224 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b) 225 { 226 return a + b; 227 } 228 229 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk) 230 { 231 return sk; 232 } 233 234 __diag_pop(); 235 236 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO); 237 238 BTF_SET_START(test_sk_kfunc_ids) 239 BTF_ID(func, bpf_kfunc_call_test1) 240 BTF_ID(func, bpf_kfunc_call_test2) 241 BTF_ID(func, bpf_kfunc_call_test3) 242 BTF_SET_END(test_sk_kfunc_ids) 243 244 bool bpf_prog_test_check_kfunc_call(u32 kfunc_id) 245 { 246 return btf_id_set_contains(&test_sk_kfunc_ids, kfunc_id); 247 } 248 249 static void *bpf_test_init(const union bpf_attr *kattr, u32 size, 250 u32 headroom, u32 tailroom) 251 { 252 void __user *data_in = u64_to_user_ptr(kattr->test.data_in); 253 u32 user_size = kattr->test.data_size_in; 254 void *data; 255 256 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom) 257 return ERR_PTR(-EINVAL); 258 259 if (user_size > size) 260 return ERR_PTR(-EMSGSIZE); 261 262 data = kzalloc(size + headroom + tailroom, GFP_USER); 263 if (!data) 264 return ERR_PTR(-ENOMEM); 265 266 if (copy_from_user(data + headroom, data_in, user_size)) { 267 kfree(data); 268 return ERR_PTR(-EFAULT); 269 } 270 271 return data; 272 } 273 274 int bpf_prog_test_run_tracing(struct bpf_prog *prog, 275 const union bpf_attr *kattr, 276 union bpf_attr __user *uattr) 277 { 278 struct bpf_fentry_test_t arg = {}; 279 u16 side_effect = 0, ret = 0; 280 int b = 2, err = -EFAULT; 281 u32 retval = 0; 282 283 if (kattr->test.flags || kattr->test.cpu) 284 return -EINVAL; 285 286 switch (prog->expected_attach_type) { 287 case BPF_TRACE_FENTRY: 288 case BPF_TRACE_FEXIT: 289 if (bpf_fentry_test1(1) != 2 || 290 bpf_fentry_test2(2, 3) != 5 || 291 bpf_fentry_test3(4, 5, 6) != 15 || 292 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 || 293 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 || 294 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 || 295 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 || 296 bpf_fentry_test8(&arg) != 0) 297 goto out; 298 break; 299 case BPF_MODIFY_RETURN: 300 ret = bpf_modify_return_test(1, &b); 301 if (b != 2) 302 side_effect = 1; 303 break; 304 default: 305 goto out; 306 } 307 308 retval = ((u32)side_effect << 16) | ret; 309 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval))) 310 goto out; 311 312 err = 0; 313 out: 314 trace_bpf_test_finish(&err); 315 return err; 316 } 317 318 struct bpf_raw_tp_test_run_info { 319 struct bpf_prog *prog; 320 void *ctx; 321 u32 retval; 322 }; 323 324 static void 325 __bpf_prog_test_run_raw_tp(void *data) 326 { 327 struct bpf_raw_tp_test_run_info *info = data; 328 329 rcu_read_lock(); 330 info->retval = bpf_prog_run(info->prog, info->ctx); 331 rcu_read_unlock(); 332 } 333 334 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 335 const union bpf_attr *kattr, 336 union bpf_attr __user *uattr) 337 { 338 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in); 339 __u32 ctx_size_in = kattr->test.ctx_size_in; 340 struct bpf_raw_tp_test_run_info info; 341 int cpu = kattr->test.cpu, err = 0; 342 int current_cpu; 343 344 /* doesn't support data_in/out, ctx_out, duration, or repeat */ 345 if (kattr->test.data_in || kattr->test.data_out || 346 kattr->test.ctx_out || kattr->test.duration || 347 kattr->test.repeat) 348 return -EINVAL; 349 350 if (ctx_size_in < prog->aux->max_ctx_offset || 351 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64)) 352 return -EINVAL; 353 354 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0) 355 return -EINVAL; 356 357 if (ctx_size_in) { 358 info.ctx = kzalloc(ctx_size_in, GFP_USER); 359 if (!info.ctx) 360 return -ENOMEM; 361 if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) { 362 err = -EFAULT; 363 goto out; 364 } 365 } else { 366 info.ctx = NULL; 367 } 368 369 info.prog = prog; 370 371 current_cpu = get_cpu(); 372 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 || 373 cpu == current_cpu) { 374 __bpf_prog_test_run_raw_tp(&info); 375 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) { 376 /* smp_call_function_single() also checks cpu_online() 377 * after csd_lock(). However, since cpu is from user 378 * space, let's do an extra quick check to filter out 379 * invalid value before smp_call_function_single(). 380 */ 381 err = -ENXIO; 382 } else { 383 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp, 384 &info, 1); 385 } 386 put_cpu(); 387 388 if (!err && 389 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32))) 390 err = -EFAULT; 391 392 out: 393 kfree(info.ctx); 394 return err; 395 } 396 397 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size) 398 { 399 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in); 400 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out); 401 u32 size = kattr->test.ctx_size_in; 402 void *data; 403 int err; 404 405 if (!data_in && !data_out) 406 return NULL; 407 408 data = kzalloc(max_size, GFP_USER); 409 if (!data) 410 return ERR_PTR(-ENOMEM); 411 412 if (data_in) { 413 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size); 414 if (err) { 415 kfree(data); 416 return ERR_PTR(err); 417 } 418 419 size = min_t(u32, max_size, size); 420 if (copy_from_user(data, data_in, size)) { 421 kfree(data); 422 return ERR_PTR(-EFAULT); 423 } 424 } 425 return data; 426 } 427 428 static int bpf_ctx_finish(const union bpf_attr *kattr, 429 union bpf_attr __user *uattr, const void *data, 430 u32 size) 431 { 432 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out); 433 int err = -EFAULT; 434 u32 copy_size = size; 435 436 if (!data || !data_out) 437 return 0; 438 439 if (copy_size > kattr->test.ctx_size_out) { 440 copy_size = kattr->test.ctx_size_out; 441 err = -ENOSPC; 442 } 443 444 if (copy_to_user(data_out, data, copy_size)) 445 goto out; 446 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size))) 447 goto out; 448 if (err != -ENOSPC) 449 err = 0; 450 out: 451 return err; 452 } 453 454 /** 455 * range_is_zero - test whether buffer is initialized 456 * @buf: buffer to check 457 * @from: check from this position 458 * @to: check up until (excluding) this position 459 * 460 * This function returns true if the there is a non-zero byte 461 * in the buf in the range [from,to). 462 */ 463 static inline bool range_is_zero(void *buf, size_t from, size_t to) 464 { 465 return !memchr_inv((u8 *)buf + from, 0, to - from); 466 } 467 468 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) 469 { 470 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; 471 472 if (!__skb) 473 return 0; 474 475 /* make sure the fields we don't use are zeroed */ 476 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark))) 477 return -EINVAL; 478 479 /* mark is allowed */ 480 481 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark), 482 offsetof(struct __sk_buff, priority))) 483 return -EINVAL; 484 485 /* priority is allowed */ 486 /* ingress_ifindex is allowed */ 487 /* ifindex is allowed */ 488 489 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex), 490 offsetof(struct __sk_buff, cb))) 491 return -EINVAL; 492 493 /* cb is allowed */ 494 495 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb), 496 offsetof(struct __sk_buff, tstamp))) 497 return -EINVAL; 498 499 /* tstamp is allowed */ 500 /* wire_len is allowed */ 501 /* gso_segs is allowed */ 502 503 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs), 504 offsetof(struct __sk_buff, gso_size))) 505 return -EINVAL; 506 507 /* gso_size is allowed */ 508 509 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size), 510 offsetof(struct __sk_buff, hwtstamp))) 511 return -EINVAL; 512 513 /* hwtstamp is allowed */ 514 515 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp), 516 sizeof(struct __sk_buff))) 517 return -EINVAL; 518 519 skb->mark = __skb->mark; 520 skb->priority = __skb->priority; 521 skb->skb_iif = __skb->ingress_ifindex; 522 skb->tstamp = __skb->tstamp; 523 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN); 524 525 if (__skb->wire_len == 0) { 526 cb->pkt_len = skb->len; 527 } else { 528 if (__skb->wire_len < skb->len || 529 __skb->wire_len > GSO_MAX_SIZE) 530 return -EINVAL; 531 cb->pkt_len = __skb->wire_len; 532 } 533 534 if (__skb->gso_segs > GSO_MAX_SEGS) 535 return -EINVAL; 536 skb_shinfo(skb)->gso_segs = __skb->gso_segs; 537 skb_shinfo(skb)->gso_size = __skb->gso_size; 538 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp; 539 540 return 0; 541 } 542 543 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb) 544 { 545 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; 546 547 if (!__skb) 548 return; 549 550 __skb->mark = skb->mark; 551 __skb->priority = skb->priority; 552 __skb->ingress_ifindex = skb->skb_iif; 553 __skb->ifindex = skb->dev->ifindex; 554 __skb->tstamp = skb->tstamp; 555 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN); 556 __skb->wire_len = cb->pkt_len; 557 __skb->gso_segs = skb_shinfo(skb)->gso_segs; 558 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp; 559 } 560 561 static struct proto bpf_dummy_proto = { 562 .name = "bpf_dummy", 563 .owner = THIS_MODULE, 564 .obj_size = sizeof(struct sock), 565 }; 566 567 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 568 union bpf_attr __user *uattr) 569 { 570 bool is_l2 = false, is_direct_pkt_access = false; 571 struct net *net = current->nsproxy->net_ns; 572 struct net_device *dev = net->loopback_dev; 573 u32 size = kattr->test.data_size_in; 574 u32 repeat = kattr->test.repeat; 575 struct __sk_buff *ctx = NULL; 576 u32 retval, duration; 577 int hh_len = ETH_HLEN; 578 struct sk_buff *skb; 579 struct sock *sk; 580 void *data; 581 int ret; 582 583 if (kattr->test.flags || kattr->test.cpu) 584 return -EINVAL; 585 586 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN, 587 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))); 588 if (IS_ERR(data)) 589 return PTR_ERR(data); 590 591 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff)); 592 if (IS_ERR(ctx)) { 593 kfree(data); 594 return PTR_ERR(ctx); 595 } 596 597 switch (prog->type) { 598 case BPF_PROG_TYPE_SCHED_CLS: 599 case BPF_PROG_TYPE_SCHED_ACT: 600 is_l2 = true; 601 fallthrough; 602 case BPF_PROG_TYPE_LWT_IN: 603 case BPF_PROG_TYPE_LWT_OUT: 604 case BPF_PROG_TYPE_LWT_XMIT: 605 is_direct_pkt_access = true; 606 break; 607 default: 608 break; 609 } 610 611 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1); 612 if (!sk) { 613 kfree(data); 614 kfree(ctx); 615 return -ENOMEM; 616 } 617 sock_init_data(NULL, sk); 618 619 skb = build_skb(data, 0); 620 if (!skb) { 621 kfree(data); 622 kfree(ctx); 623 sk_free(sk); 624 return -ENOMEM; 625 } 626 skb->sk = sk; 627 628 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 629 __skb_put(skb, size); 630 if (ctx && ctx->ifindex > 1) { 631 dev = dev_get_by_index(net, ctx->ifindex); 632 if (!dev) { 633 ret = -ENODEV; 634 goto out; 635 } 636 } 637 skb->protocol = eth_type_trans(skb, dev); 638 skb_reset_network_header(skb); 639 640 switch (skb->protocol) { 641 case htons(ETH_P_IP): 642 sk->sk_family = AF_INET; 643 if (sizeof(struct iphdr) <= skb_headlen(skb)) { 644 sk->sk_rcv_saddr = ip_hdr(skb)->saddr; 645 sk->sk_daddr = ip_hdr(skb)->daddr; 646 } 647 break; 648 #if IS_ENABLED(CONFIG_IPV6) 649 case htons(ETH_P_IPV6): 650 sk->sk_family = AF_INET6; 651 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) { 652 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr; 653 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr; 654 } 655 break; 656 #endif 657 default: 658 break; 659 } 660 661 if (is_l2) 662 __skb_push(skb, hh_len); 663 if (is_direct_pkt_access) 664 bpf_compute_data_pointers(skb); 665 ret = convert___skb_to_skb(skb, ctx); 666 if (ret) 667 goto out; 668 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false); 669 if (ret) 670 goto out; 671 if (!is_l2) { 672 if (skb_headroom(skb) < hh_len) { 673 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb)); 674 675 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) { 676 ret = -ENOMEM; 677 goto out; 678 } 679 } 680 memset(__skb_push(skb, hh_len), 0, hh_len); 681 } 682 convert_skb_to___skb(skb, ctx); 683 684 size = skb->len; 685 /* bpf program can never convert linear skb to non-linear */ 686 if (WARN_ON_ONCE(skb_is_nonlinear(skb))) 687 size = skb_headlen(skb); 688 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration); 689 if (!ret) 690 ret = bpf_ctx_finish(kattr, uattr, ctx, 691 sizeof(struct __sk_buff)); 692 out: 693 if (dev && dev != net->loopback_dev) 694 dev_put(dev); 695 kfree_skb(skb); 696 sk_free(sk); 697 kfree(ctx); 698 return ret; 699 } 700 701 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp) 702 { 703 unsigned int ingress_ifindex, rx_queue_index; 704 struct netdev_rx_queue *rxqueue; 705 struct net_device *device; 706 707 if (!xdp_md) 708 return 0; 709 710 if (xdp_md->egress_ifindex != 0) 711 return -EINVAL; 712 713 ingress_ifindex = xdp_md->ingress_ifindex; 714 rx_queue_index = xdp_md->rx_queue_index; 715 716 if (!ingress_ifindex && rx_queue_index) 717 return -EINVAL; 718 719 if (ingress_ifindex) { 720 device = dev_get_by_index(current->nsproxy->net_ns, 721 ingress_ifindex); 722 if (!device) 723 return -ENODEV; 724 725 if (rx_queue_index >= device->real_num_rx_queues) 726 goto free_dev; 727 728 rxqueue = __netif_get_rx_queue(device, rx_queue_index); 729 730 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq)) 731 goto free_dev; 732 733 xdp->rxq = &rxqueue->xdp_rxq; 734 /* The device is now tracked in the xdp->rxq for later 735 * dev_put() 736 */ 737 } 738 739 xdp->data = xdp->data_meta + xdp_md->data; 740 return 0; 741 742 free_dev: 743 dev_put(device); 744 return -EINVAL; 745 } 746 747 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md) 748 { 749 if (!xdp_md) 750 return; 751 752 xdp_md->data = xdp->data - xdp->data_meta; 753 xdp_md->data_end = xdp->data_end - xdp->data_meta; 754 755 if (xdp_md->ingress_ifindex) 756 dev_put(xdp->rxq->dev); 757 } 758 759 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 760 union bpf_attr __user *uattr) 761 { 762 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 763 u32 headroom = XDP_PACKET_HEADROOM; 764 u32 size = kattr->test.data_size_in; 765 u32 repeat = kattr->test.repeat; 766 struct netdev_rx_queue *rxqueue; 767 struct xdp_buff xdp = {}; 768 u32 retval, duration; 769 struct xdp_md *ctx; 770 u32 max_data_sz; 771 void *data; 772 int ret = -EINVAL; 773 774 if (prog->expected_attach_type == BPF_XDP_DEVMAP || 775 prog->expected_attach_type == BPF_XDP_CPUMAP) 776 return -EINVAL; 777 778 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md)); 779 if (IS_ERR(ctx)) 780 return PTR_ERR(ctx); 781 782 if (ctx) { 783 /* There can't be user provided data before the meta data */ 784 if (ctx->data_meta || ctx->data_end != size || 785 ctx->data > ctx->data_end || 786 unlikely(xdp_metalen_invalid(ctx->data))) 787 goto free_ctx; 788 /* Meta data is allocated from the headroom */ 789 headroom -= ctx->data; 790 } 791 792 /* XDP have extra tailroom as (most) drivers use full page */ 793 max_data_sz = 4096 - headroom - tailroom; 794 795 data = bpf_test_init(kattr, max_data_sz, headroom, tailroom); 796 if (IS_ERR(data)) { 797 ret = PTR_ERR(data); 798 goto free_ctx; 799 } 800 801 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0); 802 xdp_init_buff(&xdp, headroom + max_data_sz + tailroom, 803 &rxqueue->xdp_rxq); 804 xdp_prepare_buff(&xdp, data, headroom, size, true); 805 806 ret = xdp_convert_md_to_buff(ctx, &xdp); 807 if (ret) 808 goto free_data; 809 810 if (repeat > 1) 811 bpf_prog_change_xdp(NULL, prog); 812 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true); 813 /* We convert the xdp_buff back to an xdp_md before checking the return 814 * code so the reference count of any held netdevice will be decremented 815 * even if the test run failed. 816 */ 817 xdp_convert_buff_to_md(&xdp, ctx); 818 if (ret) 819 goto out; 820 821 if (xdp.data_meta != data + headroom || 822 xdp.data_end != xdp.data_meta + size) 823 size = xdp.data_end - xdp.data_meta; 824 825 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, size, retval, 826 duration); 827 if (!ret) 828 ret = bpf_ctx_finish(kattr, uattr, ctx, 829 sizeof(struct xdp_md)); 830 831 out: 832 if (repeat > 1) 833 bpf_prog_change_xdp(prog, NULL); 834 free_data: 835 kfree(data); 836 free_ctx: 837 kfree(ctx); 838 return ret; 839 } 840 841 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx) 842 { 843 /* make sure the fields we don't use are zeroed */ 844 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags))) 845 return -EINVAL; 846 847 /* flags is allowed */ 848 849 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags), 850 sizeof(struct bpf_flow_keys))) 851 return -EINVAL; 852 853 return 0; 854 } 855 856 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 857 const union bpf_attr *kattr, 858 union bpf_attr __user *uattr) 859 { 860 struct bpf_test_timer t = { NO_PREEMPT }; 861 u32 size = kattr->test.data_size_in; 862 struct bpf_flow_dissector ctx = {}; 863 u32 repeat = kattr->test.repeat; 864 struct bpf_flow_keys *user_ctx; 865 struct bpf_flow_keys flow_keys; 866 const struct ethhdr *eth; 867 unsigned int flags = 0; 868 u32 retval, duration; 869 void *data; 870 int ret; 871 872 if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR) 873 return -EINVAL; 874 875 if (kattr->test.flags || kattr->test.cpu) 876 return -EINVAL; 877 878 if (size < ETH_HLEN) 879 return -EINVAL; 880 881 data = bpf_test_init(kattr, size, 0, 0); 882 if (IS_ERR(data)) 883 return PTR_ERR(data); 884 885 eth = (struct ethhdr *)data; 886 887 if (!repeat) 888 repeat = 1; 889 890 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys)); 891 if (IS_ERR(user_ctx)) { 892 kfree(data); 893 return PTR_ERR(user_ctx); 894 } 895 if (user_ctx) { 896 ret = verify_user_bpf_flow_keys(user_ctx); 897 if (ret) 898 goto out; 899 flags = user_ctx->flags; 900 } 901 902 ctx.flow_keys = &flow_keys; 903 ctx.data = data; 904 ctx.data_end = (__u8 *)data + size; 905 906 bpf_test_timer_enter(&t); 907 do { 908 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN, 909 size, flags); 910 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration)); 911 bpf_test_timer_leave(&t); 912 913 if (ret < 0) 914 goto out; 915 916 ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys), 917 retval, duration); 918 if (!ret) 919 ret = bpf_ctx_finish(kattr, uattr, user_ctx, 920 sizeof(struct bpf_flow_keys)); 921 922 out: 923 kfree(user_ctx); 924 kfree(data); 925 return ret; 926 } 927 928 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr, 929 union bpf_attr __user *uattr) 930 { 931 struct bpf_test_timer t = { NO_PREEMPT }; 932 struct bpf_prog_array *progs = NULL; 933 struct bpf_sk_lookup_kern ctx = {}; 934 u32 repeat = kattr->test.repeat; 935 struct bpf_sk_lookup *user_ctx; 936 u32 retval, duration; 937 int ret = -EINVAL; 938 939 if (prog->type != BPF_PROG_TYPE_SK_LOOKUP) 940 return -EINVAL; 941 942 if (kattr->test.flags || kattr->test.cpu) 943 return -EINVAL; 944 945 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out || 946 kattr->test.data_size_out) 947 return -EINVAL; 948 949 if (!repeat) 950 repeat = 1; 951 952 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx)); 953 if (IS_ERR(user_ctx)) 954 return PTR_ERR(user_ctx); 955 956 if (!user_ctx) 957 return -EINVAL; 958 959 if (user_ctx->sk) 960 goto out; 961 962 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx))) 963 goto out; 964 965 if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) { 966 ret = -ERANGE; 967 goto out; 968 } 969 970 ctx.family = (u16)user_ctx->family; 971 ctx.protocol = (u16)user_ctx->protocol; 972 ctx.dport = (u16)user_ctx->local_port; 973 ctx.sport = (__force __be16)user_ctx->remote_port; 974 975 switch (ctx.family) { 976 case AF_INET: 977 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4; 978 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4; 979 break; 980 981 #if IS_ENABLED(CONFIG_IPV6) 982 case AF_INET6: 983 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6; 984 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6; 985 break; 986 #endif 987 988 default: 989 ret = -EAFNOSUPPORT; 990 goto out; 991 } 992 993 progs = bpf_prog_array_alloc(1, GFP_KERNEL); 994 if (!progs) { 995 ret = -ENOMEM; 996 goto out; 997 } 998 999 progs->items[0].prog = prog; 1000 1001 bpf_test_timer_enter(&t); 1002 do { 1003 ctx.selected_sk = NULL; 1004 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run); 1005 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration)); 1006 bpf_test_timer_leave(&t); 1007 1008 if (ret < 0) 1009 goto out; 1010 1011 user_ctx->cookie = 0; 1012 if (ctx.selected_sk) { 1013 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) { 1014 ret = -EOPNOTSUPP; 1015 goto out; 1016 } 1017 1018 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk); 1019 } 1020 1021 ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration); 1022 if (!ret) 1023 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx)); 1024 1025 out: 1026 bpf_prog_array_free(progs); 1027 kfree(user_ctx); 1028 return ret; 1029 } 1030 1031 int bpf_prog_test_run_syscall(struct bpf_prog *prog, 1032 const union bpf_attr *kattr, 1033 union bpf_attr __user *uattr) 1034 { 1035 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in); 1036 __u32 ctx_size_in = kattr->test.ctx_size_in; 1037 void *ctx = NULL; 1038 u32 retval; 1039 int err = 0; 1040 1041 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */ 1042 if (kattr->test.data_in || kattr->test.data_out || 1043 kattr->test.ctx_out || kattr->test.duration || 1044 kattr->test.repeat || kattr->test.flags) 1045 return -EINVAL; 1046 1047 if (ctx_size_in < prog->aux->max_ctx_offset || 1048 ctx_size_in > U16_MAX) 1049 return -EINVAL; 1050 1051 if (ctx_size_in) { 1052 ctx = kzalloc(ctx_size_in, GFP_USER); 1053 if (!ctx) 1054 return -ENOMEM; 1055 if (copy_from_user(ctx, ctx_in, ctx_size_in)) { 1056 err = -EFAULT; 1057 goto out; 1058 } 1059 } 1060 1061 rcu_read_lock_trace(); 1062 retval = bpf_prog_run_pin_on_cpu(prog, ctx); 1063 rcu_read_unlock_trace(); 1064 1065 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) { 1066 err = -EFAULT; 1067 goto out; 1068 } 1069 if (ctx_size_in) 1070 if (copy_to_user(ctx_in, ctx, ctx_size_in)) 1071 err = -EFAULT; 1072 out: 1073 kfree(ctx); 1074 return err; 1075 } 1076