1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Copyright (c) 2017 Facebook 3 */ 4 #include <linux/bpf.h> 5 #include <linux/btf.h> 6 #include <linux/btf_ids.h> 7 #include <linux/slab.h> 8 #include <linux/init.h> 9 #include <linux/vmalloc.h> 10 #include <linux/etherdevice.h> 11 #include <linux/filter.h> 12 #include <linux/rcupdate_trace.h> 13 #include <linux/sched/signal.h> 14 #include <net/bpf_sk_storage.h> 15 #include <net/sock.h> 16 #include <net/tcp.h> 17 #include <net/net_namespace.h> 18 #include <net/page_pool.h> 19 #include <linux/error-injection.h> 20 #include <linux/smp.h> 21 #include <linux/sock_diag.h> 22 #include <net/xdp.h> 23 24 #define CREATE_TRACE_POINTS 25 #include <trace/events/bpf_test_run.h> 26 27 struct bpf_test_timer { 28 enum { NO_PREEMPT, NO_MIGRATE } mode; 29 u32 i; 30 u64 time_start, time_spent; 31 }; 32 33 static void bpf_test_timer_enter(struct bpf_test_timer *t) 34 __acquires(rcu) 35 { 36 rcu_read_lock(); 37 if (t->mode == NO_PREEMPT) 38 preempt_disable(); 39 else 40 migrate_disable(); 41 42 t->time_start = ktime_get_ns(); 43 } 44 45 static void bpf_test_timer_leave(struct bpf_test_timer *t) 46 __releases(rcu) 47 { 48 t->time_start = 0; 49 50 if (t->mode == NO_PREEMPT) 51 preempt_enable(); 52 else 53 migrate_enable(); 54 rcu_read_unlock(); 55 } 56 57 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations, 58 u32 repeat, int *err, u32 *duration) 59 __must_hold(rcu) 60 { 61 t->i += iterations; 62 if (t->i >= repeat) { 63 /* We're done. */ 64 t->time_spent += ktime_get_ns() - t->time_start; 65 do_div(t->time_spent, t->i); 66 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent; 67 *err = 0; 68 goto reset; 69 } 70 71 if (signal_pending(current)) { 72 /* During iteration: we've been cancelled, abort. */ 73 *err = -EINTR; 74 goto reset; 75 } 76 77 if (need_resched()) { 78 /* During iteration: we need to reschedule between runs. */ 79 t->time_spent += ktime_get_ns() - t->time_start; 80 bpf_test_timer_leave(t); 81 cond_resched(); 82 bpf_test_timer_enter(t); 83 } 84 85 /* Do another round. */ 86 return true; 87 88 reset: 89 t->i = 0; 90 return false; 91 } 92 93 /* We put this struct at the head of each page with a context and frame 94 * initialised when the page is allocated, so we don't have to do this on each 95 * repetition of the test run. 96 */ 97 struct xdp_page_head { 98 struct xdp_buff orig_ctx; 99 struct xdp_buff ctx; 100 union { 101 /* ::data_hard_start starts here */ 102 DECLARE_FLEX_ARRAY(struct xdp_frame, frame); 103 DECLARE_FLEX_ARRAY(u8, data); 104 }; 105 }; 106 107 struct xdp_test_data { 108 struct xdp_buff *orig_ctx; 109 struct xdp_rxq_info rxq; 110 struct net_device *dev; 111 struct page_pool *pp; 112 struct xdp_frame **frames; 113 struct sk_buff **skbs; 114 struct xdp_mem_info mem; 115 u32 batch_size; 116 u32 frame_cnt; 117 }; 118 119 /* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE 120 * must be updated accordingly this gets changed, otherwise BPF selftests 121 * will fail. 122 */ 123 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head)) 124 #define TEST_XDP_MAX_BATCH 256 125 126 static void xdp_test_run_init_page(struct page *page, void *arg) 127 { 128 struct xdp_page_head *head = phys_to_virt(page_to_phys(page)); 129 struct xdp_buff *new_ctx, *orig_ctx; 130 u32 headroom = XDP_PACKET_HEADROOM; 131 struct xdp_test_data *xdp = arg; 132 size_t frm_len, meta_len; 133 struct xdp_frame *frm; 134 void *data; 135 136 orig_ctx = xdp->orig_ctx; 137 frm_len = orig_ctx->data_end - orig_ctx->data_meta; 138 meta_len = orig_ctx->data - orig_ctx->data_meta; 139 headroom -= meta_len; 140 141 new_ctx = &head->ctx; 142 frm = head->frame; 143 data = head->data; 144 memcpy(data + headroom, orig_ctx->data_meta, frm_len); 145 146 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq); 147 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true); 148 new_ctx->data = new_ctx->data_meta + meta_len; 149 150 xdp_update_frame_from_buff(new_ctx, frm); 151 frm->mem = new_ctx->rxq->mem; 152 153 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx)); 154 } 155 156 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx) 157 { 158 struct page_pool *pp; 159 int err = -ENOMEM; 160 struct page_pool_params pp_params = { 161 .order = 0, 162 .flags = 0, 163 .pool_size = xdp->batch_size, 164 .nid = NUMA_NO_NODE, 165 .init_callback = xdp_test_run_init_page, 166 .init_arg = xdp, 167 }; 168 169 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL); 170 if (!xdp->frames) 171 return -ENOMEM; 172 173 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL); 174 if (!xdp->skbs) 175 goto err_skbs; 176 177 pp = page_pool_create(&pp_params); 178 if (IS_ERR(pp)) { 179 err = PTR_ERR(pp); 180 goto err_pp; 181 } 182 183 /* will copy 'mem.id' into pp->xdp_mem_id */ 184 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp); 185 if (err) 186 goto err_mmodel; 187 188 xdp->pp = pp; 189 190 /* We create a 'fake' RXQ referencing the original dev, but with an 191 * xdp_mem_info pointing to our page_pool 192 */ 193 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0); 194 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL; 195 xdp->rxq.mem.id = pp->xdp_mem_id; 196 xdp->dev = orig_ctx->rxq->dev; 197 xdp->orig_ctx = orig_ctx; 198 199 return 0; 200 201 err_mmodel: 202 page_pool_destroy(pp); 203 err_pp: 204 kvfree(xdp->skbs); 205 err_skbs: 206 kvfree(xdp->frames); 207 return err; 208 } 209 210 static void xdp_test_run_teardown(struct xdp_test_data *xdp) 211 { 212 xdp_unreg_mem_model(&xdp->mem); 213 page_pool_destroy(xdp->pp); 214 kfree(xdp->frames); 215 kfree(xdp->skbs); 216 } 217 218 static bool ctx_was_changed(struct xdp_page_head *head) 219 { 220 return head->orig_ctx.data != head->ctx.data || 221 head->orig_ctx.data_meta != head->ctx.data_meta || 222 head->orig_ctx.data_end != head->ctx.data_end; 223 } 224 225 static void reset_ctx(struct xdp_page_head *head) 226 { 227 if (likely(!ctx_was_changed(head))) 228 return; 229 230 head->ctx.data = head->orig_ctx.data; 231 head->ctx.data_meta = head->orig_ctx.data_meta; 232 head->ctx.data_end = head->orig_ctx.data_end; 233 xdp_update_frame_from_buff(&head->ctx, head->frame); 234 } 235 236 static int xdp_recv_frames(struct xdp_frame **frames, int nframes, 237 struct sk_buff **skbs, 238 struct net_device *dev) 239 { 240 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC; 241 int i, n; 242 LIST_HEAD(list); 243 244 n = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, (void **)skbs); 245 if (unlikely(n == 0)) { 246 for (i = 0; i < nframes; i++) 247 xdp_return_frame(frames[i]); 248 return -ENOMEM; 249 } 250 251 for (i = 0; i < nframes; i++) { 252 struct xdp_frame *xdpf = frames[i]; 253 struct sk_buff *skb = skbs[i]; 254 255 skb = __xdp_build_skb_from_frame(xdpf, skb, dev); 256 if (!skb) { 257 xdp_return_frame(xdpf); 258 continue; 259 } 260 261 list_add_tail(&skb->list, &list); 262 } 263 netif_receive_skb_list(&list); 264 265 return 0; 266 } 267 268 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog, 269 u32 repeat) 270 { 271 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 272 int err = 0, act, ret, i, nframes = 0, batch_sz; 273 struct xdp_frame **frames = xdp->frames; 274 struct xdp_page_head *head; 275 struct xdp_frame *frm; 276 bool redirect = false; 277 struct xdp_buff *ctx; 278 struct page *page; 279 280 batch_sz = min_t(u32, repeat, xdp->batch_size); 281 282 local_bh_disable(); 283 xdp_set_return_frame_no_direct(); 284 285 for (i = 0; i < batch_sz; i++) { 286 page = page_pool_dev_alloc_pages(xdp->pp); 287 if (!page) { 288 err = -ENOMEM; 289 goto out; 290 } 291 292 head = phys_to_virt(page_to_phys(page)); 293 reset_ctx(head); 294 ctx = &head->ctx; 295 frm = head->frame; 296 xdp->frame_cnt++; 297 298 act = bpf_prog_run_xdp(prog, ctx); 299 300 /* if program changed pkt bounds we need to update the xdp_frame */ 301 if (unlikely(ctx_was_changed(head))) { 302 ret = xdp_update_frame_from_buff(ctx, frm); 303 if (ret) { 304 xdp_return_buff(ctx); 305 continue; 306 } 307 } 308 309 switch (act) { 310 case XDP_TX: 311 /* we can't do a real XDP_TX since we're not in the 312 * driver, so turn it into a REDIRECT back to the same 313 * index 314 */ 315 ri->tgt_index = xdp->dev->ifindex; 316 ri->map_id = INT_MAX; 317 ri->map_type = BPF_MAP_TYPE_UNSPEC; 318 fallthrough; 319 case XDP_REDIRECT: 320 redirect = true; 321 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog); 322 if (ret) 323 xdp_return_buff(ctx); 324 break; 325 case XDP_PASS: 326 frames[nframes++] = frm; 327 break; 328 default: 329 bpf_warn_invalid_xdp_action(NULL, prog, act); 330 fallthrough; 331 case XDP_DROP: 332 xdp_return_buff(ctx); 333 break; 334 } 335 } 336 337 out: 338 if (redirect) 339 xdp_do_flush(); 340 if (nframes) { 341 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev); 342 if (ret) 343 err = ret; 344 } 345 346 xdp_clear_return_frame_no_direct(); 347 local_bh_enable(); 348 return err; 349 } 350 351 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx, 352 u32 repeat, u32 batch_size, u32 *time) 353 354 { 355 struct xdp_test_data xdp = { .batch_size = batch_size }; 356 struct bpf_test_timer t = { .mode = NO_MIGRATE }; 357 int ret; 358 359 if (!repeat) 360 repeat = 1; 361 362 ret = xdp_test_run_setup(&xdp, ctx); 363 if (ret) 364 return ret; 365 366 bpf_test_timer_enter(&t); 367 do { 368 xdp.frame_cnt = 0; 369 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i); 370 if (unlikely(ret < 0)) 371 break; 372 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time)); 373 bpf_test_timer_leave(&t); 374 375 xdp_test_run_teardown(&xdp); 376 return ret; 377 } 378 379 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat, 380 u32 *retval, u32 *time, bool xdp) 381 { 382 struct bpf_prog_array_item item = {.prog = prog}; 383 struct bpf_run_ctx *old_ctx; 384 struct bpf_cg_run_ctx run_ctx; 385 struct bpf_test_timer t = { NO_MIGRATE }; 386 enum bpf_cgroup_storage_type stype; 387 int ret; 388 389 for_each_cgroup_storage_type(stype) { 390 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype); 391 if (IS_ERR(item.cgroup_storage[stype])) { 392 item.cgroup_storage[stype] = NULL; 393 for_each_cgroup_storage_type(stype) 394 bpf_cgroup_storage_free(item.cgroup_storage[stype]); 395 return -ENOMEM; 396 } 397 } 398 399 if (!repeat) 400 repeat = 1; 401 402 bpf_test_timer_enter(&t); 403 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 404 do { 405 run_ctx.prog_item = &item; 406 local_bh_disable(); 407 if (xdp) 408 *retval = bpf_prog_run_xdp(prog, ctx); 409 else 410 *retval = bpf_prog_run(prog, ctx); 411 local_bh_enable(); 412 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time)); 413 bpf_reset_run_ctx(old_ctx); 414 bpf_test_timer_leave(&t); 415 416 for_each_cgroup_storage_type(stype) 417 bpf_cgroup_storage_free(item.cgroup_storage[stype]); 418 419 return ret; 420 } 421 422 static int bpf_test_finish(const union bpf_attr *kattr, 423 union bpf_attr __user *uattr, const void *data, 424 struct skb_shared_info *sinfo, u32 size, 425 u32 retval, u32 duration) 426 { 427 void __user *data_out = u64_to_user_ptr(kattr->test.data_out); 428 int err = -EFAULT; 429 u32 copy_size = size; 430 431 /* Clamp copy if the user has provided a size hint, but copy the full 432 * buffer if not to retain old behaviour. 433 */ 434 if (kattr->test.data_size_out && 435 copy_size > kattr->test.data_size_out) { 436 copy_size = kattr->test.data_size_out; 437 err = -ENOSPC; 438 } 439 440 if (data_out) { 441 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size; 442 443 if (len < 0) { 444 err = -ENOSPC; 445 goto out; 446 } 447 448 if (copy_to_user(data_out, data, len)) 449 goto out; 450 451 if (sinfo) { 452 int i, offset = len; 453 u32 data_len; 454 455 for (i = 0; i < sinfo->nr_frags; i++) { 456 skb_frag_t *frag = &sinfo->frags[i]; 457 458 if (offset >= copy_size) { 459 err = -ENOSPC; 460 break; 461 } 462 463 data_len = min_t(u32, copy_size - offset, 464 skb_frag_size(frag)); 465 466 if (copy_to_user(data_out + offset, 467 skb_frag_address(frag), 468 data_len)) 469 goto out; 470 471 offset += data_len; 472 } 473 } 474 } 475 476 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size))) 477 goto out; 478 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval))) 479 goto out; 480 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration))) 481 goto out; 482 if (err != -ENOSPC) 483 err = 0; 484 out: 485 trace_bpf_test_finish(&err); 486 return err; 487 } 488 489 /* Integer types of various sizes and pointer combinations cover variety of 490 * architecture dependent calling conventions. 7+ can be supported in the 491 * future. 492 */ 493 __diag_push(); 494 __diag_ignore_all("-Wmissing-prototypes", 495 "Global functions as their definitions will be in vmlinux BTF"); 496 __bpf_kfunc int bpf_fentry_test1(int a) 497 { 498 return a + 1; 499 } 500 EXPORT_SYMBOL_GPL(bpf_fentry_test1); 501 502 int noinline bpf_fentry_test2(int a, u64 b) 503 { 504 return a + b; 505 } 506 507 int noinline bpf_fentry_test3(char a, int b, u64 c) 508 { 509 return a + b + c; 510 } 511 512 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d) 513 { 514 return (long)a + b + c + d; 515 } 516 517 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e) 518 { 519 return a + (long)b + c + d + e; 520 } 521 522 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f) 523 { 524 return a + (long)b + c + d + (long)e + f; 525 } 526 527 struct bpf_fentry_test_t { 528 struct bpf_fentry_test_t *a; 529 }; 530 531 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg) 532 { 533 return (long)arg; 534 } 535 536 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg) 537 { 538 return (long)arg->a; 539 } 540 541 __bpf_kfunc int bpf_modify_return_test(int a, int *b) 542 { 543 *b += 1; 544 return a + *b; 545 } 546 547 __bpf_kfunc u64 bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d) 548 { 549 return a + b + c + d; 550 } 551 552 __bpf_kfunc int bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b) 553 { 554 return a + b; 555 } 556 557 __bpf_kfunc struct sock *bpf_kfunc_call_test3(struct sock *sk) 558 { 559 return sk; 560 } 561 562 long noinline bpf_kfunc_call_test4(signed char a, short b, int c, long d) 563 { 564 /* Provoke the compiler to assume that the caller has sign-extended a, 565 * b and c on platforms where this is required (e.g. s390x). 566 */ 567 return (long)a + (long)b + (long)c + d; 568 } 569 570 struct prog_test_member1 { 571 int a; 572 }; 573 574 struct prog_test_member { 575 struct prog_test_member1 m; 576 int c; 577 }; 578 579 struct prog_test_ref_kfunc { 580 int a; 581 int b; 582 struct prog_test_member memb; 583 struct prog_test_ref_kfunc *next; 584 refcount_t cnt; 585 }; 586 587 static struct prog_test_ref_kfunc prog_test_struct = { 588 .a = 42, 589 .b = 108, 590 .next = &prog_test_struct, 591 .cnt = REFCOUNT_INIT(1), 592 }; 593 594 __bpf_kfunc struct prog_test_ref_kfunc * 595 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr) 596 { 597 refcount_inc(&prog_test_struct.cnt); 598 return &prog_test_struct; 599 } 600 601 __bpf_kfunc struct prog_test_member * 602 bpf_kfunc_call_memb_acquire(void) 603 { 604 WARN_ON_ONCE(1); 605 return NULL; 606 } 607 608 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p) 609 { 610 if (!p) 611 return; 612 613 refcount_dec(&p->cnt); 614 } 615 616 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p) 617 { 618 } 619 620 __bpf_kfunc void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p) 621 { 622 WARN_ON_ONCE(1); 623 } 624 625 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size) 626 { 627 if (size > 2 * sizeof(int)) 628 return NULL; 629 630 return (int *)p; 631 } 632 633 __bpf_kfunc int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p, 634 const int rdwr_buf_size) 635 { 636 return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size); 637 } 638 639 __bpf_kfunc int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p, 640 const int rdonly_buf_size) 641 { 642 return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size); 643 } 644 645 /* the next 2 ones can't be really used for testing expect to ensure 646 * that the verifier rejects the call. 647 * Acquire functions must return struct pointers, so these ones are 648 * failing. 649 */ 650 __bpf_kfunc int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p, 651 const int rdonly_buf_size) 652 { 653 return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size); 654 } 655 656 __bpf_kfunc void bpf_kfunc_call_int_mem_release(int *p) 657 { 658 } 659 660 __bpf_kfunc struct prog_test_ref_kfunc * 661 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b) 662 { 663 struct prog_test_ref_kfunc *p = READ_ONCE(*pp); 664 665 if (!p) 666 return NULL; 667 refcount_inc(&p->cnt); 668 return p; 669 } 670 671 struct prog_test_pass1 { 672 int x0; 673 struct { 674 int x1; 675 struct { 676 int x2; 677 struct { 678 int x3; 679 }; 680 }; 681 }; 682 }; 683 684 struct prog_test_pass2 { 685 int len; 686 short arr1[4]; 687 struct { 688 char arr2[4]; 689 unsigned long arr3[8]; 690 } x; 691 }; 692 693 struct prog_test_fail1 { 694 void *p; 695 int x; 696 }; 697 698 struct prog_test_fail2 { 699 int x8; 700 struct prog_test_pass1 x; 701 }; 702 703 struct prog_test_fail3 { 704 int len; 705 char arr1[2]; 706 char arr2[]; 707 }; 708 709 __bpf_kfunc void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb) 710 { 711 } 712 713 __bpf_kfunc void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p) 714 { 715 } 716 717 __bpf_kfunc void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p) 718 { 719 } 720 721 __bpf_kfunc void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p) 722 { 723 } 724 725 __bpf_kfunc void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p) 726 { 727 } 728 729 __bpf_kfunc void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p) 730 { 731 } 732 733 __bpf_kfunc void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz) 734 { 735 } 736 737 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len) 738 { 739 } 740 741 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len) 742 { 743 } 744 745 __bpf_kfunc void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p) 746 { 747 } 748 749 __bpf_kfunc void bpf_kfunc_call_test_destructive(void) 750 { 751 } 752 753 __bpf_kfunc static u32 bpf_kfunc_call_test_static_unused_arg(u32 arg, u32 unused) 754 { 755 return arg; 756 } 757 758 __diag_pop(); 759 760 BTF_SET8_START(bpf_test_modify_return_ids) 761 BTF_ID_FLAGS(func, bpf_modify_return_test) 762 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE) 763 BTF_SET8_END(bpf_test_modify_return_ids) 764 765 static const struct btf_kfunc_id_set bpf_test_modify_return_set = { 766 .owner = THIS_MODULE, 767 .set = &bpf_test_modify_return_ids, 768 }; 769 770 BTF_SET8_START(test_sk_check_kfunc_ids) 771 BTF_ID_FLAGS(func, bpf_kfunc_call_test1) 772 BTF_ID_FLAGS(func, bpf_kfunc_call_test2) 773 BTF_ID_FLAGS(func, bpf_kfunc_call_test3) 774 BTF_ID_FLAGS(func, bpf_kfunc_call_test4) 775 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL) 776 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL) 777 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE) 778 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE) 779 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE) 780 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL) 781 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL) 782 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL) 783 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE) 784 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET) 785 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx) 786 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1) 787 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2) 788 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1) 789 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2) 790 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3) 791 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1) 792 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1) 793 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2) 794 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS) 795 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE) 796 BTF_ID_FLAGS(func, bpf_kfunc_call_test_static_unused_arg) 797 BTF_SET8_END(test_sk_check_kfunc_ids) 798 799 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size, 800 u32 size, u32 headroom, u32 tailroom) 801 { 802 void __user *data_in = u64_to_user_ptr(kattr->test.data_in); 803 void *data; 804 805 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom) 806 return ERR_PTR(-EINVAL); 807 808 if (user_size > size) 809 return ERR_PTR(-EMSGSIZE); 810 811 size = SKB_DATA_ALIGN(size); 812 data = kzalloc(size + headroom + tailroom, GFP_USER); 813 if (!data) 814 return ERR_PTR(-ENOMEM); 815 816 if (copy_from_user(data + headroom, data_in, user_size)) { 817 kfree(data); 818 return ERR_PTR(-EFAULT); 819 } 820 821 return data; 822 } 823 824 int bpf_prog_test_run_tracing(struct bpf_prog *prog, 825 const union bpf_attr *kattr, 826 union bpf_attr __user *uattr) 827 { 828 struct bpf_fentry_test_t arg = {}; 829 u16 side_effect = 0, ret = 0; 830 int b = 2, err = -EFAULT; 831 u32 retval = 0; 832 833 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 834 return -EINVAL; 835 836 switch (prog->expected_attach_type) { 837 case BPF_TRACE_FENTRY: 838 case BPF_TRACE_FEXIT: 839 if (bpf_fentry_test1(1) != 2 || 840 bpf_fentry_test2(2, 3) != 5 || 841 bpf_fentry_test3(4, 5, 6) != 15 || 842 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 || 843 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 || 844 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 || 845 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 || 846 bpf_fentry_test8(&arg) != 0) 847 goto out; 848 break; 849 case BPF_MODIFY_RETURN: 850 ret = bpf_modify_return_test(1, &b); 851 if (b != 2) 852 side_effect = 1; 853 break; 854 default: 855 goto out; 856 } 857 858 retval = ((u32)side_effect << 16) | ret; 859 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval))) 860 goto out; 861 862 err = 0; 863 out: 864 trace_bpf_test_finish(&err); 865 return err; 866 } 867 868 struct bpf_raw_tp_test_run_info { 869 struct bpf_prog *prog; 870 void *ctx; 871 u32 retval; 872 }; 873 874 static void 875 __bpf_prog_test_run_raw_tp(void *data) 876 { 877 struct bpf_raw_tp_test_run_info *info = data; 878 879 rcu_read_lock(); 880 info->retval = bpf_prog_run(info->prog, info->ctx); 881 rcu_read_unlock(); 882 } 883 884 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 885 const union bpf_attr *kattr, 886 union bpf_attr __user *uattr) 887 { 888 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in); 889 __u32 ctx_size_in = kattr->test.ctx_size_in; 890 struct bpf_raw_tp_test_run_info info; 891 int cpu = kattr->test.cpu, err = 0; 892 int current_cpu; 893 894 /* doesn't support data_in/out, ctx_out, duration, or repeat */ 895 if (kattr->test.data_in || kattr->test.data_out || 896 kattr->test.ctx_out || kattr->test.duration || 897 kattr->test.repeat || kattr->test.batch_size) 898 return -EINVAL; 899 900 if (ctx_size_in < prog->aux->max_ctx_offset || 901 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64)) 902 return -EINVAL; 903 904 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0) 905 return -EINVAL; 906 907 if (ctx_size_in) { 908 info.ctx = memdup_user(ctx_in, ctx_size_in); 909 if (IS_ERR(info.ctx)) 910 return PTR_ERR(info.ctx); 911 } else { 912 info.ctx = NULL; 913 } 914 915 info.prog = prog; 916 917 current_cpu = get_cpu(); 918 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 || 919 cpu == current_cpu) { 920 __bpf_prog_test_run_raw_tp(&info); 921 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) { 922 /* smp_call_function_single() also checks cpu_online() 923 * after csd_lock(). However, since cpu is from user 924 * space, let's do an extra quick check to filter out 925 * invalid value before smp_call_function_single(). 926 */ 927 err = -ENXIO; 928 } else { 929 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp, 930 &info, 1); 931 } 932 put_cpu(); 933 934 if (!err && 935 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32))) 936 err = -EFAULT; 937 938 kfree(info.ctx); 939 return err; 940 } 941 942 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size) 943 { 944 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in); 945 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out); 946 u32 size = kattr->test.ctx_size_in; 947 void *data; 948 int err; 949 950 if (!data_in && !data_out) 951 return NULL; 952 953 data = kzalloc(max_size, GFP_USER); 954 if (!data) 955 return ERR_PTR(-ENOMEM); 956 957 if (data_in) { 958 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size); 959 if (err) { 960 kfree(data); 961 return ERR_PTR(err); 962 } 963 964 size = min_t(u32, max_size, size); 965 if (copy_from_user(data, data_in, size)) { 966 kfree(data); 967 return ERR_PTR(-EFAULT); 968 } 969 } 970 return data; 971 } 972 973 static int bpf_ctx_finish(const union bpf_attr *kattr, 974 union bpf_attr __user *uattr, const void *data, 975 u32 size) 976 { 977 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out); 978 int err = -EFAULT; 979 u32 copy_size = size; 980 981 if (!data || !data_out) 982 return 0; 983 984 if (copy_size > kattr->test.ctx_size_out) { 985 copy_size = kattr->test.ctx_size_out; 986 err = -ENOSPC; 987 } 988 989 if (copy_to_user(data_out, data, copy_size)) 990 goto out; 991 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size))) 992 goto out; 993 if (err != -ENOSPC) 994 err = 0; 995 out: 996 return err; 997 } 998 999 /** 1000 * range_is_zero - test whether buffer is initialized 1001 * @buf: buffer to check 1002 * @from: check from this position 1003 * @to: check up until (excluding) this position 1004 * 1005 * This function returns true if the there is a non-zero byte 1006 * in the buf in the range [from,to). 1007 */ 1008 static inline bool range_is_zero(void *buf, size_t from, size_t to) 1009 { 1010 return !memchr_inv((u8 *)buf + from, 0, to - from); 1011 } 1012 1013 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) 1014 { 1015 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; 1016 1017 if (!__skb) 1018 return 0; 1019 1020 /* make sure the fields we don't use are zeroed */ 1021 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark))) 1022 return -EINVAL; 1023 1024 /* mark is allowed */ 1025 1026 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark), 1027 offsetof(struct __sk_buff, priority))) 1028 return -EINVAL; 1029 1030 /* priority is allowed */ 1031 /* ingress_ifindex is allowed */ 1032 /* ifindex is allowed */ 1033 1034 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex), 1035 offsetof(struct __sk_buff, cb))) 1036 return -EINVAL; 1037 1038 /* cb is allowed */ 1039 1040 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb), 1041 offsetof(struct __sk_buff, tstamp))) 1042 return -EINVAL; 1043 1044 /* tstamp is allowed */ 1045 /* wire_len is allowed */ 1046 /* gso_segs is allowed */ 1047 1048 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs), 1049 offsetof(struct __sk_buff, gso_size))) 1050 return -EINVAL; 1051 1052 /* gso_size is allowed */ 1053 1054 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size), 1055 offsetof(struct __sk_buff, hwtstamp))) 1056 return -EINVAL; 1057 1058 /* hwtstamp is allowed */ 1059 1060 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp), 1061 sizeof(struct __sk_buff))) 1062 return -EINVAL; 1063 1064 skb->mark = __skb->mark; 1065 skb->priority = __skb->priority; 1066 skb->skb_iif = __skb->ingress_ifindex; 1067 skb->tstamp = __skb->tstamp; 1068 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN); 1069 1070 if (__skb->wire_len == 0) { 1071 cb->pkt_len = skb->len; 1072 } else { 1073 if (__skb->wire_len < skb->len || 1074 __skb->wire_len > GSO_LEGACY_MAX_SIZE) 1075 return -EINVAL; 1076 cb->pkt_len = __skb->wire_len; 1077 } 1078 1079 if (__skb->gso_segs > GSO_MAX_SEGS) 1080 return -EINVAL; 1081 skb_shinfo(skb)->gso_segs = __skb->gso_segs; 1082 skb_shinfo(skb)->gso_size = __skb->gso_size; 1083 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp; 1084 1085 return 0; 1086 } 1087 1088 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb) 1089 { 1090 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; 1091 1092 if (!__skb) 1093 return; 1094 1095 __skb->mark = skb->mark; 1096 __skb->priority = skb->priority; 1097 __skb->ingress_ifindex = skb->skb_iif; 1098 __skb->ifindex = skb->dev->ifindex; 1099 __skb->tstamp = skb->tstamp; 1100 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN); 1101 __skb->wire_len = cb->pkt_len; 1102 __skb->gso_segs = skb_shinfo(skb)->gso_segs; 1103 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp; 1104 } 1105 1106 static struct proto bpf_dummy_proto = { 1107 .name = "bpf_dummy", 1108 .owner = THIS_MODULE, 1109 .obj_size = sizeof(struct sock), 1110 }; 1111 1112 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 1113 union bpf_attr __user *uattr) 1114 { 1115 bool is_l2 = false, is_direct_pkt_access = false; 1116 struct net *net = current->nsproxy->net_ns; 1117 struct net_device *dev = net->loopback_dev; 1118 u32 size = kattr->test.data_size_in; 1119 u32 repeat = kattr->test.repeat; 1120 struct __sk_buff *ctx = NULL; 1121 u32 retval, duration; 1122 int hh_len = ETH_HLEN; 1123 struct sk_buff *skb; 1124 struct sock *sk; 1125 void *data; 1126 int ret; 1127 1128 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 1129 return -EINVAL; 1130 1131 data = bpf_test_init(kattr, kattr->test.data_size_in, 1132 size, NET_SKB_PAD + NET_IP_ALIGN, 1133 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))); 1134 if (IS_ERR(data)) 1135 return PTR_ERR(data); 1136 1137 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff)); 1138 if (IS_ERR(ctx)) { 1139 kfree(data); 1140 return PTR_ERR(ctx); 1141 } 1142 1143 switch (prog->type) { 1144 case BPF_PROG_TYPE_SCHED_CLS: 1145 case BPF_PROG_TYPE_SCHED_ACT: 1146 is_l2 = true; 1147 fallthrough; 1148 case BPF_PROG_TYPE_LWT_IN: 1149 case BPF_PROG_TYPE_LWT_OUT: 1150 case BPF_PROG_TYPE_LWT_XMIT: 1151 is_direct_pkt_access = true; 1152 break; 1153 default: 1154 break; 1155 } 1156 1157 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1); 1158 if (!sk) { 1159 kfree(data); 1160 kfree(ctx); 1161 return -ENOMEM; 1162 } 1163 sock_init_data(NULL, sk); 1164 1165 skb = slab_build_skb(data); 1166 if (!skb) { 1167 kfree(data); 1168 kfree(ctx); 1169 sk_free(sk); 1170 return -ENOMEM; 1171 } 1172 skb->sk = sk; 1173 1174 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 1175 __skb_put(skb, size); 1176 if (ctx && ctx->ifindex > 1) { 1177 dev = dev_get_by_index(net, ctx->ifindex); 1178 if (!dev) { 1179 ret = -ENODEV; 1180 goto out; 1181 } 1182 } 1183 skb->protocol = eth_type_trans(skb, dev); 1184 skb_reset_network_header(skb); 1185 1186 switch (skb->protocol) { 1187 case htons(ETH_P_IP): 1188 sk->sk_family = AF_INET; 1189 if (sizeof(struct iphdr) <= skb_headlen(skb)) { 1190 sk->sk_rcv_saddr = ip_hdr(skb)->saddr; 1191 sk->sk_daddr = ip_hdr(skb)->daddr; 1192 } 1193 break; 1194 #if IS_ENABLED(CONFIG_IPV6) 1195 case htons(ETH_P_IPV6): 1196 sk->sk_family = AF_INET6; 1197 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) { 1198 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr; 1199 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr; 1200 } 1201 break; 1202 #endif 1203 default: 1204 break; 1205 } 1206 1207 if (is_l2) 1208 __skb_push(skb, hh_len); 1209 if (is_direct_pkt_access) 1210 bpf_compute_data_pointers(skb); 1211 ret = convert___skb_to_skb(skb, ctx); 1212 if (ret) 1213 goto out; 1214 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false); 1215 if (ret) 1216 goto out; 1217 if (!is_l2) { 1218 if (skb_headroom(skb) < hh_len) { 1219 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb)); 1220 1221 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) { 1222 ret = -ENOMEM; 1223 goto out; 1224 } 1225 } 1226 memset(__skb_push(skb, hh_len), 0, hh_len); 1227 } 1228 convert_skb_to___skb(skb, ctx); 1229 1230 size = skb->len; 1231 /* bpf program can never convert linear skb to non-linear */ 1232 if (WARN_ON_ONCE(skb_is_nonlinear(skb))) 1233 size = skb_headlen(skb); 1234 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval, 1235 duration); 1236 if (!ret) 1237 ret = bpf_ctx_finish(kattr, uattr, ctx, 1238 sizeof(struct __sk_buff)); 1239 out: 1240 if (dev && dev != net->loopback_dev) 1241 dev_put(dev); 1242 kfree_skb(skb); 1243 sk_free(sk); 1244 kfree(ctx); 1245 return ret; 1246 } 1247 1248 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp) 1249 { 1250 unsigned int ingress_ifindex, rx_queue_index; 1251 struct netdev_rx_queue *rxqueue; 1252 struct net_device *device; 1253 1254 if (!xdp_md) 1255 return 0; 1256 1257 if (xdp_md->egress_ifindex != 0) 1258 return -EINVAL; 1259 1260 ingress_ifindex = xdp_md->ingress_ifindex; 1261 rx_queue_index = xdp_md->rx_queue_index; 1262 1263 if (!ingress_ifindex && rx_queue_index) 1264 return -EINVAL; 1265 1266 if (ingress_ifindex) { 1267 device = dev_get_by_index(current->nsproxy->net_ns, 1268 ingress_ifindex); 1269 if (!device) 1270 return -ENODEV; 1271 1272 if (rx_queue_index >= device->real_num_rx_queues) 1273 goto free_dev; 1274 1275 rxqueue = __netif_get_rx_queue(device, rx_queue_index); 1276 1277 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq)) 1278 goto free_dev; 1279 1280 xdp->rxq = &rxqueue->xdp_rxq; 1281 /* The device is now tracked in the xdp->rxq for later 1282 * dev_put() 1283 */ 1284 } 1285 1286 xdp->data = xdp->data_meta + xdp_md->data; 1287 return 0; 1288 1289 free_dev: 1290 dev_put(device); 1291 return -EINVAL; 1292 } 1293 1294 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md) 1295 { 1296 if (!xdp_md) 1297 return; 1298 1299 xdp_md->data = xdp->data - xdp->data_meta; 1300 xdp_md->data_end = xdp->data_end - xdp->data_meta; 1301 1302 if (xdp_md->ingress_ifindex) 1303 dev_put(xdp->rxq->dev); 1304 } 1305 1306 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 1307 union bpf_attr __user *uattr) 1308 { 1309 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES); 1310 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1311 u32 batch_size = kattr->test.batch_size; 1312 u32 retval = 0, duration, max_data_sz; 1313 u32 size = kattr->test.data_size_in; 1314 u32 headroom = XDP_PACKET_HEADROOM; 1315 u32 repeat = kattr->test.repeat; 1316 struct netdev_rx_queue *rxqueue; 1317 struct skb_shared_info *sinfo; 1318 struct xdp_buff xdp = {}; 1319 int i, ret = -EINVAL; 1320 struct xdp_md *ctx; 1321 void *data; 1322 1323 if (prog->expected_attach_type == BPF_XDP_DEVMAP || 1324 prog->expected_attach_type == BPF_XDP_CPUMAP) 1325 return -EINVAL; 1326 1327 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES) 1328 return -EINVAL; 1329 1330 if (bpf_prog_is_dev_bound(prog->aux)) 1331 return -EINVAL; 1332 1333 if (do_live) { 1334 if (!batch_size) 1335 batch_size = NAPI_POLL_WEIGHT; 1336 else if (batch_size > TEST_XDP_MAX_BATCH) 1337 return -E2BIG; 1338 1339 headroom += sizeof(struct xdp_page_head); 1340 } else if (batch_size) { 1341 return -EINVAL; 1342 } 1343 1344 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md)); 1345 if (IS_ERR(ctx)) 1346 return PTR_ERR(ctx); 1347 1348 if (ctx) { 1349 /* There can't be user provided data before the meta data */ 1350 if (ctx->data_meta || ctx->data_end != size || 1351 ctx->data > ctx->data_end || 1352 unlikely(xdp_metalen_invalid(ctx->data)) || 1353 (do_live && (kattr->test.data_out || kattr->test.ctx_out))) 1354 goto free_ctx; 1355 /* Meta data is allocated from the headroom */ 1356 headroom -= ctx->data; 1357 } 1358 1359 max_data_sz = 4096 - headroom - tailroom; 1360 if (size > max_data_sz) { 1361 /* disallow live data mode for jumbo frames */ 1362 if (do_live) 1363 goto free_ctx; 1364 size = max_data_sz; 1365 } 1366 1367 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom); 1368 if (IS_ERR(data)) { 1369 ret = PTR_ERR(data); 1370 goto free_ctx; 1371 } 1372 1373 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0); 1374 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom; 1375 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq); 1376 xdp_prepare_buff(&xdp, data, headroom, size, true); 1377 sinfo = xdp_get_shared_info_from_buff(&xdp); 1378 1379 ret = xdp_convert_md_to_buff(ctx, &xdp); 1380 if (ret) 1381 goto free_data; 1382 1383 if (unlikely(kattr->test.data_size_in > size)) { 1384 void __user *data_in = u64_to_user_ptr(kattr->test.data_in); 1385 1386 while (size < kattr->test.data_size_in) { 1387 struct page *page; 1388 skb_frag_t *frag; 1389 u32 data_len; 1390 1391 if (sinfo->nr_frags == MAX_SKB_FRAGS) { 1392 ret = -ENOMEM; 1393 goto out; 1394 } 1395 1396 page = alloc_page(GFP_KERNEL); 1397 if (!page) { 1398 ret = -ENOMEM; 1399 goto out; 1400 } 1401 1402 frag = &sinfo->frags[sinfo->nr_frags++]; 1403 __skb_frag_set_page(frag, page); 1404 1405 data_len = min_t(u32, kattr->test.data_size_in - size, 1406 PAGE_SIZE); 1407 skb_frag_size_set(frag, data_len); 1408 1409 if (copy_from_user(page_address(page), data_in + size, 1410 data_len)) { 1411 ret = -EFAULT; 1412 goto out; 1413 } 1414 sinfo->xdp_frags_size += data_len; 1415 size += data_len; 1416 } 1417 xdp_buff_set_frags_flag(&xdp); 1418 } 1419 1420 if (repeat > 1) 1421 bpf_prog_change_xdp(NULL, prog); 1422 1423 if (do_live) 1424 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration); 1425 else 1426 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true); 1427 /* We convert the xdp_buff back to an xdp_md before checking the return 1428 * code so the reference count of any held netdevice will be decremented 1429 * even if the test run failed. 1430 */ 1431 xdp_convert_buff_to_md(&xdp, ctx); 1432 if (ret) 1433 goto out; 1434 1435 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size; 1436 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size, 1437 retval, duration); 1438 if (!ret) 1439 ret = bpf_ctx_finish(kattr, uattr, ctx, 1440 sizeof(struct xdp_md)); 1441 1442 out: 1443 if (repeat > 1) 1444 bpf_prog_change_xdp(prog, NULL); 1445 free_data: 1446 for (i = 0; i < sinfo->nr_frags; i++) 1447 __free_page(skb_frag_page(&sinfo->frags[i])); 1448 kfree(data); 1449 free_ctx: 1450 kfree(ctx); 1451 return ret; 1452 } 1453 1454 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx) 1455 { 1456 /* make sure the fields we don't use are zeroed */ 1457 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags))) 1458 return -EINVAL; 1459 1460 /* flags is allowed */ 1461 1462 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags), 1463 sizeof(struct bpf_flow_keys))) 1464 return -EINVAL; 1465 1466 return 0; 1467 } 1468 1469 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1470 const union bpf_attr *kattr, 1471 union bpf_attr __user *uattr) 1472 { 1473 struct bpf_test_timer t = { NO_PREEMPT }; 1474 u32 size = kattr->test.data_size_in; 1475 struct bpf_flow_dissector ctx = {}; 1476 u32 repeat = kattr->test.repeat; 1477 struct bpf_flow_keys *user_ctx; 1478 struct bpf_flow_keys flow_keys; 1479 const struct ethhdr *eth; 1480 unsigned int flags = 0; 1481 u32 retval, duration; 1482 void *data; 1483 int ret; 1484 1485 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 1486 return -EINVAL; 1487 1488 if (size < ETH_HLEN) 1489 return -EINVAL; 1490 1491 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0); 1492 if (IS_ERR(data)) 1493 return PTR_ERR(data); 1494 1495 eth = (struct ethhdr *)data; 1496 1497 if (!repeat) 1498 repeat = 1; 1499 1500 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys)); 1501 if (IS_ERR(user_ctx)) { 1502 kfree(data); 1503 return PTR_ERR(user_ctx); 1504 } 1505 if (user_ctx) { 1506 ret = verify_user_bpf_flow_keys(user_ctx); 1507 if (ret) 1508 goto out; 1509 flags = user_ctx->flags; 1510 } 1511 1512 ctx.flow_keys = &flow_keys; 1513 ctx.data = data; 1514 ctx.data_end = (__u8 *)data + size; 1515 1516 bpf_test_timer_enter(&t); 1517 do { 1518 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN, 1519 size, flags); 1520 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration)); 1521 bpf_test_timer_leave(&t); 1522 1523 if (ret < 0) 1524 goto out; 1525 1526 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL, 1527 sizeof(flow_keys), retval, duration); 1528 if (!ret) 1529 ret = bpf_ctx_finish(kattr, uattr, user_ctx, 1530 sizeof(struct bpf_flow_keys)); 1531 1532 out: 1533 kfree(user_ctx); 1534 kfree(data); 1535 return ret; 1536 } 1537 1538 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr, 1539 union bpf_attr __user *uattr) 1540 { 1541 struct bpf_test_timer t = { NO_PREEMPT }; 1542 struct bpf_prog_array *progs = NULL; 1543 struct bpf_sk_lookup_kern ctx = {}; 1544 u32 repeat = kattr->test.repeat; 1545 struct bpf_sk_lookup *user_ctx; 1546 u32 retval, duration; 1547 int ret = -EINVAL; 1548 1549 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 1550 return -EINVAL; 1551 1552 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out || 1553 kattr->test.data_size_out) 1554 return -EINVAL; 1555 1556 if (!repeat) 1557 repeat = 1; 1558 1559 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx)); 1560 if (IS_ERR(user_ctx)) 1561 return PTR_ERR(user_ctx); 1562 1563 if (!user_ctx) 1564 return -EINVAL; 1565 1566 if (user_ctx->sk) 1567 goto out; 1568 1569 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx))) 1570 goto out; 1571 1572 if (user_ctx->local_port > U16_MAX) { 1573 ret = -ERANGE; 1574 goto out; 1575 } 1576 1577 ctx.family = (u16)user_ctx->family; 1578 ctx.protocol = (u16)user_ctx->protocol; 1579 ctx.dport = (u16)user_ctx->local_port; 1580 ctx.sport = user_ctx->remote_port; 1581 1582 switch (ctx.family) { 1583 case AF_INET: 1584 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4; 1585 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4; 1586 break; 1587 1588 #if IS_ENABLED(CONFIG_IPV6) 1589 case AF_INET6: 1590 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6; 1591 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6; 1592 break; 1593 #endif 1594 1595 default: 1596 ret = -EAFNOSUPPORT; 1597 goto out; 1598 } 1599 1600 progs = bpf_prog_array_alloc(1, GFP_KERNEL); 1601 if (!progs) { 1602 ret = -ENOMEM; 1603 goto out; 1604 } 1605 1606 progs->items[0].prog = prog; 1607 1608 bpf_test_timer_enter(&t); 1609 do { 1610 ctx.selected_sk = NULL; 1611 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run); 1612 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration)); 1613 bpf_test_timer_leave(&t); 1614 1615 if (ret < 0) 1616 goto out; 1617 1618 user_ctx->cookie = 0; 1619 if (ctx.selected_sk) { 1620 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) { 1621 ret = -EOPNOTSUPP; 1622 goto out; 1623 } 1624 1625 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk); 1626 } 1627 1628 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration); 1629 if (!ret) 1630 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx)); 1631 1632 out: 1633 bpf_prog_array_free(progs); 1634 kfree(user_ctx); 1635 return ret; 1636 } 1637 1638 int bpf_prog_test_run_syscall(struct bpf_prog *prog, 1639 const union bpf_attr *kattr, 1640 union bpf_attr __user *uattr) 1641 { 1642 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in); 1643 __u32 ctx_size_in = kattr->test.ctx_size_in; 1644 void *ctx = NULL; 1645 u32 retval; 1646 int err = 0; 1647 1648 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */ 1649 if (kattr->test.data_in || kattr->test.data_out || 1650 kattr->test.ctx_out || kattr->test.duration || 1651 kattr->test.repeat || kattr->test.flags || 1652 kattr->test.batch_size) 1653 return -EINVAL; 1654 1655 if (ctx_size_in < prog->aux->max_ctx_offset || 1656 ctx_size_in > U16_MAX) 1657 return -EINVAL; 1658 1659 if (ctx_size_in) { 1660 ctx = memdup_user(ctx_in, ctx_size_in); 1661 if (IS_ERR(ctx)) 1662 return PTR_ERR(ctx); 1663 } 1664 1665 rcu_read_lock_trace(); 1666 retval = bpf_prog_run_pin_on_cpu(prog, ctx); 1667 rcu_read_unlock_trace(); 1668 1669 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) { 1670 err = -EFAULT; 1671 goto out; 1672 } 1673 if (ctx_size_in) 1674 if (copy_to_user(ctx_in, ctx, ctx_size_in)) 1675 err = -EFAULT; 1676 out: 1677 kfree(ctx); 1678 return err; 1679 } 1680 1681 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = { 1682 .owner = THIS_MODULE, 1683 .set = &test_sk_check_kfunc_ids, 1684 }; 1685 1686 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids) 1687 BTF_ID(struct, prog_test_ref_kfunc) 1688 BTF_ID(func, bpf_kfunc_call_test_release) 1689 BTF_ID(struct, prog_test_member) 1690 BTF_ID(func, bpf_kfunc_call_memb_release) 1691 1692 static int __init bpf_prog_test_run_init(void) 1693 { 1694 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = { 1695 { 1696 .btf_id = bpf_prog_test_dtor_kfunc_ids[0], 1697 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1] 1698 }, 1699 { 1700 .btf_id = bpf_prog_test_dtor_kfunc_ids[2], 1701 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3], 1702 }, 1703 }; 1704 int ret; 1705 1706 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set); 1707 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set); 1708 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set); 1709 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set); 1710 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc, 1711 ARRAY_SIZE(bpf_prog_test_dtor_kfunc), 1712 THIS_MODULE); 1713 } 1714 late_initcall(bpf_prog_test_run_init); 1715