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