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 noinline void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p) 695 { 696 } 697 698 noinline void bpf_kfunc_call_test_destructive(void) 699 { 700 } 701 702 __diag_pop(); 703 704 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO); 705 706 BTF_SET8_START(test_sk_check_kfunc_ids) 707 BTF_ID_FLAGS(func, bpf_kfunc_call_test1) 708 BTF_ID_FLAGS(func, bpf_kfunc_call_test2) 709 BTF_ID_FLAGS(func, bpf_kfunc_call_test3) 710 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL) 711 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL) 712 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE) 713 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE) 714 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE) 715 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET) 716 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx) 717 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1) 718 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2) 719 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1) 720 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2) 721 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3) 722 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1) 723 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1) 724 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2) 725 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS) 726 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE) 727 BTF_SET8_END(test_sk_check_kfunc_ids) 728 729 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size, 730 u32 size, u32 headroom, u32 tailroom) 731 { 732 void __user *data_in = u64_to_user_ptr(kattr->test.data_in); 733 void *data; 734 735 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom) 736 return ERR_PTR(-EINVAL); 737 738 if (user_size > size) 739 return ERR_PTR(-EMSGSIZE); 740 741 data = kzalloc(size + headroom + tailroom, GFP_USER); 742 if (!data) 743 return ERR_PTR(-ENOMEM); 744 745 if (copy_from_user(data + headroom, data_in, user_size)) { 746 kfree(data); 747 return ERR_PTR(-EFAULT); 748 } 749 750 return data; 751 } 752 753 int bpf_prog_test_run_tracing(struct bpf_prog *prog, 754 const union bpf_attr *kattr, 755 union bpf_attr __user *uattr) 756 { 757 struct bpf_fentry_test_t arg = {}; 758 u16 side_effect = 0, ret = 0; 759 int b = 2, err = -EFAULT; 760 u32 retval = 0; 761 762 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 763 return -EINVAL; 764 765 switch (prog->expected_attach_type) { 766 case BPF_TRACE_FENTRY: 767 case BPF_TRACE_FEXIT: 768 if (bpf_fentry_test1(1) != 2 || 769 bpf_fentry_test2(2, 3) != 5 || 770 bpf_fentry_test3(4, 5, 6) != 15 || 771 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 || 772 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 || 773 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 || 774 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 || 775 bpf_fentry_test8(&arg) != 0) 776 goto out; 777 break; 778 case BPF_MODIFY_RETURN: 779 ret = bpf_modify_return_test(1, &b); 780 if (b != 2) 781 side_effect = 1; 782 break; 783 default: 784 goto out; 785 } 786 787 retval = ((u32)side_effect << 16) | ret; 788 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval))) 789 goto out; 790 791 err = 0; 792 out: 793 trace_bpf_test_finish(&err); 794 return err; 795 } 796 797 struct bpf_raw_tp_test_run_info { 798 struct bpf_prog *prog; 799 void *ctx; 800 u32 retval; 801 }; 802 803 static void 804 __bpf_prog_test_run_raw_tp(void *data) 805 { 806 struct bpf_raw_tp_test_run_info *info = data; 807 808 rcu_read_lock(); 809 info->retval = bpf_prog_run(info->prog, info->ctx); 810 rcu_read_unlock(); 811 } 812 813 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog, 814 const union bpf_attr *kattr, 815 union bpf_attr __user *uattr) 816 { 817 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in); 818 __u32 ctx_size_in = kattr->test.ctx_size_in; 819 struct bpf_raw_tp_test_run_info info; 820 int cpu = kattr->test.cpu, err = 0; 821 int current_cpu; 822 823 /* doesn't support data_in/out, ctx_out, duration, or repeat */ 824 if (kattr->test.data_in || kattr->test.data_out || 825 kattr->test.ctx_out || kattr->test.duration || 826 kattr->test.repeat || kattr->test.batch_size) 827 return -EINVAL; 828 829 if (ctx_size_in < prog->aux->max_ctx_offset || 830 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64)) 831 return -EINVAL; 832 833 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0) 834 return -EINVAL; 835 836 if (ctx_size_in) { 837 info.ctx = memdup_user(ctx_in, ctx_size_in); 838 if (IS_ERR(info.ctx)) 839 return PTR_ERR(info.ctx); 840 } else { 841 info.ctx = NULL; 842 } 843 844 info.prog = prog; 845 846 current_cpu = get_cpu(); 847 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 || 848 cpu == current_cpu) { 849 __bpf_prog_test_run_raw_tp(&info); 850 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) { 851 /* smp_call_function_single() also checks cpu_online() 852 * after csd_lock(). However, since cpu is from user 853 * space, let's do an extra quick check to filter out 854 * invalid value before smp_call_function_single(). 855 */ 856 err = -ENXIO; 857 } else { 858 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp, 859 &info, 1); 860 } 861 put_cpu(); 862 863 if (!err && 864 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32))) 865 err = -EFAULT; 866 867 kfree(info.ctx); 868 return err; 869 } 870 871 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size) 872 { 873 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in); 874 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out); 875 u32 size = kattr->test.ctx_size_in; 876 void *data; 877 int err; 878 879 if (!data_in && !data_out) 880 return NULL; 881 882 data = kzalloc(max_size, GFP_USER); 883 if (!data) 884 return ERR_PTR(-ENOMEM); 885 886 if (data_in) { 887 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size); 888 if (err) { 889 kfree(data); 890 return ERR_PTR(err); 891 } 892 893 size = min_t(u32, max_size, size); 894 if (copy_from_user(data, data_in, size)) { 895 kfree(data); 896 return ERR_PTR(-EFAULT); 897 } 898 } 899 return data; 900 } 901 902 static int bpf_ctx_finish(const union bpf_attr *kattr, 903 union bpf_attr __user *uattr, const void *data, 904 u32 size) 905 { 906 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out); 907 int err = -EFAULT; 908 u32 copy_size = size; 909 910 if (!data || !data_out) 911 return 0; 912 913 if (copy_size > kattr->test.ctx_size_out) { 914 copy_size = kattr->test.ctx_size_out; 915 err = -ENOSPC; 916 } 917 918 if (copy_to_user(data_out, data, copy_size)) 919 goto out; 920 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size))) 921 goto out; 922 if (err != -ENOSPC) 923 err = 0; 924 out: 925 return err; 926 } 927 928 /** 929 * range_is_zero - test whether buffer is initialized 930 * @buf: buffer to check 931 * @from: check from this position 932 * @to: check up until (excluding) this position 933 * 934 * This function returns true if the there is a non-zero byte 935 * in the buf in the range [from,to). 936 */ 937 static inline bool range_is_zero(void *buf, size_t from, size_t to) 938 { 939 return !memchr_inv((u8 *)buf + from, 0, to - from); 940 } 941 942 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) 943 { 944 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; 945 946 if (!skb->len) 947 return -EINVAL; 948 949 if (!__skb) 950 return 0; 951 952 /* make sure the fields we don't use are zeroed */ 953 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark))) 954 return -EINVAL; 955 956 /* mark is allowed */ 957 958 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark), 959 offsetof(struct __sk_buff, priority))) 960 return -EINVAL; 961 962 /* priority is allowed */ 963 /* ingress_ifindex is allowed */ 964 /* ifindex is allowed */ 965 966 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex), 967 offsetof(struct __sk_buff, cb))) 968 return -EINVAL; 969 970 /* cb is allowed */ 971 972 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb), 973 offsetof(struct __sk_buff, tstamp))) 974 return -EINVAL; 975 976 /* tstamp is allowed */ 977 /* wire_len is allowed */ 978 /* gso_segs is allowed */ 979 980 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs), 981 offsetof(struct __sk_buff, gso_size))) 982 return -EINVAL; 983 984 /* gso_size is allowed */ 985 986 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size), 987 offsetof(struct __sk_buff, hwtstamp))) 988 return -EINVAL; 989 990 /* hwtstamp is allowed */ 991 992 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp), 993 sizeof(struct __sk_buff))) 994 return -EINVAL; 995 996 skb->mark = __skb->mark; 997 skb->priority = __skb->priority; 998 skb->skb_iif = __skb->ingress_ifindex; 999 skb->tstamp = __skb->tstamp; 1000 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN); 1001 1002 if (__skb->wire_len == 0) { 1003 cb->pkt_len = skb->len; 1004 } else { 1005 if (__skb->wire_len < skb->len || 1006 __skb->wire_len > GSO_LEGACY_MAX_SIZE) 1007 return -EINVAL; 1008 cb->pkt_len = __skb->wire_len; 1009 } 1010 1011 if (__skb->gso_segs > GSO_MAX_SEGS) 1012 return -EINVAL; 1013 skb_shinfo(skb)->gso_segs = __skb->gso_segs; 1014 skb_shinfo(skb)->gso_size = __skb->gso_size; 1015 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp; 1016 1017 return 0; 1018 } 1019 1020 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb) 1021 { 1022 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; 1023 1024 if (!__skb) 1025 return; 1026 1027 __skb->mark = skb->mark; 1028 __skb->priority = skb->priority; 1029 __skb->ingress_ifindex = skb->skb_iif; 1030 __skb->ifindex = skb->dev->ifindex; 1031 __skb->tstamp = skb->tstamp; 1032 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN); 1033 __skb->wire_len = cb->pkt_len; 1034 __skb->gso_segs = skb_shinfo(skb)->gso_segs; 1035 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp; 1036 } 1037 1038 static struct proto bpf_dummy_proto = { 1039 .name = "bpf_dummy", 1040 .owner = THIS_MODULE, 1041 .obj_size = sizeof(struct sock), 1042 }; 1043 1044 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr, 1045 union bpf_attr __user *uattr) 1046 { 1047 bool is_l2 = false, is_direct_pkt_access = false; 1048 struct net *net = current->nsproxy->net_ns; 1049 struct net_device *dev = net->loopback_dev; 1050 u32 size = kattr->test.data_size_in; 1051 u32 repeat = kattr->test.repeat; 1052 struct __sk_buff *ctx = NULL; 1053 u32 retval, duration; 1054 int hh_len = ETH_HLEN; 1055 struct sk_buff *skb; 1056 struct sock *sk; 1057 void *data; 1058 int ret; 1059 1060 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 1061 return -EINVAL; 1062 1063 data = bpf_test_init(kattr, kattr->test.data_size_in, 1064 size, NET_SKB_PAD + NET_IP_ALIGN, 1065 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))); 1066 if (IS_ERR(data)) 1067 return PTR_ERR(data); 1068 1069 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff)); 1070 if (IS_ERR(ctx)) { 1071 kfree(data); 1072 return PTR_ERR(ctx); 1073 } 1074 1075 switch (prog->type) { 1076 case BPF_PROG_TYPE_SCHED_CLS: 1077 case BPF_PROG_TYPE_SCHED_ACT: 1078 is_l2 = true; 1079 fallthrough; 1080 case BPF_PROG_TYPE_LWT_IN: 1081 case BPF_PROG_TYPE_LWT_OUT: 1082 case BPF_PROG_TYPE_LWT_XMIT: 1083 is_direct_pkt_access = true; 1084 break; 1085 default: 1086 break; 1087 } 1088 1089 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1); 1090 if (!sk) { 1091 kfree(data); 1092 kfree(ctx); 1093 return -ENOMEM; 1094 } 1095 sock_init_data(NULL, sk); 1096 1097 skb = build_skb(data, 0); 1098 if (!skb) { 1099 kfree(data); 1100 kfree(ctx); 1101 sk_free(sk); 1102 return -ENOMEM; 1103 } 1104 skb->sk = sk; 1105 1106 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); 1107 __skb_put(skb, size); 1108 if (ctx && ctx->ifindex > 1) { 1109 dev = dev_get_by_index(net, ctx->ifindex); 1110 if (!dev) { 1111 ret = -ENODEV; 1112 goto out; 1113 } 1114 } 1115 skb->protocol = eth_type_trans(skb, dev); 1116 skb_reset_network_header(skb); 1117 1118 switch (skb->protocol) { 1119 case htons(ETH_P_IP): 1120 sk->sk_family = AF_INET; 1121 if (sizeof(struct iphdr) <= skb_headlen(skb)) { 1122 sk->sk_rcv_saddr = ip_hdr(skb)->saddr; 1123 sk->sk_daddr = ip_hdr(skb)->daddr; 1124 } 1125 break; 1126 #if IS_ENABLED(CONFIG_IPV6) 1127 case htons(ETH_P_IPV6): 1128 sk->sk_family = AF_INET6; 1129 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) { 1130 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr; 1131 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr; 1132 } 1133 break; 1134 #endif 1135 default: 1136 break; 1137 } 1138 1139 if (is_l2) 1140 __skb_push(skb, hh_len); 1141 if (is_direct_pkt_access) 1142 bpf_compute_data_pointers(skb); 1143 ret = convert___skb_to_skb(skb, ctx); 1144 if (ret) 1145 goto out; 1146 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false); 1147 if (ret) 1148 goto out; 1149 if (!is_l2) { 1150 if (skb_headroom(skb) < hh_len) { 1151 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb)); 1152 1153 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) { 1154 ret = -ENOMEM; 1155 goto out; 1156 } 1157 } 1158 memset(__skb_push(skb, hh_len), 0, hh_len); 1159 } 1160 convert_skb_to___skb(skb, ctx); 1161 1162 size = skb->len; 1163 /* bpf program can never convert linear skb to non-linear */ 1164 if (WARN_ON_ONCE(skb_is_nonlinear(skb))) 1165 size = skb_headlen(skb); 1166 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval, 1167 duration); 1168 if (!ret) 1169 ret = bpf_ctx_finish(kattr, uattr, ctx, 1170 sizeof(struct __sk_buff)); 1171 out: 1172 if (dev && dev != net->loopback_dev) 1173 dev_put(dev); 1174 kfree_skb(skb); 1175 sk_free(sk); 1176 kfree(ctx); 1177 return ret; 1178 } 1179 1180 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp) 1181 { 1182 unsigned int ingress_ifindex, rx_queue_index; 1183 struct netdev_rx_queue *rxqueue; 1184 struct net_device *device; 1185 1186 if (!xdp_md) 1187 return 0; 1188 1189 if (xdp_md->egress_ifindex != 0) 1190 return -EINVAL; 1191 1192 ingress_ifindex = xdp_md->ingress_ifindex; 1193 rx_queue_index = xdp_md->rx_queue_index; 1194 1195 if (!ingress_ifindex && rx_queue_index) 1196 return -EINVAL; 1197 1198 if (ingress_ifindex) { 1199 device = dev_get_by_index(current->nsproxy->net_ns, 1200 ingress_ifindex); 1201 if (!device) 1202 return -ENODEV; 1203 1204 if (rx_queue_index >= device->real_num_rx_queues) 1205 goto free_dev; 1206 1207 rxqueue = __netif_get_rx_queue(device, rx_queue_index); 1208 1209 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq)) 1210 goto free_dev; 1211 1212 xdp->rxq = &rxqueue->xdp_rxq; 1213 /* The device is now tracked in the xdp->rxq for later 1214 * dev_put() 1215 */ 1216 } 1217 1218 xdp->data = xdp->data_meta + xdp_md->data; 1219 return 0; 1220 1221 free_dev: 1222 dev_put(device); 1223 return -EINVAL; 1224 } 1225 1226 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md) 1227 { 1228 if (!xdp_md) 1229 return; 1230 1231 xdp_md->data = xdp->data - xdp->data_meta; 1232 xdp_md->data_end = xdp->data_end - xdp->data_meta; 1233 1234 if (xdp_md->ingress_ifindex) 1235 dev_put(xdp->rxq->dev); 1236 } 1237 1238 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr, 1239 union bpf_attr __user *uattr) 1240 { 1241 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES); 1242 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1243 u32 batch_size = kattr->test.batch_size; 1244 u32 retval = 0, duration, max_data_sz; 1245 u32 size = kattr->test.data_size_in; 1246 u32 headroom = XDP_PACKET_HEADROOM; 1247 u32 repeat = kattr->test.repeat; 1248 struct netdev_rx_queue *rxqueue; 1249 struct skb_shared_info *sinfo; 1250 struct xdp_buff xdp = {}; 1251 int i, ret = -EINVAL; 1252 struct xdp_md *ctx; 1253 void *data; 1254 1255 if (prog->expected_attach_type == BPF_XDP_DEVMAP || 1256 prog->expected_attach_type == BPF_XDP_CPUMAP) 1257 return -EINVAL; 1258 1259 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES) 1260 return -EINVAL; 1261 1262 if (do_live) { 1263 if (!batch_size) 1264 batch_size = NAPI_POLL_WEIGHT; 1265 else if (batch_size > TEST_XDP_MAX_BATCH) 1266 return -E2BIG; 1267 1268 headroom += sizeof(struct xdp_page_head); 1269 } else if (batch_size) { 1270 return -EINVAL; 1271 } 1272 1273 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md)); 1274 if (IS_ERR(ctx)) 1275 return PTR_ERR(ctx); 1276 1277 if (ctx) { 1278 /* There can't be user provided data before the meta data */ 1279 if (ctx->data_meta || ctx->data_end != size || 1280 ctx->data > ctx->data_end || 1281 unlikely(xdp_metalen_invalid(ctx->data)) || 1282 (do_live && (kattr->test.data_out || kattr->test.ctx_out))) 1283 goto free_ctx; 1284 /* Meta data is allocated from the headroom */ 1285 headroom -= ctx->data; 1286 } 1287 1288 max_data_sz = 4096 - headroom - tailroom; 1289 if (size > max_data_sz) { 1290 /* disallow live data mode for jumbo frames */ 1291 if (do_live) 1292 goto free_ctx; 1293 size = max_data_sz; 1294 } 1295 1296 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom); 1297 if (IS_ERR(data)) { 1298 ret = PTR_ERR(data); 1299 goto free_ctx; 1300 } 1301 1302 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0); 1303 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom; 1304 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq); 1305 xdp_prepare_buff(&xdp, data, headroom, size, true); 1306 sinfo = xdp_get_shared_info_from_buff(&xdp); 1307 1308 ret = xdp_convert_md_to_buff(ctx, &xdp); 1309 if (ret) 1310 goto free_data; 1311 1312 if (unlikely(kattr->test.data_size_in > size)) { 1313 void __user *data_in = u64_to_user_ptr(kattr->test.data_in); 1314 1315 while (size < kattr->test.data_size_in) { 1316 struct page *page; 1317 skb_frag_t *frag; 1318 u32 data_len; 1319 1320 if (sinfo->nr_frags == MAX_SKB_FRAGS) { 1321 ret = -ENOMEM; 1322 goto out; 1323 } 1324 1325 page = alloc_page(GFP_KERNEL); 1326 if (!page) { 1327 ret = -ENOMEM; 1328 goto out; 1329 } 1330 1331 frag = &sinfo->frags[sinfo->nr_frags++]; 1332 __skb_frag_set_page(frag, page); 1333 1334 data_len = min_t(u32, kattr->test.data_size_in - size, 1335 PAGE_SIZE); 1336 skb_frag_size_set(frag, data_len); 1337 1338 if (copy_from_user(page_address(page), data_in + size, 1339 data_len)) { 1340 ret = -EFAULT; 1341 goto out; 1342 } 1343 sinfo->xdp_frags_size += data_len; 1344 size += data_len; 1345 } 1346 xdp_buff_set_frags_flag(&xdp); 1347 } 1348 1349 if (repeat > 1) 1350 bpf_prog_change_xdp(NULL, prog); 1351 1352 if (do_live) 1353 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration); 1354 else 1355 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true); 1356 /* We convert the xdp_buff back to an xdp_md before checking the return 1357 * code so the reference count of any held netdevice will be decremented 1358 * even if the test run failed. 1359 */ 1360 xdp_convert_buff_to_md(&xdp, ctx); 1361 if (ret) 1362 goto out; 1363 1364 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size; 1365 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size, 1366 retval, duration); 1367 if (!ret) 1368 ret = bpf_ctx_finish(kattr, uattr, ctx, 1369 sizeof(struct xdp_md)); 1370 1371 out: 1372 if (repeat > 1) 1373 bpf_prog_change_xdp(prog, NULL); 1374 free_data: 1375 for (i = 0; i < sinfo->nr_frags; i++) 1376 __free_page(skb_frag_page(&sinfo->frags[i])); 1377 kfree(data); 1378 free_ctx: 1379 kfree(ctx); 1380 return ret; 1381 } 1382 1383 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx) 1384 { 1385 /* make sure the fields we don't use are zeroed */ 1386 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags))) 1387 return -EINVAL; 1388 1389 /* flags is allowed */ 1390 1391 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags), 1392 sizeof(struct bpf_flow_keys))) 1393 return -EINVAL; 1394 1395 return 0; 1396 } 1397 1398 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog, 1399 const union bpf_attr *kattr, 1400 union bpf_attr __user *uattr) 1401 { 1402 struct bpf_test_timer t = { NO_PREEMPT }; 1403 u32 size = kattr->test.data_size_in; 1404 struct bpf_flow_dissector ctx = {}; 1405 u32 repeat = kattr->test.repeat; 1406 struct bpf_flow_keys *user_ctx; 1407 struct bpf_flow_keys flow_keys; 1408 const struct ethhdr *eth; 1409 unsigned int flags = 0; 1410 u32 retval, duration; 1411 void *data; 1412 int ret; 1413 1414 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 1415 return -EINVAL; 1416 1417 if (size < ETH_HLEN) 1418 return -EINVAL; 1419 1420 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0); 1421 if (IS_ERR(data)) 1422 return PTR_ERR(data); 1423 1424 eth = (struct ethhdr *)data; 1425 1426 if (!repeat) 1427 repeat = 1; 1428 1429 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys)); 1430 if (IS_ERR(user_ctx)) { 1431 kfree(data); 1432 return PTR_ERR(user_ctx); 1433 } 1434 if (user_ctx) { 1435 ret = verify_user_bpf_flow_keys(user_ctx); 1436 if (ret) 1437 goto out; 1438 flags = user_ctx->flags; 1439 } 1440 1441 ctx.flow_keys = &flow_keys; 1442 ctx.data = data; 1443 ctx.data_end = (__u8 *)data + size; 1444 1445 bpf_test_timer_enter(&t); 1446 do { 1447 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN, 1448 size, flags); 1449 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration)); 1450 bpf_test_timer_leave(&t); 1451 1452 if (ret < 0) 1453 goto out; 1454 1455 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL, 1456 sizeof(flow_keys), retval, duration); 1457 if (!ret) 1458 ret = bpf_ctx_finish(kattr, uattr, user_ctx, 1459 sizeof(struct bpf_flow_keys)); 1460 1461 out: 1462 kfree(user_ctx); 1463 kfree(data); 1464 return ret; 1465 } 1466 1467 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr, 1468 union bpf_attr __user *uattr) 1469 { 1470 struct bpf_test_timer t = { NO_PREEMPT }; 1471 struct bpf_prog_array *progs = NULL; 1472 struct bpf_sk_lookup_kern ctx = {}; 1473 u32 repeat = kattr->test.repeat; 1474 struct bpf_sk_lookup *user_ctx; 1475 u32 retval, duration; 1476 int ret = -EINVAL; 1477 1478 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size) 1479 return -EINVAL; 1480 1481 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out || 1482 kattr->test.data_size_out) 1483 return -EINVAL; 1484 1485 if (!repeat) 1486 repeat = 1; 1487 1488 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx)); 1489 if (IS_ERR(user_ctx)) 1490 return PTR_ERR(user_ctx); 1491 1492 if (!user_ctx) 1493 return -EINVAL; 1494 1495 if (user_ctx->sk) 1496 goto out; 1497 1498 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx))) 1499 goto out; 1500 1501 if (user_ctx->local_port > U16_MAX) { 1502 ret = -ERANGE; 1503 goto out; 1504 } 1505 1506 ctx.family = (u16)user_ctx->family; 1507 ctx.protocol = (u16)user_ctx->protocol; 1508 ctx.dport = (u16)user_ctx->local_port; 1509 ctx.sport = user_ctx->remote_port; 1510 1511 switch (ctx.family) { 1512 case AF_INET: 1513 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4; 1514 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4; 1515 break; 1516 1517 #if IS_ENABLED(CONFIG_IPV6) 1518 case AF_INET6: 1519 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6; 1520 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6; 1521 break; 1522 #endif 1523 1524 default: 1525 ret = -EAFNOSUPPORT; 1526 goto out; 1527 } 1528 1529 progs = bpf_prog_array_alloc(1, GFP_KERNEL); 1530 if (!progs) { 1531 ret = -ENOMEM; 1532 goto out; 1533 } 1534 1535 progs->items[0].prog = prog; 1536 1537 bpf_test_timer_enter(&t); 1538 do { 1539 ctx.selected_sk = NULL; 1540 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run); 1541 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration)); 1542 bpf_test_timer_leave(&t); 1543 1544 if (ret < 0) 1545 goto out; 1546 1547 user_ctx->cookie = 0; 1548 if (ctx.selected_sk) { 1549 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) { 1550 ret = -EOPNOTSUPP; 1551 goto out; 1552 } 1553 1554 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk); 1555 } 1556 1557 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration); 1558 if (!ret) 1559 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx)); 1560 1561 out: 1562 bpf_prog_array_free(progs); 1563 kfree(user_ctx); 1564 return ret; 1565 } 1566 1567 int bpf_prog_test_run_syscall(struct bpf_prog *prog, 1568 const union bpf_attr *kattr, 1569 union bpf_attr __user *uattr) 1570 { 1571 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in); 1572 __u32 ctx_size_in = kattr->test.ctx_size_in; 1573 void *ctx = NULL; 1574 u32 retval; 1575 int err = 0; 1576 1577 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */ 1578 if (kattr->test.data_in || kattr->test.data_out || 1579 kattr->test.ctx_out || kattr->test.duration || 1580 kattr->test.repeat || kattr->test.flags || 1581 kattr->test.batch_size) 1582 return -EINVAL; 1583 1584 if (ctx_size_in < prog->aux->max_ctx_offset || 1585 ctx_size_in > U16_MAX) 1586 return -EINVAL; 1587 1588 if (ctx_size_in) { 1589 ctx = memdup_user(ctx_in, ctx_size_in); 1590 if (IS_ERR(ctx)) 1591 return PTR_ERR(ctx); 1592 } 1593 1594 rcu_read_lock_trace(); 1595 retval = bpf_prog_run_pin_on_cpu(prog, ctx); 1596 rcu_read_unlock_trace(); 1597 1598 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) { 1599 err = -EFAULT; 1600 goto out; 1601 } 1602 if (ctx_size_in) 1603 if (copy_to_user(ctx_in, ctx, ctx_size_in)) 1604 err = -EFAULT; 1605 out: 1606 kfree(ctx); 1607 return err; 1608 } 1609 1610 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = { 1611 .owner = THIS_MODULE, 1612 .set = &test_sk_check_kfunc_ids, 1613 }; 1614 1615 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids) 1616 BTF_ID(struct, prog_test_ref_kfunc) 1617 BTF_ID(func, bpf_kfunc_call_test_release) 1618 BTF_ID(struct, prog_test_member) 1619 BTF_ID(func, bpf_kfunc_call_memb_release) 1620 1621 static int __init bpf_prog_test_run_init(void) 1622 { 1623 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = { 1624 { 1625 .btf_id = bpf_prog_test_dtor_kfunc_ids[0], 1626 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1] 1627 }, 1628 { 1629 .btf_id = bpf_prog_test_dtor_kfunc_ids[2], 1630 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3], 1631 }, 1632 }; 1633 int ret; 1634 1635 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set); 1636 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set); 1637 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc, 1638 ARRAY_SIZE(bpf_prog_test_dtor_kfunc), 1639 THIS_MODULE); 1640 } 1641 late_initcall(bpf_prog_test_run_init); 1642