net/bpf/test_run.c

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