kernel/bpf/bpf_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2020 Facebook */
   3
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/filter.h>
   7 #include <linux/bpf.h>
   8 #include <linux/rcupdate_trace.h>
   9
  10 struct bpf_iter_target_info {
  11         struct list_head list;
  12         const struct bpf_iter_reg *reg_info;
  13         u32 btf_id;     /* cached value */
  14 };
  15
  16 struct bpf_iter_link {
  17         struct bpf_link link;
  18         struct bpf_iter_aux_info aux;
  19         struct bpf_iter_target_info *tinfo;
  20 };
  21
  22 struct bpf_iter_priv_data {
  23         struct bpf_iter_target_info *tinfo;
  24         const struct bpf_iter_seq_info *seq_info;
  25         struct bpf_prog *prog;
  26         u64 session_id;
  27         u64 seq_num;
  28         bool done_stop;
  29         u8 target_private[] __aligned(8);
  30 };
  31
  32 static struct list_head targets = LIST_HEAD_INIT(targets);
  33 static DEFINE_MUTEX(targets_mutex);
  34
  35 /* protect bpf_iter_link changes */
  36 static DEFINE_MUTEX(link_mutex);
  37
  38 /* incremented on every opened seq_file */
  39 static atomic64_t session_id;
  40
  41 static int prepare_seq_file(struct file *file, struct bpf_iter_link *link,
  42                             const struct bpf_iter_seq_info *seq_info);
  43
  44 static void bpf_iter_inc_seq_num(struct seq_file *seq)
  45 {
  46         struct bpf_iter_priv_data *iter_priv;
  47
  48         iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
  49                                  target_private);
  50         iter_priv->seq_num++;
  51 }
  52
  53 static void bpf_iter_dec_seq_num(struct seq_file *seq)
  54 {
  55         struct bpf_iter_priv_data *iter_priv;
  56
  57         iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
  58                                  target_private);
  59         iter_priv->seq_num--;
  60 }
  61
  62 static void bpf_iter_done_stop(struct seq_file *seq)
  63 {
  64         struct bpf_iter_priv_data *iter_priv;
  65
  66         iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
  67                                  target_private);
  68         iter_priv->done_stop = true;
  69 }
  70
  71 static inline bool bpf_iter_target_support_resched(const struct bpf_iter_target_info *tinfo)
  72 {
  73         return tinfo->reg_info->feature & BPF_ITER_RESCHED;
  74 }
  75
  76 static bool bpf_iter_support_resched(struct seq_file *seq)
  77 {
  78         struct bpf_iter_priv_data *iter_priv;
  79
  80         iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
  81                                  target_private);
  82         return bpf_iter_target_support_resched(iter_priv->tinfo);
  83 }
  84
  85 /* maximum visited objects before bailing out */
  86 #define MAX_ITER_OBJECTS        1000000
  87
  88 /* bpf_seq_read, a customized and simpler version for bpf iterator.
  89  * The following are differences from seq_read():
  90  *  . fixed buffer size (PAGE_SIZE)
  91  *  . assuming NULL ->llseek()
  92  *  . stop() may call bpf program, handling potential overflow there
  93  */
  94 static ssize_t bpf_seq_read(struct file *file, char __user *buf, size_t size,
  95                             loff_t *ppos)
  96 {
  97         struct seq_file *seq = file->private_data;
  98         size_t n, offs, copied = 0;
  99         int err = 0, num_objs = 0;
 100         bool can_resched;
 101         void *p;
 102
 103         mutex_lock(&seq->lock);
 104
 105         if (!seq->buf) {
 106                 seq->size = PAGE_SIZE << 3;
 107                 seq->buf = kvmalloc(seq->size, GFP_KERNEL);
 108                 if (!seq->buf) {
 109                         err = -ENOMEM;
 110                         goto done;
 111                 }
 112         }
 113
 114         if (seq->count) {
 115                 n = min(seq->count, size);
 116                 err = copy_to_user(buf, seq->buf + seq->from, n);
 117                 if (err) {
 118                         err = -EFAULT;
 119                         goto done;
 120                 }
 121                 seq->count -= n;
 122                 seq->from += n;
 123                 copied = n;
 124                 goto done;
 125         }
 126
 127         seq->from = 0;
 128         p = seq->op->start(seq, &seq->index);
 129         if (!p)
 130                 goto stop;
 131         if (IS_ERR(p)) {
 132                 err = PTR_ERR(p);
 133                 seq->op->stop(seq, p);
 134                 seq->count = 0;
 135                 goto done;
 136         }
 137
 138         err = seq->op->show(seq, p);
 139         if (err > 0) {
 140                 /* object is skipped, decrease seq_num, so next
 141                  * valid object can reuse the same seq_num.
 142                  */
 143                 bpf_iter_dec_seq_num(seq);
 144                 seq->count = 0;
 145         } else if (err < 0 || seq_has_overflowed(seq)) {
 146                 if (!err)
 147                         err = -E2BIG;
 148                 seq->op->stop(seq, p);
 149                 seq->count = 0;
 150                 goto done;
 151         }
 152
 153         can_resched = bpf_iter_support_resched(seq);
 154         while (1) {
 155                 loff_t pos = seq->index;
 156
 157                 num_objs++;
 158                 offs = seq->count;
 159                 p = seq->op->next(seq, p, &seq->index);
 160                 if (pos == seq->index) {
 161                         pr_info_ratelimited("buggy seq_file .next function %ps "
 162                                 "did not updated position index\n",
 163                                 seq->op->next);
 164                         seq->index++;
 165                 }
 166
 167                 if (IS_ERR_OR_NULL(p))
 168                         break;
 169
 170                 /* got a valid next object, increase seq_num */
 171                 bpf_iter_inc_seq_num(seq);
 172
 173                 if (seq->count >= size)
 174                         break;
 175
 176                 if (num_objs >= MAX_ITER_OBJECTS) {
 177                         if (offs == 0) {
 178                                 err = -EAGAIN;
 179                                 seq->op->stop(seq, p);
 180                                 goto done;
 181                         }
 182                         break;
 183                 }
 184
 185                 err = seq->op->show(seq, p);
 186                 if (err > 0) {
 187                         bpf_iter_dec_seq_num(seq);
 188                         seq->count = offs;
 189                 } else if (err < 0 || seq_has_overflowed(seq)) {
 190                         seq->count = offs;
 191                         if (offs == 0) {
 192                                 if (!err)
 193                                         err = -E2BIG;
 194                                 seq->op->stop(seq, p);
 195                                 goto done;
 196                         }
 197                         break;
 198                 }
 199
 200                 if (can_resched)
 201                         cond_resched();
 202         }
 203 stop:
 204         offs = seq->count;
 205         if (IS_ERR(p)) {
 206                 seq->op->stop(seq, NULL);
 207                 err = PTR_ERR(p);
 208                 goto done;
 209         }
 210         /* bpf program called if !p */
 211         seq->op->stop(seq, p);
 212         if (!p) {
 213                 if (!seq_has_overflowed(seq)) {
 214                         bpf_iter_done_stop(seq);
 215                 } else {
 216                         seq->count = offs;
 217                         if (offs == 0) {
 218                                 err = -E2BIG;
 219                                 goto done;
 220                         }
 221                 }
 222         }
 223
 224         n = min(seq->count, size);
 225         err = copy_to_user(buf, seq->buf, n);
 226         if (err) {
 227                 err = -EFAULT;
 228                 goto done;
 229         }
 230         copied = n;
 231         seq->count -= n;
 232         seq->from = n;
 233 done:
 234         if (!copied)
 235                 copied = err;
 236         else
 237                 *ppos += copied;
 238         mutex_unlock(&seq->lock);
 239         return copied;
 240 }
 241
 242 static const struct bpf_iter_seq_info *
 243 __get_seq_info(struct bpf_iter_link *link)
 244 {
 245         const struct bpf_iter_seq_info *seq_info;
 246
 247         if (link->aux.map) {
 248                 seq_info = link->aux.map->ops->iter_seq_info;
 249                 if (seq_info)
 250                         return seq_info;
 251         }
 252
 253         return link->tinfo->reg_info->seq_info;
 254 }
 255
 256 static int iter_open(struct inode *inode, struct file *file)
 257 {
 258         struct bpf_iter_link *link = inode->i_private;
 259
 260         return prepare_seq_file(file, link, __get_seq_info(link));
 261 }
 262
 263 static int iter_release(struct inode *inode, struct file *file)
 264 {
 265         struct bpf_iter_priv_data *iter_priv;
 266         struct seq_file *seq;
 267
 268         seq = file->private_data;
 269         if (!seq)
 270                 return 0;
 271
 272         iter_priv = container_of(seq->private, struct bpf_iter_priv_data,
 273                                  target_private);
 274
 275         if (iter_priv->seq_info->fini_seq_private)
 276                 iter_priv->seq_info->fini_seq_private(seq->private);
 277
 278         bpf_prog_put(iter_priv->prog);
 279         seq->private = iter_priv;
 280
 281         return seq_release_private(inode, file);
 282 }
 283
 284 const struct file_operations bpf_iter_fops = {
 285         .open           = iter_open,
 286         .read           = bpf_seq_read,
 287         .release        = iter_release,
 288 };
 289
 290 /* The argument reg_info will be cached in bpf_iter_target_info.
 291  * The common practice is to declare target reg_info as
 292  * a const static variable and passed as an argument to
 293  * bpf_iter_reg_target().
 294  */
 295 int bpf_iter_reg_target(const struct bpf_iter_reg *reg_info)
 296 {
 297         struct bpf_iter_target_info *tinfo;
 298
 299         tinfo = kzalloc(sizeof(*tinfo), GFP_KERNEL);
 300         if (!tinfo)
 301                 return -ENOMEM;
 302
 303         tinfo->reg_info = reg_info;
 304         INIT_LIST_HEAD(&tinfo->list);
 305
 306         mutex_lock(&targets_mutex);
 307         list_add(&tinfo->list, &targets);
 308         mutex_unlock(&targets_mutex);
 309
 310         return 0;
 311 }
 312
 313 void bpf_iter_unreg_target(const struct bpf_iter_reg *reg_info)
 314 {
 315         struct bpf_iter_target_info *tinfo;
 316         bool found = false;
 317
 318         mutex_lock(&targets_mutex);
 319         list_for_each_entry(tinfo, &targets, list) {
 320                 if (reg_info == tinfo->reg_info) {
 321                         list_del(&tinfo->list);
 322                         kfree(tinfo);
 323                         found = true;
 324                         break;
 325                 }
 326         }
 327         mutex_unlock(&targets_mutex);
 328
 329         WARN_ON(found == false);
 330 }
 331
 332 static void cache_btf_id(struct bpf_iter_target_info *tinfo,
 333                          struct bpf_prog *prog)
 334 {
 335         tinfo->btf_id = prog->aux->attach_btf_id;
 336 }
 337
 338 bool bpf_iter_prog_supported(struct bpf_prog *prog)
 339 {
 340         const char *attach_fname = prog->aux->attach_func_name;
 341         struct bpf_iter_target_info *tinfo = NULL, *iter;
 342         u32 prog_btf_id = prog->aux->attach_btf_id;
 343         const char *prefix = BPF_ITER_FUNC_PREFIX;
 344         int prefix_len = strlen(prefix);
 345
 346         if (strncmp(attach_fname, prefix, prefix_len))
 347                 return false;
 348
 349         mutex_lock(&targets_mutex);
 350         list_for_each_entry(iter, &targets, list) {
 351                 if (iter->btf_id && iter->btf_id == prog_btf_id) {
 352                         tinfo = iter;
 353                         break;
 354                 }
 355                 if (!strcmp(attach_fname + prefix_len, iter->reg_info->target)) {
 356                         cache_btf_id(iter, prog);
 357                         tinfo = iter;
 358                         break;
 359                 }
 360         }
 361         mutex_unlock(&targets_mutex);
 362
 363         if (tinfo) {
 364                 prog->aux->ctx_arg_info_size = tinfo->reg_info->ctx_arg_info_size;
 365                 prog->aux->ctx_arg_info = tinfo->reg_info->ctx_arg_info;
 366         }
 367
 368         return tinfo != NULL;
 369 }
 370
 371 const struct bpf_func_proto *
 372 bpf_iter_get_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 373 {
 374         const struct bpf_iter_target_info *tinfo;
 375         const struct bpf_func_proto *fn = NULL;
 376
 377         mutex_lock(&targets_mutex);
 378         list_for_each_entry(tinfo, &targets, list) {
 379                 if (tinfo->btf_id == prog->aux->attach_btf_id) {
 380                         const struct bpf_iter_reg *reg_info;
 381
 382                         reg_info = tinfo->reg_info;
 383                         if (reg_info->get_func_proto)
 384                                 fn = reg_info->get_func_proto(func_id, prog);
 385                         break;
 386                 }
 387         }
 388         mutex_unlock(&targets_mutex);
 389
 390         return fn;
 391 }
 392
 393 static void bpf_iter_link_release(struct bpf_link *link)
 394 {
 395         struct bpf_iter_link *iter_link =
 396                 container_of(link, struct bpf_iter_link, link);
 397
 398         if (iter_link->tinfo->reg_info->detach_target)
 399                 iter_link->tinfo->reg_info->detach_target(&iter_link->aux);
 400 }
 401
 402 static void bpf_iter_link_dealloc(struct bpf_link *link)
 403 {
 404         struct bpf_iter_link *iter_link =
 405                 container_of(link, struct bpf_iter_link, link);
 406
 407         kfree(iter_link);
 408 }
 409
 410 static int bpf_iter_link_replace(struct bpf_link *link,
 411                                  struct bpf_prog *new_prog,
 412                                  struct bpf_prog *old_prog)
 413 {
 414         int ret = 0;
 415
 416         mutex_lock(&link_mutex);
 417         if (old_prog && link->prog != old_prog) {
 418                 ret = -EPERM;
 419                 goto out_unlock;
 420         }
 421
 422         if (link->prog->type != new_prog->type ||
 423             link->prog->expected_attach_type != new_prog->expected_attach_type ||
 424             link->prog->aux->attach_btf_id != new_prog->aux->attach_btf_id) {
 425                 ret = -EINVAL;
 426                 goto out_unlock;
 427         }
 428
 429         old_prog = xchg(&link->prog, new_prog);
 430         bpf_prog_put(old_prog);
 431
 432 out_unlock:
 433         mutex_unlock(&link_mutex);
 434         return ret;
 435 }
 436
 437 static void bpf_iter_link_show_fdinfo(const struct bpf_link *link,
 438                                       struct seq_file *seq)
 439 {
 440         struct bpf_iter_link *iter_link =
 441                 container_of(link, struct bpf_iter_link, link);
 442         bpf_iter_show_fdinfo_t show_fdinfo;
 443
 444         seq_printf(seq,
 445                    "target_name:\t%s\n",
 446                    iter_link->tinfo->reg_info->target);
 447
 448         show_fdinfo = iter_link->tinfo->reg_info->show_fdinfo;
 449         if (show_fdinfo)
 450                 show_fdinfo(&iter_link->aux, seq);
 451 }
 452
 453 static int bpf_iter_link_fill_link_info(const struct bpf_link *link,
 454                                         struct bpf_link_info *info)
 455 {
 456         struct bpf_iter_link *iter_link =
 457                 container_of(link, struct bpf_iter_link, link);
 458         char __user *ubuf = u64_to_user_ptr(info->iter.target_name);
 459         bpf_iter_fill_link_info_t fill_link_info;
 460         u32 ulen = info->iter.target_name_len;
 461         const char *target_name;
 462         u32 target_len;
 463
 464         if (!ulen ^ !ubuf)
 465                 return -EINVAL;
 466
 467         target_name = iter_link->tinfo->reg_info->target;
 468         target_len =  strlen(target_name);
 469         info->iter.target_name_len = target_len + 1;
 470
 471         if (ubuf) {
 472                 if (ulen >= target_len + 1) {
 473                         if (copy_to_user(ubuf, target_name, target_len + 1))
 474                                 return -EFAULT;
 475                 } else {
 476                         char zero = '\0';
 477
 478                         if (copy_to_user(ubuf, target_name, ulen - 1))
 479                                 return -EFAULT;
 480                         if (put_user(zero, ubuf + ulen - 1))
 481                                 return -EFAULT;
 482                         return -ENOSPC;
 483                 }
 484         }
 485
 486         fill_link_info = iter_link->tinfo->reg_info->fill_link_info;
 487         if (fill_link_info)
 488                 return fill_link_info(&iter_link->aux, info);
 489
 490         return 0;
 491 }
 492
 493 static const struct bpf_link_ops bpf_iter_link_lops = {
 494         .release = bpf_iter_link_release,
 495         .dealloc = bpf_iter_link_dealloc,
 496         .update_prog = bpf_iter_link_replace,
 497         .show_fdinfo = bpf_iter_link_show_fdinfo,
 498         .fill_link_info = bpf_iter_link_fill_link_info,
 499 };
 500
 501 bool bpf_link_is_iter(struct bpf_link *link)
 502 {
 503         return link->ops == &bpf_iter_link_lops;
 504 }
 505
 506 int bpf_iter_link_attach(const union bpf_attr *attr, bpfptr_t uattr,
 507                          struct bpf_prog *prog)
 508 {
 509         struct bpf_iter_target_info *tinfo = NULL, *iter;
 510         struct bpf_link_primer link_primer;
 511         union bpf_iter_link_info linfo;
 512         struct bpf_iter_link *link;
 513         u32 prog_btf_id, linfo_len;
 514         bpfptr_t ulinfo;
 515         int err;
 516
 517         if (attr->link_create.target_fd || attr->link_create.flags)
 518                 return -EINVAL;
 519
 520         memset(&linfo, 0, sizeof(union bpf_iter_link_info));
 521
 522         ulinfo = make_bpfptr(attr->link_create.iter_info, uattr.is_kernel);
 523         linfo_len = attr->link_create.iter_info_len;
 524         if (bpfptr_is_null(ulinfo) ^ !linfo_len)
 525                 return -EINVAL;
 526
 527         if (!bpfptr_is_null(ulinfo)) {
 528                 err = bpf_check_uarg_tail_zero(ulinfo, sizeof(linfo),
 529                                                linfo_len);
 530                 if (err)
 531                         return err;
 532                 linfo_len = min_t(u32, linfo_len, sizeof(linfo));
 533                 if (copy_from_bpfptr(&linfo, ulinfo, linfo_len))
 534                         return -EFAULT;
 535         }
 536
 537         prog_btf_id = prog->aux->attach_btf_id;
 538         mutex_lock(&targets_mutex);
 539         list_for_each_entry(iter, &targets, list) {
 540                 if (iter->btf_id == prog_btf_id) {
 541                         tinfo = iter;
 542                         break;
 543                 }
 544         }
 545         mutex_unlock(&targets_mutex);
 546         if (!tinfo)
 547                 return -ENOENT;
 548
 549         /* Only allow sleepable program for resched-able iterator */
 550         if (prog->sleepable && !bpf_iter_target_support_resched(tinfo))
 551                 return -EINVAL;
 552
 553         link = kzalloc(sizeof(*link), GFP_USER | __GFP_NOWARN);
 554         if (!link)
 555                 return -ENOMEM;
 556
 557         bpf_link_init(&link->link, BPF_LINK_TYPE_ITER, &bpf_iter_link_lops, prog);
 558         link->tinfo = tinfo;
 559
 560         err = bpf_link_prime(&link->link, &link_primer);
 561         if (err) {
 562                 kfree(link);
 563                 return err;
 564         }
 565
 566         if (tinfo->reg_info->attach_target) {
 567                 err = tinfo->reg_info->attach_target(prog, &linfo, &link->aux);
 568                 if (err) {
 569                         bpf_link_cleanup(&link_primer);
 570                         return err;
 571                 }
 572         }
 573
 574         return bpf_link_settle(&link_primer);
 575 }
 576
 577 static void init_seq_meta(struct bpf_iter_priv_data *priv_data,
 578                           struct bpf_iter_target_info *tinfo,
 579                           const struct bpf_iter_seq_info *seq_info,
 580                           struct bpf_prog *prog)
 581 {
 582         priv_data->tinfo = tinfo;
 583         priv_data->seq_info = seq_info;
 584         priv_data->prog = prog;
 585         priv_data->session_id = atomic64_inc_return(&session_id);
 586         priv_data->seq_num = 0;
 587         priv_data->done_stop = false;
 588 }
 589
 590 static int prepare_seq_file(struct file *file, struct bpf_iter_link *link,
 591                             const struct bpf_iter_seq_info *seq_info)
 592 {
 593         struct bpf_iter_priv_data *priv_data;
 594         struct bpf_iter_target_info *tinfo;
 595         struct bpf_prog *prog;
 596         u32 total_priv_dsize;
 597         struct seq_file *seq;
 598         int err = 0;
 599
 600         mutex_lock(&link_mutex);
 601         prog = link->link.prog;
 602         bpf_prog_inc(prog);
 603         mutex_unlock(&link_mutex);
 604
 605         tinfo = link->tinfo;
 606         total_priv_dsize = offsetof(struct bpf_iter_priv_data, target_private) +
 607                            seq_info->seq_priv_size;
 608         priv_data = __seq_open_private(file, seq_info->seq_ops,
 609                                        total_priv_dsize);
 610         if (!priv_data) {
 611                 err = -ENOMEM;
 612                 goto release_prog;
 613         }
 614
 615         if (seq_info->init_seq_private) {
 616                 err = seq_info->init_seq_private(priv_data->target_private, &link->aux);
 617                 if (err)
 618                         goto release_seq_file;
 619         }
 620
 621         init_seq_meta(priv_data, tinfo, seq_info, prog);
 622         seq = file->private_data;
 623         seq->private = priv_data->target_private;
 624
 625         return 0;
 626
 627 release_seq_file:
 628         seq_release_private(file->f_inode, file);
 629         file->private_data = NULL;
 630 release_prog:
 631         bpf_prog_put(prog);
 632         return err;
 633 }
 634
 635 int bpf_iter_new_fd(struct bpf_link *link)
 636 {
 637         struct bpf_iter_link *iter_link;
 638         struct file *file;
 639         unsigned int flags;
 640         int err, fd;
 641
 642         if (link->ops != &bpf_iter_link_lops)
 643                 return -EINVAL;
 644
 645         flags = O_RDONLY | O_CLOEXEC;
 646         fd = get_unused_fd_flags(flags);
 647         if (fd < 0)
 648                 return fd;
 649
 650         file = anon_inode_getfile("bpf_iter", &bpf_iter_fops, NULL, flags);
 651         if (IS_ERR(file)) {
 652                 err = PTR_ERR(file);
 653                 goto free_fd;
 654         }
 655
 656         iter_link = container_of(link, struct bpf_iter_link, link);
 657         err = prepare_seq_file(file, iter_link, __get_seq_info(iter_link));
 658         if (err)
 659                 goto free_file;
 660
 661         fd_install(fd, file);
 662         return fd;
 663
 664 free_file:
 665         fput(file);
 666 free_fd:
 667         put_unused_fd(fd);
 668         return err;
 669 }
 670
 671 struct bpf_prog *bpf_iter_get_info(struct bpf_iter_meta *meta, bool in_stop)
 672 {
 673         struct bpf_iter_priv_data *iter_priv;
 674         struct seq_file *seq;
 675         void *seq_priv;
 676
 677         seq = meta->seq;
 678         if (seq->file->f_op != &bpf_iter_fops)
 679                 return NULL;
 680
 681         seq_priv = seq->private;
 682         iter_priv = container_of(seq_priv, struct bpf_iter_priv_data,
 683                                  target_private);
 684
 685         if (in_stop && iter_priv->done_stop)
 686                 return NULL;
 687
 688         meta->session_id = iter_priv->session_id;
 689         meta->seq_num = iter_priv->seq_num;
 690
 691         return iter_priv->prog;
 692 }
 693
 694 int bpf_iter_run_prog(struct bpf_prog *prog, void *ctx)
 695 {
 696         struct bpf_run_ctx run_ctx, *old_run_ctx;
 697         int ret;
 698
 699         if (prog->sleepable) {
 700                 rcu_read_lock_trace();
 701                 migrate_disable();
 702                 might_fault();
 703                 old_run_ctx = bpf_set_run_ctx(&run_ctx);
 704                 ret = bpf_prog_run(prog, ctx);
 705                 bpf_reset_run_ctx(old_run_ctx);
 706                 migrate_enable();
 707                 rcu_read_unlock_trace();
 708         } else {
 709                 rcu_read_lock();
 710                 migrate_disable();
 711                 old_run_ctx = bpf_set_run_ctx(&run_ctx);
 712                 ret = bpf_prog_run(prog, ctx);
 713                 bpf_reset_run_ctx(old_run_ctx);
 714                 migrate_enable();
 715                 rcu_read_unlock();
 716         }
 717
 718         /* bpf program can only return 0 or 1:
 719          *  0 : okay
 720          *  1 : retry the same object
 721          * The bpf_iter_run_prog() return value
 722          * will be seq_ops->show() return value.
 723          */
 724         return ret == 0 ? 0 : -EAGAIN;
 725 }
 726
 727 BPF_CALL_4(bpf_for_each_map_elem, struct bpf_map *, map, void *, callback_fn,
 728            void *, callback_ctx, u64, flags)
 729 {
 730         return map->ops->map_for_each_callback(map, callback_fn, callback_ctx, flags);
 731 }
 732
 733 const struct bpf_func_proto bpf_for_each_map_elem_proto = {
 734         .func           = bpf_for_each_map_elem,
 735         .gpl_only       = false,
 736         .ret_type       = RET_INTEGER,
 737         .arg1_type      = ARG_CONST_MAP_PTR,
 738         .arg2_type      = ARG_PTR_TO_FUNC,
 739         .arg3_type      = ARG_PTR_TO_STACK_OR_NULL,
 740         .arg4_type      = ARG_ANYTHING,
 741 };
 742
 743 BPF_CALL_4(bpf_loop, u32, nr_loops, void *, callback_fn, void *, callback_ctx,
 744            u64, flags)
 745 {
 746         bpf_callback_t callback = (bpf_callback_t)callback_fn;
 747         u64 ret;
 748         u32 i;
 749
 750         /* Note: these safety checks are also verified when bpf_loop
 751          * is inlined, be careful to modify this code in sync. See
 752          * function verifier.c:inline_bpf_loop.
 753          */
 754         if (flags)
 755                 return -EINVAL;
 756         if (nr_loops > BPF_MAX_LOOPS)
 757                 return -E2BIG;
 758
 759         for (i = 0; i < nr_loops; i++) {
 760                 ret = callback((u64)i, (u64)(long)callback_ctx, 0, 0, 0);
 761                 /* return value: 0 - continue, 1 - stop and return */
 762                 if (ret)
 763                         return i + 1;
 764         }
 765
 766         return i;
 767 }
 768
 769 const struct bpf_func_proto bpf_loop_proto = {
 770         .func           = bpf_loop,
 771         .gpl_only       = false,
 772         .ret_type       = RET_INTEGER,
 773         .arg1_type      = ARG_ANYTHING,
 774         .arg2_type      = ARG_PTR_TO_FUNC,
 775         .arg3_type      = ARG_PTR_TO_STACK_OR_NULL,
 776         .arg4_type      = ARG_ANYTHING,
 777 };
 778
 779 struct bpf_iter_num_kern {
 780         int cur; /* current value, inclusive */
 781         int end; /* final value, exclusive */
 782 } __aligned(8);
 783
 784 __bpf_kfunc_start_defs();
 785
 786 __bpf_kfunc int bpf_iter_num_new(struct bpf_iter_num *it, int start, int end)
 787 {
 788         struct bpf_iter_num_kern *s = (void *)it;
 789
 790         BUILD_BUG_ON(sizeof(struct bpf_iter_num_kern) != sizeof(struct bpf_iter_num));
 791         BUILD_BUG_ON(__alignof__(struct bpf_iter_num_kern) != __alignof__(struct bpf_iter_num));
 792
 793         /* start == end is legit, it's an empty range and we'll just get NULL
 794          * on first (and any subsequent) bpf_iter_num_next() call
 795          */
 796         if (start > end) {
 797                 s->cur = s->end = 0;
 798                 return -EINVAL;
 799         }
 800
 801         /* avoid overflows, e.g., if start == INT_MIN and end == INT_MAX */
 802         if ((s64)end - (s64)start > BPF_MAX_LOOPS) {
 803                 s->cur = s->end = 0;
 804                 return -E2BIG;
 805         }
 806
 807         /* user will call bpf_iter_num_next() first,
 808          * which will set s->cur to exactly start value;
 809          * underflow shouldn't matter
 810          */
 811         s->cur = start - 1;
 812         s->end = end;
 813
 814         return 0;
 815 }
 816
 817 __bpf_kfunc int *bpf_iter_num_next(struct bpf_iter_num* it)
 818 {
 819         struct bpf_iter_num_kern *s = (void *)it;
 820
 821         /* check failed initialization or if we are done (same behavior);
 822          * need to be careful about overflow, so convert to s64 for checks,
 823          * e.g., if s->cur == s->end == INT_MAX, we can't just do
 824          * s->cur + 1 >= s->end
 825          */
 826         if ((s64)(s->cur + 1) >= s->end) {
 827                 s->cur = s->end = 0;
 828                 return NULL;
 829         }
 830
 831         s->cur++;
 832
 833         return &s->cur;
 834 }
 835
 836 __bpf_kfunc void bpf_iter_num_destroy(struct bpf_iter_num *it)
 837 {
 838         struct bpf_iter_num_kern *s = (void *)it;
 839
 840         s->cur = s->end = 0;
 841 }
 842
 843 __bpf_kfunc_end_defs();