kernel/bpf/helpers.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/rcupdate.h>
   6 #include <linux/random.h>
   7 #include <linux/smp.h>
   8 #include <linux/topology.h>
   9 #include <linux/ktime.h>
  10 #include <linux/sched.h>
  11 #include <linux/uidgid.h>
  12 #include <linux/filter.h>
  13 #include <linux/ctype.h>
  14 #include <linux/jiffies.h>
  15 #include <linux/pid_namespace.h>
  16 #include <linux/proc_ns.h>
  17 #include <linux/security.h>
  18
  19 #include "../../lib/kstrtox.h"
  20
  21 /* If kernel subsystem is allowing eBPF programs to call this function,
  22  * inside its own verifier_ops->get_func_proto() callback it should return
  23  * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
  24  *
  25  * Different map implementations will rely on rcu in map methods
  26  * lookup/update/delete, therefore eBPF programs must run under rcu lock
  27  * if program is allowed to access maps, so check rcu_read_lock_held in
  28  * all three functions.
  29  */
  30 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
  31 {
  32         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  33         return (unsigned long) map->ops->map_lookup_elem(map, key);
  34 }
  35
  36 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
  37         .func           = bpf_map_lookup_elem,
  38         .gpl_only       = false,
  39         .pkt_access     = true,
  40         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
  41         .arg1_type      = ARG_CONST_MAP_PTR,
  42         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  43 };
  44
  45 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
  46            void *, value, u64, flags)
  47 {
  48         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  49         return map->ops->map_update_elem(map, key, value, flags);
  50 }
  51
  52 const struct bpf_func_proto bpf_map_update_elem_proto = {
  53         .func           = bpf_map_update_elem,
  54         .gpl_only       = false,
  55         .pkt_access     = true,
  56         .ret_type       = RET_INTEGER,
  57         .arg1_type      = ARG_CONST_MAP_PTR,
  58         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  59         .arg3_type      = ARG_PTR_TO_MAP_VALUE,
  60         .arg4_type      = ARG_ANYTHING,
  61 };
  62
  63 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
  64 {
  65         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  66         return map->ops->map_delete_elem(map, key);
  67 }
  68
  69 const struct bpf_func_proto bpf_map_delete_elem_proto = {
  70         .func           = bpf_map_delete_elem,
  71         .gpl_only       = false,
  72         .pkt_access     = true,
  73         .ret_type       = RET_INTEGER,
  74         .arg1_type      = ARG_CONST_MAP_PTR,
  75         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  76 };
  77
  78 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
  79 {
  80         return map->ops->map_push_elem(map, value, flags);
  81 }
  82
  83 const struct bpf_func_proto bpf_map_push_elem_proto = {
  84         .func           = bpf_map_push_elem,
  85         .gpl_only       = false,
  86         .pkt_access     = true,
  87         .ret_type       = RET_INTEGER,
  88         .arg1_type      = ARG_CONST_MAP_PTR,
  89         .arg2_type      = ARG_PTR_TO_MAP_VALUE,
  90         .arg3_type      = ARG_ANYTHING,
  91 };
  92
  93 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
  94 {
  95         return map->ops->map_pop_elem(map, value);
  96 }
  97
  98 const struct bpf_func_proto bpf_map_pop_elem_proto = {
  99         .func           = bpf_map_pop_elem,
 100         .gpl_only       = false,
 101         .ret_type       = RET_INTEGER,
 102         .arg1_type      = ARG_CONST_MAP_PTR,
 103         .arg2_type      = ARG_PTR_TO_UNINIT_MAP_VALUE,
 104 };
 105
 106 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
 107 {
 108         return map->ops->map_peek_elem(map, value);
 109 }
 110
 111 const struct bpf_func_proto bpf_map_peek_elem_proto = {
 112         .func           = bpf_map_peek_elem,
 113         .gpl_only       = false,
 114         .ret_type       = RET_INTEGER,
 115         .arg1_type      = ARG_CONST_MAP_PTR,
 116         .arg2_type      = ARG_PTR_TO_UNINIT_MAP_VALUE,
 117 };
 118
 119 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
 120         .func           = bpf_user_rnd_u32,
 121         .gpl_only       = false,
 122         .ret_type       = RET_INTEGER,
 123 };
 124
 125 BPF_CALL_0(bpf_get_smp_processor_id)
 126 {
 127         return smp_processor_id();
 128 }
 129
 130 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
 131         .func           = bpf_get_smp_processor_id,
 132         .gpl_only       = false,
 133         .ret_type       = RET_INTEGER,
 134 };
 135
 136 BPF_CALL_0(bpf_get_numa_node_id)
 137 {
 138         return numa_node_id();
 139 }
 140
 141 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
 142         .func           = bpf_get_numa_node_id,
 143         .gpl_only       = false,
 144         .ret_type       = RET_INTEGER,
 145 };
 146
 147 BPF_CALL_0(bpf_ktime_get_ns)
 148 {
 149         /* NMI safe access to clock monotonic */
 150         return ktime_get_mono_fast_ns();
 151 }
 152
 153 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
 154         .func           = bpf_ktime_get_ns,
 155         .gpl_only       = false,
 156         .ret_type       = RET_INTEGER,
 157 };
 158
 159 BPF_CALL_0(bpf_ktime_get_boot_ns)
 160 {
 161         /* NMI safe access to clock boottime */
 162         return ktime_get_boot_fast_ns();
 163 }
 164
 165 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
 166         .func           = bpf_ktime_get_boot_ns,
 167         .gpl_only       = false,
 168         .ret_type       = RET_INTEGER,
 169 };
 170
 171 BPF_CALL_0(bpf_ktime_get_coarse_ns)
 172 {
 173         return ktime_get_coarse_ns();
 174 }
 175
 176 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
 177         .func           = bpf_ktime_get_coarse_ns,
 178         .gpl_only       = false,
 179         .ret_type       = RET_INTEGER,
 180 };
 181
 182 BPF_CALL_0(bpf_get_current_pid_tgid)
 183 {
 184         struct task_struct *task = current;
 185
 186         if (unlikely(!task))
 187                 return -EINVAL;
 188
 189         return (u64) task->tgid << 32 | task->pid;
 190 }
 191
 192 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
 193         .func           = bpf_get_current_pid_tgid,
 194         .gpl_only       = false,
 195         .ret_type       = RET_INTEGER,
 196 };
 197
 198 BPF_CALL_0(bpf_get_current_uid_gid)
 199 {
 200         struct task_struct *task = current;
 201         kuid_t uid;
 202         kgid_t gid;
 203
 204         if (unlikely(!task))
 205                 return -EINVAL;
 206
 207         current_uid_gid(&uid, &gid);
 208         return (u64) from_kgid(&init_user_ns, gid) << 32 |
 209                      from_kuid(&init_user_ns, uid);
 210 }
 211
 212 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
 213         .func           = bpf_get_current_uid_gid,
 214         .gpl_only       = false,
 215         .ret_type       = RET_INTEGER,
 216 };
 217
 218 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
 219 {
 220         struct task_struct *task = current;
 221
 222         if (unlikely(!task))
 223                 goto err_clear;
 224
 225         strncpy(buf, task->comm, size);
 226
 227         /* Verifier guarantees that size > 0. For task->comm exceeding
 228          * size, guarantee that buf is %NUL-terminated. Unconditionally
 229          * done here to save the size test.
 230          */
 231         buf[size - 1] = 0;
 232         return 0;
 233 err_clear:
 234         memset(buf, 0, size);
 235         return -EINVAL;
 236 }
 237
 238 const struct bpf_func_proto bpf_get_current_comm_proto = {
 239         .func           = bpf_get_current_comm,
 240         .gpl_only       = false,
 241         .ret_type       = RET_INTEGER,
 242         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 243         .arg2_type      = ARG_CONST_SIZE,
 244 };
 245
 246 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
 247
 248 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
 249 {
 250         arch_spinlock_t *l = (void *)lock;
 251         union {
 252                 __u32 val;
 253                 arch_spinlock_t lock;
 254         } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
 255
 256         compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
 257         BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
 258         BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
 259         arch_spin_lock(l);
 260 }
 261
 262 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
 263 {
 264         arch_spinlock_t *l = (void *)lock;
 265
 266         arch_spin_unlock(l);
 267 }
 268
 269 #else
 270
 271 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
 272 {
 273         atomic_t *l = (void *)lock;
 274
 275         BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
 276         do {
 277                 atomic_cond_read_relaxed(l, !VAL);
 278         } while (atomic_xchg(l, 1));
 279 }
 280
 281 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
 282 {
 283         atomic_t *l = (void *)lock;
 284
 285         atomic_set_release(l, 0);
 286 }
 287
 288 #endif
 289
 290 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
 291
 292 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
 293 {
 294         unsigned long flags;
 295
 296         local_irq_save(flags);
 297         __bpf_spin_lock(lock);
 298         __this_cpu_write(irqsave_flags, flags);
 299         return 0;
 300 }
 301
 302 const struct bpf_func_proto bpf_spin_lock_proto = {
 303         .func           = bpf_spin_lock,
 304         .gpl_only       = false,
 305         .ret_type       = RET_VOID,
 306         .arg1_type      = ARG_PTR_TO_SPIN_LOCK,
 307 };
 308
 309 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
 310 {
 311         unsigned long flags;
 312
 313         flags = __this_cpu_read(irqsave_flags);
 314         __bpf_spin_unlock(lock);
 315         local_irq_restore(flags);
 316         return 0;
 317 }
 318
 319 const struct bpf_func_proto bpf_spin_unlock_proto = {
 320         .func           = bpf_spin_unlock,
 321         .gpl_only       = false,
 322         .ret_type       = RET_VOID,
 323         .arg1_type      = ARG_PTR_TO_SPIN_LOCK,
 324 };
 325
 326 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
 327                            bool lock_src)
 328 {
 329         struct bpf_spin_lock *lock;
 330
 331         if (lock_src)
 332                 lock = src + map->spin_lock_off;
 333         else
 334                 lock = dst + map->spin_lock_off;
 335         preempt_disable();
 336         ____bpf_spin_lock(lock);
 337         copy_map_value(map, dst, src);
 338         ____bpf_spin_unlock(lock);
 339         preempt_enable();
 340 }
 341
 342 BPF_CALL_0(bpf_jiffies64)
 343 {
 344         return get_jiffies_64();
 345 }
 346
 347 const struct bpf_func_proto bpf_jiffies64_proto = {
 348         .func           = bpf_jiffies64,
 349         .gpl_only       = false,
 350         .ret_type       = RET_INTEGER,
 351 };
 352
 353 #ifdef CONFIG_CGROUPS
 354 BPF_CALL_0(bpf_get_current_cgroup_id)
 355 {
 356         struct cgroup *cgrp = task_dfl_cgroup(current);
 357
 358         return cgroup_id(cgrp);
 359 }
 360
 361 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
 362         .func           = bpf_get_current_cgroup_id,
 363         .gpl_only       = false,
 364         .ret_type       = RET_INTEGER,
 365 };
 366
 367 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
 368 {
 369         struct cgroup *cgrp = task_dfl_cgroup(current);
 370         struct cgroup *ancestor;
 371
 372         ancestor = cgroup_ancestor(cgrp, ancestor_level);
 373         if (!ancestor)
 374                 return 0;
 375         return cgroup_id(ancestor);
 376 }
 377
 378 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
 379         .func           = bpf_get_current_ancestor_cgroup_id,
 380         .gpl_only       = false,
 381         .ret_type       = RET_INTEGER,
 382         .arg1_type      = ARG_ANYTHING,
 383 };
 384
 385 #ifdef CONFIG_CGROUP_BPF
 386 DECLARE_PER_CPU(struct bpf_cgroup_storage_info,
 387                 bpf_cgroup_storage_info[BPF_CGROUP_STORAGE_NEST_MAX]);
 388
 389 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
 390 {
 391         /* flags argument is not used now,
 392          * but provides an ability to extend the API.
 393          * verifier checks that its value is correct.
 394          */
 395         enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
 396         struct bpf_cgroup_storage *storage = NULL;
 397         void *ptr;
 398         int i;
 399
 400         for (i = BPF_CGROUP_STORAGE_NEST_MAX - 1; i >= 0; i--) {
 401                 if (likely(this_cpu_read(bpf_cgroup_storage_info[i].task) != current))
 402                         continue;
 403
 404                 storage = this_cpu_read(bpf_cgroup_storage_info[i].storage[stype]);
 405                 break;
 406         }
 407
 408         if (stype == BPF_CGROUP_STORAGE_SHARED)
 409                 ptr = &READ_ONCE(storage->buf)->data[0];
 410         else
 411                 ptr = this_cpu_ptr(storage->percpu_buf);
 412
 413         return (unsigned long)ptr;
 414 }
 415
 416 const struct bpf_func_proto bpf_get_local_storage_proto = {
 417         .func           = bpf_get_local_storage,
 418         .gpl_only       = false,
 419         .ret_type       = RET_PTR_TO_MAP_VALUE,
 420         .arg1_type      = ARG_CONST_MAP_PTR,
 421         .arg2_type      = ARG_ANYTHING,
 422 };
 423 #endif
 424
 425 #define BPF_STRTOX_BASE_MASK 0x1F
 426
 427 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
 428                           unsigned long long *res, bool *is_negative)
 429 {
 430         unsigned int base = flags & BPF_STRTOX_BASE_MASK;
 431         const char *cur_buf = buf;
 432         size_t cur_len = buf_len;
 433         unsigned int consumed;
 434         size_t val_len;
 435         char str[64];
 436
 437         if (!buf || !buf_len || !res || !is_negative)
 438                 return -EINVAL;
 439
 440         if (base != 0 && base != 8 && base != 10 && base != 16)
 441                 return -EINVAL;
 442
 443         if (flags & ~BPF_STRTOX_BASE_MASK)
 444                 return -EINVAL;
 445
 446         while (cur_buf < buf + buf_len && isspace(*cur_buf))
 447                 ++cur_buf;
 448
 449         *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
 450         if (*is_negative)
 451                 ++cur_buf;
 452
 453         consumed = cur_buf - buf;
 454         cur_len -= consumed;
 455         if (!cur_len)
 456                 return -EINVAL;
 457
 458         cur_len = min(cur_len, sizeof(str) - 1);
 459         memcpy(str, cur_buf, cur_len);
 460         str[cur_len] = '\0';
 461         cur_buf = str;
 462
 463         cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
 464         val_len = _parse_integer(cur_buf, base, res);
 465
 466         if (val_len & KSTRTOX_OVERFLOW)
 467                 return -ERANGE;
 468
 469         if (val_len == 0)
 470                 return -EINVAL;
 471
 472         cur_buf += val_len;
 473         consumed += cur_buf - str;
 474
 475         return consumed;
 476 }
 477
 478 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
 479                          long long *res)
 480 {
 481         unsigned long long _res;
 482         bool is_negative;
 483         int err;
 484
 485         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 486         if (err < 0)
 487                 return err;
 488         if (is_negative) {
 489                 if ((long long)-_res > 0)
 490                         return -ERANGE;
 491                 *res = -_res;
 492         } else {
 493                 if ((long long)_res < 0)
 494                         return -ERANGE;
 495                 *res = _res;
 496         }
 497         return err;
 498 }
 499
 500 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
 501            long *, res)
 502 {
 503         long long _res;
 504         int err;
 505
 506         err = __bpf_strtoll(buf, buf_len, flags, &_res);
 507         if (err < 0)
 508                 return err;
 509         if (_res != (long)_res)
 510                 return -ERANGE;
 511         *res = _res;
 512         return err;
 513 }
 514
 515 const struct bpf_func_proto bpf_strtol_proto = {
 516         .func           = bpf_strtol,
 517         .gpl_only       = false,
 518         .ret_type       = RET_INTEGER,
 519         .arg1_type      = ARG_PTR_TO_MEM,
 520         .arg2_type      = ARG_CONST_SIZE,
 521         .arg3_type      = ARG_ANYTHING,
 522         .arg4_type      = ARG_PTR_TO_LONG,
 523 };
 524
 525 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
 526            unsigned long *, res)
 527 {
 528         unsigned long long _res;
 529         bool is_negative;
 530         int err;
 531
 532         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 533         if (err < 0)
 534                 return err;
 535         if (is_negative)
 536                 return -EINVAL;
 537         if (_res != (unsigned long)_res)
 538                 return -ERANGE;
 539         *res = _res;
 540         return err;
 541 }
 542
 543 const struct bpf_func_proto bpf_strtoul_proto = {
 544         .func           = bpf_strtoul,
 545         .gpl_only       = false,
 546         .ret_type       = RET_INTEGER,
 547         .arg1_type      = ARG_PTR_TO_MEM,
 548         .arg2_type      = ARG_CONST_SIZE,
 549         .arg3_type      = ARG_ANYTHING,
 550         .arg4_type      = ARG_PTR_TO_LONG,
 551 };
 552 #endif
 553
 554 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
 555            struct bpf_pidns_info *, nsdata, u32, size)
 556 {
 557         struct task_struct *task = current;
 558         struct pid_namespace *pidns;
 559         int err = -EINVAL;
 560
 561         if (unlikely(size != sizeof(struct bpf_pidns_info)))
 562                 goto clear;
 563
 564         if (unlikely((u64)(dev_t)dev != dev))
 565                 goto clear;
 566
 567         if (unlikely(!task))
 568                 goto clear;
 569
 570         pidns = task_active_pid_ns(task);
 571         if (unlikely(!pidns)) {
 572                 err = -ENOENT;
 573                 goto clear;
 574         }
 575
 576         if (!ns_match(&pidns->ns, (dev_t)dev, ino))
 577                 goto clear;
 578
 579         nsdata->pid = task_pid_nr_ns(task, pidns);
 580         nsdata->tgid = task_tgid_nr_ns(task, pidns);
 581         return 0;
 582 clear:
 583         memset((void *)nsdata, 0, (size_t) size);
 584         return err;
 585 }
 586
 587 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
 588         .func           = bpf_get_ns_current_pid_tgid,
 589         .gpl_only       = false,
 590         .ret_type       = RET_INTEGER,
 591         .arg1_type      = ARG_ANYTHING,
 592         .arg2_type      = ARG_ANYTHING,
 593         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 594         .arg4_type      = ARG_CONST_SIZE,
 595 };
 596
 597 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 598         .func           = bpf_get_raw_cpu_id,
 599         .gpl_only       = false,
 600         .ret_type       = RET_INTEGER,
 601 };
 602
 603 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
 604            u64, flags, void *, data, u64, size)
 605 {
 606         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 607                 return -EINVAL;
 608
 609         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
 610 }
 611
 612 const struct bpf_func_proto bpf_event_output_data_proto =  {
 613         .func           = bpf_event_output_data,
 614         .gpl_only       = true,
 615         .ret_type       = RET_INTEGER,
 616         .arg1_type      = ARG_PTR_TO_CTX,
 617         .arg2_type      = ARG_CONST_MAP_PTR,
 618         .arg3_type      = ARG_ANYTHING,
 619         .arg4_type      = ARG_PTR_TO_MEM,
 620         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 621 };
 622
 623 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
 624            const void __user *, user_ptr)
 625 {
 626         int ret = copy_from_user(dst, user_ptr, size);
 627
 628         if (unlikely(ret)) {
 629                 memset(dst, 0, size);
 630                 ret = -EFAULT;
 631         }
 632
 633         return ret;
 634 }
 635
 636 const struct bpf_func_proto bpf_copy_from_user_proto = {
 637         .func           = bpf_copy_from_user,
 638         .gpl_only       = false,
 639         .ret_type       = RET_INTEGER,
 640         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 641         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 642         .arg3_type      = ARG_ANYTHING,
 643 };
 644
 645 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
 646 {
 647         if (cpu >= nr_cpu_ids)
 648                 return (unsigned long)NULL;
 649
 650         return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
 651 }
 652
 653 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
 654         .func           = bpf_per_cpu_ptr,
 655         .gpl_only       = false,
 656         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL,
 657         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 658         .arg2_type      = ARG_ANYTHING,
 659 };
 660
 661 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
 662 {
 663         return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
 664 }
 665
 666 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
 667         .func           = bpf_this_cpu_ptr,
 668         .gpl_only       = false,
 669         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID,
 670         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 671 };
 672
 673 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
 674                 size_t bufsz)
 675 {
 676         void __user *user_ptr = (__force void __user *)unsafe_ptr;
 677
 678         buf[0] = 0;
 679
 680         switch (fmt_ptype) {
 681         case 's':
 682 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 683                 if ((unsigned long)unsafe_ptr < TASK_SIZE)
 684                         return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 685                 fallthrough;
 686 #endif
 687         case 'k':
 688                 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
 689         case 'u':
 690                 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 691         }
 692
 693         return -EINVAL;
 694 }
 695
 696 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
 697  * arguments representation.
 698  */
 699 #define MAX_BPRINTF_BUF_LEN     512
 700
 701 /* Support executing three nested bprintf helper calls on a given CPU */
 702 #define MAX_BPRINTF_NEST_LEVEL  3
 703 struct bpf_bprintf_buffers {
 704         char tmp_bufs[MAX_BPRINTF_NEST_LEVEL][MAX_BPRINTF_BUF_LEN];
 705 };
 706 static DEFINE_PER_CPU(struct bpf_bprintf_buffers, bpf_bprintf_bufs);
 707 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
 708
 709 static int try_get_fmt_tmp_buf(char **tmp_buf)
 710 {
 711         struct bpf_bprintf_buffers *bufs;
 712         int nest_level;
 713
 714         preempt_disable();
 715         nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
 716         if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
 717                 this_cpu_dec(bpf_bprintf_nest_level);
 718                 preempt_enable();
 719                 return -EBUSY;
 720         }
 721         bufs = this_cpu_ptr(&bpf_bprintf_bufs);
 722         *tmp_buf = bufs->tmp_bufs[nest_level - 1];
 723
 724         return 0;
 725 }
 726
 727 void bpf_bprintf_cleanup(void)
 728 {
 729         if (this_cpu_read(bpf_bprintf_nest_level)) {
 730                 this_cpu_dec(bpf_bprintf_nest_level);
 731                 preempt_enable();
 732         }
 733 }
 734
 735 /*
 736  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
 737  *
 738  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
 739  *
 740  * This can be used in two ways:
 741  * - Format string verification only: when bin_args is NULL
 742  * - Arguments preparation: in addition to the above verification, it writes in
 743  *   bin_args a binary representation of arguments usable by bstr_printf where
 744  *   pointers from BPF have been sanitized.
 745  *
 746  * In argument preparation mode, if 0 is returned, safe temporary buffers are
 747  * allocated and bpf_bprintf_cleanup should be called to free them after use.
 748  */
 749 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
 750                         u32 **bin_args, u32 num_args)
 751 {
 752         char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
 753         size_t sizeof_cur_arg, sizeof_cur_ip;
 754         int err, i, num_spec = 0;
 755         u64 cur_arg;
 756         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
 757
 758         fmt_end = strnchr(fmt, fmt_size, 0);
 759         if (!fmt_end)
 760                 return -EINVAL;
 761         fmt_size = fmt_end - fmt;
 762
 763         if (bin_args) {
 764                 if (num_args && try_get_fmt_tmp_buf(&tmp_buf))
 765                         return -EBUSY;
 766
 767                 tmp_buf_end = tmp_buf + MAX_BPRINTF_BUF_LEN;
 768                 *bin_args = (u32 *)tmp_buf;
 769         }
 770
 771         for (i = 0; i < fmt_size; i++) {
 772                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
 773                         err = -EINVAL;
 774                         goto out;
 775                 }
 776
 777                 if (fmt[i] != '%')
 778                         continue;
 779
 780                 if (fmt[i + 1] == '%') {
 781                         i++;
 782                         continue;
 783                 }
 784
 785                 if (num_spec >= num_args) {
 786                         err = -EINVAL;
 787                         goto out;
 788                 }
 789
 790                 /* The string is zero-terminated so if fmt[i] != 0, we can
 791                  * always access fmt[i + 1], in the worst case it will be a 0
 792                  */
 793                 i++;
 794
 795                 /* skip optional "[0 +-][num]" width formatting field */
 796                 while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
 797                        fmt[i] == ' ')
 798                         i++;
 799                 if (fmt[i] >= '1' && fmt[i] <= '9') {
 800                         i++;
 801                         while (fmt[i] >= '0' && fmt[i] <= '9')
 802                                 i++;
 803                 }
 804
 805                 if (fmt[i] == 'p') {
 806                         sizeof_cur_arg = sizeof(long);
 807
 808                         if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
 809                             fmt[i + 2] == 's') {
 810                                 fmt_ptype = fmt[i + 1];
 811                                 i += 2;
 812                                 goto fmt_str;
 813                         }
 814
 815                         if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
 816                             ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
 817                             fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
 818                             fmt[i + 1] == 'S') {
 819                                 /* just kernel pointers */
 820                                 if (tmp_buf)
 821                                         cur_arg = raw_args[num_spec];
 822                                 i++;
 823                                 goto nocopy_fmt;
 824                         }
 825
 826                         if (fmt[i + 1] == 'B') {
 827                                 if (tmp_buf)  {
 828                                         err = snprintf(tmp_buf,
 829                                                        (tmp_buf_end - tmp_buf),
 830                                                        "%pB",
 831                                                        (void *)(long)raw_args[num_spec]);
 832                                         tmp_buf += (err + 1);
 833                                 }
 834
 835                                 i++;
 836                                 num_spec++;
 837                                 continue;
 838                         }
 839
 840                         /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
 841                         if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
 842                             (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
 843                                 err = -EINVAL;
 844                                 goto out;
 845                         }
 846
 847                         i += 2;
 848                         if (!tmp_buf)
 849                                 goto nocopy_fmt;
 850
 851                         sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
 852                         if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
 853                                 err = -ENOSPC;
 854                                 goto out;
 855                         }
 856
 857                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 858                         err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
 859                                                        sizeof_cur_ip);
 860                         if (err < 0)
 861                                 memset(cur_ip, 0, sizeof_cur_ip);
 862
 863                         /* hack: bstr_printf expects IP addresses to be
 864                          * pre-formatted as strings, ironically, the easiest way
 865                          * to do that is to call snprintf.
 866                          */
 867                         ip_spec[2] = fmt[i - 1];
 868                         ip_spec[3] = fmt[i];
 869                         err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
 870                                        ip_spec, &cur_ip);
 871
 872                         tmp_buf += err + 1;
 873                         num_spec++;
 874
 875                         continue;
 876                 } else if (fmt[i] == 's') {
 877                         fmt_ptype = fmt[i];
 878 fmt_str:
 879                         if (fmt[i + 1] != 0 &&
 880                             !isspace(fmt[i + 1]) &&
 881                             !ispunct(fmt[i + 1])) {
 882                                 err = -EINVAL;
 883                                 goto out;
 884                         }
 885
 886                         if (!tmp_buf)
 887                                 goto nocopy_fmt;
 888
 889                         if (tmp_buf_end == tmp_buf) {
 890                                 err = -ENOSPC;
 891                                 goto out;
 892                         }
 893
 894                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 895                         err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
 896                                                     fmt_ptype,
 897                                                     tmp_buf_end - tmp_buf);
 898                         if (err < 0) {
 899                                 tmp_buf[0] = '\0';
 900                                 err = 1;
 901                         }
 902
 903                         tmp_buf += err;
 904                         num_spec++;
 905
 906                         continue;
 907                 }
 908
 909                 sizeof_cur_arg = sizeof(int);
 910
 911                 if (fmt[i] == 'l') {
 912                         sizeof_cur_arg = sizeof(long);
 913                         i++;
 914                 }
 915                 if (fmt[i] == 'l') {
 916                         sizeof_cur_arg = sizeof(long long);
 917                         i++;
 918                 }
 919
 920                 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
 921                     fmt[i] != 'x' && fmt[i] != 'X') {
 922                         err = -EINVAL;
 923                         goto out;
 924                 }
 925
 926                 if (tmp_buf)
 927                         cur_arg = raw_args[num_spec];
 928 nocopy_fmt:
 929                 if (tmp_buf) {
 930                         tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
 931                         if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
 932                                 err = -ENOSPC;
 933                                 goto out;
 934                         }
 935
 936                         if (sizeof_cur_arg == 8) {
 937                                 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
 938                                 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
 939                         } else {
 940                                 *(u32 *)tmp_buf = (u32)(long)cur_arg;
 941                         }
 942                         tmp_buf += sizeof_cur_arg;
 943                 }
 944                 num_spec++;
 945         }
 946
 947         err = 0;
 948 out:
 949         if (err)
 950                 bpf_bprintf_cleanup();
 951         return err;
 952 }
 953
 954 #define MAX_SNPRINTF_VARARGS            12
 955
 956 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
 957            const void *, data, u32, data_len)
 958 {
 959         int err, num_args;
 960         u32 *bin_args;
 961
 962         if (data_len % 8 || data_len > MAX_SNPRINTF_VARARGS * 8 ||
 963             (data_len && !data))
 964                 return -EINVAL;
 965         num_args = data_len / 8;
 966
 967         /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
 968          * can safely give an unbounded size.
 969          */
 970         err = bpf_bprintf_prepare(fmt, UINT_MAX, data, &bin_args, num_args);
 971         if (err < 0)
 972                 return err;
 973
 974         err = bstr_printf(str, str_size, fmt, bin_args);
 975
 976         bpf_bprintf_cleanup();
 977
 978         return err + 1;
 979 }
 980
 981 const struct bpf_func_proto bpf_snprintf_proto = {
 982         .func           = bpf_snprintf,
 983         .gpl_only       = true,
 984         .ret_type       = RET_INTEGER,
 985         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
 986         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 987         .arg3_type      = ARG_PTR_TO_CONST_STR,
 988         .arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
 989         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 990 };
 991
 992 const struct bpf_func_proto bpf_get_current_task_proto __weak;
 993 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
 994 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
 995 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
 996 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
 997
 998 const struct bpf_func_proto *
 999 bpf_base_func_proto(enum bpf_func_id func_id)
1000 {
1001         switch (func_id) {
1002         case BPF_FUNC_map_lookup_elem:
1003                 return &bpf_map_lookup_elem_proto;
1004         case BPF_FUNC_map_update_elem:
1005                 return &bpf_map_update_elem_proto;
1006         case BPF_FUNC_map_delete_elem:
1007                 return &bpf_map_delete_elem_proto;
1008         case BPF_FUNC_map_push_elem:
1009                 return &bpf_map_push_elem_proto;
1010         case BPF_FUNC_map_pop_elem:
1011                 return &bpf_map_pop_elem_proto;
1012         case BPF_FUNC_map_peek_elem:
1013                 return &bpf_map_peek_elem_proto;
1014         case BPF_FUNC_get_prandom_u32:
1015                 return &bpf_get_prandom_u32_proto;
1016         case BPF_FUNC_get_smp_processor_id:
1017                 return &bpf_get_raw_smp_processor_id_proto;
1018         case BPF_FUNC_get_numa_node_id:
1019                 return &bpf_get_numa_node_id_proto;
1020         case BPF_FUNC_tail_call:
1021                 return &bpf_tail_call_proto;
1022         case BPF_FUNC_ktime_get_ns:
1023                 return &bpf_ktime_get_ns_proto;
1024         case BPF_FUNC_ktime_get_boot_ns:
1025                 return &bpf_ktime_get_boot_ns_proto;
1026         case BPF_FUNC_ktime_get_coarse_ns:
1027                 return &bpf_ktime_get_coarse_ns_proto;
1028         case BPF_FUNC_ringbuf_output:
1029                 return &bpf_ringbuf_output_proto;
1030         case BPF_FUNC_ringbuf_reserve:
1031                 return &bpf_ringbuf_reserve_proto;
1032         case BPF_FUNC_ringbuf_submit:
1033                 return &bpf_ringbuf_submit_proto;
1034         case BPF_FUNC_ringbuf_discard:
1035                 return &bpf_ringbuf_discard_proto;
1036         case BPF_FUNC_ringbuf_query:
1037                 return &bpf_ringbuf_query_proto;
1038         case BPF_FUNC_for_each_map_elem:
1039                 return &bpf_for_each_map_elem_proto;
1040         default:
1041                 break;
1042         }
1043
1044         if (!bpf_capable())
1045                 return NULL;
1046
1047         switch (func_id) {
1048         case BPF_FUNC_spin_lock:
1049                 return &bpf_spin_lock_proto;
1050         case BPF_FUNC_spin_unlock:
1051                 return &bpf_spin_unlock_proto;
1052         case BPF_FUNC_jiffies64:
1053                 return &bpf_jiffies64_proto;
1054         case BPF_FUNC_per_cpu_ptr:
1055                 return &bpf_per_cpu_ptr_proto;
1056         case BPF_FUNC_this_cpu_ptr:
1057                 return &bpf_this_cpu_ptr_proto;
1058         default:
1059                 break;
1060         }
1061
1062         if (!perfmon_capable())
1063                 return NULL;
1064
1065         switch (func_id) {
1066         case BPF_FUNC_trace_printk:
1067                 return bpf_get_trace_printk_proto();
1068         case BPF_FUNC_get_current_task:
1069                 return &bpf_get_current_task_proto;
1070         case BPF_FUNC_probe_read_user:
1071                 return &bpf_probe_read_user_proto;
1072         case BPF_FUNC_probe_read_kernel:
1073                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1074                        NULL : &bpf_probe_read_kernel_proto;
1075         case BPF_FUNC_probe_read_user_str:
1076                 return &bpf_probe_read_user_str_proto;
1077         case BPF_FUNC_probe_read_kernel_str:
1078                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1079                        NULL : &bpf_probe_read_kernel_str_proto;
1080         case BPF_FUNC_snprintf_btf:
1081                 return &bpf_snprintf_btf_proto;
1082         case BPF_FUNC_snprintf:
1083                 return &bpf_snprintf_proto;
1084         default:
1085                 return NULL;
1086         }
1087 }