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;
 357         u64 cgrp_id;
 358
 359         rcu_read_lock();
 360         cgrp = task_dfl_cgroup(current);
 361         cgrp_id = cgroup_id(cgrp);
 362         rcu_read_unlock();
 363
 364         return cgrp_id;
 365 }
 366
 367 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
 368         .func           = bpf_get_current_cgroup_id,
 369         .gpl_only       = false,
 370         .ret_type       = RET_INTEGER,
 371 };
 372
 373 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
 374 {
 375         struct cgroup *cgrp;
 376         struct cgroup *ancestor;
 377         u64 cgrp_id;
 378
 379         rcu_read_lock();
 380         cgrp = task_dfl_cgroup(current);
 381         ancestor = cgroup_ancestor(cgrp, ancestor_level);
 382         cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
 383         rcu_read_unlock();
 384
 385         return cgrp_id;
 386 }
 387
 388 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
 389         .func           = bpf_get_current_ancestor_cgroup_id,
 390         .gpl_only       = false,
 391         .ret_type       = RET_INTEGER,
 392         .arg1_type      = ARG_ANYTHING,
 393 };
 394
 395 #ifdef CONFIG_CGROUP_BPF
 396 DECLARE_PER_CPU(struct bpf_cgroup_storage_info,
 397                 bpf_cgroup_storage_info[BPF_CGROUP_STORAGE_NEST_MAX]);
 398
 399 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
 400 {
 401         /* flags argument is not used now,
 402          * but provides an ability to extend the API.
 403          * verifier checks that its value is correct.
 404          */
 405         enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
 406         struct bpf_cgroup_storage *storage = NULL;
 407         void *ptr;
 408         int i;
 409
 410         for (i = BPF_CGROUP_STORAGE_NEST_MAX - 1; i >= 0; i--) {
 411                 if (likely(this_cpu_read(bpf_cgroup_storage_info[i].task) != current))
 412                         continue;
 413
 414                 storage = this_cpu_read(bpf_cgroup_storage_info[i].storage[stype]);
 415                 break;
 416         }
 417
 418         if (stype == BPF_CGROUP_STORAGE_SHARED)
 419                 ptr = &READ_ONCE(storage->buf)->data[0];
 420         else
 421                 ptr = this_cpu_ptr(storage->percpu_buf);
 422
 423         return (unsigned long)ptr;
 424 }
 425
 426 const struct bpf_func_proto bpf_get_local_storage_proto = {
 427         .func           = bpf_get_local_storage,
 428         .gpl_only       = false,
 429         .ret_type       = RET_PTR_TO_MAP_VALUE,
 430         .arg1_type      = ARG_CONST_MAP_PTR,
 431         .arg2_type      = ARG_ANYTHING,
 432 };
 433 #endif
 434
 435 #define BPF_STRTOX_BASE_MASK 0x1F
 436
 437 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
 438                           unsigned long long *res, bool *is_negative)
 439 {
 440         unsigned int base = flags & BPF_STRTOX_BASE_MASK;
 441         const char *cur_buf = buf;
 442         size_t cur_len = buf_len;
 443         unsigned int consumed;
 444         size_t val_len;
 445         char str[64];
 446
 447         if (!buf || !buf_len || !res || !is_negative)
 448                 return -EINVAL;
 449
 450         if (base != 0 && base != 8 && base != 10 && base != 16)
 451                 return -EINVAL;
 452
 453         if (flags & ~BPF_STRTOX_BASE_MASK)
 454                 return -EINVAL;
 455
 456         while (cur_buf < buf + buf_len && isspace(*cur_buf))
 457                 ++cur_buf;
 458
 459         *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
 460         if (*is_negative)
 461                 ++cur_buf;
 462
 463         consumed = cur_buf - buf;
 464         cur_len -= consumed;
 465         if (!cur_len)
 466                 return -EINVAL;
 467
 468         cur_len = min(cur_len, sizeof(str) - 1);
 469         memcpy(str, cur_buf, cur_len);
 470         str[cur_len] = '\0';
 471         cur_buf = str;
 472
 473         cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
 474         val_len = _parse_integer(cur_buf, base, res);
 475
 476         if (val_len & KSTRTOX_OVERFLOW)
 477                 return -ERANGE;
 478
 479         if (val_len == 0)
 480                 return -EINVAL;
 481
 482         cur_buf += val_len;
 483         consumed += cur_buf - str;
 484
 485         return consumed;
 486 }
 487
 488 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
 489                          long long *res)
 490 {
 491         unsigned long long _res;
 492         bool is_negative;
 493         int err;
 494
 495         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 496         if (err < 0)
 497                 return err;
 498         if (is_negative) {
 499                 if ((long long)-_res > 0)
 500                         return -ERANGE;
 501                 *res = -_res;
 502         } else {
 503                 if ((long long)_res < 0)
 504                         return -ERANGE;
 505                 *res = _res;
 506         }
 507         return err;
 508 }
 509
 510 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
 511            long *, res)
 512 {
 513         long long _res;
 514         int err;
 515
 516         err = __bpf_strtoll(buf, buf_len, flags, &_res);
 517         if (err < 0)
 518                 return err;
 519         if (_res != (long)_res)
 520                 return -ERANGE;
 521         *res = _res;
 522         return err;
 523 }
 524
 525 const struct bpf_func_proto bpf_strtol_proto = {
 526         .func           = bpf_strtol,
 527         .gpl_only       = false,
 528         .ret_type       = RET_INTEGER,
 529         .arg1_type      = ARG_PTR_TO_MEM,
 530         .arg2_type      = ARG_CONST_SIZE,
 531         .arg3_type      = ARG_ANYTHING,
 532         .arg4_type      = ARG_PTR_TO_LONG,
 533 };
 534
 535 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
 536            unsigned long *, res)
 537 {
 538         unsigned long long _res;
 539         bool is_negative;
 540         int err;
 541
 542         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 543         if (err < 0)
 544                 return err;
 545         if (is_negative)
 546                 return -EINVAL;
 547         if (_res != (unsigned long)_res)
 548                 return -ERANGE;
 549         *res = _res;
 550         return err;
 551 }
 552
 553 const struct bpf_func_proto bpf_strtoul_proto = {
 554         .func           = bpf_strtoul,
 555         .gpl_only       = false,
 556         .ret_type       = RET_INTEGER,
 557         .arg1_type      = ARG_PTR_TO_MEM,
 558         .arg2_type      = ARG_CONST_SIZE,
 559         .arg3_type      = ARG_ANYTHING,
 560         .arg4_type      = ARG_PTR_TO_LONG,
 561 };
 562 #endif
 563
 564 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
 565            struct bpf_pidns_info *, nsdata, u32, size)
 566 {
 567         struct task_struct *task = current;
 568         struct pid_namespace *pidns;
 569         int err = -EINVAL;
 570
 571         if (unlikely(size != sizeof(struct bpf_pidns_info)))
 572                 goto clear;
 573
 574         if (unlikely((u64)(dev_t)dev != dev))
 575                 goto clear;
 576
 577         if (unlikely(!task))
 578                 goto clear;
 579
 580         pidns = task_active_pid_ns(task);
 581         if (unlikely(!pidns)) {
 582                 err = -ENOENT;
 583                 goto clear;
 584         }
 585
 586         if (!ns_match(&pidns->ns, (dev_t)dev, ino))
 587                 goto clear;
 588
 589         nsdata->pid = task_pid_nr_ns(task, pidns);
 590         nsdata->tgid = task_tgid_nr_ns(task, pidns);
 591         return 0;
 592 clear:
 593         memset((void *)nsdata, 0, (size_t) size);
 594         return err;
 595 }
 596
 597 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
 598         .func           = bpf_get_ns_current_pid_tgid,
 599         .gpl_only       = false,
 600         .ret_type       = RET_INTEGER,
 601         .arg1_type      = ARG_ANYTHING,
 602         .arg2_type      = ARG_ANYTHING,
 603         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 604         .arg4_type      = ARG_CONST_SIZE,
 605 };
 606
 607 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 608         .func           = bpf_get_raw_cpu_id,
 609         .gpl_only       = false,
 610         .ret_type       = RET_INTEGER,
 611 };
 612
 613 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
 614            u64, flags, void *, data, u64, size)
 615 {
 616         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 617                 return -EINVAL;
 618
 619         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
 620 }
 621
 622 const struct bpf_func_proto bpf_event_output_data_proto =  {
 623         .func           = bpf_event_output_data,
 624         .gpl_only       = true,
 625         .ret_type       = RET_INTEGER,
 626         .arg1_type      = ARG_PTR_TO_CTX,
 627         .arg2_type      = ARG_CONST_MAP_PTR,
 628         .arg3_type      = ARG_ANYTHING,
 629         .arg4_type      = ARG_PTR_TO_MEM,
 630         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 631 };
 632
 633 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
 634            const void __user *, user_ptr)
 635 {
 636         int ret = copy_from_user(dst, user_ptr, size);
 637
 638         if (unlikely(ret)) {
 639                 memset(dst, 0, size);
 640                 ret = -EFAULT;
 641         }
 642
 643         return ret;
 644 }
 645
 646 const struct bpf_func_proto bpf_copy_from_user_proto = {
 647         .func           = bpf_copy_from_user,
 648         .gpl_only       = false,
 649         .ret_type       = RET_INTEGER,
 650         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 651         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 652         .arg3_type      = ARG_ANYTHING,
 653 };
 654
 655 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
 656 {
 657         if (cpu >= nr_cpu_ids)
 658                 return (unsigned long)NULL;
 659
 660         return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
 661 }
 662
 663 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
 664         .func           = bpf_per_cpu_ptr,
 665         .gpl_only       = false,
 666         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL,
 667         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 668         .arg2_type      = ARG_ANYTHING,
 669 };
 670
 671 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
 672 {
 673         return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
 674 }
 675
 676 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
 677         .func           = bpf_this_cpu_ptr,
 678         .gpl_only       = false,
 679         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID,
 680         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 681 };
 682
 683 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
 684                 size_t bufsz)
 685 {
 686         void __user *user_ptr = (__force void __user *)unsafe_ptr;
 687
 688         buf[0] = 0;
 689
 690         switch (fmt_ptype) {
 691         case 's':
 692 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 693                 if ((unsigned long)unsafe_ptr < TASK_SIZE)
 694                         return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 695                 fallthrough;
 696 #endif
 697         case 'k':
 698                 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
 699         case 'u':
 700                 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 701         }
 702
 703         return -EINVAL;
 704 }
 705
 706 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
 707  * arguments representation.
 708  */
 709 #define MAX_BPRINTF_BUF_LEN     512
 710
 711 /* Support executing three nested bprintf helper calls on a given CPU */
 712 #define MAX_BPRINTF_NEST_LEVEL  3
 713 struct bpf_bprintf_buffers {
 714         char tmp_bufs[MAX_BPRINTF_NEST_LEVEL][MAX_BPRINTF_BUF_LEN];
 715 };
 716 static DEFINE_PER_CPU(struct bpf_bprintf_buffers, bpf_bprintf_bufs);
 717 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
 718
 719 static int try_get_fmt_tmp_buf(char **tmp_buf)
 720 {
 721         struct bpf_bprintf_buffers *bufs;
 722         int nest_level;
 723
 724         preempt_disable();
 725         nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
 726         if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
 727                 this_cpu_dec(bpf_bprintf_nest_level);
 728                 preempt_enable();
 729                 return -EBUSY;
 730         }
 731         bufs = this_cpu_ptr(&bpf_bprintf_bufs);
 732         *tmp_buf = bufs->tmp_bufs[nest_level - 1];
 733
 734         return 0;
 735 }
 736
 737 void bpf_bprintf_cleanup(void)
 738 {
 739         if (this_cpu_read(bpf_bprintf_nest_level)) {
 740                 this_cpu_dec(bpf_bprintf_nest_level);
 741                 preempt_enable();
 742         }
 743 }
 744
 745 /*
 746  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
 747  *
 748  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
 749  *
 750  * This can be used in two ways:
 751  * - Format string verification only: when bin_args is NULL
 752  * - Arguments preparation: in addition to the above verification, it writes in
 753  *   bin_args a binary representation of arguments usable by bstr_printf where
 754  *   pointers from BPF have been sanitized.
 755  *
 756  * In argument preparation mode, if 0 is returned, safe temporary buffers are
 757  * allocated and bpf_bprintf_cleanup should be called to free them after use.
 758  */
 759 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
 760                         u32 **bin_args, u32 num_args)
 761 {
 762         char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
 763         size_t sizeof_cur_arg, sizeof_cur_ip;
 764         int err, i, num_spec = 0;
 765         u64 cur_arg;
 766         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
 767
 768         fmt_end = strnchr(fmt, fmt_size, 0);
 769         if (!fmt_end)
 770                 return -EINVAL;
 771         fmt_size = fmt_end - fmt;
 772
 773         if (bin_args) {
 774                 if (num_args && try_get_fmt_tmp_buf(&tmp_buf))
 775                         return -EBUSY;
 776
 777                 tmp_buf_end = tmp_buf + MAX_BPRINTF_BUF_LEN;
 778                 *bin_args = (u32 *)tmp_buf;
 779         }
 780
 781         for (i = 0; i < fmt_size; i++) {
 782                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
 783                         err = -EINVAL;
 784                         goto out;
 785                 }
 786
 787                 if (fmt[i] != '%')
 788                         continue;
 789
 790                 if (fmt[i + 1] == '%') {
 791                         i++;
 792                         continue;
 793                 }
 794
 795                 if (num_spec >= num_args) {
 796                         err = -EINVAL;
 797                         goto out;
 798                 }
 799
 800                 /* The string is zero-terminated so if fmt[i] != 0, we can
 801                  * always access fmt[i + 1], in the worst case it will be a 0
 802                  */
 803                 i++;
 804
 805                 /* skip optional "[0 +-][num]" width formatting field */
 806                 while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
 807                        fmt[i] == ' ')
 808                         i++;
 809                 if (fmt[i] >= '1' && fmt[i] <= '9') {
 810                         i++;
 811                         while (fmt[i] >= '0' && fmt[i] <= '9')
 812                                 i++;
 813                 }
 814
 815                 if (fmt[i] == 'p') {
 816                         sizeof_cur_arg = sizeof(long);
 817
 818                         if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
 819                             fmt[i + 2] == 's') {
 820                                 fmt_ptype = fmt[i + 1];
 821                                 i += 2;
 822                                 goto fmt_str;
 823                         }
 824
 825                         if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
 826                             ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
 827                             fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
 828                             fmt[i + 1] == 'S') {
 829                                 /* just kernel pointers */
 830                                 if (tmp_buf)
 831                                         cur_arg = raw_args[num_spec];
 832                                 i++;
 833                                 goto nocopy_fmt;
 834                         }
 835
 836                         if (fmt[i + 1] == 'B') {
 837                                 if (tmp_buf)  {
 838                                         err = snprintf(tmp_buf,
 839                                                        (tmp_buf_end - tmp_buf),
 840                                                        "%pB",
 841                                                        (void *)(long)raw_args[num_spec]);
 842                                         tmp_buf += (err + 1);
 843                                 }
 844
 845                                 i++;
 846                                 num_spec++;
 847                                 continue;
 848                         }
 849
 850                         /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
 851                         if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
 852                             (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
 853                                 err = -EINVAL;
 854                                 goto out;
 855                         }
 856
 857                         i += 2;
 858                         if (!tmp_buf)
 859                                 goto nocopy_fmt;
 860
 861                         sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
 862                         if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
 863                                 err = -ENOSPC;
 864                                 goto out;
 865                         }
 866
 867                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 868                         err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
 869                                                        sizeof_cur_ip);
 870                         if (err < 0)
 871                                 memset(cur_ip, 0, sizeof_cur_ip);
 872
 873                         /* hack: bstr_printf expects IP addresses to be
 874                          * pre-formatted as strings, ironically, the easiest way
 875                          * to do that is to call snprintf.
 876                          */
 877                         ip_spec[2] = fmt[i - 1];
 878                         ip_spec[3] = fmt[i];
 879                         err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
 880                                        ip_spec, &cur_ip);
 881
 882                         tmp_buf += err + 1;
 883                         num_spec++;
 884
 885                         continue;
 886                 } else if (fmt[i] == 's') {
 887                         fmt_ptype = fmt[i];
 888 fmt_str:
 889                         if (fmt[i + 1] != 0 &&
 890                             !isspace(fmt[i + 1]) &&
 891                             !ispunct(fmt[i + 1])) {
 892                                 err = -EINVAL;
 893                                 goto out;
 894                         }
 895
 896                         if (!tmp_buf)
 897                                 goto nocopy_fmt;
 898
 899                         if (tmp_buf_end == tmp_buf) {
 900                                 err = -ENOSPC;
 901                                 goto out;
 902                         }
 903
 904                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 905                         err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
 906                                                     fmt_ptype,
 907                                                     tmp_buf_end - tmp_buf);
 908                         if (err < 0) {
 909                                 tmp_buf[0] = '\0';
 910                                 err = 1;
 911                         }
 912
 913                         tmp_buf += err;
 914                         num_spec++;
 915
 916                         continue;
 917                 }
 918
 919                 sizeof_cur_arg = sizeof(int);
 920
 921                 if (fmt[i] == 'l') {
 922                         sizeof_cur_arg = sizeof(long);
 923                         i++;
 924                 }
 925                 if (fmt[i] == 'l') {
 926                         sizeof_cur_arg = sizeof(long long);
 927                         i++;
 928                 }
 929
 930                 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
 931                     fmt[i] != 'x' && fmt[i] != 'X') {
 932                         err = -EINVAL;
 933                         goto out;
 934                 }
 935
 936                 if (tmp_buf)
 937                         cur_arg = raw_args[num_spec];
 938 nocopy_fmt:
 939                 if (tmp_buf) {
 940                         tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
 941                         if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
 942                                 err = -ENOSPC;
 943                                 goto out;
 944                         }
 945
 946                         if (sizeof_cur_arg == 8) {
 947                                 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
 948                                 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
 949                         } else {
 950                                 *(u32 *)tmp_buf = (u32)(long)cur_arg;
 951                         }
 952                         tmp_buf += sizeof_cur_arg;
 953                 }
 954                 num_spec++;
 955         }
 956
 957         err = 0;
 958 out:
 959         if (err)
 960                 bpf_bprintf_cleanup();
 961         return err;
 962 }
 963
 964 #define MAX_SNPRINTF_VARARGS            12
 965
 966 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
 967            const void *, data, u32, data_len)
 968 {
 969         int err, num_args;
 970         u32 *bin_args;
 971
 972         if (data_len % 8 || data_len > MAX_SNPRINTF_VARARGS * 8 ||
 973             (data_len && !data))
 974                 return -EINVAL;
 975         num_args = data_len / 8;
 976
 977         /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
 978          * can safely give an unbounded size.
 979          */
 980         err = bpf_bprintf_prepare(fmt, UINT_MAX, data, &bin_args, num_args);
 981         if (err < 0)
 982                 return err;
 983
 984         err = bstr_printf(str, str_size, fmt, bin_args);
 985
 986         bpf_bprintf_cleanup();
 987
 988         return err + 1;
 989 }
 990
 991 const struct bpf_func_proto bpf_snprintf_proto = {
 992         .func           = bpf_snprintf,
 993         .gpl_only       = true,
 994         .ret_type       = RET_INTEGER,
 995         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
 996         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 997         .arg3_type      = ARG_PTR_TO_CONST_STR,
 998         .arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
 999         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1000 };
1001
1002 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1003 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1004 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1005 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1006 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1007
1008 const struct bpf_func_proto *
1009 bpf_base_func_proto(enum bpf_func_id func_id)
1010 {
1011         switch (func_id) {
1012         case BPF_FUNC_map_lookup_elem:
1013                 return &bpf_map_lookup_elem_proto;
1014         case BPF_FUNC_map_update_elem:
1015                 return &bpf_map_update_elem_proto;
1016         case BPF_FUNC_map_delete_elem:
1017                 return &bpf_map_delete_elem_proto;
1018         case BPF_FUNC_map_push_elem:
1019                 return &bpf_map_push_elem_proto;
1020         case BPF_FUNC_map_pop_elem:
1021                 return &bpf_map_pop_elem_proto;
1022         case BPF_FUNC_map_peek_elem:
1023                 return &bpf_map_peek_elem_proto;
1024         case BPF_FUNC_get_prandom_u32:
1025                 return &bpf_get_prandom_u32_proto;
1026         case BPF_FUNC_get_smp_processor_id:
1027                 return &bpf_get_raw_smp_processor_id_proto;
1028         case BPF_FUNC_get_numa_node_id:
1029                 return &bpf_get_numa_node_id_proto;
1030         case BPF_FUNC_tail_call:
1031                 return &bpf_tail_call_proto;
1032         case BPF_FUNC_ktime_get_ns:
1033                 return &bpf_ktime_get_ns_proto;
1034         case BPF_FUNC_ktime_get_boot_ns:
1035                 return &bpf_ktime_get_boot_ns_proto;
1036         case BPF_FUNC_ktime_get_coarse_ns:
1037                 return &bpf_ktime_get_coarse_ns_proto;
1038         case BPF_FUNC_ringbuf_output:
1039                 return &bpf_ringbuf_output_proto;
1040         case BPF_FUNC_ringbuf_reserve:
1041                 return &bpf_ringbuf_reserve_proto;
1042         case BPF_FUNC_ringbuf_submit:
1043                 return &bpf_ringbuf_submit_proto;
1044         case BPF_FUNC_ringbuf_discard:
1045                 return &bpf_ringbuf_discard_proto;
1046         case BPF_FUNC_ringbuf_query:
1047                 return &bpf_ringbuf_query_proto;
1048         case BPF_FUNC_for_each_map_elem:
1049                 return &bpf_for_each_map_elem_proto;
1050         default:
1051                 break;
1052         }
1053
1054         if (!bpf_capable())
1055                 return NULL;
1056
1057         switch (func_id) {
1058         case BPF_FUNC_spin_lock:
1059                 return &bpf_spin_lock_proto;
1060         case BPF_FUNC_spin_unlock:
1061                 return &bpf_spin_unlock_proto;
1062         case BPF_FUNC_jiffies64:
1063                 return &bpf_jiffies64_proto;
1064         case BPF_FUNC_per_cpu_ptr:
1065                 return &bpf_per_cpu_ptr_proto;
1066         case BPF_FUNC_this_cpu_ptr:
1067                 return &bpf_this_cpu_ptr_proto;
1068         default:
1069                 break;
1070         }
1071
1072         if (!perfmon_capable())
1073                 return NULL;
1074
1075         switch (func_id) {
1076         case BPF_FUNC_trace_printk:
1077                 return bpf_get_trace_printk_proto();
1078         case BPF_FUNC_get_current_task:
1079                 return &bpf_get_current_task_proto;
1080         case BPF_FUNC_probe_read_user:
1081                 return &bpf_probe_read_user_proto;
1082         case BPF_FUNC_probe_read_kernel:
1083                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1084                        NULL : &bpf_probe_read_kernel_proto;
1085         case BPF_FUNC_probe_read_user_str:
1086                 return &bpf_probe_read_user_str_proto;
1087         case BPF_FUNC_probe_read_kernel_str:
1088                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1089                        NULL : &bpf_probe_read_kernel_str_proto;
1090         case BPF_FUNC_snprintf_btf:
1091                 return &bpf_snprintf_btf_proto;
1092         case BPF_FUNC_snprintf:
1093                 return &bpf_snprintf_proto;
1094         default:
1095                 return NULL;
1096         }
1097 }