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