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/btf.h>
   6 #include <linux/bpf-cgroup.h>
   7 #include <linux/rcupdate.h>
   8 #include <linux/random.h>
   9 #include <linux/smp.h>
  10 #include <linux/topology.h>
  11 #include <linux/ktime.h>
  12 #include <linux/sched.h>
  13 #include <linux/uidgid.h>
  14 #include <linux/filter.h>
  15 #include <linux/ctype.h>
  16 #include <linux/jiffies.h>
  17 #include <linux/pid_namespace.h>
  18 #include <linux/proc_ns.h>
  19 #include <linux/security.h>
  20 #include <linux/btf_ids.h>
  21
  22 #include "../../lib/kstrtox.h"
  23
  24 /* If kernel subsystem is allowing eBPF programs to call this function,
  25  * inside its own verifier_ops->get_func_proto() callback it should return
  26  * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
  27  *
  28  * Different map implementations will rely on rcu in map methods
  29  * lookup/update/delete, therefore eBPF programs must run under rcu lock
  30  * if program is allowed to access maps, so check rcu_read_lock_held in
  31  * all three functions.
  32  */
  33 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
  34 {
  35         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  36         return (unsigned long) map->ops->map_lookup_elem(map, key);
  37 }
  38
  39 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
  40         .func           = bpf_map_lookup_elem,
  41         .gpl_only       = false,
  42         .pkt_access     = true,
  43         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
  44         .arg1_type      = ARG_CONST_MAP_PTR,
  45         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  46 };
  47
  48 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
  49            void *, value, u64, flags)
  50 {
  51         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  52         return map->ops->map_update_elem(map, key, value, flags);
  53 }
  54
  55 const struct bpf_func_proto bpf_map_update_elem_proto = {
  56         .func           = bpf_map_update_elem,
  57         .gpl_only       = false,
  58         .pkt_access     = true,
  59         .ret_type       = RET_INTEGER,
  60         .arg1_type      = ARG_CONST_MAP_PTR,
  61         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  62         .arg3_type      = ARG_PTR_TO_MAP_VALUE,
  63         .arg4_type      = ARG_ANYTHING,
  64 };
  65
  66 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
  67 {
  68         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  69         return map->ops->map_delete_elem(map, key);
  70 }
  71
  72 const struct bpf_func_proto bpf_map_delete_elem_proto = {
  73         .func           = bpf_map_delete_elem,
  74         .gpl_only       = false,
  75         .pkt_access     = true,
  76         .ret_type       = RET_INTEGER,
  77         .arg1_type      = ARG_CONST_MAP_PTR,
  78         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  79 };
  80
  81 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
  82 {
  83         return map->ops->map_push_elem(map, value, flags);
  84 }
  85
  86 const struct bpf_func_proto bpf_map_push_elem_proto = {
  87         .func           = bpf_map_push_elem,
  88         .gpl_only       = false,
  89         .pkt_access     = true,
  90         .ret_type       = RET_INTEGER,
  91         .arg1_type      = ARG_CONST_MAP_PTR,
  92         .arg2_type      = ARG_PTR_TO_MAP_VALUE,
  93         .arg3_type      = ARG_ANYTHING,
  94 };
  95
  96 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
  97 {
  98         return map->ops->map_pop_elem(map, value);
  99 }
 100
 101 const struct bpf_func_proto bpf_map_pop_elem_proto = {
 102         .func           = bpf_map_pop_elem,
 103         .gpl_only       = false,
 104         .ret_type       = RET_INTEGER,
 105         .arg1_type      = ARG_CONST_MAP_PTR,
 106         .arg2_type      = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
 107 };
 108
 109 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
 110 {
 111         return map->ops->map_peek_elem(map, value);
 112 }
 113
 114 const struct bpf_func_proto bpf_map_peek_elem_proto = {
 115         .func           = bpf_map_peek_elem,
 116         .gpl_only       = false,
 117         .ret_type       = RET_INTEGER,
 118         .arg1_type      = ARG_CONST_MAP_PTR,
 119         .arg2_type      = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
 120 };
 121
 122 BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu)
 123 {
 124         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
 125         return (unsigned long) map->ops->map_lookup_percpu_elem(map, key, cpu);
 126 }
 127
 128 const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = {
 129         .func           = bpf_map_lookup_percpu_elem,
 130         .gpl_only       = false,
 131         .pkt_access     = true,
 132         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 133         .arg1_type      = ARG_CONST_MAP_PTR,
 134         .arg2_type      = ARG_PTR_TO_MAP_KEY,
 135         .arg3_type      = ARG_ANYTHING,
 136 };
 137
 138 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
 139         .func           = bpf_user_rnd_u32,
 140         .gpl_only       = false,
 141         .ret_type       = RET_INTEGER,
 142 };
 143
 144 BPF_CALL_0(bpf_get_smp_processor_id)
 145 {
 146         return smp_processor_id();
 147 }
 148
 149 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
 150         .func           = bpf_get_smp_processor_id,
 151         .gpl_only       = false,
 152         .ret_type       = RET_INTEGER,
 153 };
 154
 155 BPF_CALL_0(bpf_get_numa_node_id)
 156 {
 157         return numa_node_id();
 158 }
 159
 160 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
 161         .func           = bpf_get_numa_node_id,
 162         .gpl_only       = false,
 163         .ret_type       = RET_INTEGER,
 164 };
 165
 166 BPF_CALL_0(bpf_ktime_get_ns)
 167 {
 168         /* NMI safe access to clock monotonic */
 169         return ktime_get_mono_fast_ns();
 170 }
 171
 172 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
 173         .func           = bpf_ktime_get_ns,
 174         .gpl_only       = false,
 175         .ret_type       = RET_INTEGER,
 176 };
 177
 178 BPF_CALL_0(bpf_ktime_get_boot_ns)
 179 {
 180         /* NMI safe access to clock boottime */
 181         return ktime_get_boot_fast_ns();
 182 }
 183
 184 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
 185         .func           = bpf_ktime_get_boot_ns,
 186         .gpl_only       = false,
 187         .ret_type       = RET_INTEGER,
 188 };
 189
 190 BPF_CALL_0(bpf_ktime_get_coarse_ns)
 191 {
 192         return ktime_get_coarse_ns();
 193 }
 194
 195 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
 196         .func           = bpf_ktime_get_coarse_ns,
 197         .gpl_only       = false,
 198         .ret_type       = RET_INTEGER,
 199 };
 200
 201 BPF_CALL_0(bpf_get_current_pid_tgid)
 202 {
 203         struct task_struct *task = current;
 204
 205         if (unlikely(!task))
 206                 return -EINVAL;
 207
 208         return (u64) task->tgid << 32 | task->pid;
 209 }
 210
 211 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
 212         .func           = bpf_get_current_pid_tgid,
 213         .gpl_only       = false,
 214         .ret_type       = RET_INTEGER,
 215 };
 216
 217 BPF_CALL_0(bpf_get_current_uid_gid)
 218 {
 219         struct task_struct *task = current;
 220         kuid_t uid;
 221         kgid_t gid;
 222
 223         if (unlikely(!task))
 224                 return -EINVAL;
 225
 226         current_uid_gid(&uid, &gid);
 227         return (u64) from_kgid(&init_user_ns, gid) << 32 |
 228                      from_kuid(&init_user_ns, uid);
 229 }
 230
 231 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
 232         .func           = bpf_get_current_uid_gid,
 233         .gpl_only       = false,
 234         .ret_type       = RET_INTEGER,
 235 };
 236
 237 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
 238 {
 239         struct task_struct *task = current;
 240
 241         if (unlikely(!task))
 242                 goto err_clear;
 243
 244         /* Verifier guarantees that size > 0 */
 245         strscpy(buf, task->comm, size);
 246         return 0;
 247 err_clear:
 248         memset(buf, 0, size);
 249         return -EINVAL;
 250 }
 251
 252 const struct bpf_func_proto bpf_get_current_comm_proto = {
 253         .func           = bpf_get_current_comm,
 254         .gpl_only       = false,
 255         .ret_type       = RET_INTEGER,
 256         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 257         .arg2_type      = ARG_CONST_SIZE,
 258 };
 259
 260 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
 261
 262 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
 263 {
 264         arch_spinlock_t *l = (void *)lock;
 265         union {
 266                 __u32 val;
 267                 arch_spinlock_t lock;
 268         } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
 269
 270         compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
 271         BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
 272         BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
 273         arch_spin_lock(l);
 274 }
 275
 276 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
 277 {
 278         arch_spinlock_t *l = (void *)lock;
 279
 280         arch_spin_unlock(l);
 281 }
 282
 283 #else
 284
 285 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
 286 {
 287         atomic_t *l = (void *)lock;
 288
 289         BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
 290         do {
 291                 atomic_cond_read_relaxed(l, !VAL);
 292         } while (atomic_xchg(l, 1));
 293 }
 294
 295 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
 296 {
 297         atomic_t *l = (void *)lock;
 298
 299         atomic_set_release(l, 0);
 300 }
 301
 302 #endif
 303
 304 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
 305
 306 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
 307 {
 308         unsigned long flags;
 309
 310         local_irq_save(flags);
 311         __bpf_spin_lock(lock);
 312         __this_cpu_write(irqsave_flags, flags);
 313 }
 314
 315 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
 316 {
 317         __bpf_spin_lock_irqsave(lock);
 318         return 0;
 319 }
 320
 321 const struct bpf_func_proto bpf_spin_lock_proto = {
 322         .func           = bpf_spin_lock,
 323         .gpl_only       = false,
 324         .ret_type       = RET_VOID,
 325         .arg1_type      = ARG_PTR_TO_SPIN_LOCK,
 326 };
 327
 328 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
 329 {
 330         unsigned long flags;
 331
 332         flags = __this_cpu_read(irqsave_flags);
 333         __bpf_spin_unlock(lock);
 334         local_irq_restore(flags);
 335 }
 336
 337 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
 338 {
 339         __bpf_spin_unlock_irqrestore(lock);
 340         return 0;
 341 }
 342
 343 const struct bpf_func_proto bpf_spin_unlock_proto = {
 344         .func           = bpf_spin_unlock,
 345         .gpl_only       = false,
 346         .ret_type       = RET_VOID,
 347         .arg1_type      = ARG_PTR_TO_SPIN_LOCK,
 348 };
 349
 350 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
 351                            bool lock_src)
 352 {
 353         struct bpf_spin_lock *lock;
 354
 355         if (lock_src)
 356                 lock = src + map->spin_lock_off;
 357         else
 358                 lock = dst + map->spin_lock_off;
 359         preempt_disable();
 360         __bpf_spin_lock_irqsave(lock);
 361         copy_map_value(map, dst, src);
 362         __bpf_spin_unlock_irqrestore(lock);
 363         preempt_enable();
 364 }
 365
 366 BPF_CALL_0(bpf_jiffies64)
 367 {
 368         return get_jiffies_64();
 369 }
 370
 371 const struct bpf_func_proto bpf_jiffies64_proto = {
 372         .func           = bpf_jiffies64,
 373         .gpl_only       = false,
 374         .ret_type       = RET_INTEGER,
 375 };
 376
 377 #ifdef CONFIG_CGROUPS
 378 BPF_CALL_0(bpf_get_current_cgroup_id)
 379 {
 380         struct cgroup *cgrp;
 381         u64 cgrp_id;
 382
 383         rcu_read_lock();
 384         cgrp = task_dfl_cgroup(current);
 385         cgrp_id = cgroup_id(cgrp);
 386         rcu_read_unlock();
 387
 388         return cgrp_id;
 389 }
 390
 391 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
 392         .func           = bpf_get_current_cgroup_id,
 393         .gpl_only       = false,
 394         .ret_type       = RET_INTEGER,
 395 };
 396
 397 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
 398 {
 399         struct cgroup *cgrp;
 400         struct cgroup *ancestor;
 401         u64 cgrp_id;
 402
 403         rcu_read_lock();
 404         cgrp = task_dfl_cgroup(current);
 405         ancestor = cgroup_ancestor(cgrp, ancestor_level);
 406         cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
 407         rcu_read_unlock();
 408
 409         return cgrp_id;
 410 }
 411
 412 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
 413         .func           = bpf_get_current_ancestor_cgroup_id,
 414         .gpl_only       = false,
 415         .ret_type       = RET_INTEGER,
 416         .arg1_type      = ARG_ANYTHING,
 417 };
 418
 419 #ifdef CONFIG_CGROUP_BPF
 420
 421 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
 422 {
 423         /* flags argument is not used now,
 424          * but provides an ability to extend the API.
 425          * verifier checks that its value is correct.
 426          */
 427         enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
 428         struct bpf_cgroup_storage *storage;
 429         struct bpf_cg_run_ctx *ctx;
 430         void *ptr;
 431
 432         /* get current cgroup storage from BPF run context */
 433         ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
 434         storage = ctx->prog_item->cgroup_storage[stype];
 435
 436         if (stype == BPF_CGROUP_STORAGE_SHARED)
 437                 ptr = &READ_ONCE(storage->buf)->data[0];
 438         else
 439                 ptr = this_cpu_ptr(storage->percpu_buf);
 440
 441         return (unsigned long)ptr;
 442 }
 443
 444 const struct bpf_func_proto bpf_get_local_storage_proto = {
 445         .func           = bpf_get_local_storage,
 446         .gpl_only       = false,
 447         .ret_type       = RET_PTR_TO_MAP_VALUE,
 448         .arg1_type      = ARG_CONST_MAP_PTR,
 449         .arg2_type      = ARG_ANYTHING,
 450 };
 451 #endif
 452
 453 #define BPF_STRTOX_BASE_MASK 0x1F
 454
 455 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
 456                           unsigned long long *res, bool *is_negative)
 457 {
 458         unsigned int base = flags & BPF_STRTOX_BASE_MASK;
 459         const char *cur_buf = buf;
 460         size_t cur_len = buf_len;
 461         unsigned int consumed;
 462         size_t val_len;
 463         char str[64];
 464
 465         if (!buf || !buf_len || !res || !is_negative)
 466                 return -EINVAL;
 467
 468         if (base != 0 && base != 8 && base != 10 && base != 16)
 469                 return -EINVAL;
 470
 471         if (flags & ~BPF_STRTOX_BASE_MASK)
 472                 return -EINVAL;
 473
 474         while (cur_buf < buf + buf_len && isspace(*cur_buf))
 475                 ++cur_buf;
 476
 477         *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
 478         if (*is_negative)
 479                 ++cur_buf;
 480
 481         consumed = cur_buf - buf;
 482         cur_len -= consumed;
 483         if (!cur_len)
 484                 return -EINVAL;
 485
 486         cur_len = min(cur_len, sizeof(str) - 1);
 487         memcpy(str, cur_buf, cur_len);
 488         str[cur_len] = '\0';
 489         cur_buf = str;
 490
 491         cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
 492         val_len = _parse_integer(cur_buf, base, res);
 493
 494         if (val_len & KSTRTOX_OVERFLOW)
 495                 return -ERANGE;
 496
 497         if (val_len == 0)
 498                 return -EINVAL;
 499
 500         cur_buf += val_len;
 501         consumed += cur_buf - str;
 502
 503         return consumed;
 504 }
 505
 506 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
 507                          long long *res)
 508 {
 509         unsigned long long _res;
 510         bool is_negative;
 511         int err;
 512
 513         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 514         if (err < 0)
 515                 return err;
 516         if (is_negative) {
 517                 if ((long long)-_res > 0)
 518                         return -ERANGE;
 519                 *res = -_res;
 520         } else {
 521                 if ((long long)_res < 0)
 522                         return -ERANGE;
 523                 *res = _res;
 524         }
 525         return err;
 526 }
 527
 528 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
 529            long *, res)
 530 {
 531         long long _res;
 532         int err;
 533
 534         err = __bpf_strtoll(buf, buf_len, flags, &_res);
 535         if (err < 0)
 536                 return err;
 537         if (_res != (long)_res)
 538                 return -ERANGE;
 539         *res = _res;
 540         return err;
 541 }
 542
 543 const struct bpf_func_proto bpf_strtol_proto = {
 544         .func           = bpf_strtol,
 545         .gpl_only       = false,
 546         .ret_type       = RET_INTEGER,
 547         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 548         .arg2_type      = ARG_CONST_SIZE,
 549         .arg3_type      = ARG_ANYTHING,
 550         .arg4_type      = ARG_PTR_TO_LONG,
 551 };
 552
 553 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
 554            unsigned long *, res)
 555 {
 556         unsigned long long _res;
 557         bool is_negative;
 558         int err;
 559
 560         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 561         if (err < 0)
 562                 return err;
 563         if (is_negative)
 564                 return -EINVAL;
 565         if (_res != (unsigned long)_res)
 566                 return -ERANGE;
 567         *res = _res;
 568         return err;
 569 }
 570
 571 const struct bpf_func_proto bpf_strtoul_proto = {
 572         .func           = bpf_strtoul,
 573         .gpl_only       = false,
 574         .ret_type       = RET_INTEGER,
 575         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 576         .arg2_type      = ARG_CONST_SIZE,
 577         .arg3_type      = ARG_ANYTHING,
 578         .arg4_type      = ARG_PTR_TO_LONG,
 579 };
 580 #endif
 581
 582 BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
 583 {
 584         return strncmp(s1, s2, s1_sz);
 585 }
 586
 587 const struct bpf_func_proto bpf_strncmp_proto = {
 588         .func           = bpf_strncmp,
 589         .gpl_only       = false,
 590         .ret_type       = RET_INTEGER,
 591         .arg1_type      = ARG_PTR_TO_MEM,
 592         .arg2_type      = ARG_CONST_SIZE,
 593         .arg3_type      = ARG_PTR_TO_CONST_STR,
 594 };
 595
 596 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
 597            struct bpf_pidns_info *, nsdata, u32, size)
 598 {
 599         struct task_struct *task = current;
 600         struct pid_namespace *pidns;
 601         int err = -EINVAL;
 602
 603         if (unlikely(size != sizeof(struct bpf_pidns_info)))
 604                 goto clear;
 605
 606         if (unlikely((u64)(dev_t)dev != dev))
 607                 goto clear;
 608
 609         if (unlikely(!task))
 610                 goto clear;
 611
 612         pidns = task_active_pid_ns(task);
 613         if (unlikely(!pidns)) {
 614                 err = -ENOENT;
 615                 goto clear;
 616         }
 617
 618         if (!ns_match(&pidns->ns, (dev_t)dev, ino))
 619                 goto clear;
 620
 621         nsdata->pid = task_pid_nr_ns(task, pidns);
 622         nsdata->tgid = task_tgid_nr_ns(task, pidns);
 623         return 0;
 624 clear:
 625         memset((void *)nsdata, 0, (size_t) size);
 626         return err;
 627 }
 628
 629 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
 630         .func           = bpf_get_ns_current_pid_tgid,
 631         .gpl_only       = false,
 632         .ret_type       = RET_INTEGER,
 633         .arg1_type      = ARG_ANYTHING,
 634         .arg2_type      = ARG_ANYTHING,
 635         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 636         .arg4_type      = ARG_CONST_SIZE,
 637 };
 638
 639 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 640         .func           = bpf_get_raw_cpu_id,
 641         .gpl_only       = false,
 642         .ret_type       = RET_INTEGER,
 643 };
 644
 645 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
 646            u64, flags, void *, data, u64, size)
 647 {
 648         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 649                 return -EINVAL;
 650
 651         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
 652 }
 653
 654 const struct bpf_func_proto bpf_event_output_data_proto =  {
 655         .func           = bpf_event_output_data,
 656         .gpl_only       = true,
 657         .ret_type       = RET_INTEGER,
 658         .arg1_type      = ARG_PTR_TO_CTX,
 659         .arg2_type      = ARG_CONST_MAP_PTR,
 660         .arg3_type      = ARG_ANYTHING,
 661         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 662         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 663 };
 664
 665 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
 666            const void __user *, user_ptr)
 667 {
 668         int ret = copy_from_user(dst, user_ptr, size);
 669
 670         if (unlikely(ret)) {
 671                 memset(dst, 0, size);
 672                 ret = -EFAULT;
 673         }
 674
 675         return ret;
 676 }
 677
 678 const struct bpf_func_proto bpf_copy_from_user_proto = {
 679         .func           = bpf_copy_from_user,
 680         .gpl_only       = false,
 681         .ret_type       = RET_INTEGER,
 682         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 683         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 684         .arg3_type      = ARG_ANYTHING,
 685 };
 686
 687 BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
 688            const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
 689 {
 690         int ret;
 691
 692         /* flags is not used yet */
 693         if (unlikely(flags))
 694                 return -EINVAL;
 695
 696         if (unlikely(!size))
 697                 return 0;
 698
 699         ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0);
 700         if (ret == size)
 701                 return 0;
 702
 703         memset(dst, 0, size);
 704         /* Return -EFAULT for partial read */
 705         return ret < 0 ? ret : -EFAULT;
 706 }
 707
 708 const struct bpf_func_proto bpf_copy_from_user_task_proto = {
 709         .func           = bpf_copy_from_user_task,
 710         .gpl_only       = true,
 711         .ret_type       = RET_INTEGER,
 712         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 713         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 714         .arg3_type      = ARG_ANYTHING,
 715         .arg4_type      = ARG_PTR_TO_BTF_ID,
 716         .arg4_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 717         .arg5_type      = ARG_ANYTHING
 718 };
 719
 720 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
 721 {
 722         if (cpu >= nr_cpu_ids)
 723                 return (unsigned long)NULL;
 724
 725         return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
 726 }
 727
 728 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
 729         .func           = bpf_per_cpu_ptr,
 730         .gpl_only       = false,
 731         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
 732         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 733         .arg2_type      = ARG_ANYTHING,
 734 };
 735
 736 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
 737 {
 738         return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
 739 }
 740
 741 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
 742         .func           = bpf_this_cpu_ptr,
 743         .gpl_only       = false,
 744         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
 745         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 746 };
 747
 748 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
 749                 size_t bufsz)
 750 {
 751         void __user *user_ptr = (__force void __user *)unsafe_ptr;
 752
 753         buf[0] = 0;
 754
 755         switch (fmt_ptype) {
 756         case 's':
 757 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 758                 if ((unsigned long)unsafe_ptr < TASK_SIZE)
 759                         return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 760                 fallthrough;
 761 #endif
 762         case 'k':
 763                 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
 764         case 'u':
 765                 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 766         }
 767
 768         return -EINVAL;
 769 }
 770
 771 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
 772  * arguments representation.
 773  */
 774 #define MAX_BPRINTF_BUF_LEN     512
 775
 776 /* Support executing three nested bprintf helper calls on a given CPU */
 777 #define MAX_BPRINTF_NEST_LEVEL  3
 778 struct bpf_bprintf_buffers {
 779         char tmp_bufs[MAX_BPRINTF_NEST_LEVEL][MAX_BPRINTF_BUF_LEN];
 780 };
 781 static DEFINE_PER_CPU(struct bpf_bprintf_buffers, bpf_bprintf_bufs);
 782 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
 783
 784 static int try_get_fmt_tmp_buf(char **tmp_buf)
 785 {
 786         struct bpf_bprintf_buffers *bufs;
 787         int nest_level;
 788
 789         preempt_disable();
 790         nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
 791         if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
 792                 this_cpu_dec(bpf_bprintf_nest_level);
 793                 preempt_enable();
 794                 return -EBUSY;
 795         }
 796         bufs = this_cpu_ptr(&bpf_bprintf_bufs);
 797         *tmp_buf = bufs->tmp_bufs[nest_level - 1];
 798
 799         return 0;
 800 }
 801
 802 void bpf_bprintf_cleanup(void)
 803 {
 804         if (this_cpu_read(bpf_bprintf_nest_level)) {
 805                 this_cpu_dec(bpf_bprintf_nest_level);
 806                 preempt_enable();
 807         }
 808 }
 809
 810 /*
 811  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
 812  *
 813  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
 814  *
 815  * This can be used in two ways:
 816  * - Format string verification only: when bin_args is NULL
 817  * - Arguments preparation: in addition to the above verification, it writes in
 818  *   bin_args a binary representation of arguments usable by bstr_printf where
 819  *   pointers from BPF have been sanitized.
 820  *
 821  * In argument preparation mode, if 0 is returned, safe temporary buffers are
 822  * allocated and bpf_bprintf_cleanup should be called to free them after use.
 823  */
 824 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
 825                         u32 **bin_args, u32 num_args)
 826 {
 827         char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
 828         size_t sizeof_cur_arg, sizeof_cur_ip;
 829         int err, i, num_spec = 0;
 830         u64 cur_arg;
 831         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
 832
 833         fmt_end = strnchr(fmt, fmt_size, 0);
 834         if (!fmt_end)
 835                 return -EINVAL;
 836         fmt_size = fmt_end - fmt;
 837
 838         if (bin_args) {
 839                 if (num_args && try_get_fmt_tmp_buf(&tmp_buf))
 840                         return -EBUSY;
 841
 842                 tmp_buf_end = tmp_buf + MAX_BPRINTF_BUF_LEN;
 843                 *bin_args = (u32 *)tmp_buf;
 844         }
 845
 846         for (i = 0; i < fmt_size; i++) {
 847                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
 848                         err = -EINVAL;
 849                         goto out;
 850                 }
 851
 852                 if (fmt[i] != '%')
 853                         continue;
 854
 855                 if (fmt[i + 1] == '%') {
 856                         i++;
 857                         continue;
 858                 }
 859
 860                 if (num_spec >= num_args) {
 861                         err = -EINVAL;
 862                         goto out;
 863                 }
 864
 865                 /* The string is zero-terminated so if fmt[i] != 0, we can
 866                  * always access fmt[i + 1], in the worst case it will be a 0
 867                  */
 868                 i++;
 869
 870                 /* skip optional "[0 +-][num]" width formatting field */
 871                 while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
 872                        fmt[i] == ' ')
 873                         i++;
 874                 if (fmt[i] >= '1' && fmt[i] <= '9') {
 875                         i++;
 876                         while (fmt[i] >= '0' && fmt[i] <= '9')
 877                                 i++;
 878                 }
 879
 880                 if (fmt[i] == 'p') {
 881                         sizeof_cur_arg = sizeof(long);
 882
 883                         if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
 884                             fmt[i + 2] == 's') {
 885                                 fmt_ptype = fmt[i + 1];
 886                                 i += 2;
 887                                 goto fmt_str;
 888                         }
 889
 890                         if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
 891                             ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
 892                             fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
 893                             fmt[i + 1] == 'S') {
 894                                 /* just kernel pointers */
 895                                 if (tmp_buf)
 896                                         cur_arg = raw_args[num_spec];
 897                                 i++;
 898                                 goto nocopy_fmt;
 899                         }
 900
 901                         if (fmt[i + 1] == 'B') {
 902                                 if (tmp_buf)  {
 903                                         err = snprintf(tmp_buf,
 904                                                        (tmp_buf_end - tmp_buf),
 905                                                        "%pB",
 906                                                        (void *)(long)raw_args[num_spec]);
 907                                         tmp_buf += (err + 1);
 908                                 }
 909
 910                                 i++;
 911                                 num_spec++;
 912                                 continue;
 913                         }
 914
 915                         /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
 916                         if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
 917                             (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
 918                                 err = -EINVAL;
 919                                 goto out;
 920                         }
 921
 922                         i += 2;
 923                         if (!tmp_buf)
 924                                 goto nocopy_fmt;
 925
 926                         sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
 927                         if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
 928                                 err = -ENOSPC;
 929                                 goto out;
 930                         }
 931
 932                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 933                         err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
 934                                                        sizeof_cur_ip);
 935                         if (err < 0)
 936                                 memset(cur_ip, 0, sizeof_cur_ip);
 937
 938                         /* hack: bstr_printf expects IP addresses to be
 939                          * pre-formatted as strings, ironically, the easiest way
 940                          * to do that is to call snprintf.
 941                          */
 942                         ip_spec[2] = fmt[i - 1];
 943                         ip_spec[3] = fmt[i];
 944                         err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
 945                                        ip_spec, &cur_ip);
 946
 947                         tmp_buf += err + 1;
 948                         num_spec++;
 949
 950                         continue;
 951                 } else if (fmt[i] == 's') {
 952                         fmt_ptype = fmt[i];
 953 fmt_str:
 954                         if (fmt[i + 1] != 0 &&
 955                             !isspace(fmt[i + 1]) &&
 956                             !ispunct(fmt[i + 1])) {
 957                                 err = -EINVAL;
 958                                 goto out;
 959                         }
 960
 961                         if (!tmp_buf)
 962                                 goto nocopy_fmt;
 963
 964                         if (tmp_buf_end == tmp_buf) {
 965                                 err = -ENOSPC;
 966                                 goto out;
 967                         }
 968
 969                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 970                         err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
 971                                                     fmt_ptype,
 972                                                     tmp_buf_end - tmp_buf);
 973                         if (err < 0) {
 974                                 tmp_buf[0] = '\0';
 975                                 err = 1;
 976                         }
 977
 978                         tmp_buf += err;
 979                         num_spec++;
 980
 981                         continue;
 982                 } else if (fmt[i] == 'c') {
 983                         if (!tmp_buf)
 984                                 goto nocopy_fmt;
 985
 986                         if (tmp_buf_end == tmp_buf) {
 987                                 err = -ENOSPC;
 988                                 goto out;
 989                         }
 990
 991                         *tmp_buf = raw_args[num_spec];
 992                         tmp_buf++;
 993                         num_spec++;
 994
 995                         continue;
 996                 }
 997
 998                 sizeof_cur_arg = sizeof(int);
 999
1000                 if (fmt[i] == 'l') {
1001                         sizeof_cur_arg = sizeof(long);
1002                         i++;
1003                 }
1004                 if (fmt[i] == 'l') {
1005                         sizeof_cur_arg = sizeof(long long);
1006                         i++;
1007                 }
1008
1009                 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1010                     fmt[i] != 'x' && fmt[i] != 'X') {
1011                         err = -EINVAL;
1012                         goto out;
1013                 }
1014
1015                 if (tmp_buf)
1016                         cur_arg = raw_args[num_spec];
1017 nocopy_fmt:
1018                 if (tmp_buf) {
1019                         tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1020                         if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1021                                 err = -ENOSPC;
1022                                 goto out;
1023                         }
1024
1025                         if (sizeof_cur_arg == 8) {
1026                                 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
1027                                 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1028                         } else {
1029                                 *(u32 *)tmp_buf = (u32)(long)cur_arg;
1030                         }
1031                         tmp_buf += sizeof_cur_arg;
1032                 }
1033                 num_spec++;
1034         }
1035
1036         err = 0;
1037 out:
1038         if (err)
1039                 bpf_bprintf_cleanup();
1040         return err;
1041 }
1042
1043 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1044            const void *, data, u32, data_len)
1045 {
1046         int err, num_args;
1047         u32 *bin_args;
1048
1049         if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1050             (data_len && !data))
1051                 return -EINVAL;
1052         num_args = data_len / 8;
1053
1054         /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1055          * can safely give an unbounded size.
1056          */
1057         err = bpf_bprintf_prepare(fmt, UINT_MAX, data, &bin_args, num_args);
1058         if (err < 0)
1059                 return err;
1060
1061         err = bstr_printf(str, str_size, fmt, bin_args);
1062
1063         bpf_bprintf_cleanup();
1064
1065         return err + 1;
1066 }
1067
1068 const struct bpf_func_proto bpf_snprintf_proto = {
1069         .func           = bpf_snprintf,
1070         .gpl_only       = true,
1071         .ret_type       = RET_INTEGER,
1072         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1073         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1074         .arg3_type      = ARG_PTR_TO_CONST_STR,
1075         .arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1076         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1077 };
1078
1079 /* BPF map elements can contain 'struct bpf_timer'.
1080  * Such map owns all of its BPF timers.
1081  * 'struct bpf_timer' is allocated as part of map element allocation
1082  * and it's zero initialized.
1083  * That space is used to keep 'struct bpf_timer_kern'.
1084  * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1085  * remembers 'struct bpf_map *' pointer it's part of.
1086  * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1087  * bpf_timer_start() arms the timer.
1088  * If user space reference to a map goes to zero at this point
1089  * ops->map_release_uref callback is responsible for cancelling the timers,
1090  * freeing their memory, and decrementing prog's refcnts.
1091  * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1092  * Inner maps can contain bpf timers as well. ops->map_release_uref is
1093  * freeing the timers when inner map is replaced or deleted by user space.
1094  */
1095 struct bpf_hrtimer {
1096         struct hrtimer timer;
1097         struct bpf_map *map;
1098         struct bpf_prog *prog;
1099         void __rcu *callback_fn;
1100         void *value;
1101 };
1102
1103 /* the actual struct hidden inside uapi struct bpf_timer */
1104 struct bpf_timer_kern {
1105         struct bpf_hrtimer *timer;
1106         /* bpf_spin_lock is used here instead of spinlock_t to make
1107          * sure that it always fits into space reserved by struct bpf_timer
1108          * regardless of LOCKDEP and spinlock debug flags.
1109          */
1110         struct bpf_spin_lock lock;
1111 } __attribute__((aligned(8)));
1112
1113 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1114
1115 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1116 {
1117         struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1118         struct bpf_map *map = t->map;
1119         void *value = t->value;
1120         bpf_callback_t callback_fn;
1121         void *key;
1122         u32 idx;
1123
1124         BTF_TYPE_EMIT(struct bpf_timer);
1125         callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1126         if (!callback_fn)
1127                 goto out;
1128
1129         /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1130          * cannot be preempted by another bpf_timer_cb() on the same cpu.
1131          * Remember the timer this callback is servicing to prevent
1132          * deadlock if callback_fn() calls bpf_timer_cancel() or
1133          * bpf_map_delete_elem() on the same timer.
1134          */
1135         this_cpu_write(hrtimer_running, t);
1136         if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1137                 struct bpf_array *array = container_of(map, struct bpf_array, map);
1138
1139                 /* compute the key */
1140                 idx = ((char *)value - array->value) / array->elem_size;
1141                 key = &idx;
1142         } else { /* hash or lru */
1143                 key = value - round_up(map->key_size, 8);
1144         }
1145
1146         callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1147         /* The verifier checked that return value is zero. */
1148
1149         this_cpu_write(hrtimer_running, NULL);
1150 out:
1151         return HRTIMER_NORESTART;
1152 }
1153
1154 BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1155            u64, flags)
1156 {
1157         clockid_t clockid = flags & (MAX_CLOCKS - 1);
1158         struct bpf_hrtimer *t;
1159         int ret = 0;
1160
1161         BUILD_BUG_ON(MAX_CLOCKS != 16);
1162         BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1163         BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1164
1165         if (in_nmi())
1166                 return -EOPNOTSUPP;
1167
1168         if (flags >= MAX_CLOCKS ||
1169             /* similar to timerfd except _ALARM variants are not supported */
1170             (clockid != CLOCK_MONOTONIC &&
1171              clockid != CLOCK_REALTIME &&
1172              clockid != CLOCK_BOOTTIME))
1173                 return -EINVAL;
1174         __bpf_spin_lock_irqsave(&timer->lock);
1175         t = timer->timer;
1176         if (t) {
1177                 ret = -EBUSY;
1178                 goto out;
1179         }
1180         if (!atomic64_read(&map->usercnt)) {
1181                 /* maps with timers must be either held by user space
1182                  * or pinned in bpffs.
1183                  */
1184                 ret = -EPERM;
1185                 goto out;
1186         }
1187         /* allocate hrtimer via map_kmalloc to use memcg accounting */
1188         t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
1189         if (!t) {
1190                 ret = -ENOMEM;
1191                 goto out;
1192         }
1193         t->value = (void *)timer - map->timer_off;
1194         t->map = map;
1195         t->prog = NULL;
1196         rcu_assign_pointer(t->callback_fn, NULL);
1197         hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1198         t->timer.function = bpf_timer_cb;
1199         timer->timer = t;
1200 out:
1201         __bpf_spin_unlock_irqrestore(&timer->lock);
1202         return ret;
1203 }
1204
1205 static const struct bpf_func_proto bpf_timer_init_proto = {
1206         .func           = bpf_timer_init,
1207         .gpl_only       = true,
1208         .ret_type       = RET_INTEGER,
1209         .arg1_type      = ARG_PTR_TO_TIMER,
1210         .arg2_type      = ARG_CONST_MAP_PTR,
1211         .arg3_type      = ARG_ANYTHING,
1212 };
1213
1214 BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1215            struct bpf_prog_aux *, aux)
1216 {
1217         struct bpf_prog *prev, *prog = aux->prog;
1218         struct bpf_hrtimer *t;
1219         int ret = 0;
1220
1221         if (in_nmi())
1222                 return -EOPNOTSUPP;
1223         __bpf_spin_lock_irqsave(&timer->lock);
1224         t = timer->timer;
1225         if (!t) {
1226                 ret = -EINVAL;
1227                 goto out;
1228         }
1229         if (!atomic64_read(&t->map->usercnt)) {
1230                 /* maps with timers must be either held by user space
1231                  * or pinned in bpffs. Otherwise timer might still be
1232                  * running even when bpf prog is detached and user space
1233                  * is gone, since map_release_uref won't ever be called.
1234                  */
1235                 ret = -EPERM;
1236                 goto out;
1237         }
1238         prev = t->prog;
1239         if (prev != prog) {
1240                 /* Bump prog refcnt once. Every bpf_timer_set_callback()
1241                  * can pick different callback_fn-s within the same prog.
1242                  */
1243                 prog = bpf_prog_inc_not_zero(prog);
1244                 if (IS_ERR(prog)) {
1245                         ret = PTR_ERR(prog);
1246                         goto out;
1247                 }
1248                 if (prev)
1249                         /* Drop prev prog refcnt when swapping with new prog */
1250                         bpf_prog_put(prev);
1251                 t->prog = prog;
1252         }
1253         rcu_assign_pointer(t->callback_fn, callback_fn);
1254 out:
1255         __bpf_spin_unlock_irqrestore(&timer->lock);
1256         return ret;
1257 }
1258
1259 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1260         .func           = bpf_timer_set_callback,
1261         .gpl_only       = true,
1262         .ret_type       = RET_INTEGER,
1263         .arg1_type      = ARG_PTR_TO_TIMER,
1264         .arg2_type      = ARG_PTR_TO_FUNC,
1265 };
1266
1267 BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1268 {
1269         struct bpf_hrtimer *t;
1270         int ret = 0;
1271
1272         if (in_nmi())
1273                 return -EOPNOTSUPP;
1274         if (flags)
1275                 return -EINVAL;
1276         __bpf_spin_lock_irqsave(&timer->lock);
1277         t = timer->timer;
1278         if (!t || !t->prog) {
1279                 ret = -EINVAL;
1280                 goto out;
1281         }
1282         hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT);
1283 out:
1284         __bpf_spin_unlock_irqrestore(&timer->lock);
1285         return ret;
1286 }
1287
1288 static const struct bpf_func_proto bpf_timer_start_proto = {
1289         .func           = bpf_timer_start,
1290         .gpl_only       = true,
1291         .ret_type       = RET_INTEGER,
1292         .arg1_type      = ARG_PTR_TO_TIMER,
1293         .arg2_type      = ARG_ANYTHING,
1294         .arg3_type      = ARG_ANYTHING,
1295 };
1296
1297 static void drop_prog_refcnt(struct bpf_hrtimer *t)
1298 {
1299         struct bpf_prog *prog = t->prog;
1300
1301         if (prog) {
1302                 bpf_prog_put(prog);
1303                 t->prog = NULL;
1304                 rcu_assign_pointer(t->callback_fn, NULL);
1305         }
1306 }
1307
1308 BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1309 {
1310         struct bpf_hrtimer *t;
1311         int ret = 0;
1312
1313         if (in_nmi())
1314                 return -EOPNOTSUPP;
1315         __bpf_spin_lock_irqsave(&timer->lock);
1316         t = timer->timer;
1317         if (!t) {
1318                 ret = -EINVAL;
1319                 goto out;
1320         }
1321         if (this_cpu_read(hrtimer_running) == t) {
1322                 /* If bpf callback_fn is trying to bpf_timer_cancel()
1323                  * its own timer the hrtimer_cancel() will deadlock
1324                  * since it waits for callback_fn to finish
1325                  */
1326                 ret = -EDEADLK;
1327                 goto out;
1328         }
1329         drop_prog_refcnt(t);
1330 out:
1331         __bpf_spin_unlock_irqrestore(&timer->lock);
1332         /* Cancel the timer and wait for associated callback to finish
1333          * if it was running.
1334          */
1335         ret = ret ?: hrtimer_cancel(&t->timer);
1336         return ret;
1337 }
1338
1339 static const struct bpf_func_proto bpf_timer_cancel_proto = {
1340         .func           = bpf_timer_cancel,
1341         .gpl_only       = true,
1342         .ret_type       = RET_INTEGER,
1343         .arg1_type      = ARG_PTR_TO_TIMER,
1344 };
1345
1346 /* This function is called by map_delete/update_elem for individual element and
1347  * by ops->map_release_uref when the user space reference to a map reaches zero.
1348  */
1349 void bpf_timer_cancel_and_free(void *val)
1350 {
1351         struct bpf_timer_kern *timer = val;
1352         struct bpf_hrtimer *t;
1353
1354         /* Performance optimization: read timer->timer without lock first. */
1355         if (!READ_ONCE(timer->timer))
1356                 return;
1357
1358         __bpf_spin_lock_irqsave(&timer->lock);
1359         /* re-read it under lock */
1360         t = timer->timer;
1361         if (!t)
1362                 goto out;
1363         drop_prog_refcnt(t);
1364         /* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1365          * this timer, since it won't be initialized.
1366          */
1367         timer->timer = NULL;
1368 out:
1369         __bpf_spin_unlock_irqrestore(&timer->lock);
1370         if (!t)
1371                 return;
1372         /* Cancel the timer and wait for callback to complete if it was running.
1373          * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1374          * right after for both preallocated and non-preallocated maps.
1375          * The timer->timer = NULL was already done and no code path can
1376          * see address 't' anymore.
1377          *
1378          * Check that bpf_map_delete/update_elem() wasn't called from timer
1379          * callback_fn. In such case don't call hrtimer_cancel() (since it will
1380          * deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1381          * return -1). Though callback_fn is still running on this cpu it's
1382          * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1383          * from 't'. The bpf subprog callback_fn won't be able to access 't',
1384          * since timer->timer = NULL was already done. The timer will be
1385          * effectively cancelled because bpf_timer_cb() will return
1386          * HRTIMER_NORESTART.
1387          */
1388         if (this_cpu_read(hrtimer_running) != t)
1389                 hrtimer_cancel(&t->timer);
1390         kfree(t);
1391 }
1392
1393 BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
1394 {
1395         unsigned long *kptr = map_value;
1396
1397         return xchg(kptr, (unsigned long)ptr);
1398 }
1399
1400 /* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1401  * helper is determined dynamically by the verifier.
1402  */
1403 #define BPF_PTR_POISON ((void *)((0xeB9FUL << 2) + POISON_POINTER_DELTA))
1404
1405 const struct bpf_func_proto bpf_kptr_xchg_proto = {
1406         .func         = bpf_kptr_xchg,
1407         .gpl_only     = false,
1408         .ret_type     = RET_PTR_TO_BTF_ID_OR_NULL,
1409         .ret_btf_id   = BPF_PTR_POISON,
1410         .arg1_type    = ARG_PTR_TO_KPTR,
1411         .arg2_type    = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1412         .arg2_btf_id  = BPF_PTR_POISON,
1413 };
1414
1415 /* Since the upper 8 bits of dynptr->size is reserved, the
1416  * maximum supported size is 2^24 - 1.
1417  */
1418 #define DYNPTR_MAX_SIZE ((1UL << 24) - 1)
1419 #define DYNPTR_TYPE_SHIFT       28
1420 #define DYNPTR_SIZE_MASK        0xFFFFFF
1421 #define DYNPTR_RDONLY_BIT       BIT(31)
1422
1423 static bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr)
1424 {
1425         return ptr->size & DYNPTR_RDONLY_BIT;
1426 }
1427
1428 static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1429 {
1430         ptr->size |= type << DYNPTR_TYPE_SHIFT;
1431 }
1432
1433 static u32 bpf_dynptr_get_size(struct bpf_dynptr_kern *ptr)
1434 {
1435         return ptr->size & DYNPTR_SIZE_MASK;
1436 }
1437
1438 int bpf_dynptr_check_size(u32 size)
1439 {
1440         return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1441 }
1442
1443 void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1444                      enum bpf_dynptr_type type, u32 offset, u32 size)
1445 {
1446         ptr->data = data;
1447         ptr->offset = offset;
1448         ptr->size = size;
1449         bpf_dynptr_set_type(ptr, type);
1450 }
1451
1452 void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1453 {
1454         memset(ptr, 0, sizeof(*ptr));
1455 }
1456
1457 static int bpf_dynptr_check_off_len(struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1458 {
1459         u32 size = bpf_dynptr_get_size(ptr);
1460
1461         if (len > size || offset > size - len)
1462                 return -E2BIG;
1463
1464         return 0;
1465 }
1466
1467 BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1468 {
1469         int err;
1470
1471         err = bpf_dynptr_check_size(size);
1472         if (err)
1473                 goto error;
1474
1475         /* flags is currently unsupported */
1476         if (flags) {
1477                 err = -EINVAL;
1478                 goto error;
1479         }
1480
1481         bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size);
1482
1483         return 0;
1484
1485 error:
1486         bpf_dynptr_set_null(ptr);
1487         return err;
1488 }
1489
1490 const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1491         .func           = bpf_dynptr_from_mem,
1492         .gpl_only       = false,
1493         .ret_type       = RET_INTEGER,
1494         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1495         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1496         .arg3_type      = ARG_ANYTHING,
1497         .arg4_type      = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT,
1498 };
1499
1500 BPF_CALL_4(bpf_dynptr_read, void *, dst, u32, len, struct bpf_dynptr_kern *, src, u32, offset)
1501 {
1502         int err;
1503
1504         if (!src->data)
1505                 return -EINVAL;
1506
1507         err = bpf_dynptr_check_off_len(src, offset, len);
1508         if (err)
1509                 return err;
1510
1511         memcpy(dst, src->data + src->offset + offset, len);
1512
1513         return 0;
1514 }
1515
1516 const struct bpf_func_proto bpf_dynptr_read_proto = {
1517         .func           = bpf_dynptr_read,
1518         .gpl_only       = false,
1519         .ret_type       = RET_INTEGER,
1520         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1521         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1522         .arg3_type      = ARG_PTR_TO_DYNPTR,
1523         .arg4_type      = ARG_ANYTHING,
1524 };
1525
1526 BPF_CALL_4(bpf_dynptr_write, struct bpf_dynptr_kern *, dst, u32, offset, void *, src, u32, len)
1527 {
1528         int err;
1529
1530         if (!dst->data || bpf_dynptr_is_rdonly(dst))
1531                 return -EINVAL;
1532
1533         err = bpf_dynptr_check_off_len(dst, offset, len);
1534         if (err)
1535                 return err;
1536
1537         memcpy(dst->data + dst->offset + offset, src, len);
1538
1539         return 0;
1540 }
1541
1542 const struct bpf_func_proto bpf_dynptr_write_proto = {
1543         .func           = bpf_dynptr_write,
1544         .gpl_only       = false,
1545         .ret_type       = RET_INTEGER,
1546         .arg1_type      = ARG_PTR_TO_DYNPTR,
1547         .arg2_type      = ARG_ANYTHING,
1548         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1549         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
1550 };
1551
1552 BPF_CALL_3(bpf_dynptr_data, struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1553 {
1554         int err;
1555
1556         if (!ptr->data)
1557                 return 0;
1558
1559         err = bpf_dynptr_check_off_len(ptr, offset, len);
1560         if (err)
1561                 return 0;
1562
1563         if (bpf_dynptr_is_rdonly(ptr))
1564                 return 0;
1565
1566         return (unsigned long)(ptr->data + ptr->offset + offset);
1567 }
1568
1569 const struct bpf_func_proto bpf_dynptr_data_proto = {
1570         .func           = bpf_dynptr_data,
1571         .gpl_only       = false,
1572         .ret_type       = RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1573         .arg1_type      = ARG_PTR_TO_DYNPTR,
1574         .arg2_type      = ARG_ANYTHING,
1575         .arg3_type      = ARG_CONST_ALLOC_SIZE_OR_ZERO,
1576 };
1577
1578 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1579 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1580 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1581 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1582 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1583 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1584 const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1585
1586 const struct bpf_func_proto *
1587 bpf_base_func_proto(enum bpf_func_id func_id)
1588 {
1589         switch (func_id) {
1590         case BPF_FUNC_map_lookup_elem:
1591                 return &bpf_map_lookup_elem_proto;
1592         case BPF_FUNC_map_update_elem:
1593                 return &bpf_map_update_elem_proto;
1594         case BPF_FUNC_map_delete_elem:
1595                 return &bpf_map_delete_elem_proto;
1596         case BPF_FUNC_map_push_elem:
1597                 return &bpf_map_push_elem_proto;
1598         case BPF_FUNC_map_pop_elem:
1599                 return &bpf_map_pop_elem_proto;
1600         case BPF_FUNC_map_peek_elem:
1601                 return &bpf_map_peek_elem_proto;
1602         case BPF_FUNC_map_lookup_percpu_elem:
1603                 return &bpf_map_lookup_percpu_elem_proto;
1604         case BPF_FUNC_get_prandom_u32:
1605                 return &bpf_get_prandom_u32_proto;
1606         case BPF_FUNC_get_smp_processor_id:
1607                 return &bpf_get_raw_smp_processor_id_proto;
1608         case BPF_FUNC_get_numa_node_id:
1609                 return &bpf_get_numa_node_id_proto;
1610         case BPF_FUNC_tail_call:
1611                 return &bpf_tail_call_proto;
1612         case BPF_FUNC_ktime_get_ns:
1613                 return &bpf_ktime_get_ns_proto;
1614         case BPF_FUNC_ktime_get_boot_ns:
1615                 return &bpf_ktime_get_boot_ns_proto;
1616         case BPF_FUNC_ringbuf_output:
1617                 return &bpf_ringbuf_output_proto;
1618         case BPF_FUNC_ringbuf_reserve:
1619                 return &bpf_ringbuf_reserve_proto;
1620         case BPF_FUNC_ringbuf_submit:
1621                 return &bpf_ringbuf_submit_proto;
1622         case BPF_FUNC_ringbuf_discard:
1623                 return &bpf_ringbuf_discard_proto;
1624         case BPF_FUNC_ringbuf_query:
1625                 return &bpf_ringbuf_query_proto;
1626         case BPF_FUNC_ringbuf_reserve_dynptr:
1627                 return &bpf_ringbuf_reserve_dynptr_proto;
1628         case BPF_FUNC_ringbuf_submit_dynptr:
1629                 return &bpf_ringbuf_submit_dynptr_proto;
1630         case BPF_FUNC_ringbuf_discard_dynptr:
1631                 return &bpf_ringbuf_discard_dynptr_proto;
1632         case BPF_FUNC_for_each_map_elem:
1633                 return &bpf_for_each_map_elem_proto;
1634         case BPF_FUNC_loop:
1635                 return &bpf_loop_proto;
1636         case BPF_FUNC_strncmp:
1637                 return &bpf_strncmp_proto;
1638         case BPF_FUNC_dynptr_from_mem:
1639                 return &bpf_dynptr_from_mem_proto;
1640         case BPF_FUNC_dynptr_read:
1641                 return &bpf_dynptr_read_proto;
1642         case BPF_FUNC_dynptr_write:
1643                 return &bpf_dynptr_write_proto;
1644         case BPF_FUNC_dynptr_data:
1645                 return &bpf_dynptr_data_proto;
1646         default:
1647                 break;
1648         }
1649
1650         if (!bpf_capable())
1651                 return NULL;
1652
1653         switch (func_id) {
1654         case BPF_FUNC_spin_lock:
1655                 return &bpf_spin_lock_proto;
1656         case BPF_FUNC_spin_unlock:
1657                 return &bpf_spin_unlock_proto;
1658         case BPF_FUNC_jiffies64:
1659                 return &bpf_jiffies64_proto;
1660         case BPF_FUNC_per_cpu_ptr:
1661                 return &bpf_per_cpu_ptr_proto;
1662         case BPF_FUNC_this_cpu_ptr:
1663                 return &bpf_this_cpu_ptr_proto;
1664         case BPF_FUNC_timer_init:
1665                 return &bpf_timer_init_proto;
1666         case BPF_FUNC_timer_set_callback:
1667                 return &bpf_timer_set_callback_proto;
1668         case BPF_FUNC_timer_start:
1669                 return &bpf_timer_start_proto;
1670         case BPF_FUNC_timer_cancel:
1671                 return &bpf_timer_cancel_proto;
1672         case BPF_FUNC_kptr_xchg:
1673                 return &bpf_kptr_xchg_proto;
1674         default:
1675                 break;
1676         }
1677
1678         if (!perfmon_capable())
1679                 return NULL;
1680
1681         switch (func_id) {
1682         case BPF_FUNC_trace_printk:
1683                 return bpf_get_trace_printk_proto();
1684         case BPF_FUNC_get_current_task:
1685                 return &bpf_get_current_task_proto;
1686         case BPF_FUNC_get_current_task_btf:
1687                 return &bpf_get_current_task_btf_proto;
1688         case BPF_FUNC_probe_read_user:
1689                 return &bpf_probe_read_user_proto;
1690         case BPF_FUNC_probe_read_kernel:
1691                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1692                        NULL : &bpf_probe_read_kernel_proto;
1693         case BPF_FUNC_probe_read_user_str:
1694                 return &bpf_probe_read_user_str_proto;
1695         case BPF_FUNC_probe_read_kernel_str:
1696                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1697                        NULL : &bpf_probe_read_kernel_str_proto;
1698         case BPF_FUNC_snprintf_btf:
1699                 return &bpf_snprintf_btf_proto;
1700         case BPF_FUNC_snprintf:
1701                 return &bpf_snprintf_proto;
1702         case BPF_FUNC_task_pt_regs:
1703                 return &bpf_task_pt_regs_proto;
1704         case BPF_FUNC_trace_vprintk:
1705                 return bpf_get_trace_vprintk_proto();
1706         default:
1707                 return NULL;
1708         }
1709 }