kernel/bpf/helpers.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/rcupdate.h>
   6 #include <linux/random.h>
   7 #include <linux/smp.h>
   8 #include <linux/topology.h>
   9 #include <linux/ktime.h>
  10 #include <linux/sched.h>
  11 #include <linux/uidgid.h>
  12 #include <linux/filter.h>
  13 #include <linux/ctype.h>
  14 #include <linux/jiffies.h>
  15 #include <linux/pid_namespace.h>
  16 #include <linux/proc_ns.h>
  17 #include <linux/security.h>
  18
  19 #include "../../lib/kstrtox.h"
  20
  21 /* If kernel subsystem is allowing eBPF programs to call this function,
  22  * inside its own verifier_ops->get_func_proto() callback it should return
  23  * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
  24  *
  25  * Different map implementations will rely on rcu in map methods
  26  * lookup/update/delete, therefore eBPF programs must run under rcu lock
  27  * if program is allowed to access maps, so check rcu_read_lock_held in
  28  * all three functions.
  29  */
  30 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
  31 {
  32         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  33         return (unsigned long) map->ops->map_lookup_elem(map, key);
  34 }
  35
  36 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
  37         .func           = bpf_map_lookup_elem,
  38         .gpl_only       = false,
  39         .pkt_access     = true,
  40         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
  41         .arg1_type      = ARG_CONST_MAP_PTR,
  42         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  43 };
  44
  45 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
  46            void *, value, u64, flags)
  47 {
  48         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  49         return map->ops->map_update_elem(map, key, value, flags);
  50 }
  51
  52 const struct bpf_func_proto bpf_map_update_elem_proto = {
  53         .func           = bpf_map_update_elem,
  54         .gpl_only       = false,
  55         .pkt_access     = true,
  56         .ret_type       = RET_INTEGER,
  57         .arg1_type      = ARG_CONST_MAP_PTR,
  58         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  59         .arg3_type      = ARG_PTR_TO_MAP_VALUE,
  60         .arg4_type      = ARG_ANYTHING,
  61 };
  62
  63 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
  64 {
  65         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
  66         return map->ops->map_delete_elem(map, key);
  67 }
  68
  69 const struct bpf_func_proto bpf_map_delete_elem_proto = {
  70         .func           = bpf_map_delete_elem,
  71         .gpl_only       = false,
  72         .pkt_access     = true,
  73         .ret_type       = RET_INTEGER,
  74         .arg1_type      = ARG_CONST_MAP_PTR,
  75         .arg2_type      = ARG_PTR_TO_MAP_KEY,
  76 };
  77
  78 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
  79 {
  80         return map->ops->map_push_elem(map, value, flags);
  81 }
  82
  83 const struct bpf_func_proto bpf_map_push_elem_proto = {
  84         .func           = bpf_map_push_elem,
  85         .gpl_only       = false,
  86         .pkt_access     = true,
  87         .ret_type       = RET_INTEGER,
  88         .arg1_type      = ARG_CONST_MAP_PTR,
  89         .arg2_type      = ARG_PTR_TO_MAP_VALUE,
  90         .arg3_type      = ARG_ANYTHING,
  91 };
  92
  93 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
  94 {
  95         return map->ops->map_pop_elem(map, value);
  96 }
  97
  98 const struct bpf_func_proto bpf_map_pop_elem_proto = {
  99         .func           = bpf_map_pop_elem,
 100         .gpl_only       = false,
 101         .ret_type       = RET_INTEGER,
 102         .arg1_type      = ARG_CONST_MAP_PTR,
 103         .arg2_type      = ARG_PTR_TO_UNINIT_MAP_VALUE,
 104 };
 105
 106 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
 107 {
 108         return map->ops->map_peek_elem(map, value);
 109 }
 110
 111 const struct bpf_func_proto bpf_map_peek_elem_proto = {
 112         .func           = bpf_map_peek_elem,
 113         .gpl_only       = false,
 114         .ret_type       = RET_INTEGER,
 115         .arg1_type      = ARG_CONST_MAP_PTR,
 116         .arg2_type      = ARG_PTR_TO_UNINIT_MAP_VALUE,
 117 };
 118
 119 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
 120         .func           = bpf_user_rnd_u32,
 121         .gpl_only       = false,
 122         .ret_type       = RET_INTEGER,
 123 };
 124
 125 BPF_CALL_0(bpf_get_smp_processor_id)
 126 {
 127         return smp_processor_id();
 128 }
 129
 130 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
 131         .func           = bpf_get_smp_processor_id,
 132         .gpl_only       = false,
 133         .ret_type       = RET_INTEGER,
 134 };
 135
 136 BPF_CALL_0(bpf_get_numa_node_id)
 137 {
 138         return numa_node_id();
 139 }
 140
 141 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
 142         .func           = bpf_get_numa_node_id,
 143         .gpl_only       = false,
 144         .ret_type       = RET_INTEGER,
 145 };
 146
 147 BPF_CALL_0(bpf_ktime_get_ns)
 148 {
 149         /* NMI safe access to clock monotonic */
 150         return ktime_get_mono_fast_ns();
 151 }
 152
 153 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
 154         .func           = bpf_ktime_get_ns,
 155         .gpl_only       = false,
 156         .ret_type       = RET_INTEGER,
 157 };
 158
 159 BPF_CALL_0(bpf_ktime_get_boot_ns)
 160 {
 161         /* NMI safe access to clock boottime */
 162         return ktime_get_boot_fast_ns();
 163 }
 164
 165 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
 166         .func           = bpf_ktime_get_boot_ns,
 167         .gpl_only       = false,
 168         .ret_type       = RET_INTEGER,
 169 };
 170
 171 BPF_CALL_0(bpf_ktime_get_coarse_ns)
 172 {
 173         return ktime_get_coarse_ns();
 174 }
 175
 176 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
 177         .func           = bpf_ktime_get_coarse_ns,
 178         .gpl_only       = false,
 179         .ret_type       = RET_INTEGER,
 180 };
 181
 182 BPF_CALL_0(bpf_get_current_pid_tgid)
 183 {
 184         struct task_struct *task = current;
 185
 186         if (unlikely(!task))
 187                 return -EINVAL;
 188
 189         return (u64) task->tgid << 32 | task->pid;
 190 }
 191
 192 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
 193         .func           = bpf_get_current_pid_tgid,
 194         .gpl_only       = false,
 195         .ret_type       = RET_INTEGER,
 196 };
 197
 198 BPF_CALL_0(bpf_get_current_uid_gid)
 199 {
 200         struct task_struct *task = current;
 201         kuid_t uid;
 202         kgid_t gid;
 203
 204         if (unlikely(!task))
 205                 return -EINVAL;
 206
 207         current_uid_gid(&uid, &gid);
 208         return (u64) from_kgid(&init_user_ns, gid) << 32 |
 209                      from_kuid(&init_user_ns, uid);
 210 }
 211
 212 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
 213         .func           = bpf_get_current_uid_gid,
 214         .gpl_only       = false,
 215         .ret_type       = RET_INTEGER,
 216 };
 217
 218 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
 219 {
 220         struct task_struct *task = current;
 221
 222         if (unlikely(!task))
 223                 goto err_clear;
 224
 225         strncpy(buf, task->comm, size);
 226
 227         /* Verifier guarantees that size > 0. For task->comm exceeding
 228          * size, guarantee that buf is %NUL-terminated. Unconditionally
 229          * done here to save the size test.
 230          */
 231         buf[size - 1] = 0;
 232         return 0;
 233 err_clear:
 234         memset(buf, 0, size);
 235         return -EINVAL;
 236 }
 237
 238 const struct bpf_func_proto bpf_get_current_comm_proto = {
 239         .func           = bpf_get_current_comm,
 240         .gpl_only       = false,
 241         .ret_type       = RET_INTEGER,
 242         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 243         .arg2_type      = ARG_CONST_SIZE,
 244 };
 245
 246 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
 247
 248 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
 249 {
 250         arch_spinlock_t *l = (void *)lock;
 251         union {
 252                 __u32 val;
 253                 arch_spinlock_t lock;
 254         } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
 255
 256         compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
 257         BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
 258         BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
 259         arch_spin_lock(l);
 260 }
 261
 262 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
 263 {
 264         arch_spinlock_t *l = (void *)lock;
 265
 266         arch_spin_unlock(l);
 267 }
 268
 269 #else
 270
 271 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
 272 {
 273         atomic_t *l = (void *)lock;
 274
 275         BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
 276         do {
 277                 atomic_cond_read_relaxed(l, !VAL);
 278         } while (atomic_xchg(l, 1));
 279 }
 280
 281 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
 282 {
 283         atomic_t *l = (void *)lock;
 284
 285         atomic_set_release(l, 0);
 286 }
 287
 288 #endif
 289
 290 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
 291
 292 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
 293 {
 294         unsigned long flags;
 295
 296         local_irq_save(flags);
 297         __bpf_spin_lock(lock);
 298         __this_cpu_write(irqsave_flags, flags);
 299 }
 300
 301 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
 302 {
 303         __bpf_spin_lock_irqsave(lock);
 304         return 0;
 305 }
 306
 307 const struct bpf_func_proto bpf_spin_lock_proto = {
 308         .func           = bpf_spin_lock,
 309         .gpl_only       = false,
 310         .ret_type       = RET_VOID,
 311         .arg1_type      = ARG_PTR_TO_SPIN_LOCK,
 312 };
 313
 314 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
 315 {
 316         unsigned long flags;
 317
 318         flags = __this_cpu_read(irqsave_flags);
 319         __bpf_spin_unlock(lock);
 320         local_irq_restore(flags);
 321 }
 322
 323 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
 324 {
 325         __bpf_spin_unlock_irqrestore(lock);
 326         return 0;
 327 }
 328
 329 const struct bpf_func_proto bpf_spin_unlock_proto = {
 330         .func           = bpf_spin_unlock,
 331         .gpl_only       = false,
 332         .ret_type       = RET_VOID,
 333         .arg1_type      = ARG_PTR_TO_SPIN_LOCK,
 334 };
 335
 336 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
 337                            bool lock_src)
 338 {
 339         struct bpf_spin_lock *lock;
 340
 341         if (lock_src)
 342                 lock = src + map->spin_lock_off;
 343         else
 344                 lock = dst + map->spin_lock_off;
 345         preempt_disable();
 346         __bpf_spin_lock_irqsave(lock);
 347         copy_map_value(map, dst, src);
 348         __bpf_spin_unlock_irqrestore(lock);
 349         preempt_enable();
 350 }
 351
 352 BPF_CALL_0(bpf_jiffies64)
 353 {
 354         return get_jiffies_64();
 355 }
 356
 357 const struct bpf_func_proto bpf_jiffies64_proto = {
 358         .func           = bpf_jiffies64,
 359         .gpl_only       = false,
 360         .ret_type       = RET_INTEGER,
 361 };
 362
 363 #ifdef CONFIG_CGROUPS
 364 BPF_CALL_0(bpf_get_current_cgroup_id)
 365 {
 366         struct cgroup *cgrp;
 367         u64 cgrp_id;
 368
 369         rcu_read_lock();
 370         cgrp = task_dfl_cgroup(current);
 371         cgrp_id = cgroup_id(cgrp);
 372         rcu_read_unlock();
 373
 374         return cgrp_id;
 375 }
 376
 377 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
 378         .func           = bpf_get_current_cgroup_id,
 379         .gpl_only       = false,
 380         .ret_type       = RET_INTEGER,
 381 };
 382
 383 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
 384 {
 385         struct cgroup *cgrp;
 386         struct cgroup *ancestor;
 387         u64 cgrp_id;
 388
 389         rcu_read_lock();
 390         cgrp = task_dfl_cgroup(current);
 391         ancestor = cgroup_ancestor(cgrp, ancestor_level);
 392         cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
 393         rcu_read_unlock();
 394
 395         return cgrp_id;
 396 }
 397
 398 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
 399         .func           = bpf_get_current_ancestor_cgroup_id,
 400         .gpl_only       = false,
 401         .ret_type       = RET_INTEGER,
 402         .arg1_type      = ARG_ANYTHING,
 403 };
 404
 405 #ifdef CONFIG_CGROUP_BPF
 406
 407 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
 408 {
 409         /* flags argument is not used now,
 410          * but provides an ability to extend the API.
 411          * verifier checks that its value is correct.
 412          */
 413         enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
 414         struct bpf_cgroup_storage *storage;
 415         struct bpf_cg_run_ctx *ctx;
 416         void *ptr;
 417
 418         /* get current cgroup storage from BPF run context */
 419         ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
 420         storage = ctx->prog_item->cgroup_storage[stype];
 421
 422         if (stype == BPF_CGROUP_STORAGE_SHARED)
 423                 ptr = &READ_ONCE(storage->buf)->data[0];
 424         else
 425                 ptr = this_cpu_ptr(storage->percpu_buf);
 426
 427         return (unsigned long)ptr;
 428 }
 429
 430 const struct bpf_func_proto bpf_get_local_storage_proto = {
 431         .func           = bpf_get_local_storage,
 432         .gpl_only       = false,
 433         .ret_type       = RET_PTR_TO_MAP_VALUE,
 434         .arg1_type      = ARG_CONST_MAP_PTR,
 435         .arg2_type      = ARG_ANYTHING,
 436 };
 437 #endif
 438
 439 #define BPF_STRTOX_BASE_MASK 0x1F
 440
 441 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
 442                           unsigned long long *res, bool *is_negative)
 443 {
 444         unsigned int base = flags & BPF_STRTOX_BASE_MASK;
 445         const char *cur_buf = buf;
 446         size_t cur_len = buf_len;
 447         unsigned int consumed;
 448         size_t val_len;
 449         char str[64];
 450
 451         if (!buf || !buf_len || !res || !is_negative)
 452                 return -EINVAL;
 453
 454         if (base != 0 && base != 8 && base != 10 && base != 16)
 455                 return -EINVAL;
 456
 457         if (flags & ~BPF_STRTOX_BASE_MASK)
 458                 return -EINVAL;
 459
 460         while (cur_buf < buf + buf_len && isspace(*cur_buf))
 461                 ++cur_buf;
 462
 463         *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
 464         if (*is_negative)
 465                 ++cur_buf;
 466
 467         consumed = cur_buf - buf;
 468         cur_len -= consumed;
 469         if (!cur_len)
 470                 return -EINVAL;
 471
 472         cur_len = min(cur_len, sizeof(str) - 1);
 473         memcpy(str, cur_buf, cur_len);
 474         str[cur_len] = '\0';
 475         cur_buf = str;
 476
 477         cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
 478         val_len = _parse_integer(cur_buf, base, res);
 479
 480         if (val_len & KSTRTOX_OVERFLOW)
 481                 return -ERANGE;
 482
 483         if (val_len == 0)
 484                 return -EINVAL;
 485
 486         cur_buf += val_len;
 487         consumed += cur_buf - str;
 488
 489         return consumed;
 490 }
 491
 492 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
 493                          long long *res)
 494 {
 495         unsigned long long _res;
 496         bool is_negative;
 497         int err;
 498
 499         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 500         if (err < 0)
 501                 return err;
 502         if (is_negative) {
 503                 if ((long long)-_res > 0)
 504                         return -ERANGE;
 505                 *res = -_res;
 506         } else {
 507                 if ((long long)_res < 0)
 508                         return -ERANGE;
 509                 *res = _res;
 510         }
 511         return err;
 512 }
 513
 514 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
 515            long *, res)
 516 {
 517         long long _res;
 518         int err;
 519
 520         err = __bpf_strtoll(buf, buf_len, flags, &_res);
 521         if (err < 0)
 522                 return err;
 523         if (_res != (long)_res)
 524                 return -ERANGE;
 525         *res = _res;
 526         return err;
 527 }
 528
 529 const struct bpf_func_proto bpf_strtol_proto = {
 530         .func           = bpf_strtol,
 531         .gpl_only       = false,
 532         .ret_type       = RET_INTEGER,
 533         .arg1_type      = ARG_PTR_TO_MEM,
 534         .arg2_type      = ARG_CONST_SIZE,
 535         .arg3_type      = ARG_ANYTHING,
 536         .arg4_type      = ARG_PTR_TO_LONG,
 537 };
 538
 539 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
 540            unsigned long *, res)
 541 {
 542         unsigned long long _res;
 543         bool is_negative;
 544         int err;
 545
 546         err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
 547         if (err < 0)
 548                 return err;
 549         if (is_negative)
 550                 return -EINVAL;
 551         if (_res != (unsigned long)_res)
 552                 return -ERANGE;
 553         *res = _res;
 554         return err;
 555 }
 556
 557 const struct bpf_func_proto bpf_strtoul_proto = {
 558         .func           = bpf_strtoul,
 559         .gpl_only       = false,
 560         .ret_type       = RET_INTEGER,
 561         .arg1_type      = ARG_PTR_TO_MEM,
 562         .arg2_type      = ARG_CONST_SIZE,
 563         .arg3_type      = ARG_ANYTHING,
 564         .arg4_type      = ARG_PTR_TO_LONG,
 565 };
 566 #endif
 567
 568 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
 569            struct bpf_pidns_info *, nsdata, u32, size)
 570 {
 571         struct task_struct *task = current;
 572         struct pid_namespace *pidns;
 573         int err = -EINVAL;
 574
 575         if (unlikely(size != sizeof(struct bpf_pidns_info)))
 576                 goto clear;
 577
 578         if (unlikely((u64)(dev_t)dev != dev))
 579                 goto clear;
 580
 581         if (unlikely(!task))
 582                 goto clear;
 583
 584         pidns = task_active_pid_ns(task);
 585         if (unlikely(!pidns)) {
 586                 err = -ENOENT;
 587                 goto clear;
 588         }
 589
 590         if (!ns_match(&pidns->ns, (dev_t)dev, ino))
 591                 goto clear;
 592
 593         nsdata->pid = task_pid_nr_ns(task, pidns);
 594         nsdata->tgid = task_tgid_nr_ns(task, pidns);
 595         return 0;
 596 clear:
 597         memset((void *)nsdata, 0, (size_t) size);
 598         return err;
 599 }
 600
 601 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
 602         .func           = bpf_get_ns_current_pid_tgid,
 603         .gpl_only       = false,
 604         .ret_type       = RET_INTEGER,
 605         .arg1_type      = ARG_ANYTHING,
 606         .arg2_type      = ARG_ANYTHING,
 607         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 608         .arg4_type      = ARG_CONST_SIZE,
 609 };
 610
 611 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 612         .func           = bpf_get_raw_cpu_id,
 613         .gpl_only       = false,
 614         .ret_type       = RET_INTEGER,
 615 };
 616
 617 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
 618            u64, flags, void *, data, u64, size)
 619 {
 620         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 621                 return -EINVAL;
 622
 623         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
 624 }
 625
 626 const struct bpf_func_proto bpf_event_output_data_proto =  {
 627         .func           = bpf_event_output_data,
 628         .gpl_only       = true,
 629         .ret_type       = RET_INTEGER,
 630         .arg1_type      = ARG_PTR_TO_CTX,
 631         .arg2_type      = ARG_CONST_MAP_PTR,
 632         .arg3_type      = ARG_ANYTHING,
 633         .arg4_type      = ARG_PTR_TO_MEM,
 634         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 635 };
 636
 637 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
 638            const void __user *, user_ptr)
 639 {
 640         int ret = copy_from_user(dst, user_ptr, size);
 641
 642         if (unlikely(ret)) {
 643                 memset(dst, 0, size);
 644                 ret = -EFAULT;
 645         }
 646
 647         return ret;
 648 }
 649
 650 const struct bpf_func_proto bpf_copy_from_user_proto = {
 651         .func           = bpf_copy_from_user,
 652         .gpl_only       = false,
 653         .ret_type       = RET_INTEGER,
 654         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 655         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 656         .arg3_type      = ARG_ANYTHING,
 657 };
 658
 659 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
 660 {
 661         if (cpu >= nr_cpu_ids)
 662                 return (unsigned long)NULL;
 663
 664         return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
 665 }
 666
 667 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
 668         .func           = bpf_per_cpu_ptr,
 669         .gpl_only       = false,
 670         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL,
 671         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 672         .arg2_type      = ARG_ANYTHING,
 673 };
 674
 675 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
 676 {
 677         return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
 678 }
 679
 680 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
 681         .func           = bpf_this_cpu_ptr,
 682         .gpl_only       = false,
 683         .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID,
 684         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
 685 };
 686
 687 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
 688                 size_t bufsz)
 689 {
 690         void __user *user_ptr = (__force void __user *)unsafe_ptr;
 691
 692         buf[0] = 0;
 693
 694         switch (fmt_ptype) {
 695         case 's':
 696 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 697                 if ((unsigned long)unsafe_ptr < TASK_SIZE)
 698                         return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 699                 fallthrough;
 700 #endif
 701         case 'k':
 702                 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
 703         case 'u':
 704                 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
 705         }
 706
 707         return -EINVAL;
 708 }
 709
 710 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
 711  * arguments representation.
 712  */
 713 #define MAX_BPRINTF_BUF_LEN     512
 714
 715 /* Support executing three nested bprintf helper calls on a given CPU */
 716 #define MAX_BPRINTF_NEST_LEVEL  3
 717 struct bpf_bprintf_buffers {
 718         char tmp_bufs[MAX_BPRINTF_NEST_LEVEL][MAX_BPRINTF_BUF_LEN];
 719 };
 720 static DEFINE_PER_CPU(struct bpf_bprintf_buffers, bpf_bprintf_bufs);
 721 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
 722
 723 static int try_get_fmt_tmp_buf(char **tmp_buf)
 724 {
 725         struct bpf_bprintf_buffers *bufs;
 726         int nest_level;
 727
 728         preempt_disable();
 729         nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
 730         if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
 731                 this_cpu_dec(bpf_bprintf_nest_level);
 732                 preempt_enable();
 733                 return -EBUSY;
 734         }
 735         bufs = this_cpu_ptr(&bpf_bprintf_bufs);
 736         *tmp_buf = bufs->tmp_bufs[nest_level - 1];
 737
 738         return 0;
 739 }
 740
 741 void bpf_bprintf_cleanup(void)
 742 {
 743         if (this_cpu_read(bpf_bprintf_nest_level)) {
 744                 this_cpu_dec(bpf_bprintf_nest_level);
 745                 preempt_enable();
 746         }
 747 }
 748
 749 /*
 750  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
 751  *
 752  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
 753  *
 754  * This can be used in two ways:
 755  * - Format string verification only: when bin_args is NULL
 756  * - Arguments preparation: in addition to the above verification, it writes in
 757  *   bin_args a binary representation of arguments usable by bstr_printf where
 758  *   pointers from BPF have been sanitized.
 759  *
 760  * In argument preparation mode, if 0 is returned, safe temporary buffers are
 761  * allocated and bpf_bprintf_cleanup should be called to free them after use.
 762  */
 763 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
 764                         u32 **bin_args, u32 num_args)
 765 {
 766         char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
 767         size_t sizeof_cur_arg, sizeof_cur_ip;
 768         int err, i, num_spec = 0;
 769         u64 cur_arg;
 770         char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
 771
 772         fmt_end = strnchr(fmt, fmt_size, 0);
 773         if (!fmt_end)
 774                 return -EINVAL;
 775         fmt_size = fmt_end - fmt;
 776
 777         if (bin_args) {
 778                 if (num_args && try_get_fmt_tmp_buf(&tmp_buf))
 779                         return -EBUSY;
 780
 781                 tmp_buf_end = tmp_buf + MAX_BPRINTF_BUF_LEN;
 782                 *bin_args = (u32 *)tmp_buf;
 783         }
 784
 785         for (i = 0; i < fmt_size; i++) {
 786                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
 787                         err = -EINVAL;
 788                         goto out;
 789                 }
 790
 791                 if (fmt[i] != '%')
 792                         continue;
 793
 794                 if (fmt[i + 1] == '%') {
 795                         i++;
 796                         continue;
 797                 }
 798
 799                 if (num_spec >= num_args) {
 800                         err = -EINVAL;
 801                         goto out;
 802                 }
 803
 804                 /* The string is zero-terminated so if fmt[i] != 0, we can
 805                  * always access fmt[i + 1], in the worst case it will be a 0
 806                  */
 807                 i++;
 808
 809                 /* skip optional "[0 +-][num]" width formatting field */
 810                 while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
 811                        fmt[i] == ' ')
 812                         i++;
 813                 if (fmt[i] >= '1' && fmt[i] <= '9') {
 814                         i++;
 815                         while (fmt[i] >= '0' && fmt[i] <= '9')
 816                                 i++;
 817                 }
 818
 819                 if (fmt[i] == 'p') {
 820                         sizeof_cur_arg = sizeof(long);
 821
 822                         if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
 823                             fmt[i + 2] == 's') {
 824                                 fmt_ptype = fmt[i + 1];
 825                                 i += 2;
 826                                 goto fmt_str;
 827                         }
 828
 829                         if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
 830                             ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
 831                             fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
 832                             fmt[i + 1] == 'S') {
 833                                 /* just kernel pointers */
 834                                 if (tmp_buf)
 835                                         cur_arg = raw_args[num_spec];
 836                                 i++;
 837                                 goto nocopy_fmt;
 838                         }
 839
 840                         if (fmt[i + 1] == 'B') {
 841                                 if (tmp_buf)  {
 842                                         err = snprintf(tmp_buf,
 843                                                        (tmp_buf_end - tmp_buf),
 844                                                        "%pB",
 845                                                        (void *)(long)raw_args[num_spec]);
 846                                         tmp_buf += (err + 1);
 847                                 }
 848
 849                                 i++;
 850                                 num_spec++;
 851                                 continue;
 852                         }
 853
 854                         /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
 855                         if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
 856                             (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
 857                                 err = -EINVAL;
 858                                 goto out;
 859                         }
 860
 861                         i += 2;
 862                         if (!tmp_buf)
 863                                 goto nocopy_fmt;
 864
 865                         sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
 866                         if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
 867                                 err = -ENOSPC;
 868                                 goto out;
 869                         }
 870
 871                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 872                         err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
 873                                                        sizeof_cur_ip);
 874                         if (err < 0)
 875                                 memset(cur_ip, 0, sizeof_cur_ip);
 876
 877                         /* hack: bstr_printf expects IP addresses to be
 878                          * pre-formatted as strings, ironically, the easiest way
 879                          * to do that is to call snprintf.
 880                          */
 881                         ip_spec[2] = fmt[i - 1];
 882                         ip_spec[3] = fmt[i];
 883                         err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
 884                                        ip_spec, &cur_ip);
 885
 886                         tmp_buf += err + 1;
 887                         num_spec++;
 888
 889                         continue;
 890                 } else if (fmt[i] == 's') {
 891                         fmt_ptype = fmt[i];
 892 fmt_str:
 893                         if (fmt[i + 1] != 0 &&
 894                             !isspace(fmt[i + 1]) &&
 895                             !ispunct(fmt[i + 1])) {
 896                                 err = -EINVAL;
 897                                 goto out;
 898                         }
 899
 900                         if (!tmp_buf)
 901                                 goto nocopy_fmt;
 902
 903                         if (tmp_buf_end == tmp_buf) {
 904                                 err = -ENOSPC;
 905                                 goto out;
 906                         }
 907
 908                         unsafe_ptr = (char *)(long)raw_args[num_spec];
 909                         err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
 910                                                     fmt_ptype,
 911                                                     tmp_buf_end - tmp_buf);
 912                         if (err < 0) {
 913                                 tmp_buf[0] = '\0';
 914                                 err = 1;
 915                         }
 916
 917                         tmp_buf += err;
 918                         num_spec++;
 919
 920                         continue;
 921                 } else if (fmt[i] == 'c') {
 922                         if (!tmp_buf)
 923                                 goto nocopy_fmt;
 924
 925                         if (tmp_buf_end == tmp_buf) {
 926                                 err = -ENOSPC;
 927                                 goto out;
 928                         }
 929
 930                         *tmp_buf = raw_args[num_spec];
 931                         tmp_buf++;
 932                         num_spec++;
 933
 934                         continue;
 935                 }
 936
 937                 sizeof_cur_arg = sizeof(int);
 938
 939                 if (fmt[i] == 'l') {
 940                         sizeof_cur_arg = sizeof(long);
 941                         i++;
 942                 }
 943                 if (fmt[i] == 'l') {
 944                         sizeof_cur_arg = sizeof(long long);
 945                         i++;
 946                 }
 947
 948                 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
 949                     fmt[i] != 'x' && fmt[i] != 'X') {
 950                         err = -EINVAL;
 951                         goto out;
 952                 }
 953
 954                 if (tmp_buf)
 955                         cur_arg = raw_args[num_spec];
 956 nocopy_fmt:
 957                 if (tmp_buf) {
 958                         tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
 959                         if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
 960                                 err = -ENOSPC;
 961                                 goto out;
 962                         }
 963
 964                         if (sizeof_cur_arg == 8) {
 965                                 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
 966                                 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
 967                         } else {
 968                                 *(u32 *)tmp_buf = (u32)(long)cur_arg;
 969                         }
 970                         tmp_buf += sizeof_cur_arg;
 971                 }
 972                 num_spec++;
 973         }
 974
 975         err = 0;
 976 out:
 977         if (err)
 978                 bpf_bprintf_cleanup();
 979         return err;
 980 }
 981
 982 #define MAX_SNPRINTF_VARARGS            12
 983
 984 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
 985            const void *, data, u32, data_len)
 986 {
 987         int err, num_args;
 988         u32 *bin_args;
 989
 990         if (data_len % 8 || data_len > MAX_SNPRINTF_VARARGS * 8 ||
 991             (data_len && !data))
 992                 return -EINVAL;
 993         num_args = data_len / 8;
 994
 995         /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
 996          * can safely give an unbounded size.
 997          */
 998         err = bpf_bprintf_prepare(fmt, UINT_MAX, data, &bin_args, num_args);
 999         if (err < 0)
1000                 return err;
1001
1002         err = bstr_printf(str, str_size, fmt, bin_args);
1003
1004         bpf_bprintf_cleanup();
1005
1006         return err + 1;
1007 }
1008
1009 const struct bpf_func_proto bpf_snprintf_proto = {
1010         .func           = bpf_snprintf,
1011         .gpl_only       = true,
1012         .ret_type       = RET_INTEGER,
1013         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1014         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1015         .arg3_type      = ARG_PTR_TO_CONST_STR,
1016         .arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
1017         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1018 };
1019
1020 /* BPF map elements can contain 'struct bpf_timer'.
1021  * Such map owns all of its BPF timers.
1022  * 'struct bpf_timer' is allocated as part of map element allocation
1023  * and it's zero initialized.
1024  * That space is used to keep 'struct bpf_timer_kern'.
1025  * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1026  * remembers 'struct bpf_map *' pointer it's part of.
1027  * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1028  * bpf_timer_start() arms the timer.
1029  * If user space reference to a map goes to zero at this point
1030  * ops->map_release_uref callback is responsible for cancelling the timers,
1031  * freeing their memory, and decrementing prog's refcnts.
1032  * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1033  * Inner maps can contain bpf timers as well. ops->map_release_uref is
1034  * freeing the timers when inner map is replaced or deleted by user space.
1035  */
1036 struct bpf_hrtimer {
1037         struct hrtimer timer;
1038         struct bpf_map *map;
1039         struct bpf_prog *prog;
1040         void __rcu *callback_fn;
1041         void *value;
1042 };
1043
1044 /* the actual struct hidden inside uapi struct bpf_timer */
1045 struct bpf_timer_kern {
1046         struct bpf_hrtimer *timer;
1047         /* bpf_spin_lock is used here instead of spinlock_t to make
1048          * sure that it always fits into space resereved by struct bpf_timer
1049          * regardless of LOCKDEP and spinlock debug flags.
1050          */
1051         struct bpf_spin_lock lock;
1052 } __attribute__((aligned(8)));
1053
1054 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1055
1056 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1057 {
1058         struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1059         struct bpf_map *map = t->map;
1060         void *value = t->value;
1061         void *callback_fn;
1062         void *key;
1063         u32 idx;
1064
1065         callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1066         if (!callback_fn)
1067                 goto out;
1068
1069         /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1070          * cannot be preempted by another bpf_timer_cb() on the same cpu.
1071          * Remember the timer this callback is servicing to prevent
1072          * deadlock if callback_fn() calls bpf_timer_cancel() or
1073          * bpf_map_delete_elem() on the same timer.
1074          */
1075         this_cpu_write(hrtimer_running, t);
1076         if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1077                 struct bpf_array *array = container_of(map, struct bpf_array, map);
1078
1079                 /* compute the key */
1080                 idx = ((char *)value - array->value) / array->elem_size;
1081                 key = &idx;
1082         } else { /* hash or lru */
1083                 key = value - round_up(map->key_size, 8);
1084         }
1085
1086         BPF_CAST_CALL(callback_fn)((u64)(long)map, (u64)(long)key,
1087                                    (u64)(long)value, 0, 0);
1088         /* The verifier checked that return value is zero. */
1089
1090         this_cpu_write(hrtimer_running, NULL);
1091 out:
1092         return HRTIMER_NORESTART;
1093 }
1094
1095 BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1096            u64, flags)
1097 {
1098         clockid_t clockid = flags & (MAX_CLOCKS - 1);
1099         struct bpf_hrtimer *t;
1100         int ret = 0;
1101
1102         BUILD_BUG_ON(MAX_CLOCKS != 16);
1103         BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1104         BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1105
1106         if (in_nmi())
1107                 return -EOPNOTSUPP;
1108
1109         if (flags >= MAX_CLOCKS ||
1110             /* similar to timerfd except _ALARM variants are not supported */
1111             (clockid != CLOCK_MONOTONIC &&
1112              clockid != CLOCK_REALTIME &&
1113              clockid != CLOCK_BOOTTIME))
1114                 return -EINVAL;
1115         __bpf_spin_lock_irqsave(&timer->lock);
1116         t = timer->timer;
1117         if (t) {
1118                 ret = -EBUSY;
1119                 goto out;
1120         }
1121         if (!atomic64_read(&map->usercnt)) {
1122                 /* maps with timers must be either held by user space
1123                  * or pinned in bpffs.
1124                  */
1125                 ret = -EPERM;
1126                 goto out;
1127         }
1128         /* allocate hrtimer via map_kmalloc to use memcg accounting */
1129         t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
1130         if (!t) {
1131                 ret = -ENOMEM;
1132                 goto out;
1133         }
1134         t->value = (void *)timer - map->timer_off;
1135         t->map = map;
1136         t->prog = NULL;
1137         rcu_assign_pointer(t->callback_fn, NULL);
1138         hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1139         t->timer.function = bpf_timer_cb;
1140         timer->timer = t;
1141 out:
1142         __bpf_spin_unlock_irqrestore(&timer->lock);
1143         return ret;
1144 }
1145
1146 static const struct bpf_func_proto bpf_timer_init_proto = {
1147         .func           = bpf_timer_init,
1148         .gpl_only       = true,
1149         .ret_type       = RET_INTEGER,
1150         .arg1_type      = ARG_PTR_TO_TIMER,
1151         .arg2_type      = ARG_CONST_MAP_PTR,
1152         .arg3_type      = ARG_ANYTHING,
1153 };
1154
1155 BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1156            struct bpf_prog_aux *, aux)
1157 {
1158         struct bpf_prog *prev, *prog = aux->prog;
1159         struct bpf_hrtimer *t;
1160         int ret = 0;
1161
1162         if (in_nmi())
1163                 return -EOPNOTSUPP;
1164         __bpf_spin_lock_irqsave(&timer->lock);
1165         t = timer->timer;
1166         if (!t) {
1167                 ret = -EINVAL;
1168                 goto out;
1169         }
1170         if (!atomic64_read(&t->map->usercnt)) {
1171                 /* maps with timers must be either held by user space
1172                  * or pinned in bpffs. Otherwise timer might still be
1173                  * running even when bpf prog is detached and user space
1174                  * is gone, since map_release_uref won't ever be called.
1175                  */
1176                 ret = -EPERM;
1177                 goto out;
1178         }
1179         prev = t->prog;
1180         if (prev != prog) {
1181                 /* Bump prog refcnt once. Every bpf_timer_set_callback()
1182                  * can pick different callback_fn-s within the same prog.
1183                  */
1184                 prog = bpf_prog_inc_not_zero(prog);
1185                 if (IS_ERR(prog)) {
1186                         ret = PTR_ERR(prog);
1187                         goto out;
1188                 }
1189                 if (prev)
1190                         /* Drop prev prog refcnt when swapping with new prog */
1191                         bpf_prog_put(prev);
1192                 t->prog = prog;
1193         }
1194         rcu_assign_pointer(t->callback_fn, callback_fn);
1195 out:
1196         __bpf_spin_unlock_irqrestore(&timer->lock);
1197         return ret;
1198 }
1199
1200 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1201         .func           = bpf_timer_set_callback,
1202         .gpl_only       = true,
1203         .ret_type       = RET_INTEGER,
1204         .arg1_type      = ARG_PTR_TO_TIMER,
1205         .arg2_type      = ARG_PTR_TO_FUNC,
1206 };
1207
1208 BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1209 {
1210         struct bpf_hrtimer *t;
1211         int ret = 0;
1212
1213         if (in_nmi())
1214                 return -EOPNOTSUPP;
1215         if (flags)
1216                 return -EINVAL;
1217         __bpf_spin_lock_irqsave(&timer->lock);
1218         t = timer->timer;
1219         if (!t || !t->prog) {
1220                 ret = -EINVAL;
1221                 goto out;
1222         }
1223         hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT);
1224 out:
1225         __bpf_spin_unlock_irqrestore(&timer->lock);
1226         return ret;
1227 }
1228
1229 static const struct bpf_func_proto bpf_timer_start_proto = {
1230         .func           = bpf_timer_start,
1231         .gpl_only       = true,
1232         .ret_type       = RET_INTEGER,
1233         .arg1_type      = ARG_PTR_TO_TIMER,
1234         .arg2_type      = ARG_ANYTHING,
1235         .arg3_type      = ARG_ANYTHING,
1236 };
1237
1238 static void drop_prog_refcnt(struct bpf_hrtimer *t)
1239 {
1240         struct bpf_prog *prog = t->prog;
1241
1242         if (prog) {
1243                 bpf_prog_put(prog);
1244                 t->prog = NULL;
1245                 rcu_assign_pointer(t->callback_fn, NULL);
1246         }
1247 }
1248
1249 BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1250 {
1251         struct bpf_hrtimer *t;
1252         int ret = 0;
1253
1254         if (in_nmi())
1255                 return -EOPNOTSUPP;
1256         __bpf_spin_lock_irqsave(&timer->lock);
1257         t = timer->timer;
1258         if (!t) {
1259                 ret = -EINVAL;
1260                 goto out;
1261         }
1262         if (this_cpu_read(hrtimer_running) == t) {
1263                 /* If bpf callback_fn is trying to bpf_timer_cancel()
1264                  * its own timer the hrtimer_cancel() will deadlock
1265                  * since it waits for callback_fn to finish
1266                  */
1267                 ret = -EDEADLK;
1268                 goto out;
1269         }
1270         drop_prog_refcnt(t);
1271 out:
1272         __bpf_spin_unlock_irqrestore(&timer->lock);
1273         /* Cancel the timer and wait for associated callback to finish
1274          * if it was running.
1275          */
1276         ret = ret ?: hrtimer_cancel(&t->timer);
1277         return ret;
1278 }
1279
1280 static const struct bpf_func_proto bpf_timer_cancel_proto = {
1281         .func           = bpf_timer_cancel,
1282         .gpl_only       = true,
1283         .ret_type       = RET_INTEGER,
1284         .arg1_type      = ARG_PTR_TO_TIMER,
1285 };
1286
1287 /* This function is called by map_delete/update_elem for individual element and
1288  * by ops->map_release_uref when the user space reference to a map reaches zero.
1289  */
1290 void bpf_timer_cancel_and_free(void *val)
1291 {
1292         struct bpf_timer_kern *timer = val;
1293         struct bpf_hrtimer *t;
1294
1295         /* Performance optimization: read timer->timer without lock first. */
1296         if (!READ_ONCE(timer->timer))
1297                 return;
1298
1299         __bpf_spin_lock_irqsave(&timer->lock);
1300         /* re-read it under lock */
1301         t = timer->timer;
1302         if (!t)
1303                 goto out;
1304         drop_prog_refcnt(t);
1305         /* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1306          * this timer, since it won't be initialized.
1307          */
1308         timer->timer = NULL;
1309 out:
1310         __bpf_spin_unlock_irqrestore(&timer->lock);
1311         if (!t)
1312                 return;
1313         /* Cancel the timer and wait for callback to complete if it was running.
1314          * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1315          * right after for both preallocated and non-preallocated maps.
1316          * The timer->timer = NULL was already done and no code path can
1317          * see address 't' anymore.
1318          *
1319          * Check that bpf_map_delete/update_elem() wasn't called from timer
1320          * callback_fn. In such case don't call hrtimer_cancel() (since it will
1321          * deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1322          * return -1). Though callback_fn is still running on this cpu it's
1323          * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1324          * from 't'. The bpf subprog callback_fn won't be able to access 't',
1325          * since timer->timer = NULL was already done. The timer will be
1326          * effectively cancelled because bpf_timer_cb() will return
1327          * HRTIMER_NORESTART.
1328          */
1329         if (this_cpu_read(hrtimer_running) != t)
1330                 hrtimer_cancel(&t->timer);
1331         kfree(t);
1332 }
1333
1334 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1335 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1336 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1337 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1338 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1339 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1340 const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1341
1342 const struct bpf_func_proto *
1343 bpf_base_func_proto(enum bpf_func_id func_id)
1344 {
1345         switch (func_id) {
1346         case BPF_FUNC_map_lookup_elem:
1347                 return &bpf_map_lookup_elem_proto;
1348         case BPF_FUNC_map_update_elem:
1349                 return &bpf_map_update_elem_proto;
1350         case BPF_FUNC_map_delete_elem:
1351                 return &bpf_map_delete_elem_proto;
1352         case BPF_FUNC_map_push_elem:
1353                 return &bpf_map_push_elem_proto;
1354         case BPF_FUNC_map_pop_elem:
1355                 return &bpf_map_pop_elem_proto;
1356         case BPF_FUNC_map_peek_elem:
1357                 return &bpf_map_peek_elem_proto;
1358         case BPF_FUNC_get_prandom_u32:
1359                 return &bpf_get_prandom_u32_proto;
1360         case BPF_FUNC_get_smp_processor_id:
1361                 return &bpf_get_raw_smp_processor_id_proto;
1362         case BPF_FUNC_get_numa_node_id:
1363                 return &bpf_get_numa_node_id_proto;
1364         case BPF_FUNC_tail_call:
1365                 return &bpf_tail_call_proto;
1366         case BPF_FUNC_ktime_get_ns:
1367                 return &bpf_ktime_get_ns_proto;
1368         case BPF_FUNC_ktime_get_boot_ns:
1369                 return &bpf_ktime_get_boot_ns_proto;
1370         case BPF_FUNC_ktime_get_coarse_ns:
1371                 return &bpf_ktime_get_coarse_ns_proto;
1372         case BPF_FUNC_ringbuf_output:
1373                 return &bpf_ringbuf_output_proto;
1374         case BPF_FUNC_ringbuf_reserve:
1375                 return &bpf_ringbuf_reserve_proto;
1376         case BPF_FUNC_ringbuf_submit:
1377                 return &bpf_ringbuf_submit_proto;
1378         case BPF_FUNC_ringbuf_discard:
1379                 return &bpf_ringbuf_discard_proto;
1380         case BPF_FUNC_ringbuf_query:
1381                 return &bpf_ringbuf_query_proto;
1382         case BPF_FUNC_for_each_map_elem:
1383                 return &bpf_for_each_map_elem_proto;
1384         default:
1385                 break;
1386         }
1387
1388         if (!bpf_capable())
1389                 return NULL;
1390
1391         switch (func_id) {
1392         case BPF_FUNC_spin_lock:
1393                 return &bpf_spin_lock_proto;
1394         case BPF_FUNC_spin_unlock:
1395                 return &bpf_spin_unlock_proto;
1396         case BPF_FUNC_jiffies64:
1397                 return &bpf_jiffies64_proto;
1398         case BPF_FUNC_per_cpu_ptr:
1399                 return &bpf_per_cpu_ptr_proto;
1400         case BPF_FUNC_this_cpu_ptr:
1401                 return &bpf_this_cpu_ptr_proto;
1402         case BPF_FUNC_timer_init:
1403                 return &bpf_timer_init_proto;
1404         case BPF_FUNC_timer_set_callback:
1405                 return &bpf_timer_set_callback_proto;
1406         case BPF_FUNC_timer_start:
1407                 return &bpf_timer_start_proto;
1408         case BPF_FUNC_timer_cancel:
1409                 return &bpf_timer_cancel_proto;
1410         default:
1411                 break;
1412         }
1413
1414         if (!perfmon_capable())
1415                 return NULL;
1416
1417         switch (func_id) {
1418         case BPF_FUNC_trace_printk:
1419                 return bpf_get_trace_printk_proto();
1420         case BPF_FUNC_get_current_task:
1421                 return &bpf_get_current_task_proto;
1422         case BPF_FUNC_get_current_task_btf:
1423                 return &bpf_get_current_task_btf_proto;
1424         case BPF_FUNC_probe_read_user:
1425                 return &bpf_probe_read_user_proto;
1426         case BPF_FUNC_probe_read_kernel:
1427                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1428                        NULL : &bpf_probe_read_kernel_proto;
1429         case BPF_FUNC_probe_read_user_str:
1430                 return &bpf_probe_read_user_str_proto;
1431         case BPF_FUNC_probe_read_kernel_str:
1432                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1433                        NULL : &bpf_probe_read_kernel_str_proto;
1434         case BPF_FUNC_snprintf_btf:
1435                 return &bpf_snprintf_btf_proto;
1436         case BPF_FUNC_snprintf:
1437                 return &bpf_snprintf_proto;
1438         case BPF_FUNC_task_pt_regs:
1439                 return &bpf_task_pt_regs_proto;
1440         default:
1441                 return NULL;
1442         }
1443 }