kernel/trace/bpf_trace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   3  * Copyright (c) 2016 Facebook
   4  */
   5 #include <linux/kernel.h>
   6 #include <linux/types.h>
   7 #include <linux/slab.h>
   8 #include <linux/bpf.h>
   9 #include <linux/bpf_perf_event.h>
  10 #include <linux/btf.h>
  11 #include <linux/filter.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/ctype.h>
  14 #include <linux/kprobes.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/btf_ids.h>
  19 #include <linux/bpf_lsm.h>
  20
  21 #include <net/bpf_sk_storage.h>
  22
  23 #include <uapi/linux/bpf.h>
  24 #include <uapi/linux/btf.h>
  25
  26 #include <asm/tlb.h>
  27
  28 #include "trace_probe.h"
  29 #include "trace.h"
  30
  31 #define CREATE_TRACE_POINTS
  32 #include "bpf_trace.h"
  33
  34 #define bpf_event_rcu_dereference(p)                                    \
  35         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  36
  37 #ifdef CONFIG_MODULES
  38 struct bpf_trace_module {
  39         struct module *module;
  40         struct list_head list;
  41 };
  42
  43 static LIST_HEAD(bpf_trace_modules);
  44 static DEFINE_MUTEX(bpf_module_mutex);
  45
  46 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  47 {
  48         struct bpf_raw_event_map *btp, *ret = NULL;
  49         struct bpf_trace_module *btm;
  50         unsigned int i;
  51
  52         mutex_lock(&bpf_module_mutex);
  53         list_for_each_entry(btm, &bpf_trace_modules, list) {
  54                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  55                         btp = &btm->module->bpf_raw_events[i];
  56                         if (!strcmp(btp->tp->name, name)) {
  57                                 if (try_module_get(btm->module))
  58                                         ret = btp;
  59                                 goto out;
  60                         }
  61                 }
  62         }
  63 out:
  64         mutex_unlock(&bpf_module_mutex);
  65         return ret;
  66 }
  67 #else
  68 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  69 {
  70         return NULL;
  71 }
  72 #endif /* CONFIG_MODULES */
  73
  74 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  75 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  76
  77 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
  78                                   u64 flags, const struct btf **btf,
  79                                   s32 *btf_id);
  80
  81 /**
  82  * trace_call_bpf - invoke BPF program
  83  * @call: tracepoint event
  84  * @ctx: opaque context pointer
  85  *
  86  * kprobe handlers execute BPF programs via this helper.
  87  * Can be used from static tracepoints in the future.
  88  *
  89  * Return: BPF programs always return an integer which is interpreted by
  90  * kprobe handler as:
  91  * 0 - return from kprobe (event is filtered out)
  92  * 1 - store kprobe event into ring buffer
  93  * Other values are reserved and currently alias to 1
  94  */
  95 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
  96 {
  97         unsigned int ret;
  98
  99         cant_sleep();
 100
 101         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 102                 /*
 103                  * since some bpf program is already running on this cpu,
 104                  * don't call into another bpf program (same or different)
 105                  * and don't send kprobe event into ring-buffer,
 106                  * so return zero here
 107                  */
 108                 ret = 0;
 109                 goto out;
 110         }
 111
 112         /*
 113          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 114          * to all call sites, we did a bpf_prog_array_valid() there to check
 115          * whether call->prog_array is empty or not, which is
 116          * a heuristic to speed up execution.
 117          *
 118          * If bpf_prog_array_valid() fetched prog_array was
 119          * non-NULL, we go into trace_call_bpf() and do the actual
 120          * proper rcu_dereference() under RCU lock.
 121          * If it turns out that prog_array is NULL then, we bail out.
 122          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 123          * was NULL, you'll skip the prog_array with the risk of missing
 124          * out of events when it was updated in between this and the
 125          * rcu_dereference() which is accepted risk.
 126          */
 127         ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
 128
 129  out:
 130         __this_cpu_dec(bpf_prog_active);
 131
 132         return ret;
 133 }
 134
 135 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 136 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 137 {
 138         regs_set_return_value(regs, rc);
 139         override_function_with_return(regs);
 140         return 0;
 141 }
 142
 143 static const struct bpf_func_proto bpf_override_return_proto = {
 144         .func           = bpf_override_return,
 145         .gpl_only       = true,
 146         .ret_type       = RET_INTEGER,
 147         .arg1_type      = ARG_PTR_TO_CTX,
 148         .arg2_type      = ARG_ANYTHING,
 149 };
 150 #endif
 151
 152 static __always_inline int
 153 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 154 {
 155         int ret;
 156
 157         ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 158         if (unlikely(ret < 0))
 159                 memset(dst, 0, size);
 160         return ret;
 161 }
 162
 163 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 164            const void __user *, unsafe_ptr)
 165 {
 166         return bpf_probe_read_user_common(dst, size, unsafe_ptr);
 167 }
 168
 169 const struct bpf_func_proto bpf_probe_read_user_proto = {
 170         .func           = bpf_probe_read_user,
 171         .gpl_only       = true,
 172         .ret_type       = RET_INTEGER,
 173         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 174         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 175         .arg3_type      = ARG_ANYTHING,
 176 };
 177
 178 static __always_inline int
 179 bpf_probe_read_user_str_common(void *dst, u32 size,
 180                                const void __user *unsafe_ptr)
 181 {
 182         int ret;
 183
 184         /*
 185          * NB: We rely on strncpy_from_user() not copying junk past the NUL
 186          * terminator into `dst`.
 187          *
 188          * strncpy_from_user() does long-sized strides in the fast path. If the
 189          * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
 190          * then there could be junk after the NUL in `dst`. If user takes `dst`
 191          * and keys a hash map with it, then semantically identical strings can
 192          * occupy multiple entries in the map.
 193          */
 194         ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 195         if (unlikely(ret < 0))
 196                 memset(dst, 0, size);
 197         return ret;
 198 }
 199
 200 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 201            const void __user *, unsafe_ptr)
 202 {
 203         return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
 204 }
 205
 206 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 207         .func           = bpf_probe_read_user_str,
 208         .gpl_only       = true,
 209         .ret_type       = RET_INTEGER,
 210         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 211         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 212         .arg3_type      = ARG_ANYTHING,
 213 };
 214
 215 static __always_inline int
 216 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
 217 {
 218         int ret = security_locked_down(LOCKDOWN_BPF_READ);
 219
 220         if (unlikely(ret < 0))
 221                 goto fail;
 222         ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
 223         if (unlikely(ret < 0))
 224                 goto fail;
 225         return ret;
 226 fail:
 227         memset(dst, 0, size);
 228         return ret;
 229 }
 230
 231 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 232            const void *, unsafe_ptr)
 233 {
 234         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 235 }
 236
 237 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 238         .func           = bpf_probe_read_kernel,
 239         .gpl_only       = true,
 240         .ret_type       = RET_INTEGER,
 241         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 242         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 243         .arg3_type      = ARG_ANYTHING,
 244 };
 245
 246 static __always_inline int
 247 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
 248 {
 249         int ret = security_locked_down(LOCKDOWN_BPF_READ);
 250
 251         if (unlikely(ret < 0))
 252                 goto fail;
 253
 254         /*
 255          * The strncpy_from_kernel_nofault() call will likely not fill the
 256          * entire buffer, but that's okay in this circumstance as we're probing
 257          * arbitrary memory anyway similar to bpf_probe_read_*() and might
 258          * as well probe the stack. Thus, memory is explicitly cleared
 259          * only in error case, so that improper users ignoring return
 260          * code altogether don't copy garbage; otherwise length of string
 261          * is returned that can be used for bpf_perf_event_output() et al.
 262          */
 263         ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
 264         if (unlikely(ret < 0))
 265                 goto fail;
 266
 267         return ret;
 268 fail:
 269         memset(dst, 0, size);
 270         return ret;
 271 }
 272
 273 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 274            const void *, unsafe_ptr)
 275 {
 276         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 277 }
 278
 279 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 280         .func           = bpf_probe_read_kernel_str,
 281         .gpl_only       = true,
 282         .ret_type       = RET_INTEGER,
 283         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 284         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 285         .arg3_type      = ARG_ANYTHING,
 286 };
 287
 288 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 289 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 290            const void *, unsafe_ptr)
 291 {
 292         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 293                 return bpf_probe_read_user_common(dst, size,
 294                                 (__force void __user *)unsafe_ptr);
 295         }
 296         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 297 }
 298
 299 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 300         .func           = bpf_probe_read_compat,
 301         .gpl_only       = true,
 302         .ret_type       = RET_INTEGER,
 303         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 304         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 305         .arg3_type      = ARG_ANYTHING,
 306 };
 307
 308 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 309            const void *, unsafe_ptr)
 310 {
 311         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 312                 return bpf_probe_read_user_str_common(dst, size,
 313                                 (__force void __user *)unsafe_ptr);
 314         }
 315         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 316 }
 317
 318 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 319         .func           = bpf_probe_read_compat_str,
 320         .gpl_only       = true,
 321         .ret_type       = RET_INTEGER,
 322         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 323         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 324         .arg3_type      = ARG_ANYTHING,
 325 };
 326 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 327
 328 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 329            u32, size)
 330 {
 331         /*
 332          * Ensure we're in user context which is safe for the helper to
 333          * run. This helper has no business in a kthread.
 334          *
 335          * access_ok() should prevent writing to non-user memory, but in
 336          * some situations (nommu, temporary switch, etc) access_ok() does
 337          * not provide enough validation, hence the check on KERNEL_DS.
 338          *
 339          * nmi_uaccess_okay() ensures the probe is not run in an interim
 340          * state, when the task or mm are switched. This is specifically
 341          * required to prevent the use of temporary mm.
 342          */
 343
 344         if (unlikely(in_interrupt() ||
 345                      current->flags & (PF_KTHREAD | PF_EXITING)))
 346                 return -EPERM;
 347         if (unlikely(uaccess_kernel()))
 348                 return -EPERM;
 349         if (unlikely(!nmi_uaccess_okay()))
 350                 return -EPERM;
 351
 352         return copy_to_user_nofault(unsafe_ptr, src, size);
 353 }
 354
 355 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 356         .func           = bpf_probe_write_user,
 357         .gpl_only       = true,
 358         .ret_type       = RET_INTEGER,
 359         .arg1_type      = ARG_ANYTHING,
 360         .arg2_type      = ARG_PTR_TO_MEM,
 361         .arg3_type      = ARG_CONST_SIZE,
 362 };
 363
 364 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 365 {
 366         if (!capable(CAP_SYS_ADMIN))
 367                 return NULL;
 368
 369         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 370                             current->comm, task_pid_nr(current));
 371
 372         return &bpf_probe_write_user_proto;
 373 }
 374
 375 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
 376
 377 #define MAX_TRACE_PRINTK_VARARGS        3
 378 #define BPF_TRACE_PRINTK_SIZE           1024
 379
 380 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 381            u64, arg2, u64, arg3)
 382 {
 383         u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
 384         u32 *bin_args;
 385         static char buf[BPF_TRACE_PRINTK_SIZE];
 386         unsigned long flags;
 387         int ret;
 388
 389         ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
 390                                   MAX_TRACE_PRINTK_VARARGS);
 391         if (ret < 0)
 392                 return ret;
 393
 394         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 395         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 396
 397         trace_bpf_trace_printk(buf);
 398         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 399
 400         bpf_bprintf_cleanup();
 401
 402         return ret;
 403 }
 404
 405 static const struct bpf_func_proto bpf_trace_printk_proto = {
 406         .func           = bpf_trace_printk,
 407         .gpl_only       = true,
 408         .ret_type       = RET_INTEGER,
 409         .arg1_type      = ARG_PTR_TO_MEM,
 410         .arg2_type      = ARG_CONST_SIZE,
 411 };
 412
 413 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 414 {
 415         /*
 416          * This program might be calling bpf_trace_printk,
 417          * so enable the associated bpf_trace/bpf_trace_printk event.
 418          * Repeat this each time as it is possible a user has
 419          * disabled bpf_trace_printk events.  By loading a program
 420          * calling bpf_trace_printk() however the user has expressed
 421          * the intent to see such events.
 422          */
 423         if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 424                 pr_warn_ratelimited("could not enable bpf_trace_printk events");
 425
 426         return &bpf_trace_printk_proto;
 427 }
 428
 429 #define MAX_SEQ_PRINTF_VARARGS          12
 430
 431 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 432            const void *, data, u32, data_len)
 433 {
 434         int err, num_args;
 435         u32 *bin_args;
 436
 437         if (data_len & 7 || data_len > MAX_SEQ_PRINTF_VARARGS * 8 ||
 438             (data_len && !data))
 439                 return -EINVAL;
 440         num_args = data_len / 8;
 441
 442         err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 443         if (err < 0)
 444                 return err;
 445
 446         seq_bprintf(m, fmt, bin_args);
 447
 448         bpf_bprintf_cleanup();
 449
 450         return seq_has_overflowed(m) ? -EOVERFLOW : 0;
 451 }
 452
 453 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
 454
 455 static const struct bpf_func_proto bpf_seq_printf_proto = {
 456         .func           = bpf_seq_printf,
 457         .gpl_only       = true,
 458         .ret_type       = RET_INTEGER,
 459         .arg1_type      = ARG_PTR_TO_BTF_ID,
 460         .arg1_btf_id    = &btf_seq_file_ids[0],
 461         .arg2_type      = ARG_PTR_TO_MEM,
 462         .arg3_type      = ARG_CONST_SIZE,
 463         .arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
 464         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 465 };
 466
 467 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
 468 {
 469         return seq_write(m, data, len) ? -EOVERFLOW : 0;
 470 }
 471
 472 static const struct bpf_func_proto bpf_seq_write_proto = {
 473         .func           = bpf_seq_write,
 474         .gpl_only       = true,
 475         .ret_type       = RET_INTEGER,
 476         .arg1_type      = ARG_PTR_TO_BTF_ID,
 477         .arg1_btf_id    = &btf_seq_file_ids[0],
 478         .arg2_type      = ARG_PTR_TO_MEM,
 479         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 480 };
 481
 482 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
 483            u32, btf_ptr_size, u64, flags)
 484 {
 485         const struct btf *btf;
 486         s32 btf_id;
 487         int ret;
 488
 489         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 490         if (ret)
 491                 return ret;
 492
 493         return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
 494 }
 495
 496 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
 497         .func           = bpf_seq_printf_btf,
 498         .gpl_only       = true,
 499         .ret_type       = RET_INTEGER,
 500         .arg1_type      = ARG_PTR_TO_BTF_ID,
 501         .arg1_btf_id    = &btf_seq_file_ids[0],
 502         .arg2_type      = ARG_PTR_TO_MEM,
 503         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 504         .arg4_type      = ARG_ANYTHING,
 505 };
 506
 507 static __always_inline int
 508 get_map_perf_counter(struct bpf_map *map, u64 flags,
 509                      u64 *value, u64 *enabled, u64 *running)
 510 {
 511         struct bpf_array *array = container_of(map, struct bpf_array, map);
 512         unsigned int cpu = smp_processor_id();
 513         u64 index = flags & BPF_F_INDEX_MASK;
 514         struct bpf_event_entry *ee;
 515
 516         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 517                 return -EINVAL;
 518         if (index == BPF_F_CURRENT_CPU)
 519                 index = cpu;
 520         if (unlikely(index >= array->map.max_entries))
 521                 return -E2BIG;
 522
 523         ee = READ_ONCE(array->ptrs[index]);
 524         if (!ee)
 525                 return -ENOENT;
 526
 527         return perf_event_read_local(ee->event, value, enabled, running);
 528 }
 529
 530 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 531 {
 532         u64 value = 0;
 533         int err;
 534
 535         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 536         /*
 537          * this api is ugly since we miss [-22..-2] range of valid
 538          * counter values, but that's uapi
 539          */
 540         if (err)
 541                 return err;
 542         return value;
 543 }
 544
 545 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 546         .func           = bpf_perf_event_read,
 547         .gpl_only       = true,
 548         .ret_type       = RET_INTEGER,
 549         .arg1_type      = ARG_CONST_MAP_PTR,
 550         .arg2_type      = ARG_ANYTHING,
 551 };
 552
 553 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 554            struct bpf_perf_event_value *, buf, u32, size)
 555 {
 556         int err = -EINVAL;
 557
 558         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 559                 goto clear;
 560         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 561                                    &buf->running);
 562         if (unlikely(err))
 563                 goto clear;
 564         return 0;
 565 clear:
 566         memset(buf, 0, size);
 567         return err;
 568 }
 569
 570 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 571         .func           = bpf_perf_event_read_value,
 572         .gpl_only       = true,
 573         .ret_type       = RET_INTEGER,
 574         .arg1_type      = ARG_CONST_MAP_PTR,
 575         .arg2_type      = ARG_ANYTHING,
 576         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 577         .arg4_type      = ARG_CONST_SIZE,
 578 };
 579
 580 static __always_inline u64
 581 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 582                         u64 flags, struct perf_sample_data *sd)
 583 {
 584         struct bpf_array *array = container_of(map, struct bpf_array, map);
 585         unsigned int cpu = smp_processor_id();
 586         u64 index = flags & BPF_F_INDEX_MASK;
 587         struct bpf_event_entry *ee;
 588         struct perf_event *event;
 589
 590         if (index == BPF_F_CURRENT_CPU)
 591                 index = cpu;
 592         if (unlikely(index >= array->map.max_entries))
 593                 return -E2BIG;
 594
 595         ee = READ_ONCE(array->ptrs[index]);
 596         if (!ee)
 597                 return -ENOENT;
 598
 599         event = ee->event;
 600         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 601                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 602                 return -EINVAL;
 603
 604         if (unlikely(event->oncpu != cpu))
 605                 return -EOPNOTSUPP;
 606
 607         return perf_event_output(event, sd, regs);
 608 }
 609
 610 /*
 611  * Support executing tracepoints in normal, irq, and nmi context that each call
 612  * bpf_perf_event_output
 613  */
 614 struct bpf_trace_sample_data {
 615         struct perf_sample_data sds[3];
 616 };
 617
 618 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 619 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 620 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 621            u64, flags, void *, data, u64, size)
 622 {
 623         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 624         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 625         struct perf_raw_record raw = {
 626                 .frag = {
 627                         .size = size,
 628                         .data = data,
 629                 },
 630         };
 631         struct perf_sample_data *sd;
 632         int err;
 633
 634         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 635                 err = -EBUSY;
 636                 goto out;
 637         }
 638
 639         sd = &sds->sds[nest_level - 1];
 640
 641         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 642                 err = -EINVAL;
 643                 goto out;
 644         }
 645
 646         perf_sample_data_init(sd, 0, 0);
 647         sd->raw = &raw;
 648
 649         err = __bpf_perf_event_output(regs, map, flags, sd);
 650
 651 out:
 652         this_cpu_dec(bpf_trace_nest_level);
 653         return err;
 654 }
 655
 656 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 657         .func           = bpf_perf_event_output,
 658         .gpl_only       = true,
 659         .ret_type       = RET_INTEGER,
 660         .arg1_type      = ARG_PTR_TO_CTX,
 661         .arg2_type      = ARG_CONST_MAP_PTR,
 662         .arg3_type      = ARG_ANYTHING,
 663         .arg4_type      = ARG_PTR_TO_MEM,
 664         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 665 };
 666
 667 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 668 struct bpf_nested_pt_regs {
 669         struct pt_regs regs[3];
 670 };
 671 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 672 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 673
 674 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 675                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 676 {
 677         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 678         struct perf_raw_frag frag = {
 679                 .copy           = ctx_copy,
 680                 .size           = ctx_size,
 681                 .data           = ctx,
 682         };
 683         struct perf_raw_record raw = {
 684                 .frag = {
 685                         {
 686                                 .next   = ctx_size ? &frag : NULL,
 687                         },
 688                         .size   = meta_size,
 689                         .data   = meta,
 690                 },
 691         };
 692         struct perf_sample_data *sd;
 693         struct pt_regs *regs;
 694         u64 ret;
 695
 696         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 697                 ret = -EBUSY;
 698                 goto out;
 699         }
 700         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 701         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 702
 703         perf_fetch_caller_regs(regs);
 704         perf_sample_data_init(sd, 0, 0);
 705         sd->raw = &raw;
 706
 707         ret = __bpf_perf_event_output(regs, map, flags, sd);
 708 out:
 709         this_cpu_dec(bpf_event_output_nest_level);
 710         return ret;
 711 }
 712
 713 BPF_CALL_0(bpf_get_current_task)
 714 {
 715         return (long) current;
 716 }
 717
 718 const struct bpf_func_proto bpf_get_current_task_proto = {
 719         .func           = bpf_get_current_task,
 720         .gpl_only       = true,
 721         .ret_type       = RET_INTEGER,
 722 };
 723
 724 BPF_CALL_0(bpf_get_current_task_btf)
 725 {
 726         return (unsigned long) current;
 727 }
 728
 729 BTF_ID_LIST_SINGLE(bpf_get_current_btf_ids, struct, task_struct)
 730
 731 static const struct bpf_func_proto bpf_get_current_task_btf_proto = {
 732         .func           = bpf_get_current_task_btf,
 733         .gpl_only       = true,
 734         .ret_type       = RET_PTR_TO_BTF_ID,
 735         .ret_btf_id     = &bpf_get_current_btf_ids[0],
 736 };
 737
 738 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 739 {
 740         struct bpf_array *array = container_of(map, struct bpf_array, map);
 741         struct cgroup *cgrp;
 742
 743         if (unlikely(idx >= array->map.max_entries))
 744                 return -E2BIG;
 745
 746         cgrp = READ_ONCE(array->ptrs[idx]);
 747         if (unlikely(!cgrp))
 748                 return -EAGAIN;
 749
 750         return task_under_cgroup_hierarchy(current, cgrp);
 751 }
 752
 753 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 754         .func           = bpf_current_task_under_cgroup,
 755         .gpl_only       = false,
 756         .ret_type       = RET_INTEGER,
 757         .arg1_type      = ARG_CONST_MAP_PTR,
 758         .arg2_type      = ARG_ANYTHING,
 759 };
 760
 761 struct send_signal_irq_work {
 762         struct irq_work irq_work;
 763         struct task_struct *task;
 764         u32 sig;
 765         enum pid_type type;
 766 };
 767
 768 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 769
 770 static void do_bpf_send_signal(struct irq_work *entry)
 771 {
 772         struct send_signal_irq_work *work;
 773
 774         work = container_of(entry, struct send_signal_irq_work, irq_work);
 775         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 776 }
 777
 778 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 779 {
 780         struct send_signal_irq_work *work = NULL;
 781
 782         /* Similar to bpf_probe_write_user, task needs to be
 783          * in a sound condition and kernel memory access be
 784          * permitted in order to send signal to the current
 785          * task.
 786          */
 787         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 788                 return -EPERM;
 789         if (unlikely(uaccess_kernel()))
 790                 return -EPERM;
 791         if (unlikely(!nmi_uaccess_okay()))
 792                 return -EPERM;
 793
 794         if (irqs_disabled()) {
 795                 /* Do an early check on signal validity. Otherwise,
 796                  * the error is lost in deferred irq_work.
 797                  */
 798                 if (unlikely(!valid_signal(sig)))
 799                         return -EINVAL;
 800
 801                 work = this_cpu_ptr(&send_signal_work);
 802                 if (irq_work_is_busy(&work->irq_work))
 803                         return -EBUSY;
 804
 805                 /* Add the current task, which is the target of sending signal,
 806                  * to the irq_work. The current task may change when queued
 807                  * irq works get executed.
 808                  */
 809                 work->task = current;
 810                 work->sig = sig;
 811                 work->type = type;
 812                 irq_work_queue(&work->irq_work);
 813                 return 0;
 814         }
 815
 816         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 817 }
 818
 819 BPF_CALL_1(bpf_send_signal, u32, sig)
 820 {
 821         return bpf_send_signal_common(sig, PIDTYPE_TGID);
 822 }
 823
 824 static const struct bpf_func_proto bpf_send_signal_proto = {
 825         .func           = bpf_send_signal,
 826         .gpl_only       = false,
 827         .ret_type       = RET_INTEGER,
 828         .arg1_type      = ARG_ANYTHING,
 829 };
 830
 831 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 832 {
 833         return bpf_send_signal_common(sig, PIDTYPE_PID);
 834 }
 835
 836 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 837         .func           = bpf_send_signal_thread,
 838         .gpl_only       = false,
 839         .ret_type       = RET_INTEGER,
 840         .arg1_type      = ARG_ANYTHING,
 841 };
 842
 843 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 844 {
 845         long len;
 846         char *p;
 847
 848         if (!sz)
 849                 return 0;
 850
 851         p = d_path(path, buf, sz);
 852         if (IS_ERR(p)) {
 853                 len = PTR_ERR(p);
 854         } else {
 855                 len = buf + sz - p;
 856                 memmove(buf, p, len);
 857         }
 858
 859         return len;
 860 }
 861
 862 BTF_SET_START(btf_allowlist_d_path)
 863 #ifdef CONFIG_SECURITY
 864 BTF_ID(func, security_file_permission)
 865 BTF_ID(func, security_inode_getattr)
 866 BTF_ID(func, security_file_open)
 867 #endif
 868 #ifdef CONFIG_SECURITY_PATH
 869 BTF_ID(func, security_path_truncate)
 870 #endif
 871 BTF_ID(func, vfs_truncate)
 872 BTF_ID(func, vfs_fallocate)
 873 BTF_ID(func, dentry_open)
 874 BTF_ID(func, vfs_getattr)
 875 BTF_ID(func, filp_close)
 876 BTF_SET_END(btf_allowlist_d_path)
 877
 878 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 879 {
 880         if (prog->type == BPF_PROG_TYPE_TRACING &&
 881             prog->expected_attach_type == BPF_TRACE_ITER)
 882                 return true;
 883
 884         if (prog->type == BPF_PROG_TYPE_LSM)
 885                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 886
 887         return btf_id_set_contains(&btf_allowlist_d_path,
 888                                    prog->aux->attach_btf_id);
 889 }
 890
 891 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 892
 893 static const struct bpf_func_proto bpf_d_path_proto = {
 894         .func           = bpf_d_path,
 895         .gpl_only       = false,
 896         .ret_type       = RET_INTEGER,
 897         .arg1_type      = ARG_PTR_TO_BTF_ID,
 898         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
 899         .arg2_type      = ARG_PTR_TO_MEM,
 900         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 901         .allowed        = bpf_d_path_allowed,
 902 };
 903
 904 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
 905                          BTF_F_PTR_RAW | BTF_F_ZERO)
 906
 907 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 908                                   u64 flags, const struct btf **btf,
 909                                   s32 *btf_id)
 910 {
 911         const struct btf_type *t;
 912
 913         if (unlikely(flags & ~(BTF_F_ALL)))
 914                 return -EINVAL;
 915
 916         if (btf_ptr_size != sizeof(struct btf_ptr))
 917                 return -EINVAL;
 918
 919         *btf = bpf_get_btf_vmlinux();
 920
 921         if (IS_ERR_OR_NULL(*btf))
 922                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 923
 924         if (ptr->type_id > 0)
 925                 *btf_id = ptr->type_id;
 926         else
 927                 return -EINVAL;
 928
 929         if (*btf_id > 0)
 930                 t = btf_type_by_id(*btf, *btf_id);
 931         if (*btf_id <= 0 || !t)
 932                 return -ENOENT;
 933
 934         return 0;
 935 }
 936
 937 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
 938            u32, btf_ptr_size, u64, flags)
 939 {
 940         const struct btf *btf;
 941         s32 btf_id;
 942         int ret;
 943
 944         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 945         if (ret)
 946                 return ret;
 947
 948         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
 949                                       flags);
 950 }
 951
 952 const struct bpf_func_proto bpf_snprintf_btf_proto = {
 953         .func           = bpf_snprintf_btf,
 954         .gpl_only       = false,
 955         .ret_type       = RET_INTEGER,
 956         .arg1_type      = ARG_PTR_TO_MEM,
 957         .arg2_type      = ARG_CONST_SIZE,
 958         .arg3_type      = ARG_PTR_TO_MEM,
 959         .arg4_type      = ARG_CONST_SIZE,
 960         .arg5_type      = ARG_ANYTHING,
 961 };
 962
 963 const struct bpf_func_proto *
 964 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 965 {
 966         switch (func_id) {
 967         case BPF_FUNC_map_lookup_elem:
 968                 return &bpf_map_lookup_elem_proto;
 969         case BPF_FUNC_map_update_elem:
 970                 return &bpf_map_update_elem_proto;
 971         case BPF_FUNC_map_delete_elem:
 972                 return &bpf_map_delete_elem_proto;
 973         case BPF_FUNC_map_push_elem:
 974                 return &bpf_map_push_elem_proto;
 975         case BPF_FUNC_map_pop_elem:
 976                 return &bpf_map_pop_elem_proto;
 977         case BPF_FUNC_map_peek_elem:
 978                 return &bpf_map_peek_elem_proto;
 979         case BPF_FUNC_ktime_get_ns:
 980                 return &bpf_ktime_get_ns_proto;
 981         case BPF_FUNC_ktime_get_boot_ns:
 982                 return &bpf_ktime_get_boot_ns_proto;
 983         case BPF_FUNC_ktime_get_coarse_ns:
 984                 return &bpf_ktime_get_coarse_ns_proto;
 985         case BPF_FUNC_tail_call:
 986                 return &bpf_tail_call_proto;
 987         case BPF_FUNC_get_current_pid_tgid:
 988                 return &bpf_get_current_pid_tgid_proto;
 989         case BPF_FUNC_get_current_task:
 990                 return &bpf_get_current_task_proto;
 991         case BPF_FUNC_get_current_task_btf:
 992                 return &bpf_get_current_task_btf_proto;
 993         case BPF_FUNC_get_current_uid_gid:
 994                 return &bpf_get_current_uid_gid_proto;
 995         case BPF_FUNC_get_current_comm:
 996                 return &bpf_get_current_comm_proto;
 997         case BPF_FUNC_trace_printk:
 998                 return bpf_get_trace_printk_proto();
 999         case BPF_FUNC_get_smp_processor_id:
1000                 return &bpf_get_smp_processor_id_proto;
1001         case BPF_FUNC_get_numa_node_id:
1002                 return &bpf_get_numa_node_id_proto;
1003         case BPF_FUNC_perf_event_read:
1004                 return &bpf_perf_event_read_proto;
1005         case BPF_FUNC_probe_write_user:
1006                 return bpf_get_probe_write_proto();
1007         case BPF_FUNC_current_task_under_cgroup:
1008                 return &bpf_current_task_under_cgroup_proto;
1009         case BPF_FUNC_get_prandom_u32:
1010                 return &bpf_get_prandom_u32_proto;
1011         case BPF_FUNC_probe_read_user:
1012                 return &bpf_probe_read_user_proto;
1013         case BPF_FUNC_probe_read_kernel:
1014                 return &bpf_probe_read_kernel_proto;
1015         case BPF_FUNC_probe_read_user_str:
1016                 return &bpf_probe_read_user_str_proto;
1017         case BPF_FUNC_probe_read_kernel_str:
1018                 return &bpf_probe_read_kernel_str_proto;
1019 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1020         case BPF_FUNC_probe_read:
1021                 return &bpf_probe_read_compat_proto;
1022         case BPF_FUNC_probe_read_str:
1023                 return &bpf_probe_read_compat_str_proto;
1024 #endif
1025 #ifdef CONFIG_CGROUPS
1026         case BPF_FUNC_get_current_cgroup_id:
1027                 return &bpf_get_current_cgroup_id_proto;
1028 #endif
1029         case BPF_FUNC_send_signal:
1030                 return &bpf_send_signal_proto;
1031         case BPF_FUNC_send_signal_thread:
1032                 return &bpf_send_signal_thread_proto;
1033         case BPF_FUNC_perf_event_read_value:
1034                 return &bpf_perf_event_read_value_proto;
1035         case BPF_FUNC_get_ns_current_pid_tgid:
1036                 return &bpf_get_ns_current_pid_tgid_proto;
1037         case BPF_FUNC_ringbuf_output:
1038                 return &bpf_ringbuf_output_proto;
1039         case BPF_FUNC_ringbuf_reserve:
1040                 return &bpf_ringbuf_reserve_proto;
1041         case BPF_FUNC_ringbuf_submit:
1042                 return &bpf_ringbuf_submit_proto;
1043         case BPF_FUNC_ringbuf_discard:
1044                 return &bpf_ringbuf_discard_proto;
1045         case BPF_FUNC_ringbuf_query:
1046                 return &bpf_ringbuf_query_proto;
1047         case BPF_FUNC_jiffies64:
1048                 return &bpf_jiffies64_proto;
1049         case BPF_FUNC_get_task_stack:
1050                 return &bpf_get_task_stack_proto;
1051         case BPF_FUNC_copy_from_user:
1052                 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1053         case BPF_FUNC_snprintf_btf:
1054                 return &bpf_snprintf_btf_proto;
1055         case BPF_FUNC_per_cpu_ptr:
1056                 return &bpf_per_cpu_ptr_proto;
1057         case BPF_FUNC_this_cpu_ptr:
1058                 return &bpf_this_cpu_ptr_proto;
1059         case BPF_FUNC_task_storage_get:
1060                 return &bpf_task_storage_get_proto;
1061         case BPF_FUNC_task_storage_delete:
1062                 return &bpf_task_storage_delete_proto;
1063         case BPF_FUNC_for_each_map_elem:
1064                 return &bpf_for_each_map_elem_proto;
1065         case BPF_FUNC_snprintf:
1066                 return &bpf_snprintf_proto;
1067         default:
1068                 return NULL;
1069         }
1070 }
1071
1072 static const struct bpf_func_proto *
1073 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1074 {
1075         switch (func_id) {
1076         case BPF_FUNC_perf_event_output:
1077                 return &bpf_perf_event_output_proto;
1078         case BPF_FUNC_get_stackid:
1079                 return &bpf_get_stackid_proto;
1080         case BPF_FUNC_get_stack:
1081                 return &bpf_get_stack_proto;
1082 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1083         case BPF_FUNC_override_return:
1084                 return &bpf_override_return_proto;
1085 #endif
1086         default:
1087                 return bpf_tracing_func_proto(func_id, prog);
1088         }
1089 }
1090
1091 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1092 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1093                                         const struct bpf_prog *prog,
1094                                         struct bpf_insn_access_aux *info)
1095 {
1096         if (off < 0 || off >= sizeof(struct pt_regs))
1097                 return false;
1098         if (type != BPF_READ)
1099                 return false;
1100         if (off % size != 0)
1101                 return false;
1102         /*
1103          * Assertion for 32 bit to make sure last 8 byte access
1104          * (BPF_DW) to the last 4 byte member is disallowed.
1105          */
1106         if (off + size > sizeof(struct pt_regs))
1107                 return false;
1108
1109         return true;
1110 }
1111
1112 const struct bpf_verifier_ops kprobe_verifier_ops = {
1113         .get_func_proto  = kprobe_prog_func_proto,
1114         .is_valid_access = kprobe_prog_is_valid_access,
1115 };
1116
1117 const struct bpf_prog_ops kprobe_prog_ops = {
1118 };
1119
1120 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1121            u64, flags, void *, data, u64, size)
1122 {
1123         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1124
1125         /*
1126          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1127          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1128          * from there and call the same bpf_perf_event_output() helper inline.
1129          */
1130         return ____bpf_perf_event_output(regs, map, flags, data, size);
1131 }
1132
1133 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1134         .func           = bpf_perf_event_output_tp,
1135         .gpl_only       = true,
1136         .ret_type       = RET_INTEGER,
1137         .arg1_type      = ARG_PTR_TO_CTX,
1138         .arg2_type      = ARG_CONST_MAP_PTR,
1139         .arg3_type      = ARG_ANYTHING,
1140         .arg4_type      = ARG_PTR_TO_MEM,
1141         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1142 };
1143
1144 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1145            u64, flags)
1146 {
1147         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1148
1149         /*
1150          * Same comment as in bpf_perf_event_output_tp(), only that this time
1151          * the other helper's function body cannot be inlined due to being
1152          * external, thus we need to call raw helper function.
1153          */
1154         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1155                                flags, 0, 0);
1156 }
1157
1158 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1159         .func           = bpf_get_stackid_tp,
1160         .gpl_only       = true,
1161         .ret_type       = RET_INTEGER,
1162         .arg1_type      = ARG_PTR_TO_CTX,
1163         .arg2_type      = ARG_CONST_MAP_PTR,
1164         .arg3_type      = ARG_ANYTHING,
1165 };
1166
1167 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1168            u64, flags)
1169 {
1170         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1171
1172         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1173                              (unsigned long) size, flags, 0);
1174 }
1175
1176 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1177         .func           = bpf_get_stack_tp,
1178         .gpl_only       = true,
1179         .ret_type       = RET_INTEGER,
1180         .arg1_type      = ARG_PTR_TO_CTX,
1181         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1182         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1183         .arg4_type      = ARG_ANYTHING,
1184 };
1185
1186 static const struct bpf_func_proto *
1187 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1188 {
1189         switch (func_id) {
1190         case BPF_FUNC_perf_event_output:
1191                 return &bpf_perf_event_output_proto_tp;
1192         case BPF_FUNC_get_stackid:
1193                 return &bpf_get_stackid_proto_tp;
1194         case BPF_FUNC_get_stack:
1195                 return &bpf_get_stack_proto_tp;
1196         default:
1197                 return bpf_tracing_func_proto(func_id, prog);
1198         }
1199 }
1200
1201 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1202                                     const struct bpf_prog *prog,
1203                                     struct bpf_insn_access_aux *info)
1204 {
1205         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1206                 return false;
1207         if (type != BPF_READ)
1208                 return false;
1209         if (off % size != 0)
1210                 return false;
1211
1212         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1213         return true;
1214 }
1215
1216 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1217         .get_func_proto  = tp_prog_func_proto,
1218         .is_valid_access = tp_prog_is_valid_access,
1219 };
1220
1221 const struct bpf_prog_ops tracepoint_prog_ops = {
1222 };
1223
1224 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1225            struct bpf_perf_event_value *, buf, u32, size)
1226 {
1227         int err = -EINVAL;
1228
1229         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1230                 goto clear;
1231         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1232                                     &buf->running);
1233         if (unlikely(err))
1234                 goto clear;
1235         return 0;
1236 clear:
1237         memset(buf, 0, size);
1238         return err;
1239 }
1240
1241 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1242          .func           = bpf_perf_prog_read_value,
1243          .gpl_only       = true,
1244          .ret_type       = RET_INTEGER,
1245          .arg1_type      = ARG_PTR_TO_CTX,
1246          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1247          .arg3_type      = ARG_CONST_SIZE,
1248 };
1249
1250 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1251            void *, buf, u32, size, u64, flags)
1252 {
1253 #ifndef CONFIG_X86
1254         return -ENOENT;
1255 #else
1256         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1257         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1258         u32 to_copy;
1259
1260         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1261                 return -EINVAL;
1262
1263         if (unlikely(!br_stack))
1264                 return -EINVAL;
1265
1266         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1267                 return br_stack->nr * br_entry_size;
1268
1269         if (!buf || (size % br_entry_size != 0))
1270                 return -EINVAL;
1271
1272         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1273         memcpy(buf, br_stack->entries, to_copy);
1274
1275         return to_copy;
1276 #endif
1277 }
1278
1279 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1280         .func           = bpf_read_branch_records,
1281         .gpl_only       = true,
1282         .ret_type       = RET_INTEGER,
1283         .arg1_type      = ARG_PTR_TO_CTX,
1284         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1285         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1286         .arg4_type      = ARG_ANYTHING,
1287 };
1288
1289 static const struct bpf_func_proto *
1290 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1291 {
1292         switch (func_id) {
1293         case BPF_FUNC_perf_event_output:
1294                 return &bpf_perf_event_output_proto_tp;
1295         case BPF_FUNC_get_stackid:
1296                 return &bpf_get_stackid_proto_pe;
1297         case BPF_FUNC_get_stack:
1298                 return &bpf_get_stack_proto_pe;
1299         case BPF_FUNC_perf_prog_read_value:
1300                 return &bpf_perf_prog_read_value_proto;
1301         case BPF_FUNC_read_branch_records:
1302                 return &bpf_read_branch_records_proto;
1303         default:
1304                 return bpf_tracing_func_proto(func_id, prog);
1305         }
1306 }
1307
1308 /*
1309  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1310  * to avoid potential recursive reuse issue when/if tracepoints are added
1311  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1312  *
1313  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1314  * in normal, irq, and nmi context.
1315  */
1316 struct bpf_raw_tp_regs {
1317         struct pt_regs regs[3];
1318 };
1319 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1320 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1321 static struct pt_regs *get_bpf_raw_tp_regs(void)
1322 {
1323         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1324         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1325
1326         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1327                 this_cpu_dec(bpf_raw_tp_nest_level);
1328                 return ERR_PTR(-EBUSY);
1329         }
1330
1331         return &tp_regs->regs[nest_level - 1];
1332 }
1333
1334 static void put_bpf_raw_tp_regs(void)
1335 {
1336         this_cpu_dec(bpf_raw_tp_nest_level);
1337 }
1338
1339 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1340            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1341 {
1342         struct pt_regs *regs = get_bpf_raw_tp_regs();
1343         int ret;
1344
1345         if (IS_ERR(regs))
1346                 return PTR_ERR(regs);
1347
1348         perf_fetch_caller_regs(regs);
1349         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1350
1351         put_bpf_raw_tp_regs();
1352         return ret;
1353 }
1354
1355 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1356         .func           = bpf_perf_event_output_raw_tp,
1357         .gpl_only       = true,
1358         .ret_type       = RET_INTEGER,
1359         .arg1_type      = ARG_PTR_TO_CTX,
1360         .arg2_type      = ARG_CONST_MAP_PTR,
1361         .arg3_type      = ARG_ANYTHING,
1362         .arg4_type      = ARG_PTR_TO_MEM,
1363         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1364 };
1365
1366 extern const struct bpf_func_proto bpf_skb_output_proto;
1367 extern const struct bpf_func_proto bpf_xdp_output_proto;
1368
1369 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1370            struct bpf_map *, map, u64, flags)
1371 {
1372         struct pt_regs *regs = get_bpf_raw_tp_regs();
1373         int ret;
1374
1375         if (IS_ERR(regs))
1376                 return PTR_ERR(regs);
1377
1378         perf_fetch_caller_regs(regs);
1379         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1380         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1381                               flags, 0, 0);
1382         put_bpf_raw_tp_regs();
1383         return ret;
1384 }
1385
1386 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1387         .func           = bpf_get_stackid_raw_tp,
1388         .gpl_only       = true,
1389         .ret_type       = RET_INTEGER,
1390         .arg1_type      = ARG_PTR_TO_CTX,
1391         .arg2_type      = ARG_CONST_MAP_PTR,
1392         .arg3_type      = ARG_ANYTHING,
1393 };
1394
1395 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1396            void *, buf, u32, size, u64, flags)
1397 {
1398         struct pt_regs *regs = get_bpf_raw_tp_regs();
1399         int ret;
1400
1401         if (IS_ERR(regs))
1402                 return PTR_ERR(regs);
1403
1404         perf_fetch_caller_regs(regs);
1405         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1406                             (unsigned long) size, flags, 0);
1407         put_bpf_raw_tp_regs();
1408         return ret;
1409 }
1410
1411 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1412         .func           = bpf_get_stack_raw_tp,
1413         .gpl_only       = true,
1414         .ret_type       = RET_INTEGER,
1415         .arg1_type      = ARG_PTR_TO_CTX,
1416         .arg2_type      = ARG_PTR_TO_MEM,
1417         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1418         .arg4_type      = ARG_ANYTHING,
1419 };
1420
1421 static const struct bpf_func_proto *
1422 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1423 {
1424         switch (func_id) {
1425         case BPF_FUNC_perf_event_output:
1426                 return &bpf_perf_event_output_proto_raw_tp;
1427         case BPF_FUNC_get_stackid:
1428                 return &bpf_get_stackid_proto_raw_tp;
1429         case BPF_FUNC_get_stack:
1430                 return &bpf_get_stack_proto_raw_tp;
1431         default:
1432                 return bpf_tracing_func_proto(func_id, prog);
1433         }
1434 }
1435
1436 const struct bpf_func_proto *
1437 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1438 {
1439         switch (func_id) {
1440 #ifdef CONFIG_NET
1441         case BPF_FUNC_skb_output:
1442                 return &bpf_skb_output_proto;
1443         case BPF_FUNC_xdp_output:
1444                 return &bpf_xdp_output_proto;
1445         case BPF_FUNC_skc_to_tcp6_sock:
1446                 return &bpf_skc_to_tcp6_sock_proto;
1447         case BPF_FUNC_skc_to_tcp_sock:
1448                 return &bpf_skc_to_tcp_sock_proto;
1449         case BPF_FUNC_skc_to_tcp_timewait_sock:
1450                 return &bpf_skc_to_tcp_timewait_sock_proto;
1451         case BPF_FUNC_skc_to_tcp_request_sock:
1452                 return &bpf_skc_to_tcp_request_sock_proto;
1453         case BPF_FUNC_skc_to_udp6_sock:
1454                 return &bpf_skc_to_udp6_sock_proto;
1455         case BPF_FUNC_sk_storage_get:
1456                 return &bpf_sk_storage_get_tracing_proto;
1457         case BPF_FUNC_sk_storage_delete:
1458                 return &bpf_sk_storage_delete_tracing_proto;
1459         case BPF_FUNC_sock_from_file:
1460                 return &bpf_sock_from_file_proto;
1461         case BPF_FUNC_get_socket_cookie:
1462                 return &bpf_get_socket_ptr_cookie_proto;
1463 #endif
1464         case BPF_FUNC_seq_printf:
1465                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1466                        &bpf_seq_printf_proto :
1467                        NULL;
1468         case BPF_FUNC_seq_write:
1469                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1470                        &bpf_seq_write_proto :
1471                        NULL;
1472         case BPF_FUNC_seq_printf_btf:
1473                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1474                        &bpf_seq_printf_btf_proto :
1475                        NULL;
1476         case BPF_FUNC_d_path:
1477                 return &bpf_d_path_proto;
1478         default:
1479                 return raw_tp_prog_func_proto(func_id, prog);
1480         }
1481 }
1482
1483 static bool raw_tp_prog_is_valid_access(int off, int size,
1484                                         enum bpf_access_type type,
1485                                         const struct bpf_prog *prog,
1486                                         struct bpf_insn_access_aux *info)
1487 {
1488         if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1489                 return false;
1490         if (type != BPF_READ)
1491                 return false;
1492         if (off % size != 0)
1493                 return false;
1494         return true;
1495 }
1496
1497 static bool tracing_prog_is_valid_access(int off, int size,
1498                                          enum bpf_access_type type,
1499                                          const struct bpf_prog *prog,
1500                                          struct bpf_insn_access_aux *info)
1501 {
1502         if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1503                 return false;
1504         if (type != BPF_READ)
1505                 return false;
1506         if (off % size != 0)
1507                 return false;
1508         return btf_ctx_access(off, size, type, prog, info);
1509 }
1510
1511 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1512                                      const union bpf_attr *kattr,
1513                                      union bpf_attr __user *uattr)
1514 {
1515         return -ENOTSUPP;
1516 }
1517
1518 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1519         .get_func_proto  = raw_tp_prog_func_proto,
1520         .is_valid_access = raw_tp_prog_is_valid_access,
1521 };
1522
1523 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1524 #ifdef CONFIG_NET
1525         .test_run = bpf_prog_test_run_raw_tp,
1526 #endif
1527 };
1528
1529 const struct bpf_verifier_ops tracing_verifier_ops = {
1530         .get_func_proto  = tracing_prog_func_proto,
1531         .is_valid_access = tracing_prog_is_valid_access,
1532 };
1533
1534 const struct bpf_prog_ops tracing_prog_ops = {
1535         .test_run = bpf_prog_test_run_tracing,
1536 };
1537
1538 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1539                                                  enum bpf_access_type type,
1540                                                  const struct bpf_prog *prog,
1541                                                  struct bpf_insn_access_aux *info)
1542 {
1543         if (off == 0) {
1544                 if (size != sizeof(u64) || type != BPF_READ)
1545                         return false;
1546                 info->reg_type = PTR_TO_TP_BUFFER;
1547         }
1548         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1549 }
1550
1551 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1552         .get_func_proto  = raw_tp_prog_func_proto,
1553         .is_valid_access = raw_tp_writable_prog_is_valid_access,
1554 };
1555
1556 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1557 };
1558
1559 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1560                                     const struct bpf_prog *prog,
1561                                     struct bpf_insn_access_aux *info)
1562 {
1563         const int size_u64 = sizeof(u64);
1564
1565         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1566                 return false;
1567         if (type != BPF_READ)
1568                 return false;
1569         if (off % size != 0) {
1570                 if (sizeof(unsigned long) != 4)
1571                         return false;
1572                 if (size != 8)
1573                         return false;
1574                 if (off % size != 4)
1575                         return false;
1576         }
1577
1578         switch (off) {
1579         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1580                 bpf_ctx_record_field_size(info, size_u64);
1581                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1582                         return false;
1583                 break;
1584         case bpf_ctx_range(struct bpf_perf_event_data, addr):
1585                 bpf_ctx_record_field_size(info, size_u64);
1586                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1587                         return false;
1588                 break;
1589         default:
1590                 if (size != sizeof(long))
1591                         return false;
1592         }
1593
1594         return true;
1595 }
1596
1597 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1598                                       const struct bpf_insn *si,
1599                                       struct bpf_insn *insn_buf,
1600                                       struct bpf_prog *prog, u32 *target_size)
1601 {
1602         struct bpf_insn *insn = insn_buf;
1603
1604         switch (si->off) {
1605         case offsetof(struct bpf_perf_event_data, sample_period):
1606                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1607                                                        data), si->dst_reg, si->src_reg,
1608                                       offsetof(struct bpf_perf_event_data_kern, data));
1609                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1610                                       bpf_target_off(struct perf_sample_data, period, 8,
1611                                                      target_size));
1612                 break;
1613         case offsetof(struct bpf_perf_event_data, addr):
1614                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1615                                                        data), si->dst_reg, si->src_reg,
1616                                       offsetof(struct bpf_perf_event_data_kern, data));
1617                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1618                                       bpf_target_off(struct perf_sample_data, addr, 8,
1619                                                      target_size));
1620                 break;
1621         default:
1622                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1623                                                        regs), si->dst_reg, si->src_reg,
1624                                       offsetof(struct bpf_perf_event_data_kern, regs));
1625                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1626                                       si->off);
1627                 break;
1628         }
1629
1630         return insn - insn_buf;
1631 }
1632
1633 const struct bpf_verifier_ops perf_event_verifier_ops = {
1634         .get_func_proto         = pe_prog_func_proto,
1635         .is_valid_access        = pe_prog_is_valid_access,
1636         .convert_ctx_access     = pe_prog_convert_ctx_access,
1637 };
1638
1639 const struct bpf_prog_ops perf_event_prog_ops = {
1640 };
1641
1642 static DEFINE_MUTEX(bpf_event_mutex);
1643
1644 #define BPF_TRACE_MAX_PROGS 64
1645
1646 int perf_event_attach_bpf_prog(struct perf_event *event,
1647                                struct bpf_prog *prog)
1648 {
1649         struct bpf_prog_array *old_array;
1650         struct bpf_prog_array *new_array;
1651         int ret = -EEXIST;
1652
1653         /*
1654          * Kprobe override only works if they are on the function entry,
1655          * and only if they are on the opt-in list.
1656          */
1657         if (prog->kprobe_override &&
1658             (!trace_kprobe_on_func_entry(event->tp_event) ||
1659              !trace_kprobe_error_injectable(event->tp_event)))
1660                 return -EINVAL;
1661
1662         mutex_lock(&bpf_event_mutex);
1663
1664         if (event->prog)
1665                 goto unlock;
1666
1667         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1668         if (old_array &&
1669             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1670                 ret = -E2BIG;
1671                 goto unlock;
1672         }
1673
1674         ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1675         if (ret < 0)
1676                 goto unlock;
1677
1678         /* set the new array to event->tp_event and set event->prog */
1679         event->prog = prog;
1680         rcu_assign_pointer(event->tp_event->prog_array, new_array);
1681         bpf_prog_array_free(old_array);
1682
1683 unlock:
1684         mutex_unlock(&bpf_event_mutex);
1685         return ret;
1686 }
1687
1688 void perf_event_detach_bpf_prog(struct perf_event *event)
1689 {
1690         struct bpf_prog_array *old_array;
1691         struct bpf_prog_array *new_array;
1692         int ret;
1693
1694         mutex_lock(&bpf_event_mutex);
1695
1696         if (!event->prog)
1697                 goto unlock;
1698
1699         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1700         ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1701         if (ret == -ENOENT)
1702                 goto unlock;
1703         if (ret < 0) {
1704                 bpf_prog_array_delete_safe(old_array, event->prog);
1705         } else {
1706                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1707                 bpf_prog_array_free(old_array);
1708         }
1709
1710         bpf_prog_put(event->prog);
1711         event->prog = NULL;
1712
1713 unlock:
1714         mutex_unlock(&bpf_event_mutex);
1715 }
1716
1717 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1718 {
1719         struct perf_event_query_bpf __user *uquery = info;
1720         struct perf_event_query_bpf query = {};
1721         struct bpf_prog_array *progs;
1722         u32 *ids, prog_cnt, ids_len;
1723         int ret;
1724
1725         if (!perfmon_capable())
1726                 return -EPERM;
1727         if (event->attr.type != PERF_TYPE_TRACEPOINT)
1728                 return -EINVAL;
1729         if (copy_from_user(&query, uquery, sizeof(query)))
1730                 return -EFAULT;
1731
1732         ids_len = query.ids_len;
1733         if (ids_len > BPF_TRACE_MAX_PROGS)
1734                 return -E2BIG;
1735         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1736         if (!ids)
1737                 return -ENOMEM;
1738         /*
1739          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1740          * is required when user only wants to check for uquery->prog_cnt.
1741          * There is no need to check for it since the case is handled
1742          * gracefully in bpf_prog_array_copy_info.
1743          */
1744
1745         mutex_lock(&bpf_event_mutex);
1746         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1747         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1748         mutex_unlock(&bpf_event_mutex);
1749
1750         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1751             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1752                 ret = -EFAULT;
1753
1754         kfree(ids);
1755         return ret;
1756 }
1757
1758 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1759 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1760
1761 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1762 {
1763         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1764
1765         for (; btp < __stop__bpf_raw_tp; btp++) {
1766                 if (!strcmp(btp->tp->name, name))
1767                         return btp;
1768         }
1769
1770         return bpf_get_raw_tracepoint_module(name);
1771 }
1772
1773 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1774 {
1775         struct module *mod;
1776
1777         preempt_disable();
1778         mod = __module_address((unsigned long)btp);
1779         module_put(mod);
1780         preempt_enable();
1781 }
1782
1783 static __always_inline
1784 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1785 {
1786         cant_sleep();
1787         rcu_read_lock();
1788         (void) BPF_PROG_RUN(prog, args);
1789         rcu_read_unlock();
1790 }
1791
1792 #define UNPACK(...)                     __VA_ARGS__
1793 #define REPEAT_1(FN, DL, X, ...)        FN(X)
1794 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1795 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1796 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1797 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1798 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1799 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1800 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1801 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1802 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1803 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1804 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1805 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
1806
1807 #define SARG(X)         u64 arg##X
1808 #define COPY(X)         args[X] = arg##X
1809
1810 #define __DL_COM        (,)
1811 #define __DL_SEM        (;)
1812
1813 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1814
1815 #define BPF_TRACE_DEFN_x(x)                                             \
1816         void bpf_trace_run##x(struct bpf_prog *prog,                    \
1817                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
1818         {                                                               \
1819                 u64 args[x];                                            \
1820                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
1821                 __bpf_trace_run(prog, args);                            \
1822         }                                                               \
1823         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1824 BPF_TRACE_DEFN_x(1);
1825 BPF_TRACE_DEFN_x(2);
1826 BPF_TRACE_DEFN_x(3);
1827 BPF_TRACE_DEFN_x(4);
1828 BPF_TRACE_DEFN_x(5);
1829 BPF_TRACE_DEFN_x(6);
1830 BPF_TRACE_DEFN_x(7);
1831 BPF_TRACE_DEFN_x(8);
1832 BPF_TRACE_DEFN_x(9);
1833 BPF_TRACE_DEFN_x(10);
1834 BPF_TRACE_DEFN_x(11);
1835 BPF_TRACE_DEFN_x(12);
1836
1837 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1838 {
1839         struct tracepoint *tp = btp->tp;
1840
1841         /*
1842          * check that program doesn't access arguments beyond what's
1843          * available in this tracepoint
1844          */
1845         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1846                 return -EINVAL;
1847
1848         if (prog->aux->max_tp_access > btp->writable_size)
1849                 return -EINVAL;
1850
1851         return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1852 }
1853
1854 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1855 {
1856         return __bpf_probe_register(btp, prog);
1857 }
1858
1859 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1860 {
1861         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1862 }
1863
1864 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1865                             u32 *fd_type, const char **buf,
1866                             u64 *probe_offset, u64 *probe_addr)
1867 {
1868         bool is_tracepoint, is_syscall_tp;
1869         struct bpf_prog *prog;
1870         int flags, err = 0;
1871
1872         prog = event->prog;
1873         if (!prog)
1874                 return -ENOENT;
1875
1876         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1877         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1878                 return -EOPNOTSUPP;
1879
1880         *prog_id = prog->aux->id;
1881         flags = event->tp_event->flags;
1882         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1883         is_syscall_tp = is_syscall_trace_event(event->tp_event);
1884
1885         if (is_tracepoint || is_syscall_tp) {
1886                 *buf = is_tracepoint ? event->tp_event->tp->name
1887                                      : event->tp_event->name;
1888                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1889                 *probe_offset = 0x0;
1890                 *probe_addr = 0x0;
1891         } else {
1892                 /* kprobe/uprobe */
1893                 err = -EOPNOTSUPP;
1894 #ifdef CONFIG_KPROBE_EVENTS
1895                 if (flags & TRACE_EVENT_FL_KPROBE)
1896                         err = bpf_get_kprobe_info(event, fd_type, buf,
1897                                                   probe_offset, probe_addr,
1898                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
1899 #endif
1900 #ifdef CONFIG_UPROBE_EVENTS
1901                 if (flags & TRACE_EVENT_FL_UPROBE)
1902                         err = bpf_get_uprobe_info(event, fd_type, buf,
1903                                                   probe_offset,
1904                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
1905 #endif
1906         }
1907
1908         return err;
1909 }
1910
1911 static int __init send_signal_irq_work_init(void)
1912 {
1913         int cpu;
1914         struct send_signal_irq_work *work;
1915
1916         for_each_possible_cpu(cpu) {
1917                 work = per_cpu_ptr(&send_signal_work, cpu);
1918                 init_irq_work(&work->irq_work, do_bpf_send_signal);
1919         }
1920         return 0;
1921 }
1922
1923 subsys_initcall(send_signal_irq_work_init);
1924
1925 #ifdef CONFIG_MODULES
1926 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1927                             void *module)
1928 {
1929         struct bpf_trace_module *btm, *tmp;
1930         struct module *mod = module;
1931         int ret = 0;
1932
1933         if (mod->num_bpf_raw_events == 0 ||
1934             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1935                 goto out;
1936
1937         mutex_lock(&bpf_module_mutex);
1938
1939         switch (op) {
1940         case MODULE_STATE_COMING:
1941                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1942                 if (btm) {
1943                         btm->module = module;
1944                         list_add(&btm->list, &bpf_trace_modules);
1945                 } else {
1946                         ret = -ENOMEM;
1947                 }
1948                 break;
1949         case MODULE_STATE_GOING:
1950                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1951                         if (btm->module == module) {
1952                                 list_del(&btm->list);
1953                                 kfree(btm);
1954                                 break;
1955                         }
1956                 }
1957                 break;
1958         }
1959
1960         mutex_unlock(&bpf_module_mutex);
1961
1962 out:
1963         return notifier_from_errno(ret);
1964 }
1965
1966 static struct notifier_block bpf_module_nb = {
1967         .notifier_call = bpf_event_notify,
1968 };
1969
1970 static int __init bpf_event_init(void)
1971 {
1972         register_module_notifier(&bpf_module_nb);
1973         return 0;
1974 }
1975
1976 fs_initcall(bpf_event_init);
1977 #endif /* CONFIG_MODULES */