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