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/filter.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/ctype.h>
  13 #include <linux/kprobes.h>
  14 #include <linux/syscalls.h>
  15 #include <linux/error-injection.h>
  16
  17 #include <asm/tlb.h>
  18
  19 #include "trace_probe.h"
  20 #include "trace.h"
  21
  22 #define bpf_event_rcu_dereference(p)                                    \
  23         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  24
  25 #ifdef CONFIG_MODULES
  26 struct bpf_trace_module {
  27         struct module *module;
  28         struct list_head list;
  29 };
  30
  31 static LIST_HEAD(bpf_trace_modules);
  32 static DEFINE_MUTEX(bpf_module_mutex);
  33
  34 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  35 {
  36         struct bpf_raw_event_map *btp, *ret = NULL;
  37         struct bpf_trace_module *btm;
  38         unsigned int i;
  39
  40         mutex_lock(&bpf_module_mutex);
  41         list_for_each_entry(btm, &bpf_trace_modules, list) {
  42                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  43                         btp = &btm->module->bpf_raw_events[i];
  44                         if (!strcmp(btp->tp->name, name)) {
  45                                 if (try_module_get(btm->module))
  46                                         ret = btp;
  47                                 goto out;
  48                         }
  49                 }
  50         }
  51 out:
  52         mutex_unlock(&bpf_module_mutex);
  53         return ret;
  54 }
  55 #else
  56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  57 {
  58         return NULL;
  59 }
  60 #endif /* CONFIG_MODULES */
  61
  62 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  63 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  64
  65 /**
  66  * trace_call_bpf - invoke BPF program
  67  * @call: tracepoint event
  68  * @ctx: opaque context pointer
  69  *
  70  * kprobe handlers execute BPF programs via this helper.
  71  * Can be used from static tracepoints in the future.
  72  *
  73  * Return: BPF programs always return an integer which is interpreted by
  74  * kprobe handler as:
  75  * 0 - return from kprobe (event is filtered out)
  76  * 1 - store kprobe event into ring buffer
  77  * Other values are reserved and currently alias to 1
  78  */
  79 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
  80 {
  81         unsigned int ret;
  82
  83         if (in_nmi()) /* not supported yet */
  84                 return 1;
  85
  86         cant_sleep();
  87
  88         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
  89                 /*
  90                  * since some bpf program is already running on this cpu,
  91                  * don't call into another bpf program (same or different)
  92                  * and don't send kprobe event into ring-buffer,
  93                  * so return zero here
  94                  */
  95                 ret = 0;
  96                 goto out;
  97         }
  98
  99         /*
 100          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 101          * to all call sites, we did a bpf_prog_array_valid() there to check
 102          * whether call->prog_array is empty or not, which is
 103          * a heurisitc to speed up execution.
 104          *
 105          * If bpf_prog_array_valid() fetched prog_array was
 106          * non-NULL, we go into trace_call_bpf() and do the actual
 107          * proper rcu_dereference() under RCU lock.
 108          * If it turns out that prog_array is NULL then, we bail out.
 109          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 110          * was NULL, you'll skip the prog_array with the risk of missing
 111          * out of events when it was updated in between this and the
 112          * rcu_dereference() which is accepted risk.
 113          */
 114         ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
 115
 116  out:
 117         __this_cpu_dec(bpf_prog_active);
 118
 119         return ret;
 120 }
 121
 122 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 123 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 124 {
 125         regs_set_return_value(regs, rc);
 126         override_function_with_return(regs);
 127         return 0;
 128 }
 129
 130 static const struct bpf_func_proto bpf_override_return_proto = {
 131         .func           = bpf_override_return,
 132         .gpl_only       = true,
 133         .ret_type       = RET_INTEGER,
 134         .arg1_type      = ARG_PTR_TO_CTX,
 135         .arg2_type      = ARG_ANYTHING,
 136 };
 137 #endif
 138
 139 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 140            const void __user *, unsafe_ptr)
 141 {
 142         int ret = probe_user_read(dst, unsafe_ptr, size);
 143
 144         if (unlikely(ret < 0))
 145                 memset(dst, 0, size);
 146
 147         return ret;
 148 }
 149
 150 const struct bpf_func_proto bpf_probe_read_user_proto = {
 151         .func           = bpf_probe_read_user,
 152         .gpl_only       = true,
 153         .ret_type       = RET_INTEGER,
 154         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 155         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 156         .arg3_type      = ARG_ANYTHING,
 157 };
 158
 159 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 160            const void __user *, unsafe_ptr)
 161 {
 162         int ret = strncpy_from_unsafe_user(dst, unsafe_ptr, size);
 163
 164         if (unlikely(ret < 0))
 165                 memset(dst, 0, size);
 166
 167         return ret;
 168 }
 169
 170 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 171         .func           = bpf_probe_read_user_str,
 172         .gpl_only       = true,
 173         .ret_type       = RET_INTEGER,
 174         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 175         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 176         .arg3_type      = ARG_ANYTHING,
 177 };
 178
 179 static __always_inline int
 180 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr,
 181                              const bool compat)
 182 {
 183         int ret = security_locked_down(LOCKDOWN_BPF_READ);
 184
 185         if (unlikely(ret < 0))
 186                 goto out;
 187         ret = compat ? probe_kernel_read(dst, unsafe_ptr, size) :
 188               probe_kernel_read_strict(dst, unsafe_ptr, size);
 189         if (unlikely(ret < 0))
 190 out:
 191                 memset(dst, 0, size);
 192         return ret;
 193 }
 194
 195 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 196            const void *, unsafe_ptr)
 197 {
 198         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, false);
 199 }
 200
 201 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 202         .func           = bpf_probe_read_kernel,
 203         .gpl_only       = true,
 204         .ret_type       = RET_INTEGER,
 205         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 206         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 207         .arg3_type      = ARG_ANYTHING,
 208 };
 209
 210 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 211            const void *, unsafe_ptr)
 212 {
 213         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, true);
 214 }
 215
 216 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 217         .func           = bpf_probe_read_compat,
 218         .gpl_only       = true,
 219         .ret_type       = RET_INTEGER,
 220         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 221         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 222         .arg3_type      = ARG_ANYTHING,
 223 };
 224
 225 static __always_inline int
 226 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr,
 227                                  const bool compat)
 228 {
 229         int ret = security_locked_down(LOCKDOWN_BPF_READ);
 230
 231         if (unlikely(ret < 0))
 232                 goto out;
 233         /*
 234          * The strncpy_from_unsafe_*() call will likely not fill the entire
 235          * buffer, but that's okay in this circumstance as we're probing
 236          * arbitrary memory anyway similar to bpf_probe_read_*() and might
 237          * as well probe the stack. Thus, memory is explicitly cleared
 238          * only in error case, so that improper users ignoring return
 239          * code altogether don't copy garbage; otherwise length of string
 240          * is returned that can be used for bpf_perf_event_output() et al.
 241          */
 242         ret = compat ? strncpy_from_unsafe(dst, unsafe_ptr, size) :
 243               strncpy_from_unsafe_strict(dst, unsafe_ptr, size);
 244         if (unlikely(ret < 0))
 245 out:
 246                 memset(dst, 0, size);
 247         return ret;
 248 }
 249
 250 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 251            const void *, unsafe_ptr)
 252 {
 253         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, false);
 254 }
 255
 256 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 257         .func           = bpf_probe_read_kernel_str,
 258         .gpl_only       = true,
 259         .ret_type       = RET_INTEGER,
 260         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 261         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 262         .arg3_type      = ARG_ANYTHING,
 263 };
 264
 265 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 266            const void *, unsafe_ptr)
 267 {
 268         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, true);
 269 }
 270
 271 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 272         .func           = bpf_probe_read_compat_str,
 273         .gpl_only       = true,
 274         .ret_type       = RET_INTEGER,
 275         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 276         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 277         .arg3_type      = ARG_ANYTHING,
 278 };
 279
 280 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 281            u32, size)
 282 {
 283         /*
 284          * Ensure we're in user context which is safe for the helper to
 285          * run. This helper has no business in a kthread.
 286          *
 287          * access_ok() should prevent writing to non-user memory, but in
 288          * some situations (nommu, temporary switch, etc) access_ok() does
 289          * not provide enough validation, hence the check on KERNEL_DS.
 290          *
 291          * nmi_uaccess_okay() ensures the probe is not run in an interim
 292          * state, when the task or mm are switched. This is specifically
 293          * required to prevent the use of temporary mm.
 294          */
 295
 296         if (unlikely(in_interrupt() ||
 297                      current->flags & (PF_KTHREAD | PF_EXITING)))
 298                 return -EPERM;
 299         if (unlikely(uaccess_kernel()))
 300                 return -EPERM;
 301         if (unlikely(!nmi_uaccess_okay()))
 302                 return -EPERM;
 303
 304         return probe_user_write(unsafe_ptr, src, size);
 305 }
 306
 307 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 308         .func           = bpf_probe_write_user,
 309         .gpl_only       = true,
 310         .ret_type       = RET_INTEGER,
 311         .arg1_type      = ARG_ANYTHING,
 312         .arg2_type      = ARG_PTR_TO_MEM,
 313         .arg3_type      = ARG_CONST_SIZE,
 314 };
 315
 316 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 317 {
 318         if (!capable(CAP_SYS_ADMIN))
 319                 return NULL;
 320
 321         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 322                             current->comm, task_pid_nr(current));
 323
 324         return &bpf_probe_write_user_proto;
 325 }
 326
 327 /*
 328  * Only limited trace_printk() conversion specifiers allowed:
 329  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pks %pus %s
 330  */
 331 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 332            u64, arg2, u64, arg3)
 333 {
 334         int i, mod[3] = {}, fmt_cnt = 0;
 335         char buf[64], fmt_ptype;
 336         void *unsafe_ptr = NULL;
 337         bool str_seen = false;
 338
 339         /*
 340          * bpf_check()->check_func_arg()->check_stack_boundary()
 341          * guarantees that fmt points to bpf program stack,
 342          * fmt_size bytes of it were initialized and fmt_size > 0
 343          */
 344         if (fmt[--fmt_size] != 0)
 345                 return -EINVAL;
 346
 347         /* check format string for allowed specifiers */
 348         for (i = 0; i < fmt_size; i++) {
 349                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 350                         return -EINVAL;
 351
 352                 if (fmt[i] != '%')
 353                         continue;
 354
 355                 if (fmt_cnt >= 3)
 356                         return -EINVAL;
 357
 358                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 359                 i++;
 360                 if (fmt[i] == 'l') {
 361                         mod[fmt_cnt]++;
 362                         i++;
 363                 } else if (fmt[i] == 'p') {
 364                         mod[fmt_cnt]++;
 365                         if ((fmt[i + 1] == 'k' ||
 366                              fmt[i + 1] == 'u') &&
 367                             fmt[i + 2] == 's') {
 368                                 fmt_ptype = fmt[i + 1];
 369                                 i += 2;
 370                                 goto fmt_str;
 371                         }
 372
 373                         /* disallow any further format extensions */
 374                         if (fmt[i + 1] != 0 &&
 375                             !isspace(fmt[i + 1]) &&
 376                             !ispunct(fmt[i + 1]))
 377                                 return -EINVAL;
 378
 379                         goto fmt_next;
 380                 } else if (fmt[i] == 's') {
 381                         mod[fmt_cnt]++;
 382                         fmt_ptype = fmt[i];
 383 fmt_str:
 384                         if (str_seen)
 385                                 /* allow only one '%s' per fmt string */
 386                                 return -EINVAL;
 387                         str_seen = true;
 388
 389                         if (fmt[i + 1] != 0 &&
 390                             !isspace(fmt[i + 1]) &&
 391                             !ispunct(fmt[i + 1]))
 392                                 return -EINVAL;
 393
 394                         switch (fmt_cnt) {
 395                         case 0:
 396                                 unsafe_ptr = (void *)(long)arg1;
 397                                 arg1 = (long)buf;
 398                                 break;
 399                         case 1:
 400                                 unsafe_ptr = (void *)(long)arg2;
 401                                 arg2 = (long)buf;
 402                                 break;
 403                         case 2:
 404                                 unsafe_ptr = (void *)(long)arg3;
 405                                 arg3 = (long)buf;
 406                                 break;
 407                         }
 408
 409                         buf[0] = 0;
 410                         switch (fmt_ptype) {
 411                         case 's':
 412 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 413                                 strncpy_from_unsafe(buf, unsafe_ptr,
 414                                                     sizeof(buf));
 415                                 break;
 416 #endif
 417                         case 'k':
 418                                 strncpy_from_unsafe_strict(buf, unsafe_ptr,
 419                                                            sizeof(buf));
 420                                 break;
 421                         case 'u':
 422                                 strncpy_from_unsafe_user(buf,
 423                                         (__force void __user *)unsafe_ptr,
 424                                                          sizeof(buf));
 425                                 break;
 426                         }
 427                         goto fmt_next;
 428                 }
 429
 430                 if (fmt[i] == 'l') {
 431                         mod[fmt_cnt]++;
 432                         i++;
 433                 }
 434
 435                 if (fmt[i] != 'i' && fmt[i] != 'd' &&
 436                     fmt[i] != 'u' && fmt[i] != 'x')
 437                         return -EINVAL;
 438 fmt_next:
 439                 fmt_cnt++;
 440         }
 441
 442 /* Horrid workaround for getting va_list handling working with different
 443  * argument type combinations generically for 32 and 64 bit archs.
 444  */
 445 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
 446 #define __BPF_TP(...)                                                   \
 447         __trace_printk(0 /* Fake ip */,                                 \
 448                        fmt, ##__VA_ARGS__)
 449
 450 #define __BPF_ARG1_TP(...)                                              \
 451         ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))        \
 452           ? __BPF_TP(arg1, ##__VA_ARGS__)                               \
 453           : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))    \
 454               ? __BPF_TP((long)arg1, ##__VA_ARGS__)                     \
 455               : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
 456
 457 #define __BPF_ARG2_TP(...)                                              \
 458         ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))        \
 459           ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)                          \
 460           : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))    \
 461               ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)                \
 462               : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
 463
 464 #define __BPF_ARG3_TP(...)                                              \
 465         ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))        \
 466           ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)                          \
 467           : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))    \
 468               ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)                \
 469               : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
 470
 471         return __BPF_TP_EMIT();
 472 }
 473
 474 static const struct bpf_func_proto bpf_trace_printk_proto = {
 475         .func           = bpf_trace_printk,
 476         .gpl_only       = true,
 477         .ret_type       = RET_INTEGER,
 478         .arg1_type      = ARG_PTR_TO_MEM,
 479         .arg2_type      = ARG_CONST_SIZE,
 480 };
 481
 482 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 483 {
 484         /*
 485          * this program might be calling bpf_trace_printk,
 486          * so allocate per-cpu printk buffers
 487          */
 488         trace_printk_init_buffers();
 489
 490         return &bpf_trace_printk_proto;
 491 }
 492
 493 #define MAX_SEQ_PRINTF_VARARGS          12
 494 #define MAX_SEQ_PRINTF_MAX_MEMCPY       6
 495 #define MAX_SEQ_PRINTF_STR_LEN          128
 496
 497 struct bpf_seq_printf_buf {
 498         char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
 499 };
 500 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
 501 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
 502
 503 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 504            const void *, data, u32, data_len)
 505 {
 506         int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
 507         int i, buf_used, copy_size, num_args;
 508         u64 params[MAX_SEQ_PRINTF_VARARGS];
 509         struct bpf_seq_printf_buf *bufs;
 510         const u64 *args = data;
 511
 512         buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
 513         if (WARN_ON_ONCE(buf_used > 1)) {
 514                 err = -EBUSY;
 515                 goto out;
 516         }
 517
 518         bufs = this_cpu_ptr(&bpf_seq_printf_buf);
 519
 520         /*
 521          * bpf_check()->check_func_arg()->check_stack_boundary()
 522          * guarantees that fmt points to bpf program stack,
 523          * fmt_size bytes of it were initialized and fmt_size > 0
 524          */
 525         if (fmt[--fmt_size] != 0)
 526                 goto out;
 527
 528         if (data_len & 7)
 529                 goto out;
 530
 531         for (i = 0; i < fmt_size; i++) {
 532                 if (fmt[i] == '%') {
 533                         if (fmt[i + 1] == '%')
 534                                 i++;
 535                         else if (!data || !data_len)
 536                                 goto out;
 537                 }
 538         }
 539
 540         num_args = data_len / 8;
 541
 542         /* check format string for allowed specifiers */
 543         for (i = 0; i < fmt_size; i++) {
 544                 /* only printable ascii for now. */
 545                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
 546                         err = -EINVAL;
 547                         goto out;
 548                 }
 549
 550                 if (fmt[i] != '%')
 551                         continue;
 552
 553                 if (fmt[i + 1] == '%') {
 554                         i++;
 555                         continue;
 556                 }
 557
 558                 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
 559                         err = -E2BIG;
 560                         goto out;
 561                 }
 562
 563                 if (fmt_cnt >= num_args) {
 564                         err = -EINVAL;
 565                         goto out;
 566                 }
 567
 568                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 569                 i++;
 570
 571                 /* skip optional "[0 +-][num]" width formating field */
 572                 while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
 573                        fmt[i] == ' ')
 574                         i++;
 575                 if (fmt[i] >= '1' && fmt[i] <= '9') {
 576                         i++;
 577                         while (fmt[i] >= '0' && fmt[i] <= '9')
 578                                 i++;
 579                 }
 580
 581                 if (fmt[i] == 's') {
 582                         /* try our best to copy */
 583                         if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
 584                                 err = -E2BIG;
 585                                 goto out;
 586                         }
 587
 588                         err = strncpy_from_unsafe(bufs->buf[memcpy_cnt],
 589                                                   (void *) (long) args[fmt_cnt],
 590                                                   MAX_SEQ_PRINTF_STR_LEN);
 591                         if (err < 0)
 592                                 bufs->buf[memcpy_cnt][0] = '\0';
 593                         params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
 594
 595                         fmt_cnt++;
 596                         memcpy_cnt++;
 597                         continue;
 598                 }
 599
 600                 if (fmt[i] == 'p') {
 601                         if (fmt[i + 1] == 0 ||
 602                             fmt[i + 1] == 'K' ||
 603                             fmt[i + 1] == 'x') {
 604                                 /* just kernel pointers */
 605                                 params[fmt_cnt] = args[fmt_cnt];
 606                                 fmt_cnt++;
 607                                 continue;
 608                         }
 609
 610                         /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
 611                         if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
 612                                 err = -EINVAL;
 613                                 goto out;
 614                         }
 615                         if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
 616                                 err = -EINVAL;
 617                                 goto out;
 618                         }
 619
 620                         if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
 621                                 err = -E2BIG;
 622                                 goto out;
 623                         }
 624
 625
 626                         copy_size = (fmt[i + 2] == '4') ? 4 : 16;
 627
 628                         err = probe_kernel_read(bufs->buf[memcpy_cnt],
 629                                                 (void *) (long) args[fmt_cnt],
 630                                                 copy_size);
 631                         if (err < 0)
 632                                 memset(bufs->buf[memcpy_cnt], 0, copy_size);
 633                         params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
 634
 635                         i += 2;
 636                         fmt_cnt++;
 637                         memcpy_cnt++;
 638                         continue;
 639                 }
 640
 641                 if (fmt[i] == 'l') {
 642                         i++;
 643                         if (fmt[i] == 'l')
 644                                 i++;
 645                 }
 646
 647                 if (fmt[i] != 'i' && fmt[i] != 'd' &&
 648                     fmt[i] != 'u' && fmt[i] != 'x') {
 649                         err = -EINVAL;
 650                         goto out;
 651                 }
 652
 653                 params[fmt_cnt] = args[fmt_cnt];
 654                 fmt_cnt++;
 655         }
 656
 657         /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
 658          * all of them to seq_printf().
 659          */
 660         seq_printf(m, fmt, params[0], params[1], params[2], params[3],
 661                    params[4], params[5], params[6], params[7], params[8],
 662                    params[9], params[10], params[11]);
 663
 664         err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
 665 out:
 666         this_cpu_dec(bpf_seq_printf_buf_used);
 667         return err;
 668 }
 669
 670 static int bpf_seq_printf_btf_ids[5];
 671 static const struct bpf_func_proto bpf_seq_printf_proto = {
 672         .func           = bpf_seq_printf,
 673         .gpl_only       = true,
 674         .ret_type       = RET_INTEGER,
 675         .arg1_type      = ARG_PTR_TO_BTF_ID,
 676         .arg2_type      = ARG_PTR_TO_MEM,
 677         .arg3_type      = ARG_CONST_SIZE,
 678         .arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
 679         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 680         .btf_id         = bpf_seq_printf_btf_ids,
 681 };
 682
 683 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
 684 {
 685         return seq_write(m, data, len) ? -EOVERFLOW : 0;
 686 }
 687
 688 static int bpf_seq_write_btf_ids[5];
 689 static const struct bpf_func_proto bpf_seq_write_proto = {
 690         .func           = bpf_seq_write,
 691         .gpl_only       = true,
 692         .ret_type       = RET_INTEGER,
 693         .arg1_type      = ARG_PTR_TO_BTF_ID,
 694         .arg2_type      = ARG_PTR_TO_MEM,
 695         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 696         .btf_id         = bpf_seq_write_btf_ids,
 697 };
 698
 699 static __always_inline int
 700 get_map_perf_counter(struct bpf_map *map, u64 flags,
 701                      u64 *value, u64 *enabled, u64 *running)
 702 {
 703         struct bpf_array *array = container_of(map, struct bpf_array, map);
 704         unsigned int cpu = smp_processor_id();
 705         u64 index = flags & BPF_F_INDEX_MASK;
 706         struct bpf_event_entry *ee;
 707
 708         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 709                 return -EINVAL;
 710         if (index == BPF_F_CURRENT_CPU)
 711                 index = cpu;
 712         if (unlikely(index >= array->map.max_entries))
 713                 return -E2BIG;
 714
 715         ee = READ_ONCE(array->ptrs[index]);
 716         if (!ee)
 717                 return -ENOENT;
 718
 719         return perf_event_read_local(ee->event, value, enabled, running);
 720 }
 721
 722 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 723 {
 724         u64 value = 0;
 725         int err;
 726
 727         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 728         /*
 729          * this api is ugly since we miss [-22..-2] range of valid
 730          * counter values, but that's uapi
 731          */
 732         if (err)
 733                 return err;
 734         return value;
 735 }
 736
 737 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 738         .func           = bpf_perf_event_read,
 739         .gpl_only       = true,
 740         .ret_type       = RET_INTEGER,
 741         .arg1_type      = ARG_CONST_MAP_PTR,
 742         .arg2_type      = ARG_ANYTHING,
 743 };
 744
 745 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 746            struct bpf_perf_event_value *, buf, u32, size)
 747 {
 748         int err = -EINVAL;
 749
 750         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 751                 goto clear;
 752         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 753                                    &buf->running);
 754         if (unlikely(err))
 755                 goto clear;
 756         return 0;
 757 clear:
 758         memset(buf, 0, size);
 759         return err;
 760 }
 761
 762 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 763         .func           = bpf_perf_event_read_value,
 764         .gpl_only       = true,
 765         .ret_type       = RET_INTEGER,
 766         .arg1_type      = ARG_CONST_MAP_PTR,
 767         .arg2_type      = ARG_ANYTHING,
 768         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 769         .arg4_type      = ARG_CONST_SIZE,
 770 };
 771
 772 static __always_inline u64
 773 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 774                         u64 flags, struct perf_sample_data *sd)
 775 {
 776         struct bpf_array *array = container_of(map, struct bpf_array, map);
 777         unsigned int cpu = smp_processor_id();
 778         u64 index = flags & BPF_F_INDEX_MASK;
 779         struct bpf_event_entry *ee;
 780         struct perf_event *event;
 781
 782         if (index == BPF_F_CURRENT_CPU)
 783                 index = cpu;
 784         if (unlikely(index >= array->map.max_entries))
 785                 return -E2BIG;
 786
 787         ee = READ_ONCE(array->ptrs[index]);
 788         if (!ee)
 789                 return -ENOENT;
 790
 791         event = ee->event;
 792         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 793                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 794                 return -EINVAL;
 795
 796         if (unlikely(event->oncpu != cpu))
 797                 return -EOPNOTSUPP;
 798
 799         return perf_event_output(event, sd, regs);
 800 }
 801
 802 /*
 803  * Support executing tracepoints in normal, irq, and nmi context that each call
 804  * bpf_perf_event_output
 805  */
 806 struct bpf_trace_sample_data {
 807         struct perf_sample_data sds[3];
 808 };
 809
 810 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 811 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 812 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 813            u64, flags, void *, data, u64, size)
 814 {
 815         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 816         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 817         struct perf_raw_record raw = {
 818                 .frag = {
 819                         .size = size,
 820                         .data = data,
 821                 },
 822         };
 823         struct perf_sample_data *sd;
 824         int err;
 825
 826         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 827                 err = -EBUSY;
 828                 goto out;
 829         }
 830
 831         sd = &sds->sds[nest_level - 1];
 832
 833         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 834                 err = -EINVAL;
 835                 goto out;
 836         }
 837
 838         perf_sample_data_init(sd, 0, 0);
 839         sd->raw = &raw;
 840
 841         err = __bpf_perf_event_output(regs, map, flags, sd);
 842
 843 out:
 844         this_cpu_dec(bpf_trace_nest_level);
 845         return err;
 846 }
 847
 848 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 849         .func           = bpf_perf_event_output,
 850         .gpl_only       = true,
 851         .ret_type       = RET_INTEGER,
 852         .arg1_type      = ARG_PTR_TO_CTX,
 853         .arg2_type      = ARG_CONST_MAP_PTR,
 854         .arg3_type      = ARG_ANYTHING,
 855         .arg4_type      = ARG_PTR_TO_MEM,
 856         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 857 };
 858
 859 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 860 struct bpf_nested_pt_regs {
 861         struct pt_regs regs[3];
 862 };
 863 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 864 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 865
 866 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 867                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 868 {
 869         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 870         struct perf_raw_frag frag = {
 871                 .copy           = ctx_copy,
 872                 .size           = ctx_size,
 873                 .data           = ctx,
 874         };
 875         struct perf_raw_record raw = {
 876                 .frag = {
 877                         {
 878                                 .next   = ctx_size ? &frag : NULL,
 879                         },
 880                         .size   = meta_size,
 881                         .data   = meta,
 882                 },
 883         };
 884         struct perf_sample_data *sd;
 885         struct pt_regs *regs;
 886         u64 ret;
 887
 888         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 889                 ret = -EBUSY;
 890                 goto out;
 891         }
 892         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 893         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 894
 895         perf_fetch_caller_regs(regs);
 896         perf_sample_data_init(sd, 0, 0);
 897         sd->raw = &raw;
 898
 899         ret = __bpf_perf_event_output(regs, map, flags, sd);
 900 out:
 901         this_cpu_dec(bpf_event_output_nest_level);
 902         return ret;
 903 }
 904
 905 BPF_CALL_0(bpf_get_current_task)
 906 {
 907         return (long) current;
 908 }
 909
 910 const struct bpf_func_proto bpf_get_current_task_proto = {
 911         .func           = bpf_get_current_task,
 912         .gpl_only       = true,
 913         .ret_type       = RET_INTEGER,
 914 };
 915
 916 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 917 {
 918         struct bpf_array *array = container_of(map, struct bpf_array, map);
 919         struct cgroup *cgrp;
 920
 921         if (unlikely(idx >= array->map.max_entries))
 922                 return -E2BIG;
 923
 924         cgrp = READ_ONCE(array->ptrs[idx]);
 925         if (unlikely(!cgrp))
 926                 return -EAGAIN;
 927
 928         return task_under_cgroup_hierarchy(current, cgrp);
 929 }
 930
 931 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 932         .func           = bpf_current_task_under_cgroup,
 933         .gpl_only       = false,
 934         .ret_type       = RET_INTEGER,
 935         .arg1_type      = ARG_CONST_MAP_PTR,
 936         .arg2_type      = ARG_ANYTHING,
 937 };
 938
 939 struct send_signal_irq_work {
 940         struct irq_work irq_work;
 941         struct task_struct *task;
 942         u32 sig;
 943         enum pid_type type;
 944 };
 945
 946 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 947
 948 static void do_bpf_send_signal(struct irq_work *entry)
 949 {
 950         struct send_signal_irq_work *work;
 951
 952         work = container_of(entry, struct send_signal_irq_work, irq_work);
 953         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 954 }
 955
 956 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 957 {
 958         struct send_signal_irq_work *work = NULL;
 959
 960         /* Similar to bpf_probe_write_user, task needs to be
 961          * in a sound condition and kernel memory access be
 962          * permitted in order to send signal to the current
 963          * task.
 964          */
 965         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 966                 return -EPERM;
 967         if (unlikely(uaccess_kernel()))
 968                 return -EPERM;
 969         if (unlikely(!nmi_uaccess_okay()))
 970                 return -EPERM;
 971
 972         if (irqs_disabled()) {
 973                 /* Do an early check on signal validity. Otherwise,
 974                  * the error is lost in deferred irq_work.
 975                  */
 976                 if (unlikely(!valid_signal(sig)))
 977                         return -EINVAL;
 978
 979                 work = this_cpu_ptr(&send_signal_work);
 980                 if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
 981                         return -EBUSY;
 982
 983                 /* Add the current task, which is the target of sending signal,
 984                  * to the irq_work. The current task may change when queued
 985                  * irq works get executed.
 986                  */
 987                 work->task = current;
 988                 work->sig = sig;
 989                 work->type = type;
 990                 irq_work_queue(&work->irq_work);
 991                 return 0;
 992         }
 993
 994         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 995 }
 996
 997 BPF_CALL_1(bpf_send_signal, u32, sig)
 998 {
 999         return bpf_send_signal_common(sig, PIDTYPE_TGID);
1000 }
1001
1002 static const struct bpf_func_proto bpf_send_signal_proto = {
1003         .func           = bpf_send_signal,
1004         .gpl_only       = false,
1005         .ret_type       = RET_INTEGER,
1006         .arg1_type      = ARG_ANYTHING,
1007 };
1008
1009 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1010 {
1011         return bpf_send_signal_common(sig, PIDTYPE_PID);
1012 }
1013
1014 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1015         .func           = bpf_send_signal_thread,
1016         .gpl_only       = false,
1017         .ret_type       = RET_INTEGER,
1018         .arg1_type      = ARG_ANYTHING,
1019 };
1020
1021 const struct bpf_func_proto *
1022 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1023 {
1024         switch (func_id) {
1025         case BPF_FUNC_map_lookup_elem:
1026                 return &bpf_map_lookup_elem_proto;
1027         case BPF_FUNC_map_update_elem:
1028                 return &bpf_map_update_elem_proto;
1029         case BPF_FUNC_map_delete_elem:
1030                 return &bpf_map_delete_elem_proto;
1031         case BPF_FUNC_map_push_elem:
1032                 return &bpf_map_push_elem_proto;
1033         case BPF_FUNC_map_pop_elem:
1034                 return &bpf_map_pop_elem_proto;
1035         case BPF_FUNC_map_peek_elem:
1036                 return &bpf_map_peek_elem_proto;
1037         case BPF_FUNC_ktime_get_ns:
1038                 return &bpf_ktime_get_ns_proto;
1039         case BPF_FUNC_ktime_get_boot_ns:
1040                 return &bpf_ktime_get_boot_ns_proto;
1041         case BPF_FUNC_tail_call:
1042                 return &bpf_tail_call_proto;
1043         case BPF_FUNC_get_current_pid_tgid:
1044                 return &bpf_get_current_pid_tgid_proto;
1045         case BPF_FUNC_get_current_task:
1046                 return &bpf_get_current_task_proto;
1047         case BPF_FUNC_get_current_uid_gid:
1048                 return &bpf_get_current_uid_gid_proto;
1049         case BPF_FUNC_get_current_comm:
1050                 return &bpf_get_current_comm_proto;
1051         case BPF_FUNC_trace_printk:
1052                 return bpf_get_trace_printk_proto();
1053         case BPF_FUNC_get_smp_processor_id:
1054                 return &bpf_get_smp_processor_id_proto;
1055         case BPF_FUNC_get_numa_node_id:
1056                 return &bpf_get_numa_node_id_proto;
1057         case BPF_FUNC_perf_event_read:
1058                 return &bpf_perf_event_read_proto;
1059         case BPF_FUNC_probe_write_user:
1060                 return bpf_get_probe_write_proto();
1061         case BPF_FUNC_current_task_under_cgroup:
1062                 return &bpf_current_task_under_cgroup_proto;
1063         case BPF_FUNC_get_prandom_u32:
1064                 return &bpf_get_prandom_u32_proto;
1065         case BPF_FUNC_probe_read_user:
1066                 return &bpf_probe_read_user_proto;
1067         case BPF_FUNC_probe_read_kernel:
1068                 return &bpf_probe_read_kernel_proto;
1069         case BPF_FUNC_probe_read_user_str:
1070                 return &bpf_probe_read_user_str_proto;
1071         case BPF_FUNC_probe_read_kernel_str:
1072                 return &bpf_probe_read_kernel_str_proto;
1073 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1074         case BPF_FUNC_probe_read:
1075                 return &bpf_probe_read_compat_proto;
1076         case BPF_FUNC_probe_read_str:
1077                 return &bpf_probe_read_compat_str_proto;
1078 #endif
1079 #ifdef CONFIG_CGROUPS
1080         case BPF_FUNC_get_current_cgroup_id:
1081                 return &bpf_get_current_cgroup_id_proto;
1082 #endif
1083         case BPF_FUNC_send_signal:
1084                 return &bpf_send_signal_proto;
1085         case BPF_FUNC_send_signal_thread:
1086                 return &bpf_send_signal_thread_proto;
1087         case BPF_FUNC_perf_event_read_value:
1088                 return &bpf_perf_event_read_value_proto;
1089         case BPF_FUNC_get_ns_current_pid_tgid:
1090                 return &bpf_get_ns_current_pid_tgid_proto;
1091         default:
1092                 return NULL;
1093         }
1094 }
1095
1096 static const struct bpf_func_proto *
1097 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1098 {
1099         switch (func_id) {
1100         case BPF_FUNC_perf_event_output:
1101                 return &bpf_perf_event_output_proto;
1102         case BPF_FUNC_get_stackid:
1103                 return &bpf_get_stackid_proto;
1104         case BPF_FUNC_get_stack:
1105                 return &bpf_get_stack_proto;
1106 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1107         case BPF_FUNC_override_return:
1108                 return &bpf_override_return_proto;
1109 #endif
1110         default:
1111                 return bpf_tracing_func_proto(func_id, prog);
1112         }
1113 }
1114
1115 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1116 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1117                                         const struct bpf_prog *prog,
1118                                         struct bpf_insn_access_aux *info)
1119 {
1120         if (off < 0 || off >= sizeof(struct pt_regs))
1121                 return false;
1122         if (type != BPF_READ)
1123                 return false;
1124         if (off % size != 0)
1125                 return false;
1126         /*
1127          * Assertion for 32 bit to make sure last 8 byte access
1128          * (BPF_DW) to the last 4 byte member is disallowed.
1129          */
1130         if (off + size > sizeof(struct pt_regs))
1131                 return false;
1132
1133         return true;
1134 }
1135
1136 const struct bpf_verifier_ops kprobe_verifier_ops = {
1137         .get_func_proto  = kprobe_prog_func_proto,
1138         .is_valid_access = kprobe_prog_is_valid_access,
1139 };
1140
1141 const struct bpf_prog_ops kprobe_prog_ops = {
1142 };
1143
1144 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1145            u64, flags, void *, data, u64, size)
1146 {
1147         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1148
1149         /*
1150          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1151          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1152          * from there and call the same bpf_perf_event_output() helper inline.
1153          */
1154         return ____bpf_perf_event_output(regs, map, flags, data, size);
1155 }
1156
1157 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1158         .func           = bpf_perf_event_output_tp,
1159         .gpl_only       = true,
1160         .ret_type       = RET_INTEGER,
1161         .arg1_type      = ARG_PTR_TO_CTX,
1162         .arg2_type      = ARG_CONST_MAP_PTR,
1163         .arg3_type      = ARG_ANYTHING,
1164         .arg4_type      = ARG_PTR_TO_MEM,
1165         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1166 };
1167
1168 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1169            u64, flags)
1170 {
1171         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1172
1173         /*
1174          * Same comment as in bpf_perf_event_output_tp(), only that this time
1175          * the other helper's function body cannot be inlined due to being
1176          * external, thus we need to call raw helper function.
1177          */
1178         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1179                                flags, 0, 0);
1180 }
1181
1182 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1183         .func           = bpf_get_stackid_tp,
1184         .gpl_only       = true,
1185         .ret_type       = RET_INTEGER,
1186         .arg1_type      = ARG_PTR_TO_CTX,
1187         .arg2_type      = ARG_CONST_MAP_PTR,
1188         .arg3_type      = ARG_ANYTHING,
1189 };
1190
1191 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1192            u64, flags)
1193 {
1194         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1195
1196         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1197                              (unsigned long) size, flags, 0);
1198 }
1199
1200 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1201         .func           = bpf_get_stack_tp,
1202         .gpl_only       = true,
1203         .ret_type       = RET_INTEGER,
1204         .arg1_type      = ARG_PTR_TO_CTX,
1205         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1206         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1207         .arg4_type      = ARG_ANYTHING,
1208 };
1209
1210 static const struct bpf_func_proto *
1211 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1212 {
1213         switch (func_id) {
1214         case BPF_FUNC_perf_event_output:
1215                 return &bpf_perf_event_output_proto_tp;
1216         case BPF_FUNC_get_stackid:
1217                 return &bpf_get_stackid_proto_tp;
1218         case BPF_FUNC_get_stack:
1219                 return &bpf_get_stack_proto_tp;
1220         default:
1221                 return bpf_tracing_func_proto(func_id, prog);
1222         }
1223 }
1224
1225 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1226                                     const struct bpf_prog *prog,
1227                                     struct bpf_insn_access_aux *info)
1228 {
1229         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1230                 return false;
1231         if (type != BPF_READ)
1232                 return false;
1233         if (off % size != 0)
1234                 return false;
1235
1236         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1237         return true;
1238 }
1239
1240 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1241         .get_func_proto  = tp_prog_func_proto,
1242         .is_valid_access = tp_prog_is_valid_access,
1243 };
1244
1245 const struct bpf_prog_ops tracepoint_prog_ops = {
1246 };
1247
1248 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1249            struct bpf_perf_event_value *, buf, u32, size)
1250 {
1251         int err = -EINVAL;
1252
1253         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1254                 goto clear;
1255         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1256                                     &buf->running);
1257         if (unlikely(err))
1258                 goto clear;
1259         return 0;
1260 clear:
1261         memset(buf, 0, size);
1262         return err;
1263 }
1264
1265 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1266          .func           = bpf_perf_prog_read_value,
1267          .gpl_only       = true,
1268          .ret_type       = RET_INTEGER,
1269          .arg1_type      = ARG_PTR_TO_CTX,
1270          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1271          .arg3_type      = ARG_CONST_SIZE,
1272 };
1273
1274 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1275            void *, buf, u32, size, u64, flags)
1276 {
1277 #ifndef CONFIG_X86
1278         return -ENOENT;
1279 #else
1280         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1281         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1282         u32 to_copy;
1283
1284         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1285                 return -EINVAL;
1286
1287         if (unlikely(!br_stack))
1288                 return -EINVAL;
1289
1290         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1291                 return br_stack->nr * br_entry_size;
1292
1293         if (!buf || (size % br_entry_size != 0))
1294                 return -EINVAL;
1295
1296         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1297         memcpy(buf, br_stack->entries, to_copy);
1298
1299         return to_copy;
1300 #endif
1301 }
1302
1303 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1304         .func           = bpf_read_branch_records,
1305         .gpl_only       = true,
1306         .ret_type       = RET_INTEGER,
1307         .arg1_type      = ARG_PTR_TO_CTX,
1308         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1309         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1310         .arg4_type      = ARG_ANYTHING,
1311 };
1312
1313 static const struct bpf_func_proto *
1314 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1315 {
1316         switch (func_id) {
1317         case BPF_FUNC_perf_event_output:
1318                 return &bpf_perf_event_output_proto_tp;
1319         case BPF_FUNC_get_stackid:
1320                 return &bpf_get_stackid_proto_tp;
1321         case BPF_FUNC_get_stack:
1322                 return &bpf_get_stack_proto_tp;
1323         case BPF_FUNC_perf_prog_read_value:
1324                 return &bpf_perf_prog_read_value_proto;
1325         case BPF_FUNC_read_branch_records:
1326                 return &bpf_read_branch_records_proto;
1327         default:
1328                 return bpf_tracing_func_proto(func_id, prog);
1329         }
1330 }
1331
1332 /*
1333  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1334  * to avoid potential recursive reuse issue when/if tracepoints are added
1335  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1336  *
1337  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1338  * in normal, irq, and nmi context.
1339  */
1340 struct bpf_raw_tp_regs {
1341         struct pt_regs regs[3];
1342 };
1343 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1344 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1345 static struct pt_regs *get_bpf_raw_tp_regs(void)
1346 {
1347         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1348         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1349
1350         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1351                 this_cpu_dec(bpf_raw_tp_nest_level);
1352                 return ERR_PTR(-EBUSY);
1353         }
1354
1355         return &tp_regs->regs[nest_level - 1];
1356 }
1357
1358 static void put_bpf_raw_tp_regs(void)
1359 {
1360         this_cpu_dec(bpf_raw_tp_nest_level);
1361 }
1362
1363 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1364            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1365 {
1366         struct pt_regs *regs = get_bpf_raw_tp_regs();
1367         int ret;
1368
1369         if (IS_ERR(regs))
1370                 return PTR_ERR(regs);
1371
1372         perf_fetch_caller_regs(regs);
1373         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1374
1375         put_bpf_raw_tp_regs();
1376         return ret;
1377 }
1378
1379 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1380         .func           = bpf_perf_event_output_raw_tp,
1381         .gpl_only       = true,
1382         .ret_type       = RET_INTEGER,
1383         .arg1_type      = ARG_PTR_TO_CTX,
1384         .arg2_type      = ARG_CONST_MAP_PTR,
1385         .arg3_type      = ARG_ANYTHING,
1386         .arg4_type      = ARG_PTR_TO_MEM,
1387         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1388 };
1389
1390 extern const struct bpf_func_proto bpf_skb_output_proto;
1391 extern const struct bpf_func_proto bpf_xdp_output_proto;
1392
1393 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1394            struct bpf_map *, map, u64, flags)
1395 {
1396         struct pt_regs *regs = get_bpf_raw_tp_regs();
1397         int ret;
1398
1399         if (IS_ERR(regs))
1400                 return PTR_ERR(regs);
1401
1402         perf_fetch_caller_regs(regs);
1403         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1404         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1405                               flags, 0, 0);
1406         put_bpf_raw_tp_regs();
1407         return ret;
1408 }
1409
1410 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1411         .func           = bpf_get_stackid_raw_tp,
1412         .gpl_only       = true,
1413         .ret_type       = RET_INTEGER,
1414         .arg1_type      = ARG_PTR_TO_CTX,
1415         .arg2_type      = ARG_CONST_MAP_PTR,
1416         .arg3_type      = ARG_ANYTHING,
1417 };
1418
1419 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1420            void *, buf, u32, size, u64, flags)
1421 {
1422         struct pt_regs *regs = get_bpf_raw_tp_regs();
1423         int ret;
1424
1425         if (IS_ERR(regs))
1426                 return PTR_ERR(regs);
1427
1428         perf_fetch_caller_regs(regs);
1429         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1430                             (unsigned long) size, flags, 0);
1431         put_bpf_raw_tp_regs();
1432         return ret;
1433 }
1434
1435 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1436         .func           = bpf_get_stack_raw_tp,
1437         .gpl_only       = true,
1438         .ret_type       = RET_INTEGER,
1439         .arg1_type      = ARG_PTR_TO_CTX,
1440         .arg2_type      = ARG_PTR_TO_MEM,
1441         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1442         .arg4_type      = ARG_ANYTHING,
1443 };
1444
1445 static const struct bpf_func_proto *
1446 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1447 {
1448         switch (func_id) {
1449         case BPF_FUNC_perf_event_output:
1450                 return &bpf_perf_event_output_proto_raw_tp;
1451         case BPF_FUNC_get_stackid:
1452                 return &bpf_get_stackid_proto_raw_tp;
1453         case BPF_FUNC_get_stack:
1454                 return &bpf_get_stack_proto_raw_tp;
1455         default:
1456                 return bpf_tracing_func_proto(func_id, prog);
1457         }
1458 }
1459
1460 static const struct bpf_func_proto *
1461 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1462 {
1463         switch (func_id) {
1464 #ifdef CONFIG_NET
1465         case BPF_FUNC_skb_output:
1466                 return &bpf_skb_output_proto;
1467         case BPF_FUNC_xdp_output:
1468                 return &bpf_xdp_output_proto;
1469 #endif
1470         case BPF_FUNC_seq_printf:
1471                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1472                        &bpf_seq_printf_proto :
1473                        NULL;
1474         case BPF_FUNC_seq_write:
1475                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1476                        &bpf_seq_write_proto :
1477                        NULL;
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 };
1525
1526 const struct bpf_verifier_ops tracing_verifier_ops = {
1527         .get_func_proto  = tracing_prog_func_proto,
1528         .is_valid_access = tracing_prog_is_valid_access,
1529 };
1530
1531 const struct bpf_prog_ops tracing_prog_ops = {
1532         .test_run = bpf_prog_test_run_tracing,
1533 };
1534
1535 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1536                                                  enum bpf_access_type type,
1537                                                  const struct bpf_prog *prog,
1538                                                  struct bpf_insn_access_aux *info)
1539 {
1540         if (off == 0) {
1541                 if (size != sizeof(u64) || type != BPF_READ)
1542                         return false;
1543                 info->reg_type = PTR_TO_TP_BUFFER;
1544         }
1545         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1546 }
1547
1548 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1549         .get_func_proto  = raw_tp_prog_func_proto,
1550         .is_valid_access = raw_tp_writable_prog_is_valid_access,
1551 };
1552
1553 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1554 };
1555
1556 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1557                                     const struct bpf_prog *prog,
1558                                     struct bpf_insn_access_aux *info)
1559 {
1560         const int size_u64 = sizeof(u64);
1561
1562         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1563                 return false;
1564         if (type != BPF_READ)
1565                 return false;
1566         if (off % size != 0) {
1567                 if (sizeof(unsigned long) != 4)
1568                         return false;
1569                 if (size != 8)
1570                         return false;
1571                 if (off % size != 4)
1572                         return false;
1573         }
1574
1575         switch (off) {
1576         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
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         case bpf_ctx_range(struct bpf_perf_event_data, addr):
1582                 bpf_ctx_record_field_size(info, size_u64);
1583                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1584                         return false;
1585                 break;
1586         default:
1587                 if (size != sizeof(long))
1588                         return false;
1589         }
1590
1591         return true;
1592 }
1593
1594 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1595                                       const struct bpf_insn *si,
1596                                       struct bpf_insn *insn_buf,
1597                                       struct bpf_prog *prog, u32 *target_size)
1598 {
1599         struct bpf_insn *insn = insn_buf;
1600
1601         switch (si->off) {
1602         case offsetof(struct bpf_perf_event_data, sample_period):
1603                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1604                                                        data), si->dst_reg, si->src_reg,
1605                                       offsetof(struct bpf_perf_event_data_kern, data));
1606                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1607                                       bpf_target_off(struct perf_sample_data, period, 8,
1608                                                      target_size));
1609                 break;
1610         case offsetof(struct bpf_perf_event_data, addr):
1611                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1612                                                        data), si->dst_reg, si->src_reg,
1613                                       offsetof(struct bpf_perf_event_data_kern, data));
1614                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1615                                       bpf_target_off(struct perf_sample_data, addr, 8,
1616                                                      target_size));
1617                 break;
1618         default:
1619                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1620                                                        regs), si->dst_reg, si->src_reg,
1621                                       offsetof(struct bpf_perf_event_data_kern, regs));
1622                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1623                                       si->off);
1624                 break;
1625         }
1626
1627         return insn - insn_buf;
1628 }
1629
1630 const struct bpf_verifier_ops perf_event_verifier_ops = {
1631         .get_func_proto         = pe_prog_func_proto,
1632         .is_valid_access        = pe_prog_is_valid_access,
1633         .convert_ctx_access     = pe_prog_convert_ctx_access,
1634 };
1635
1636 const struct bpf_prog_ops perf_event_prog_ops = {
1637 };
1638
1639 static DEFINE_MUTEX(bpf_event_mutex);
1640
1641 #define BPF_TRACE_MAX_PROGS 64
1642
1643 int perf_event_attach_bpf_prog(struct perf_event *event,
1644                                struct bpf_prog *prog)
1645 {
1646         struct bpf_prog_array *old_array;
1647         struct bpf_prog_array *new_array;
1648         int ret = -EEXIST;
1649
1650         /*
1651          * Kprobe override only works if they are on the function entry,
1652          * and only if they are on the opt-in list.
1653          */
1654         if (prog->kprobe_override &&
1655             (!trace_kprobe_on_func_entry(event->tp_event) ||
1656              !trace_kprobe_error_injectable(event->tp_event)))
1657                 return -EINVAL;
1658
1659         mutex_lock(&bpf_event_mutex);
1660
1661         if (event->prog)
1662                 goto unlock;
1663
1664         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1665         if (old_array &&
1666             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1667                 ret = -E2BIG;
1668                 goto unlock;
1669         }
1670
1671         ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1672         if (ret < 0)
1673                 goto unlock;
1674
1675         /* set the new array to event->tp_event and set event->prog */
1676         event->prog = prog;
1677         rcu_assign_pointer(event->tp_event->prog_array, new_array);
1678         bpf_prog_array_free(old_array);
1679
1680 unlock:
1681         mutex_unlock(&bpf_event_mutex);
1682         return ret;
1683 }
1684
1685 void perf_event_detach_bpf_prog(struct perf_event *event)
1686 {
1687         struct bpf_prog_array *old_array;
1688         struct bpf_prog_array *new_array;
1689         int ret;
1690
1691         mutex_lock(&bpf_event_mutex);
1692
1693         if (!event->prog)
1694                 goto unlock;
1695
1696         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1697         ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1698         if (ret == -ENOENT)
1699                 goto unlock;
1700         if (ret < 0) {
1701                 bpf_prog_array_delete_safe(old_array, event->prog);
1702         } else {
1703                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1704                 bpf_prog_array_free(old_array);
1705         }
1706
1707         bpf_prog_put(event->prog);
1708         event->prog = NULL;
1709
1710 unlock:
1711         mutex_unlock(&bpf_event_mutex);
1712 }
1713
1714 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1715 {
1716         struct perf_event_query_bpf __user *uquery = info;
1717         struct perf_event_query_bpf query = {};
1718         struct bpf_prog_array *progs;
1719         u32 *ids, prog_cnt, ids_len;
1720         int ret;
1721
1722         if (!capable(CAP_SYS_ADMIN))
1723                 return -EPERM;
1724         if (event->attr.type != PERF_TYPE_TRACEPOINT)
1725                 return -EINVAL;
1726         if (copy_from_user(&query, uquery, sizeof(query)))
1727                 return -EFAULT;
1728
1729         ids_len = query.ids_len;
1730         if (ids_len > BPF_TRACE_MAX_PROGS)
1731                 return -E2BIG;
1732         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1733         if (!ids)
1734                 return -ENOMEM;
1735         /*
1736          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1737          * is required when user only wants to check for uquery->prog_cnt.
1738          * There is no need to check for it since the case is handled
1739          * gracefully in bpf_prog_array_copy_info.
1740          */
1741
1742         mutex_lock(&bpf_event_mutex);
1743         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1744         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1745         mutex_unlock(&bpf_event_mutex);
1746
1747         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1748             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1749                 ret = -EFAULT;
1750
1751         kfree(ids);
1752         return ret;
1753 }
1754
1755 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1756 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1757
1758 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1759 {
1760         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1761
1762         for (; btp < __stop__bpf_raw_tp; btp++) {
1763                 if (!strcmp(btp->tp->name, name))
1764                         return btp;
1765         }
1766
1767         return bpf_get_raw_tracepoint_module(name);
1768 }
1769
1770 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1771 {
1772         struct module *mod = __module_address((unsigned long)btp);
1773
1774         if (mod)
1775                 module_put(mod);
1776 }
1777
1778 static __always_inline
1779 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1780 {
1781         cant_sleep();
1782         rcu_read_lock();
1783         (void) BPF_PROG_RUN(prog, args);
1784         rcu_read_unlock();
1785 }
1786
1787 #define UNPACK(...)                     __VA_ARGS__
1788 #define REPEAT_1(FN, DL, X, ...)        FN(X)
1789 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1790 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1791 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1792 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1793 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1794 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1795 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1796 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1797 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1798 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1799 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1800 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
1801
1802 #define SARG(X)         u64 arg##X
1803 #define COPY(X)         args[X] = arg##X
1804
1805 #define __DL_COM        (,)
1806 #define __DL_SEM        (;)
1807
1808 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1809
1810 #define BPF_TRACE_DEFN_x(x)                                             \
1811         void bpf_trace_run##x(struct bpf_prog *prog,                    \
1812                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
1813         {                                                               \
1814                 u64 args[x];                                            \
1815                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
1816                 __bpf_trace_run(prog, args);                            \
1817         }                                                               \
1818         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1819 BPF_TRACE_DEFN_x(1);
1820 BPF_TRACE_DEFN_x(2);
1821 BPF_TRACE_DEFN_x(3);
1822 BPF_TRACE_DEFN_x(4);
1823 BPF_TRACE_DEFN_x(5);
1824 BPF_TRACE_DEFN_x(6);
1825 BPF_TRACE_DEFN_x(7);
1826 BPF_TRACE_DEFN_x(8);
1827 BPF_TRACE_DEFN_x(9);
1828 BPF_TRACE_DEFN_x(10);
1829 BPF_TRACE_DEFN_x(11);
1830 BPF_TRACE_DEFN_x(12);
1831
1832 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1833 {
1834         struct tracepoint *tp = btp->tp;
1835
1836         /*
1837          * check that program doesn't access arguments beyond what's
1838          * available in this tracepoint
1839          */
1840         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1841                 return -EINVAL;
1842
1843         if (prog->aux->max_tp_access > btp->writable_size)
1844                 return -EINVAL;
1845
1846         return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1847 }
1848
1849 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1850 {
1851         return __bpf_probe_register(btp, prog);
1852 }
1853
1854 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1855 {
1856         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1857 }
1858
1859 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1860                             u32 *fd_type, const char **buf,
1861                             u64 *probe_offset, u64 *probe_addr)
1862 {
1863         bool is_tracepoint, is_syscall_tp;
1864         struct bpf_prog *prog;
1865         int flags, err = 0;
1866
1867         prog = event->prog;
1868         if (!prog)
1869                 return -ENOENT;
1870
1871         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1872         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1873                 return -EOPNOTSUPP;
1874
1875         *prog_id = prog->aux->id;
1876         flags = event->tp_event->flags;
1877         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1878         is_syscall_tp = is_syscall_trace_event(event->tp_event);
1879
1880         if (is_tracepoint || is_syscall_tp) {
1881                 *buf = is_tracepoint ? event->tp_event->tp->name
1882                                      : event->tp_event->name;
1883                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1884                 *probe_offset = 0x0;
1885                 *probe_addr = 0x0;
1886         } else {
1887                 /* kprobe/uprobe */
1888                 err = -EOPNOTSUPP;
1889 #ifdef CONFIG_KPROBE_EVENTS
1890                 if (flags & TRACE_EVENT_FL_KPROBE)
1891                         err = bpf_get_kprobe_info(event, fd_type, buf,
1892                                                   probe_offset, probe_addr,
1893                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
1894 #endif
1895 #ifdef CONFIG_UPROBE_EVENTS
1896                 if (flags & TRACE_EVENT_FL_UPROBE)
1897                         err = bpf_get_uprobe_info(event, fd_type, buf,
1898                                                   probe_offset,
1899                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
1900 #endif
1901         }
1902
1903         return err;
1904 }
1905
1906 static int __init send_signal_irq_work_init(void)
1907 {
1908         int cpu;
1909         struct send_signal_irq_work *work;
1910
1911         for_each_possible_cpu(cpu) {
1912                 work = per_cpu_ptr(&send_signal_work, cpu);
1913                 init_irq_work(&work->irq_work, do_bpf_send_signal);
1914         }
1915         return 0;
1916 }
1917
1918 subsys_initcall(send_signal_irq_work_init);
1919
1920 #ifdef CONFIG_MODULES
1921 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1922                             void *module)
1923 {
1924         struct bpf_trace_module *btm, *tmp;
1925         struct module *mod = module;
1926
1927         if (mod->num_bpf_raw_events == 0 ||
1928             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1929                 return 0;
1930
1931         mutex_lock(&bpf_module_mutex);
1932
1933         switch (op) {
1934         case MODULE_STATE_COMING:
1935                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1936                 if (btm) {
1937                         btm->module = module;
1938                         list_add(&btm->list, &bpf_trace_modules);
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         return 0;
1955 }
1956
1957 static struct notifier_block bpf_module_nb = {
1958         .notifier_call = bpf_event_notify,
1959 };
1960
1961 static int __init bpf_event_init(void)
1962 {
1963         register_module_notifier(&bpf_module_nb);
1964         return 0;
1965 }
1966
1967 fs_initcall(bpf_event_init);
1968 #endif /* CONFIG_MODULES */